]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
16 years ago - Check whether we've recorded this tick in ts_ticks on another cpu in
jeff [Sat, 19 Jul 2008 05:13:47 +0000 (05:13 +0000)]
 - Check whether we've recorded this tick in ts_ticks on another cpu in
   sched_tick() to prevent multiple increments for one tick.  This pushes
   the value out of range and breaks priority calculation.

Reviewed by: kib
Found by: pho/nokia
Sponsored by: Nokia
MFC after: 3 days

16 years agoDisable SSP on arm for the time being.
cognet [Sat, 19 Jul 2008 00:19:16 +0000 (00:19 +0000)]
Disable SSP on arm for the time being.
The segfaults when using SSP seem to be a gcc bug, a patch is available
in the gcc bugzilla, and will be imported once it's committed
into the official gcc tree.

16 years agoStaticify symbols that will not be used in other
delphij [Sat, 19 Jul 2008 00:13:26 +0000 (00:13 +0000)]
Staticify symbols that will not be used in other
places.

16 years agoUse %zd for size_t. With this gcore(1) is WARNS=6 clean.
delphij [Fri, 18 Jul 2008 23:37:05 +0000 (23:37 +0000)]
Use %zd for size_t.  With this gcore(1) is WARNS=6 clean.

16 years agoIndent the else path, reduce diff against OpenBSD.
delphij [Fri, 18 Jul 2008 23:35:31 +0000 (23:35 +0000)]
Indent the else path, reduce diff against OpenBSD.

16 years agoCorrect an error in pmap_change_attr()'s initial loop that verifies that the
alc [Fri, 18 Jul 2008 22:05:51 +0000 (22:05 +0000)]
Correct an error in pmap_change_attr()'s initial loop that verifies that the
given range of addresses are mapped.  Previously, the loop was testing the
same address every time.

Submitted by: Magesh Dhasayyan

16 years agoSimplify pmap_extract()'s control flow, making it more like the related
alc [Fri, 18 Jul 2008 20:07:50 +0000 (20:07 +0000)]
Simplify pmap_extract()'s control flow, making it more like the related
functions pmap_extract_and_hold() and pmap_kextract().

16 years agoEnhance arena_chunk_map_t to directly support run coalescing, and use
jasone [Fri, 18 Jul 2008 19:35:44 +0000 (19:35 +0000)]
Enhance arena_chunk_map_t to directly support run coalescing, and use
the chunk map instead of red-black trees where possible.  Remove the
red-black trees and node objects that are obsoleted by this change.  The
net result is a ~1-2% memory savings, and a substantial allocation speed
improvement.

16 years agoEliminate stale comments from kmem_malloc().
alc [Fri, 18 Jul 2008 17:41:31 +0000 (17:41 +0000)]
Eliminate stale comments from kmem_malloc().

16 years agoFix a longstanding bug, from Otto Moerbeck:
kevlo [Fri, 18 Jul 2008 15:05:14 +0000 (15:05 +0000)]
Fix a longstanding bug, from Otto Moerbeck:
if we're reducing a rule that has an empty
right hand side and the yacc stackpointer is pointing at the very
end of the allocated stack, we end up accessing the stack out of
bounds by the implicit $$ = $1 action

Obtained from: OpenBSD

16 years agoRemove unnessasary cast
kevlo [Fri, 18 Jul 2008 14:55:22 +0000 (14:55 +0000)]
Remove unnessasary cast

16 years agoAdd a man page for the DNS accept filter.
dwmalone [Fri, 18 Jul 2008 14:46:01 +0000 (14:46 +0000)]
Add a man page for the DNS accept filter.

16 years agoAdd an accept filter for TCP based DNS requests. It waits until the
dwmalone [Fri, 18 Jul 2008 14:44:51 +0000 (14:44 +0000)]
Add an accept filter for TCP based DNS requests. It waits until the
whole first request is present before returning from accept.

16 years agoEliminate use of the global ripsrc which was being used to pass address
rwatson [Fri, 18 Jul 2008 10:47:07 +0000 (10:47 +0000)]
Eliminate use of the global ripsrc which was being used to pass address
information from rip_input() to rip_append().  Instead, pass the source
address for an IP datagram to rip_append() using a stack-allocated
sockaddr_in, similar to udp_input() and udp_append().

Prior to the move to rwlocks for inpcbinfo, this was not a problem, as
use of the global was synchronized using the ripcbinfo mutex, but with
read-locking there is the potential for a race during concurrent
receive.

This problem is not present in the IPv6 raw IP socket code, which
already used a stack variable for the address.

Spotted by: mav
MFC after: 1 week (before inpcbinfo rwlock changes)

16 years agorevert local change
kmacy [Fri, 18 Jul 2008 07:10:33 +0000 (07:10 +0000)]
revert local change

16 years agorevert change from local tree
kmacy [Fri, 18 Jul 2008 07:07:57 +0000 (07:07 +0000)]
revert change from local tree

16 years agonew vendor PHY support
kmacy [Fri, 18 Jul 2008 07:01:51 +0000 (07:01 +0000)]
new vendor PHY support

16 years agorevert changes accidentally included in last commit
kmacy [Fri, 18 Jul 2008 06:22:57 +0000 (06:22 +0000)]
revert changes accidentally included in last commit

16 years agoEliminate unused global variables. (These global variables became fields of
alc [Fri, 18 Jul 2008 06:14:36 +0000 (06:14 +0000)]
Eliminate unused global variables.  (These global variables became fields of
struct kva_md_info many years ago.)

16 years agoimport vendor fixes to cxgb
kmacy [Fri, 18 Jul 2008 06:12:31 +0000 (06:12 +0000)]
import vendor fixes to cxgb

16 years agoCorrect 1000Mbps link handling logic for JMC250. This should make
yongari [Fri, 18 Jul 2008 04:20:48 +0000 (04:20 +0000)]
Correct 1000Mbps link handling logic for JMC250. This should make
jme(4) run on 1000Mbps link.

16 years agoSort the .PATH entries to give a more reasonable order of precedence:
das [Fri, 18 Jul 2008 02:18:34 +0000 (02:18 +0000)]
Sort the .PATH entries to give a more reasonable order of precedence:
     1. architecture-specific files
     2. long double format-specific files
     3. bsdsrc
     4. src
     5. man
The original order was virtually the opposite of this.

This should not cause any functional changes at this time. The
difference is only significant when one wants to override, say, a
generic foo.c with a more specialized foo.c (as opposed to foo.S).

16 years agoUse DELAY() instead of pause if waiting time is less than 1ms.
yongari [Fri, 18 Jul 2008 01:00:54 +0000 (01:00 +0000)]
Use DELAY() instead of pause if waiting time is less than 1ms.
This will fix driver hang if hz < 1000.

Pointed out by: thompsa

16 years agoFix a benign typo that would give out an incorrect warning message.
luoqi [Thu, 17 Jul 2008 22:40:23 +0000 (22:40 +0000)]
Fix a benign typo that would give out an incorrect warning message.
Change a get-or-set sequence on OHCI_COMMAND_STATUS register which
is "write to set" to a simple set.

16 years agoMake it more clear what it means that SA_RESTART is set for signal
simon [Thu, 17 Jul 2008 21:54:23 +0000 (21:54 +0000)]
Make it more clear what it means that SA_RESTART is set for signal
handlers added with signal(3).

Submitted by: gnn (slightly modified by me)
MFC after: 1 week

16 years ago- dns queries might go also over TCP, so allow it.
danger [Thu, 17 Jul 2008 20:00:18 +0000 (20:00 +0000)]
- dns queries might go also over TCP, so allow it.

Approved by: rink
MFC after: 1 week

16 years agoPair the VOP_OPEN call from do_execve() with the reciprocal VOP_CLOSE.
kib [Thu, 17 Jul 2008 16:44:07 +0000 (16:44 +0000)]
Pair the VOP_OPEN call from do_execve() with the reciprocal VOP_CLOSE.
This was unnoticed because local filesystems usually do nothing
non-trivial in the close vop.

Reported and tested by: Rick Macklem
MFC after: 2 weeks

16 years agoClean up mxge's use of callouts as pointed out by jhb,
gallatin [Thu, 17 Jul 2008 15:46:35 +0000 (15:46 +0000)]
Clean up mxge's use of callouts as pointed out by jhb,
and handle NIC hardware watchdog resets.

- remove buggy code at the top of mxge_tick() which tried
  to detect a race which is already detected in the kernel's
  callout code.

- move callout_stop() and callout_reset() into mxge_close()
  mxge_open() rather than doing the callout manipulation
  all over the place.

- use callout_drain(), rather than callout_stop() to prevent
  a potential race between mxge_tick() and mxge_detach()
  which could lead to softclock using a destroyed mutex

- restructure the mxge_tick() and mxge_watchdog_reset()
  routines to avoid resetting a callout, and then
  immediately stopping it if the watchdog reset routine
  is called, and fails.

- enable the driver to handle NIC hardware watchdog
  resets by restoring the NIC's PCI config space, which is
  lost when the NIC hardware watchdog triggers.

Reviewed by: jhb (previus version)

16 years agoDon't set PACKAGE_BUILDING while installing packages.
flz [Thu, 17 Jul 2008 15:08:26 +0000 (15:08 +0000)]
Don't set PACKAGE_BUILDING while installing packages.

Submitted by: erwin
Discussed on: http://lists.freebsd.org/pipermail/freebsd-ports/2008-June/049074.html
MFC after: 1 week

16 years agoBefore updating the password database, the pw(8) utility first performs a
jhb [Thu, 17 Jul 2008 13:47:59 +0000 (13:47 +0000)]
Before updating the password database, the pw(8) utility first performs a
sanity check by invoking "pwd_mkdb -C".  However, if this failed it
silently returned success.  Fix this so it fails the update operation
instead.

MFC after: 1 week

16 years agoAdd the shutdown KEYWORD to those scripts that start persistent services
dougb [Wed, 16 Jul 2008 19:50:29 +0000 (19:50 +0000)]
Add the shutdown KEYWORD to those scripts that start persistent services
to allow them to do a "clean" shutdown.

I purposely avoided making changes to network-related stuff since the
system shutting down is pretty conclusive, and there may be complicated
dependencies on the network that I would rather not try to unravel.

I also skipped kerberos-related stuff for the reasons above, and
because I have no way to test it.

16 years agoAs previously discussed, add the svn:executable property to all scripts
dougb [Wed, 16 Jul 2008 19:22:48 +0000 (19:22 +0000)]
As previously discussed, add the svn:executable property to all scripts

16 years agoMove the TCSA* definitions out of _KERNEL. They are processed in libc.
ed [Wed, 16 Jul 2008 12:36:39 +0000 (12:36 +0000)]
Move the TCSA* definitions out of _KERNEL. They are processed in libc.

The tcsetattr() routine already converts the TCSA* arguments to their
respective TIOCSETA* ioctl's in the C library. There is no need to have
these definitions inside the kernel.

Approved by: philip (mentor, implicit)

16 years agoSort the ioctl's in <sys/ttycom.h> by number.
ed [Wed, 16 Jul 2008 11:23:15 +0000 (11:23 +0000)]
Sort the ioctl's in <sys/ttycom.h> by number.

I think one of the reasons why we have so many conflicts in the TTY
ioctl category, is because the ioctl's aren't ordered logically. This
commit only sorts them by number. The comments may still be inaccurate.

Approved by: philip (mentor)

16 years agoRemove OTTYDISC, NETLDISC and NTTYDISC definitions.
ed [Wed, 16 Jul 2008 11:20:04 +0000 (11:20 +0000)]
Remove OTTYDISC, NETLDISC and NTTYDISC definitions.

When I ported most applications away from <sgtty.h>, I noticed none of
them were actually using these definitions. I kept them in place,
because I didn't want to touch tools like pstat(8) and stty(1).

In preparation for the MPSAFE TTY layer, remove these definitions. This
doesn't have any impact with respect to binary compatibility (see
tty_conf.c).

We couldn now add an #error to <sys/ioctl_compat.h> when included
outside the kernel. Unfortunately, kdump's mkioctls includes this file
unconditionally.

Approved by: philip (mentor)

16 years agoFix error in comment.
rwatson [Wed, 16 Jul 2008 10:55:50 +0000 (10:55 +0000)]
Fix error in comment.

MFC after: 3 weeks

16 years agoFix a multicast handling regression on VT6105M introduced in
yongari [Wed, 16 Jul 2008 08:35:29 +0000 (08:35 +0000)]
Fix a multicast handling regression on VT6105M introduced in
vr(4) overhauling(r177050).

It seems that filtering multicast addresses with multicast CAM
entries require accessing 'CAM enable bit' for each CAM entry.
Subsequent accessing multicast CAM control register without
toggling the 'CAM enable bit' seem to no effects.
In order to fix that separate CAM setup from CAM mask configuration
and CAM entry modification. While I'm here add VLAN CAM filtering
feature which will be enabled in future(FreeBSD now can receive
VLAN id insertion/removal event from vlan(4) on the fly).

For VT6105M hardware, explicitly disable VLAN hardware tag
insertion/stripping and enable VLAN CAM filtering for VLAN id 0.
This shall make non-VLAN frames set VR_RXSTAT_VIDHIT bit in Rx
status word.

Added multicast/VLAN CAM address definition to header file.

PR: kern/125010, kern/125024
MFC after: 1 week

16 years agoFix VR_RXSTAT_RX_OK bit definition which lasted for more than 9
yongari [Wed, 16 Jul 2008 08:02:23 +0000 (08:02 +0000)]
Fix VR_RXSTAT_RX_OK bit definition which lasted for more than 9
years. All datasheet I have indicates the bit 15 is the
VR_RXSTAT_RX_OK. The bit 14 is reserved for all Rhine family
except VT6105M. VT6105M uses that bit to indicate a VLAN frame
with matching CAM VLAN id.
Use the VR_RXSTAT_RX_OK instead of VR_RXSTAT_RXERR when vr(4)
checks the validity of received frame.
This should fix occasional dropping frames on VT6105M.

Tested by: Goran Lowkrantz ( goran.lowkrantz at ismobile dot com )
MFC after: 1 week

16 years agoInvoke err() with a format string rather than directly with a passed
rwatson [Tue, 15 Jul 2008 16:07:34 +0000 (16:07 +0000)]
Invoke err() with a format string rather than directly with a passed
command line argument.

Submitted by: Alfredo Ortega <alfredo at coresecurity dot com>
Thanks to: Core Security Technologies
MFC after: 3 days

16 years agoMerge last of a series of rwlock conversion changes to UDP, which
rwatson [Tue, 15 Jul 2008 15:38:47 +0000 (15:38 +0000)]
Merge last of a series of rwlock conversion changes to UDP, which
completes the move to a fully parallel UDP transmit path by using
global read, rather than write, locking of inpcbinfo in further
semi-connected cases:

- Add macros to allow try-locking of inpcb and inpcbinfo.
- Always acquire an incpcb read lock in udp_output(), which stablizes the
  local inpcb address and port bindings in order to determine what further
  locking is required:
  - If the inpcb is currently not bound (at all) and are implicitly
    connecting, we require inpcbinfo and inpcb write locks, so drop the
    read lock and re-acquire.
  - If the inpcb is bound for at least one of the port or address, but an
    explicit source or destination is requested, trylock the inpcbinfo
    lock, and if that fails, drop the inpcb lock, lock the global lock,
    and relock the inpcb lock.
  - Otherwise, no further locking is required (common case).
- Update comments.

In practice, this means that the vast majority of consumers of UDP sockets
will not acquire any exclusive locks at the socket or UDP levels of the
network stack.  This leads to a marked performance improvement in several
important workloads, including BIND, nsd, and memcached over UDP, as well
as significant improvements in pps microbenchmarks.

The plan is to MFC all of the rwlock changes to RELENG_7 once they have
settled for a weeks in the tree.

Tested by: ps, kris (older revision), bde
MFC after: 3 weeks

16 years agoFix commment in typo.
rpaulo [Tue, 15 Jul 2008 10:32:35 +0000 (10:32 +0000)]
Fix commment in typo.

M    tcp_output.c

16 years agoUpdate the definition of modspecific structure
danger [Tue, 15 Jul 2008 10:06:37 +0000 (10:06 +0000)]
Update the definition of modspecific structure

PR: docs/125630

16 years agoUpdate bus_dmamem_alloc()'s first call to malloc() such that M_WAITOK is
alc [Tue, 15 Jul 2008 03:34:49 +0000 (03:34 +0000)]
Update bus_dmamem_alloc()'s first call to malloc() such that M_WAITOK is
specified when appropriate.

Reviewed by: scottl

16 years agoAdd quirk for Dell D630 laptops.
delphij [Tue, 15 Jul 2008 02:34:44 +0000 (02:34 +0000)]
Add quirk for Dell D630 laptops.

Tested by: Quake Lee <quakelee geekcn org>,
Robert Noland <rnoland 2hip net>
MFC after: 1 week
Approved by: ariff

16 years agoAllow injecting big packets via bpf(4) up to min(MTU, 16K-byte).
jkim [Mon, 14 Jul 2008 22:41:48 +0000 (22:41 +0000)]
Allow injecting big packets via bpf(4) up to min(MTU, 16K-byte).

MFC after: 1 week

16 years agoMatch the implementation of the inline function from libkern.h.
obrien [Mon, 14 Jul 2008 21:36:02 +0000 (21:36 +0000)]
Match the implementation of the inline function from libkern.h.

16 years agoFix carp(4) panics that can occur during carp interface configuration.
eri [Mon, 14 Jul 2008 20:11:51 +0000 (20:11 +0000)]
Fix carp(4) panics that can occur during carp interface configuration.

Approved by: mlaier (mentor)
Reported by: Scott Ullrich
MFC after: 1 week

16 years agoUnmangle an E.W.Dijkstra quote.
phk [Mon, 14 Jul 2008 18:45:06 +0000 (18:45 +0000)]
Unmangle an E.W.Dijkstra quote.

16 years agoAdd event notification at attach/detach so the NIC
jfv [Mon, 14 Jul 2008 18:40:21 +0000 (18:40 +0000)]
Add event notification at attach/detach so the NIC
is able to detect it and do hardware filtering.

16 years agoAdd an event handler to the vlan driver so the NIC driver
jfv [Mon, 14 Jul 2008 18:38:52 +0000 (18:38 +0000)]
Add an event handler to the vlan driver so the NIC driver
becomes aware of it, and gets the VLAN ID. This will allow
the easy use of VLAN hardware filtering by adapters that
support it.

16 years agoManpage for the et(4) driver.
brueffer [Mon, 14 Jul 2008 18:15:43 +0000 (18:15 +0000)]
Manpage for the et(4) driver.

Obtained from: DragonFly

16 years agoFill in the string portion of the bluetooth stack version sysctl.
trhodes [Mon, 14 Jul 2008 13:45:05 +0000 (13:45 +0000)]
Fill in the string portion of the bluetooth stack version sysctl.

Approved by: emax

16 years ago- Enhance volume handling
gahr [Mon, 14 Jul 2008 13:22:09 +0000 (13:22 +0000)]
- Enhance volume handling

PR: 125054
Submitted by: gahr
Approved by: cperciva

16 years agoChange the character prefixed to the svn version to "r" since that seems
dougb [Sun, 13 Jul 2008 20:08:38 +0000 (20:08 +0000)]
Change the character prefixed to the svn version to "r" since that seems
to be how they are commonly referred to.

16 years agoHandle a race between pmap_kextract() and pmap_promote_pde(). This race
alc [Sun, 13 Jul 2008 18:19:53 +0000 (18:19 +0000)]
Handle a race between pmap_kextract() and pmap_promote_pde().  This race
caused ZFS to crash when restoring a snapshot with superpage promotion
enabled.

Reported by: kris

16 years agoAdd an obsolete header after pecoff removal.
antoine [Sun, 13 Jul 2008 17:17:35 +0000 (17:17 +0000)]
Add an obsolete header after pecoff removal.

Approved by: rwatson (mentor)

16 years agoStaticize M_STACK.
antoine [Sun, 13 Jul 2008 17:15:05 +0000 (17:15 +0000)]
Staticize M_STACK.

Approved by: rwatson (mentor)
MFC after: 1 month

16 years agoThe libc acl_valid(3) function validates the contents of a POSIX.1e ACL.
rwatson [Sun, 13 Jul 2008 16:37:51 +0000 (16:37 +0000)]
The libc acl_valid(3) function validates the contents of a POSIX.1e ACL.
This change removes the requirement that an ACL contain no ACL_USER
entries with a uid the same as those of a file, or ACL_GROUP entries
with a gid the same as those of a file.  This requirement is not in the
specification, and not enforced by the kernel's ACL implementation.

Reported by: Iustin Pop <iusty at k1024 dot org>
MFC after: 1 week

16 years agoMore explicitly mention that we support the 2920C with the ahc driver.
remko [Sun, 13 Jul 2008 13:56:21 +0000 (13:56 +0000)]
More explicitly mention that we support the 2920C with the ahc driver.

PR: 125535
Submitted by: Bob Bishop <rb at gid dot co dot uk>
MFC after: 3 days

16 years agoUpdate the ktr_header structure, which changed over time.
remko [Sun, 13 Jul 2008 13:42:52 +0000 (13:42 +0000)]
Update the ktr_header structure, which changed over time.

PR: 125546
Submitted by: Mateusz Guzik <mjguzik at gmail dot com>
MFC after: 3 days

16 years agoMake uart(4) the default serial port driver on i386 and amd64.
ed [Sun, 13 Jul 2008 07:20:14 +0000 (07:20 +0000)]
Make uart(4) the default serial port driver on i386 and amd64.

The uart(4) driver has the advantage of supporting a wider variety of
hardware on a greater amount of platforms. This driver has already been
the standard on platforms such as ia64, powerpc and sparc64.

I've decided not to change anything on pc98. I'd rather let people from
the pc98 team look at this.

Approved by: philip (mentor), marcel

16 years agofix multicast hash register definition
ticso [Sat, 12 Jul 2008 23:40:07 +0000 (23:40 +0000)]
fix multicast hash register definition

16 years agoRefine the changes made in SVN rev 180430. Specifically, instantiate a new
alc [Sat, 12 Jul 2008 21:24:42 +0000 (21:24 +0000)]
Refine the changes made in SVN rev 180430.  Specifically, instantiate a new
page table page only if the 2MB page mapping has been used.  Also, refactor
some assertions.

16 years agoIn nmount(), if we see "update" in the mount options,
rodrigc [Sat, 12 Jul 2008 20:12:40 +0000 (20:12 +0000)]
In nmount(), if we see "update" in the mount options,
set MNT_UPDATE in fsflags, and delete the
"update" option from the global mount options.

MNT_UPDATE is a command, and not a property of a mount
that should persist after the command is executed.

We need to do similar things for MNT_FORCE and MNT_RELOAD.

All mount flags are prefixed by MNT_..... it would
be nice if flags which were commands were named differently
from flags which are persistent properties of a mount.
This was not such a big deal in the pre-nmount() days,
but with nmount() it is more important.

Requested by: yar
MFC after: 2 weeks

16 years agoIn order to apply pmap_demote_pde() to a page directory entry (PDE) from the
alc [Sat, 12 Jul 2008 18:43:57 +0000 (18:43 +0000)]
In order to apply pmap_demote_pde() to a page directory entry (PDE) from the
direct map, the PDE must have PG_M and PG_A preset.

Noticed by: Magesh Dhasayyan

16 years agoStrongly discourage the use of the query-source option, and explain why.
dougb [Sat, 12 Jul 2008 10:00:36 +0000 (10:00 +0000)]
Strongly discourage the use of the query-source option, and explain why.

Give a better example if a user absolutely must use this option, and
suggest they pick something from the ephemeral port range rather than
port 53. This means that the example will not work if it is merely
uncommented, but this will hopefully encourage users to read the comment.

16 years agoMerge from vendor/bind9/dist as of the 9.4.2-P1 import, including
dougb [Sat, 12 Jul 2008 09:38:35 +0000 (09:38 +0000)]
Merge from vendor/bind9/dist as of the 9.4.2-P1 import, including
the patch from ISC for lib/bind9/check.c and deletion of unused
files in lib/bind.

This version will by default randomize the UDP query source port
(and sequence number of course) for every query.

In order to take advantage of this randomization users MUST have an
appropriate firewall configuration to allow UDP queries to be sent and
answers to be received on random ports; and users MUST NOT specify a
port number using the query-source[-v6] options.

The avoid-v[46]-udp-ports options exist for users who wish to eliminate
certain port numbers from being chosen by named for this purpose. See
the ARM Chatper 6 for more information.

Also please note, this issue applies only to UDP query ports. A random
ephemeral port is always chosen for TCP queries.

This issue applies primarily to name servers whose main purpose is to
resolve random queries (sometimes referred to as "caching" servers, or
more properly as "resolving" servers), although even an "authoritative"
name server will make some queries, primarily at startup time.

All users of BIND are strongly encouraged to upgrade to the latest
version, and to utilize the source port randomization feature.

This update addresses issues raised in:
http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2008-1447
http://www.kb.cert.org/vuls/id/800113
http://tools.ietf.org/html/draft-ietf-dnsext-forgery-resilience

16 years agoThe vendor area is the proper home for these files now.
dougb [Sat, 12 Jul 2008 08:46:21 +0000 (08:46 +0000)]
The vendor area is the proper home for these files now.

16 years agoA number of significant enhancements to the ciss driver:
scottl [Fri, 11 Jul 2008 21:20:51 +0000 (21:20 +0000)]
A number of significant enhancements to the ciss driver:

1.  The FreeBSD driver was setting an interrupt coalesce delay of 1000us
for reasons that I can only speculate on.  This was hurting everything
from lame sequential I/O "benchmarks" to legitimate filesystem metadata
operations that relied on serialized barrier writes.  One of my
filesystem tests went from 35s to complete down to 6s.

2.  Implemented the Performant transport method.  Without the fix in
(1), I saw almost no difference.  With it, my filesystem tests showed
another 5-10% improvement in speed.  It was hard to measure CPU
utilization in any meaningful way, so it's not clear if there was a
benefit there, though there should have been since the interrupt handler
was reduced from 2 or more PCI reads down to 1.

3.  Implemented MSI-X.  Without any docs on this, I was just taking a
guess, and it appears to only work with the Performant method.  This
could be a programming or understanding mistake on my part.  While this
by itself made almost no difference to performance since the Performant
method already eliminated most of the synchronous reads over the PCI
bus, it did allow the CISS hardware to stop sharing its interrupt with
the USB hardware, which in turn allowed the driver to become decoupled
from the Giant-locked USB driver stack.  This increased performance by
almost 20%.  The MSI-X setup was done with 4 vectors allocated, but only
1 vector used since the performant method was told to only use 1 of 4
queues.  Fiddling with this might make it work with the simpleq method,
not sure.  I did not implement MSI since I have no MSI-specific hardware
in my test lab.

4.  Improved the locking in the driver, trimmed some data structures.
This didn't improve test times in any measurable way, but it does look
like it gave a minor improvement to CPU usage when many
processes/threads were doing I/O in parallel.  Again, this was hard to
accurately test.

16 years agoDon't leak DMA map if not freed.
delphij [Fri, 11 Jul 2008 18:26:12 +0000 (18:26 +0000)]
Don't leak DMA map if not freed.

Submitted by: kevlo

16 years agoDust off old code for support of USB isochronous transfers.
emax [Fri, 11 Jul 2008 17:13:43 +0000 (17:13 +0000)]
Dust off old code for support of USB isochronous transfers.
USB isochronous transfer support is required for Bluetooth SCO.
While i'm here change u_int to uint and update TODO.
This should produce no visible changes unless the device is
broken (or really old).

MFC after: 3 months

16 years ago- Fix a logic error when updating plex configuration.
lulf [Fri, 11 Jul 2008 16:46:29 +0000 (16:46 +0000)]
- Fix a logic error when updating plex configuration.

Approved by: pjd (mentor)

16 years agoPut the _cpuset* symbols in FBSDprivate_1.0 instead of trying to put
brooks [Fri, 11 Jul 2008 15:17:06 +0000 (15:17 +0000)]
Put the _cpuset* symbols in FBSDprivate_1.0 instead of trying to put
nonexistant __cpuset* there.

16 years agoImprove readability and cscope searches a little bit by not using the
obrien [Fri, 11 Jul 2008 14:48:28 +0000 (14:48 +0000)]
Improve readability and cscope searches a little bit by not using the
same variable name in closely related (but not conflicting) contexts.

16 years agoUse the VM_ALLOC_INTERRUPT for the page requests when allocating memory
kib [Fri, 11 Jul 2008 11:27:42 +0000 (11:27 +0000)]
Use the VM_ALLOC_INTERRUPT for the page requests when allocating memory
for the bio for swapout write. It allows the page allocator to drain
free page list deeper. As result, a deadlock where pageout deamon sleeps
waiting for bio to be allocated for swapout is no more reproducable in
practice.

Alan said that M_USE_RESERVE shall be ressurrected and used there, but
until this is implemented, M_NOWAIT does exactly what is needed.

Tested by: pho, kris
Reviewed by: alc
No objections from: phk
MFC after: 2 weeks (RELENG_7 only)

16 years agoMake it atomic for the devfs_populate_loop() to see the setting of
kib [Fri, 11 Jul 2008 11:22:19 +0000 (11:22 +0000)]
Make it atomic for the devfs_populate_loop() to see the setting of
SI_ALIAS flag and initialization of the si_parent when alias is created.
Assert that supplied parent device is not NULL.

Both situations could cause NULL dereference in the
devfs_populate_loop() when creating a symlink for SI_ALIAS'ed device.
Namely, cdp->cdp_c.si_parent may be NULL.

Reported by: mav
MFC after: 2 weeks

16 years agoThe pfctl(8) program is already pretty verbose, so don't print extra
mtm [Fri, 11 Jul 2008 08:11:49 +0000 (08:11 +0000)]
The pfctl(8) program is already pretty verbose, so don't print extra
information in quiet mode.

16 years agoRevert r180431.
obrien [Fri, 11 Jul 2008 01:10:40 +0000 (01:10 +0000)]
Revert r180431.
r180431 broke the AMD64 build (the only arch using kern/link_elf_obj.c)

16 years agoDetect ATA controllers in the Macbook3.
remko [Thu, 10 Jul 2008 21:36:53 +0000 (21:36 +0000)]
Detect ATA controllers in the Macbook3.

PR: 118135
Submitted by: ed
Approved by: imp (mentor, implicit)
MFC after: 1 week

16 years agostyle(9): put parentheses around return values.
brooks [Thu, 10 Jul 2008 19:54:34 +0000 (19:54 +0000)]
style(9): put parentheses around return values.

16 years agoFix -o mntudp after the conversion to default TCP mounts
kris [Thu, 10 Jul 2008 17:53:10 +0000 (17:53 +0000)]
Fix -o mntudp after the conversion to default TCP mounts

Submitted by: danfe
MFC after: 1 week

16 years agoRegen
brooks [Thu, 10 Jul 2008 17:46:58 +0000 (17:46 +0000)]
Regen

16 years agoid_t is a 64-bit integer and thus is passed as two arguments like off_t is.
brooks [Thu, 10 Jul 2008 17:45:57 +0000 (17:45 +0000)]
id_t is a 64-bit integer and thus is passed as two arguments like off_t is.
As a result, those arguments must be recombined before calling the real
syscal implementation.  This change fixes 32-bit compatibility for
cpuset_getid(), cpuset_setid(), cpuset_getaffinity(), and
cpuset_setaffinity().

16 years agoAllow 'elf_file_t' to be used in a wider scope.
obrien [Thu, 10 Jul 2008 16:35:57 +0000 (16:35 +0000)]
Allow 'elf_file_t' to be used in a wider scope.

16 years agoExtend pmap_demote_pde() to include the ability to instantiate a new page
alc [Thu, 10 Jul 2008 16:22:24 +0000 (16:22 +0000)]
Extend pmap_demote_pde() to include the ability to instantiate a new page
table page where none existed before.

16 years agoSlightly rearrange validation of UDP arguments and jail processing in
rwatson [Thu, 10 Jul 2008 16:20:18 +0000 (16:20 +0000)]
Slightly rearrange validation of UDP arguments and jail processing in
udp_output() so that argument validation occurs before jail processing.

Add additional comments explaining what's going on when we process
addresses and binding during udp_output().

MFC after: 3 weeks

16 years agoUse 'CSCOPE_ARCHDIR' to change the default architecture directories to
obrien [Thu, 10 Jul 2008 16:03:24 +0000 (16:03 +0000)]
Use 'CSCOPE_ARCHDIR' to change the default architecture directories to
cscope.  After the addition of sys/modules/dtrace/dtrace, setting
'ARCHDIR' in /etc/src.conf breaks the build.

16 years agoPass the ucred along into in{,6}_pcblookup_local for upcoming
bz [Thu, 10 Jul 2008 13:31:11 +0000 (13:31 +0000)]
Pass the ucred along into in{,6}_pcblookup_local for upcoming
prison checks.

Reviewed by: rwatson

16 years agoadd __unused to succeed at WARNS=6
charnier [Thu, 10 Jul 2008 13:26:46 +0000 (13:26 +0000)]
add __unused to succeed at WARNS=6

16 years agoFor consistency take lport as u_short in in{,6}_pcblookup_local.
bz [Thu, 10 Jul 2008 13:23:22 +0000 (13:23 +0000)]
For consistency take lport as u_short in in{,6}_pcblookup_local.
All callers either pass in an u_short or u_int16_t.

Reviewed by: rwatson

16 years agoBack-out my previous change. See the thread at
danger [Thu, 10 Jul 2008 11:24:41 +0000 (11:24 +0000)]
Back-out my previous change. See the thread at
http://lists.freebsd.org/pipermail/cvs-all/2008-July/263779.html

16 years agoGarbage collect a variable which is assigned a value once but otherwise
cperciva [Thu, 10 Jul 2008 09:50:55 +0000 (09:50 +0000)]
Garbage collect a variable which is assigned a value once but otherwise
is never used.

Found by: LLVM/Clang Static Analyzer

16 years agoApply the MAC label to an outgoing UDP packet when other inpcb properties are
rwatson [Thu, 10 Jul 2008 09:45:28 +0000 (09:45 +0000)]
Apply the MAC label to an outgoing UDP packet when other inpcb properties are
processed, meaning that we avoid the cost of MAC label assignment if we're
going to drop the packet due to mbuf exhaustion, etc.

MFC after: 3 weeks

16 years agothe driver don't support AD-HOC mode currently.
weongyo [Thu, 10 Jul 2008 08:44:02 +0000 (08:44 +0000)]
the driver don't support AD-HOC mode currently.

16 years agoMerge gnu cpio 2.6 -> 2.8 changes. Unfortunately, we have massive
peter [Thu, 10 Jul 2008 02:08:00 +0000 (02:08 +0000)]
Merge gnu cpio 2.6 -> 2.8 changes.  Unfortunately, we have massive
conflicts due to radically different approaches to security and bug fixes.
In some cases I re-started from the vendor version and reimplemented our
patches.  Fortunately, this is not enabled by default in -current.

16 years agoImport heavily trimmed gnu cpio-2.8.
peter [Thu, 10 Jul 2008 00:52:03 +0000 (00:52 +0000)]
Import heavily trimmed gnu cpio-2.8.
(2.9 switched to gpl3 a few weeks later)

16 years agoGet in some basic infrastructure for Bluetooth SCO support.
emax [Thu, 10 Jul 2008 00:15:29 +0000 (00:15 +0000)]
Get in some basic infrastructure for Bluetooth SCO support.

MFC after: 3 months

16 years agoFlatten cvs2svn generated import tree.
peter [Wed, 9 Jul 2008 23:21:55 +0000 (23:21 +0000)]
Flatten cvs2svn generated import tree.

16 years agocvs2svn managed to overlay two different path trees. fix.
peter [Wed, 9 Jul 2008 23:07:03 +0000 (23:07 +0000)]
cvs2svn managed to overlay two different path trees. fix.

16 years agoAdd a new program to the multicast test suite. The mcgrab program
gnn [Wed, 9 Jul 2008 22:33:46 +0000 (22:33 +0000)]
Add a new program to the multicast test suite.  The mcgrab program
is used to grab and hold some number of multicast addresses in order
to test what happens when an interface goes over the number of multicast
addresses it can filter in hardware.