]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
15 years agoAdd a comment about what may be happening when we get certain
Warner Losh [Mon, 1 Jun 2009 16:22:01 +0000 (16:22 +0000)]
Add a comment about what may be happening when we get certain
messages.  No change to actual code.

15 years agoRegenerate generated syscall files following changes to struct sysent in
Robert Watson [Mon, 1 Jun 2009 16:14:38 +0000 (16:14 +0000)]
Regenerate generated syscall files following changes to struct sysent in
r193234.

15 years agoAdd 'sy_flags', a currently unused per-syscall entry flags field that will
Robert Watson [Mon, 1 Jun 2009 16:13:06 +0000 (16:13 +0000)]
Add 'sy_flags', a currently unused per-syscall entry flags field that will
see future use in 9-CURRENT and 8-STABLE for features such as the
capability-mode enable flag and pay-as-you-audit.

Discussed with: jhb, sson

15 years agoUpdate UPDATING for NETISR2 merge, fix a typo in another UPDATING entry.
Robert Watson [Mon, 1 Jun 2009 16:00:36 +0000 (16:00 +0000)]
Update UPDATING for NETISR2 merge, fix a typo in another UPDATING entry.

15 years agoConvert the two dimensional array to be malloced and introduce
Bjoern A. Zeeb [Mon, 1 Jun 2009 15:49:42 +0000 (15:49 +0000)]
Convert the two dimensional array to be malloced and introduce
an accessor function to get the correct rnh pointer back.

Update netstat to get the correct pointer using kvm_read()
as well.

This not only fixes the ABI problem depending on the kernel
option but also permits the tunable to overwrite the kernel
option at boot time up to MAXFIBS, enlarging the number of
FIBs without having to recompile. So people could just use
GENERIC now.

Reviewed by: julian, rwatson, zec
X-MFC: not possible

15 years agoMerge fixes from p4:
Bruce M Simpson [Mon, 1 Jun 2009 15:30:18 +0000 (15:30 +0000)]
Merge fixes from p4:
 * Tighten v1 query input processing.
 * Borrow changes from MLDv2 for how general queries are processed.
   * Do address field validation upfront before accepting input.
   * Do NOT switch protocol version if old querier present timer active.
 * Always clear IGMPv3 state in igmp_v3_cancel_link_timers().
 * Update comments.

Tested by: deeptech71 at gmail dot com

15 years agoGarbage collect NETISR_POLL and NETISR_POLLMORE, which are no longer
Robert Watson [Mon, 1 Jun 2009 15:03:58 +0000 (15:03 +0000)]
Garbage collect NETISR_POLL and NETISR_POLLMORE, which are no longer
required for options DEVICE_POLLING.

De-fragment the NETISR_ constant space and lower NETISR_MAXPROT from
32 to 16 -- when sizing queue arrays using this compile-time constant,
significant amounts of memory are saved.

Warn on the console when tunable values for netisr are automatically
adjusted during boot due to exceeding limits, invalid values, or as a
result of DEVICE_POLLING.

15 years agoAdd a missing parameter when displaying GPT partitions with an unknown
John Baldwin [Mon, 1 Jun 2009 14:20:13 +0000 (14:20 +0000)]
Add a missing parameter when displaying GPT partitions with an unknown
UUID.

Submitted by: Pawel Worach  pawel.worach | gmail
MFC after: 1 week

15 years agoComment out old Realtek ALC883 quirk, that was disabling phantop power on
Alexander Motin [Mon, 1 Jun 2009 13:13:47 +0000 (13:13 +0000)]
Comment out old Realtek ALC883 quirk, that was disabling phantop power on
mic inputs. I have no idea what for it was made that time, but now I have
several reports that it should be removed to make microphones work. If
this quirk is still required for some systems then they should be identified
and specified explicitly.

15 years agouse explicit 'unsigned int' instead of just the implicit-style 'unsigned' to make...
Ralf S. Engelschall [Mon, 1 Jun 2009 11:38:38 +0000 (11:38 +0000)]
use explicit 'unsigned int' instead of just the implicit-style 'unsigned' to make linting tools (e.g. FlexeLint) happy, too

15 years agoalign coding style with style(9) to avoid misunderstandings
Ralf S. Engelschall [Mon, 1 Jun 2009 11:11:46 +0000 (11:11 +0000)]
align coding style with style(9) to avoid misunderstandings

15 years agocorrectly test for __GNUC__ macro (non-GCC compilers do not have it defined at all)
Ralf S. Engelschall [Mon, 1 Jun 2009 11:02:09 +0000 (11:02 +0000)]
correctly test for __GNUC__ macro (non-GCC compilers do not have it defined at all)

15 years agobe more type correct and align local ckmalloc() with its underlying malloc(3) by...
Ralf S. Engelschall [Mon, 1 Jun 2009 10:50:17 +0000 (10:50 +0000)]
be more type correct and align local ckmalloc() with its underlying malloc(3) by using a "size_t" instead of an "int" argument

15 years agoReimplement the netisr framework in order to support parallel netisr
Robert Watson [Mon, 1 Jun 2009 10:41:38 +0000 (10:41 +0000)]
Reimplement the netisr framework in order to support parallel netisr
threads:

- Support up to one netisr thread per CPU, each processings its own
  workstream, or set of per-protocol queues.  Threads may be bound
  to specific CPUs, or allowed to migrate, based on a global policy.

  In the future it would be desirable to support topology-centric
  policies, such as "one netisr per package".

- Allow each protocol to advertise an ordering policy, which can
  currently be one of:

  NETISR_POLICY_SOURCE: packets must maintain ordering with respect to
    an implicit or explicit source (such as an interface or socket).

  NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work,
    as well as allowing protocols to provide a flow generation function
    for mbufs without flow identifers (m2flow).  Falls back on
    NETISR_POLICY_SOURCE if now flow ID is available.

  NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for
    each packet handled by netisr (m2cpuid).

- Provide utility functions for querying the number of workstreams
  being used, as well as a mapping function from workstream to CPU ID,
  which protocols may use in work placement decisions.

- Add explicit interfaces to get and set per-protocol queue limits, and
  get and clear drop counters, which query data or apply changes across
  all workstreams.

- Add a more extensible netisr registration interface, in which
  protocols declare 'struct netisr_handler' structures for each
  registered NETISR_ type.  These include name, handler function,
  optional mbuf to flow ID function, optional mbuf to CPU ID function,
  queue limit, and ordering policy.  Padding is present to allow these
  to be expanded in the future.  If no queue limit is declared, then
  a default is used.

- Queue limits are now per-workstream, and raised from the previous
  IFQ_MAXLEN default of 50 to 256.

- All protocols are updated to use the new registration interface, and
  with the exception of netnatm, default queue limits.  Most protocols
  register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use
  NETISR_POLICY_FLOW, and will therefore take advantage of driver-
  generated flow IDs if present.

- Formalize a non-packet based interface between interface polling and
  the netisr, rather than having polling pretend to be two protocols.
  Provide two explicit hooks in the netisr worker for start and end
  events for runs: netisr_poll() and netisr_pollmore(), as well as a
  function, netisr_sched_poll(), to allow the polling code to schedule
  netisr execution.  DEVICE_POLLING still embeds single-netisr
  assumptions in its implementation, so for now if it is compiled into
  the kernel, a single and un-bound netisr thread is enforced
  regardless of tunable configuration.

In the default configuration, the new netisr implementation maintains
the same basic assumptions as the previous implementation: a single,
un-bound worker thread processes all deferred work, and direct dispatch
is enabled by default wherever possible.

Performance measurement shows a marginal performance improvement over
the old implementation due to the use of batched dequeue.

An rmlock is used to synchronize use and registration/unregistration
using the framework; currently, synchronized use is disabled
(replicating current netisr policy) due to a measurable 3%-6% hit in
ping-pong micro-benchmarking.  It will be enabled once further rmlock
optimization has taken place.  However, in practice, netisrs are
rarely registered or unregistered at runtime.

A new man page for netisr will follow, but since one doesn't currently
exist, it hasn't been updated.

This change is not appropriate for MFC, although the polling shutdown
handler should be merged to 7-STABLE.

Bump __FreeBSD_version.

Reviewed by: bz

15 years agoDocument IP_BINDANY IP socket option.
Pawel Jakub Dawidek [Mon, 1 Jun 2009 10:30:52 +0000 (10:30 +0000)]
Document IP_BINDANY IP socket option.

Reviewed by: brueffer

15 years ago- Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent
Pawel Jakub Dawidek [Mon, 1 Jun 2009 10:30:00 +0000 (10:30 +0000)]
- Rename IP_NONLOCALOK IP socket option to IP_BINDANY, to be more consistent
  with OpenBSD (and BSD/OS originally). We can't easly do it SOL_SOCKET option
  as there is no more space for more SOL_SOCKET options, but this option also
  fits better as an IP socket option, it seems.
- Implement this functionality also for IPv6 and RAW IP sockets.
- Always compile it in (don't use additional kernel options).
- Remove sysctl to turn this functionality on and off.
- Introduce new privilege - PRIV_NETINET_BINDANY, which allows to use this
  functionality (currently only unjail root can use it).

Discussed with: julian, adrian, jhb, rwatson, kmacy

15 years agoDocument EINVAL for bind(2).
Pawel Jakub Dawidek [Mon, 1 Jun 2009 09:32:12 +0000 (09:32 +0000)]
Document EINVAL for bind(2).

Reviewed by: rwatson
Obtained from: SuSv3

15 years ago- Add missing data argument to printf.
Ulf Lilleengen [Mon, 1 Jun 2009 09:25:32 +0000 (09:25 +0000)]
- Add missing data argument to printf.

Submitted by: Pawel Worach <pawel.worach -AT- gmail.com>
MFC after: 1 week

15 years agoUse the "flag" word consistently.
Edward Tomasz Napierala [Mon, 1 Jun 2009 07:48:27 +0000 (07:48 +0000)]
Use the "flag" word consistently.

Submitted by: Ben Kaduk <minimarmot at gmail.com>

15 years agoAccording to Intel documentation (307013), 3Gbps mode is supported on
Xin LI [Mon, 1 Jun 2009 07:05:52 +0000 (07:05 +0000)]
According to Intel documentation (307013), 3Gbps mode is supported on
Desktop chipsets only for ICH7 series, so mark all ICH7M as ATA_SA150
instead of ATA_SA300.

15 years agoo) Remove some references to long-unsupported old-style config(8) directives.
Juli Mallett [Mon, 1 Jun 2009 06:52:03 +0000 (06:52 +0000)]
o) Remove some references to long-unsupported old-style config(8) directives.
o) Borrow da(4) language about autoconfiguration for ch(4).

15 years agoo) Restructure tcpdrop(8) to provide a facility to try to drop all established
Juli Mallett [Mon, 1 Jun 2009 06:49:09 +0000 (06:49 +0000)]
o) Restructure tcpdrop(8) to provide a facility to try to drop all established
   connections.  Including a flag to instead output a sequence of tcpdrop(8)
   invocations that would accomplish the same thing, which is convenient for
   scripting.
o) Make tcpdrop complain if the addresses given to it are entirely in different
   address families, rather than failing silently.
o) When cross-referencing httpd(8), do not explicitly specify the apache2 port,
   since the example in question is generic.

15 years agoLocal hack to get the build going again while ISC works on a more
Doug Barton [Mon, 1 Jun 2009 06:31:04 +0000 (06:31 +0000)]
Local hack to get the build going again while ISC works on a more
permanent solution for 9.6.1-release.

"My suggestion is to remove the whole attribute construct.
It only suppresses a warning when a function is unused. In this case
the function is defined as inline, so it's not causing a warning when
not used."

Submitted by: marcel

15 years agoEliminate a comment describing code that was deleted over eight years ago.
Alan Cox [Mon, 1 Jun 2009 06:12:08 +0000 (06:12 +0000)]
Eliminate a comment describing code that was deleted over eight years ago.
Move another comment to its proper place.  Fix a typo in a third comment.

15 years agoEliminate the warning that "Values of network_interfaces other than
Doug Barton [Mon, 1 Jun 2009 05:37:13 +0000 (05:37 +0000)]
Eliminate the warning that "Values of network_interfaces other than
AUTO are deprecated.' There is no good reason to deprecate them, and
setting this to different values can be useful for custom solutions
and/or one-off configuration problems.

15 years agoMake the pf and ipfw firewalls start before netif, just like ipfilter
Doug Barton [Mon, 1 Jun 2009 05:35:03 +0000 (05:35 +0000)]
Make the pf and ipfw firewalls start before netif, just like ipfilter
already does. This eliminates a logical inconsistency, and a small
window where the system is open after the network comes up.

15 years agoSubstitute ypset for ypbind in REQUIRE lines. If you use ypset it has to
Doug Barton [Mon, 1 Jun 2009 04:55:13 +0000 (04:55 +0000)]
Substitute ypset for ypbind in REQUIRE lines. If you use ypset it has to
happen right after ypbind, and before anything that uses NIS. The only
change in rcorder accomplished by this patch is make that happen.

PR: conf/117555
Submitted by: John Marshall <john@rwsrv05.mby.riverwillow.net.au>

15 years agoconnect urtw(4) to the amd64/i386 build that it's not tested on the big
Weongyo Jeong [Mon, 1 Jun 2009 02:37:06 +0000 (02:37 +0000)]
connect urtw(4) to the amd64/i386 build that it's not tested on the big
endian machines yet.

15 years agoZyXEL G-202 has zd1211b chipset, not zd1211.
Weongyo Jeong [Mon, 1 Jun 2009 01:51:37 +0000 (01:51 +0000)]
ZyXEL G-202 has zd1211b chipset, not zd1211.

Tested by: Samuel Boivie <samuel at boivie.org>

15 years agosys/boot/common.c
Craig Rodrigues [Mon, 1 Jun 2009 01:02:30 +0000 (01:02 +0000)]
sys/boot/common.c
=================
Extend the loader to parse the root file system mount options in /etc/fstab,
and set a new loader variable vfs.root.mountfrom.options with these options.
The root mount options must be a comma-delimited string, as specified in
/etc/fstab.
Only set the vfs.root.mountfrom.options variable if it has not been
set in the environment.

sys/kern/vfs_mount.c
====================
When mounting the root file system, pass the mount options
specified in vfs.root.mountfrom.options, but filter out "rw" and "noro",
since the initial mount of the root file system must be done as "ro".
While we are here, try to add a few hints to the mountroot prompt
to give users and idea what might of gone wrong during mounting
of the root file system.

Reviewed by: jhb (an earlier patch)

15 years agoCode for parsing nmount options in kernel was merged
Craig Rodrigues [Mon, 1 Jun 2009 00:40:39 +0000 (00:40 +0000)]
Code for parsing nmount options in kernel was merged
to stable/7 branch in r190315.  So only resort to fallback_mount()
could which passes struct nfs_args to kernel in kernel versions
less than 702100.

15 years agoAdd myself to the list of ports committers
Alexander Logvinov [Sun, 31 May 2009 22:33:53 +0000 (22:33 +0000)]
Add myself to the list of ports committers

Approved by: tabthorpe (mentor)

15 years agoFix minor issues in libstand.
Ed Schouten [Sun, 31 May 2009 21:29:07 +0000 (21:29 +0000)]
Fix minor issues in libstand.

- Don't call tftp_makereq() with too many arguments.
- Don't forget to close one of the comments.

Submitted by: Pawel Worach

15 years agoSeveral cleanups to ipfwpcap(8).
Ed Schouten [Sun, 31 May 2009 20:59:20 +0000 (20:59 +0000)]
Several cleanups to ipfwpcap(8).

- Enable WARNS?=6.
- Include missing headers.
- Mark prog and pidfile as static. Remove unneeded initializer.
- Use ANSI prototypes.
- Remove unneeded fp variable.
- snprintf() guarantees the buffer to be null terminated. Remove
  unneeded - 1 and bzero call.
- Remove unneeded casting.

Submitted by: Pawel Worach, Christoph Mallon

15 years agonfs_write() can use the recently introduced vfs_bio_set_valid() instead of
Alan Cox [Sun, 31 May 2009 20:18:02 +0000 (20:18 +0000)]
nfs_write() can use the recently introduced vfs_bio_set_valid() instead of
vfs_bio_set_validclean(), thereby avoiding the page queues lock.

Garbage collect vfs_bio_set_validclean().  Nothing uses it any longer.

15 years agoUnbreak the build. Add missed probes.
Konstantin Belousov [Sun, 31 May 2009 20:16:06 +0000 (20:16 +0000)]
Unbreak the build. Add missed probes.

Reviewed by: rwatson
Pointy hat to: me

15 years agosh: Make read's timeout (-t) apply to the entire line, not only the first
Jilles Tjoelker [Sun, 31 May 2009 19:37:06 +0000 (19:37 +0000)]
sh: Make read's timeout (-t) apply to the entire line, not only the first
character.

This avoids using non-standard behaviour of the old (upto FreeBSD 7) TTY
layer: it reprocesses the input queue when switching to canonical mode. The
new TTY layer does not provide this functionality and so read -t worked
very poorly (first character is not echoed, cannot be backspaced but is
still read).

This also agrees with what most other shells with read -t do.

PR: bin/129566
Reviewed by: stefanf
Approved by: ed (mentor)

15 years agoRestore support for bell pitch/duration.
Ed Schouten [Sun, 31 May 2009 19:35:41 +0000 (19:35 +0000)]
Restore support for bell pitch/duration.

Because we only support a single argument to tf_param, use 16 bits for
the pitch and 16 bits for the duration. While there, make the argument
unsigned. There isn't a single param call that needs a signed integer.

Submitted by: danfe (modified)

15 years agoAdd tests for r193169.
Stefan Farfeleder [Sun, 31 May 2009 17:23:27 +0000 (17:23 +0000)]
Add tests for r193169.

15 years agoUnbreak buildworld.
Marko Zec [Sun, 31 May 2009 15:41:46 +0000 (15:41 +0000)]
Unbreak buildworld.

(not waiting for an approval from mentor (julian) due to emergency)

15 years agoUnlock the pseudofs vnode before calling fill method for pfs_readlink().
Konstantin Belousov [Sun, 31 May 2009 15:01:50 +0000 (15:01 +0000)]
Unlock the pseudofs vnode before calling fill method for pfs_readlink().
The fill code may need to lock another vnode, e.g. procfs file
implementation.

Reviewed by: des
Tested by: pho
MFC after: 2 weeks

15 years agoImplement the bypass routine for VOP_VPTOCNP in nullfs.
Konstantin Belousov [Sun, 31 May 2009 14:58:43 +0000 (14:58 +0000)]
Implement the bypass routine for VOP_VPTOCNP in nullfs.
Among other things, this makes procfs <pid>/file working for executables
started from nullfs mount.

Tested by: pho
PR: 94269, 104938

15 years agoEliminate code duplication in vn_fullpath1() around the cache lookups
Konstantin Belousov [Sun, 31 May 2009 14:57:43 +0000 (14:57 +0000)]
Eliminate code duplication in vn_fullpath1() around the cache lookups
and calls to vn_vptocnp() by moving more of the common code to
vn_vptocnp(). Rename vn_vptocnp() to vn_vptocnp_locked() to signify that
cache is locked around the call.

Do not track buffer position by both the pointer and offset, use only
buflen to record the start of the free space.

Export vn_vptocnp() for external consumers as a wrapper around
vn_vptocnp_locked() that locks the cache and handles hold counts.

Tested by: pho

15 years agoDo not drop vnode interlock in null_checkvp(). null_lock() verifies that
Konstantin Belousov [Sun, 31 May 2009 14:54:20 +0000 (14:54 +0000)]
Do not drop vnode interlock in null_checkvp(). null_lock() verifies that
v_data is not-null before calling NULLVPTOLOWERVP(), and dropping the
interlock allows for reclaim to clean v_data and free the memory.

While there, remove unneeded semicolons and convert the infinite loops
to panics. I have a will to remove null_checkvp() altogether, or leave
it as a trivial stub, but not now.

Reported and tested by: pho

15 years agoLock the real null vnode lock before substitution of vp->v_vnlock.
Konstantin Belousov [Sun, 31 May 2009 14:52:45 +0000 (14:52 +0000)]
Lock the real null vnode lock before substitution of vp->v_vnlock.
This should not really matter for correctness, since vp->v_lock is
not locked before the call, and null_lock() holds the interlock,
but makes the control flow for reclaim more clear.

Tested by: pho

15 years agoAdd a NO_SYNCHRONIZE_CACHE quirk for an AIPTEK2
Daniel Eischen [Sun, 31 May 2009 14:48:51 +0000 (14:48 +0000)]
Add a NO_SYNCHRONIZE_CACHE quirk for an AIPTEK2
part identified as Sunplus Technology Inc.  This
happens to sit in a Rosewill RX81U-ES-25A 2.5" SATA
to USB 2.0 external enclosure.

Reviewed by:    Hans Petter Selasky

15 years agoFix the eval command in combination with set -e. Before this change the shell
Stefan Farfeleder [Sun, 31 May 2009 12:36:14 +0000 (12:36 +0000)]
Fix the eval command in combination with set -e.  Before this change the shell
would always terminate if eval returned with a non-zero exit status regardless
if the status was actually tested.  Unfortunately a new file-scope variable
is needed, the alternative would only be to add a new parameter to all
built-ins.

PR: 134881

15 years agoImplement a variation of the socketpair() syscall which takes a flags
Dmitry Chagin [Sun, 31 May 2009 12:16:31 +0000 (12:16 +0000)]
Implement a variation of the socketpair() syscall which takes a flags
in addition to the type argument.

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

15 years agoSplit native socketpair() syscall onto kern_socketpair() which should
Dmitry Chagin [Sun, 31 May 2009 12:12:38 +0000 (12:12 +0000)]
Split native socketpair() syscall onto kern_socketpair() which should
be used by kernel consumers and socketpair() itself.

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

15 years agoIntroduce an interm userland-kernel API for creating vnets and
Marko Zec [Sun, 31 May 2009 12:10:04 +0000 (12:10 +0000)]
Introduce an interm userland-kernel API for creating vnets and
assigning ifnets from one vnet to another.  Deletion of vnets is not
yet supported.

The interface is implemented as an ioctl extension so that no syscalls
had to be introduced.  This should be acceptable given that the new
interface will be used for a short / interim period only, until the
new jail management framwork gains the capability of managing vnets.
This method for managing vimages / vnets has been in use for the past
7 years without any observable issues.

The userland tool to be used in conjunction with the interim API can be
found in p4: //depot/projects/vimage-commit2/src/usr.sbin/vimage/... and
will most probably never get commited to svn.

While here, bump copyright notices in kern_vimage.c and vimage.h to
cover work done in year 2009.

Approved by: julian (mentor)
Discussed with: bz, rwatson

15 years agoMove new socket flags handling into a separate function as Linux
Dmitry Chagin [Sun, 31 May 2009 12:04:01 +0000 (12:04 +0000)]
Move new socket flags handling into a separate function as Linux
introduced more syscalls which uses these flags.

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

15 years agoRemove empty lines.
Dmitry Chagin [Sun, 31 May 2009 12:00:16 +0000 (12:00 +0000)]
Remove empty lines.

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

15 years agoAllow the bootfs property to be set for raidz pools on FreeBSD.
Doug Rabson [Sun, 31 May 2009 11:59:32 +0000 (11:59 +0000)]
Allow the bootfs property to be set for raidz pools on FreeBSD.

Reviewed by: pjd

15 years agoUnbreak options VIMAGE kernel builds.
Marko Zec [Sun, 31 May 2009 11:57:51 +0000 (11:57 +0000)]
Unbreak options VIMAGE kernel builds.

Approved by: julian (mentor)

15 years agoTeach powerd how to query the PMU AC line state on PowerPC.
Nathan Whitehorn [Sun, 31 May 2009 10:27:24 +0000 (10:27 +0000)]
Teach powerd how to query the PMU AC line state on PowerPC.

15 years agoProvide an analogous sysctl to hw.acpi.acline (dev.pmu.0.acline) to
Nathan Whitehorn [Sun, 31 May 2009 10:02:20 +0000 (10:02 +0000)]
Provide an analogous sysctl to hw.acpi.acline (dev.pmu.0.acline) to
determine whether the computer is plugged in to mains power.

15 years agoUpgrade audit(4) from experimental to production status for FreeBSD 8.0.
Robert Watson [Sun, 31 May 2009 09:03:14 +0000 (09:03 +0000)]
Upgrade audit(4) from experimental to production status for FreeBSD 8.0.
While there remain some incomplete aspects of the implementation (such
as incomplete auditing of some system calls), the implementation has
been burned in for a few years, as well as in GENERIC for a few years.

Obtained from: TrustedBSD Project

15 years agoIntroduce support for cpufreq on PowerPC with the dynamic frequency
Nathan Whitehorn [Sun, 31 May 2009 09:01:23 +0000 (09:01 +0000)]
Introduce support for cpufreq on PowerPC with the dynamic frequency
switching capabilities of the MPC7447A and MPC7448.

15 years agoProvide a new CPU device driver ivar to report the nominal speed of the
Nathan Whitehorn [Sun, 31 May 2009 08:59:15 +0000 (08:59 +0000)]
Provide a new CPU device driver ivar to report the nominal speed of the
CPU, if available. This is meant to solve the issue of cpufreq misreporting
speeds on CPUs that boot in a reduced power mode and have only relative
speed control.

15 years agoFix the MP IPI code to differentiate between bitmapped IPIs and function IPIs.
Adrian Chadd [Sun, 31 May 2009 08:11:39 +0000 (08:11 +0000)]
Fix the MP IPI code to differentiate between bitmapped IPIs and function IPIs.

This attempts to fix the IPI handling code to correctly differentiate
between bitmapped IPIs and function IPIs. The Xen IPIs were on low numbers
which clashed with the bitmapped IPIs.

This commit bumps those IPI numbers up to 240 and above (just like in the i386
code) and fiddles with the ipi_vectors[] logic to call the correct function.

This still isn't "right". Specifically, the IPI code may work fine for TLB
shootdown events but the rendezvous/lazypmap IPIs are thrown by calling ipi_*()
routines which don't set the call_func stuff (function id, addr1, addr2) that
the TLB shootdown events are. So the Xen SMP support is still broken.

PR: 135069

15 years agoRemove some unused code in ipi_selected() .
Adrian Chadd [Sun, 31 May 2009 07:25:24 +0000 (07:25 +0000)]
Remove some unused code in ipi_selected() .

The code path this was copied from (sys/i386/i386/mp_machdep.c:ipi_selected())
handles bitmap'ed IPIs and normal IPIs via separate notification paths. Xen
SMP handles them the same way.

15 years agoUpdate BIND to version 9.6.1rc1. This version has better performance and
Doug Barton [Sun, 31 May 2009 05:44:21 +0000 (05:44 +0000)]
Update BIND to version 9.6.1rc1. This version has better performance and
lots of new features compared to 9.4.x, including:

Full NSEC3 support
Automatic zone re-signing
New update-policy methods tcp-self and 6to4-self
DHCID support.
More detailed statistics counters including those supported in BIND 8.
Faster ACL processing.
Efficient LRU cache-cleaning mechanism.
NSID support.

15 years agoUpdate BIND to version 9.6.1rc1. This version has better performance and
Doug Barton [Sun, 31 May 2009 05:42:58 +0000 (05:42 +0000)]
Update BIND to version 9.6.1rc1. This version has better performance and
lots of new features compared to 9.4.x, including:

Full NSEC3 support
Automatic zone re-signing
New update-policy methods tcp-self and 6to4-self
DHCID support.
More detailed statistics counters including those supported in BIND 8.
Faster ACL processing.
Efficient LRU cache-cleaning mechanism.
NSID support.

15 years agoUse GCC's __SOFTFP__ to test whether we're being compiled
Marcel Moolenaar [Sun, 31 May 2009 02:03:40 +0000 (02:03 +0000)]
Use GCC's __SOFTFP__ to test whether we're being compiled
with softfloat or not. Now -msoft-float can be overridden
more easily.

15 years agoMark the cascaded AT interrupt handler as MP safe to avoid having
Marcel Moolenaar [Sun, 31 May 2009 01:56:06 +0000 (01:56 +0000)]
Mark the cascaded AT interrupt handler as MP safe to avoid having
it grab Giant. The next step would be to make it a filter.

15 years agoVendor import of BIND 9.6.1rc1
Doug Barton [Sun, 31 May 2009 00:11:36 +0000 (00:11 +0000)]
Vendor import of BIND 9.6.1rc1

15 years agoUpdate relative to the BIND 9.6.1rc1 import
Doug Barton [Sun, 31 May 2009 00:03:41 +0000 (00:03 +0000)]
Update relative to the BIND 9.6.1rc1 import

15 years agoRemove the now invalid (and possibly unused) debug.mpsafevfs
Attilio Rao [Sat, 30 May 2009 23:52:23 +0000 (23:52 +0000)]
Remove the now invalid (and possibly unused) debug.mpsafevfs
sysctl/tunable.

Reviewed by: emaste
Sponsored by: Sandvine Incorporated

15 years agoIn preparation for the BIND 9.6.1rc1 import, remove this file.
Doug Barton [Sat, 30 May 2009 23:51:02 +0000 (23:51 +0000)]
In preparation for the BIND 9.6.1rc1 import, remove this file.
We don't use it.

15 years agoIn preparation for the BIND 9.6.1rc1 import, remove this directory.
Doug Barton [Sat, 30 May 2009 23:50:12 +0000 (23:50 +0000)]
In preparation for the BIND 9.6.1rc1 import, remove this directory.
The libbind library is no longer distributed as part of the main
BIND package, and we never built it in any case.

15 years agoIn preparation for the BIND 9.6.1rc1 import, remove these two directories.
Doug Barton [Sat, 30 May 2009 23:48:09 +0000 (23:48 +0000)]
In preparation for the BIND 9.6.1rc1 import, remove these two directories.
We do not install these files so there is little use to keeping them in
the tree, and the drafts directory in particular is the source of a lot
of churn for each new version.

15 years agoCrank the debug level necessary to display the "Label foo is removed"
Doug Barton [Sat, 30 May 2009 22:31:52 +0000 (22:31 +0000)]
Crank the debug level necessary to display the "Label foo is removed"
and "Label for provider ..." messages up from 0 to 1.

15 years agofix xdrmem_control to be safe in an if statement
Kip Macy [Sat, 30 May 2009 22:23:58 +0000 (22:23 +0000)]
fix xdrmem_control to be safe in an if statement
fix zfs to depend on krpc
remove xdr from zfs makefile

Submitted by: dchagin@freebsd.org

15 years agoEliminate a stale comment and the two remaining uses of the "register"
Alan Cox [Sat, 30 May 2009 22:15:55 +0000 (22:15 +0000)]
Eliminate a stale comment and the two remaining uses of the "register"
keyword in this file.

15 years agoAdd a check to v_type == VREG for the recently modified code that
Rick Macklem [Sat, 30 May 2009 22:11:12 +0000 (22:11 +0000)]
Add a check to v_type == VREG for the recently modified code that
does NFSv4 Closes in the experimental client's VOP_INACTIVE().
I also replaced a bunch of ap->a_vp with a local copy of vp,
because I thought that made it more readable.

Approved by: kib (mentor)

15 years agoAdd assertions in two places where a page's valid or dirty bits are changed.
Alan Cox [Sat, 30 May 2009 22:06:58 +0000 (22:06 +0000)]
Add assertions in two places where a page's valid or dirty bits are changed.

15 years agog_part_ebr.c includes opt_geom.h
Marcel Moolenaar [Sat, 30 May 2009 22:04:18 +0000 (22:04 +0000)]
g_part_ebr.c includes opt_geom.h

15 years agoSmall cleanup, add (spurious) quotation marks around the value
Doug Barton [Sat, 30 May 2009 21:51:38 +0000 (21:51 +0000)]
Small cleanup, add (spurious) quotation marks around the value
for name= to make these scripts consistent with the rest.

15 years agoNow that the last of the *.sh scripts are gone from the base,
Doug Barton [Sat, 30 May 2009 21:41:54 +0000 (21:41 +0000)]
Now that the last of the *.sh scripts are gone from the base,
emit a warning if come across one.

15 years agodistribute sysctl decls so global variables can be made static
Sam Leffler [Sat, 30 May 2009 20:11:23 +0000 (20:11 +0000)]
distribute sysctl decls so global variables can be made static

15 years agoo assert TDMA_MAXSLOTS is 2 so noone tries to blindly increase it
Sam Leffler [Sat, 30 May 2009 19:57:31 +0000 (19:57 +0000)]
o assert TDMA_MAXSLOTS is 2 so noone tries to blindly increase it
o add safety belt in vdetach for failed state block allocation
o fix dynamic change to tdma config; ERESTART may not result in
  kicking the state machine so we need to explicitly mark the
  beacon for update

Sponsored by:

15 years agoRemoval of early.sh
Doug Barton [Sat, 30 May 2009 19:39:57 +0000 (19:39 +0000)]
Removal of early.sh

15 years agoAs previously advertised, remove this script prior to the 8.0 branch.
Doug Barton [Sat, 30 May 2009 19:38:51 +0000 (19:38 +0000)]
As previously advertised, remove this script prior to the 8.0 branch.

15 years agoFix NETIF_DEBUG compilation.
Marcel Moolenaar [Sat, 30 May 2009 19:28:38 +0000 (19:28 +0000)]
Fix NETIF_DEBUG compilation.

15 years agowork around snapshot shutdown race reported by Henri Hennebert
Kip Macy [Sat, 30 May 2009 19:26:35 +0000 (19:26 +0000)]
work around snapshot shutdown race reported by Henri Hennebert

15 years agoPrint the returned port number when RPC_DEBUG is defined.
Marcel Moolenaar [Sat, 30 May 2009 19:23:09 +0000 (19:23 +0000)]
Print the returned port number when RPC_DEBUG is defined.
This improves debugging.

15 years agoUnbreak build.
Attilio Rao [Sat, 30 May 2009 18:39:22 +0000 (18:39 +0000)]
Unbreak build.

Pointy hat to: attilio

15 years agomaintain existing style
Sam Leffler [Sat, 30 May 2009 18:23:55 +0000 (18:23 +0000)]
maintain existing style

15 years agoFix return values appropriately.
Attilio Rao [Sat, 30 May 2009 17:56:19 +0000 (17:56 +0000)]
Fix return values appropriately.

Tested by: zec

15 years agoInclude libmd and libcrypto in DPADD
Tim Kientzle [Sat, 30 May 2009 17:26:55 +0000 (17:26 +0000)]
Include libmd and libcrypto in DPADD

15 years agos/rk_npkts/rx_npkts
Attilio Rao [Sat, 30 May 2009 17:25:14 +0000 (17:25 +0000)]
s/rk_npkts/rx_npkts

Reported by: zec

15 years agoEven though I'm not quite sure that the call_func stuff will work properly
Adrian Chadd [Sat, 30 May 2009 15:20:25 +0000 (15:20 +0000)]
Even though I'm not quite sure that the call_func stuff will work properly
in all the places/cases IPI messages will be generated, at least be consistent
with how the call_data pointer is assigned and cleared (ie, all done inside
the spinlock.

Ensure that its NULL before continuing, just to try and identify situations
where things are going horribly wrong.

15 years agoWhen user_frac in the polling subsystem is low it is going to busy the
Attilio Rao [Sat, 30 May 2009 15:14:44 +0000 (15:14 +0000)]
When user_frac in the polling subsystem is low it is going to busy the
CPU for too long period than necessary.  Additively, interfaces are kept
polled (in the tick) even if no more packets are available.
In order to avoid such situations a new generic mechanism can be
implemented in proactive way, keeping track of the time spent on any
packet and fragmenting the time for any tick, stopping the processing
as soon as possible.

In order to implement such mechanism, the polling handler needs to
change, returning the number of packets processed.
While the intended logic is not part of this patch, the polling KPI is
broken by this commit, adding an int return value and the new flag
IFCAP_POLLING_NOCOUNT (which will signal that the return value is
meaningless for the installed handler and checking should be skipped).

Bump __FreeBSD_version in order to signal such situation.

Reviewed by: emaste
Sponsored by: Sandvine Incorporated

15 years agoDon't schedule a CALL_FUNCTION_VECTOR software IPI if the IPI was signaled
Adrian Chadd [Sat, 30 May 2009 14:59:08 +0000 (14:59 +0000)]
Don't schedule a CALL_FUNCTION_VECTOR software IPI if the IPI was signaled
via the bitmap (and thus sent via RESCHEDULE_VECTOR.)

15 years agoBump __FreeBSD_version after addition of VOP_ACCESSX(9).
Edward Tomasz Napierala [Sat, 30 May 2009 14:01:01 +0000 (14:01 +0000)]
Bump __FreeBSD_version after addition of VOP_ACCESSX(9).

15 years agoAdd VOP_ACCESSX, which can be used to query for newly added V*
Edward Tomasz Napierala [Sat, 30 May 2009 13:59:05 +0000 (13:59 +0000)]
Add VOP_ACCESSX, which can be used to query for newly added V*
permissions, such as VWRITE_ACL.  For a filsystems that don't
implement it, there is a default implementation, which works
as a wrapper around VOP_ACCESS.

Reviewed by: rwatson@

15 years agoAdds missing sysctl to manage the vtag_time_wait time. This will
Randall Stewart [Sat, 30 May 2009 11:14:41 +0000 (11:14 +0000)]
Adds missing sysctl to manage the vtag_time_wait time. This will
even allow disabling time-wait all together if you set the value
to 0 (not advisable actually). The default remains the same
i.e. 60 seconds.

15 years agoFix a small memory leak from the nr-sack code - the mapping array
Randall Stewart [Sat, 30 May 2009 10:56:27 +0000 (10:56 +0000)]
Fix a small memory leak from the nr-sack code - the mapping array
was not being freed at term of association. Also get rid of
the MICHAELS_EXP code.

15 years agoMake sctp_uio user to kernel structure match the
Randall Stewart [Sat, 30 May 2009 10:50:40 +0000 (10:50 +0000)]
Make sctp_uio user to kernel structure match the
socket-api draft. Two fields were uint32_t when they
should have been uint16_t.

Reported by Jonathan Leighton at U-del.

15 years agorm, find -delete: fix removing symlinks with uchg/uappnd set.
Jilles Tjoelker [Sat, 30 May 2009 10:42:19 +0000 (10:42 +0000)]
rm, find -delete: fix removing symlinks with uchg/uappnd set.

Formerly, this tried to clear the flags on the symlink's target
instead of the symlink itself.

As before, this only happens for root or for the unlink(1) variant of rm.

PR: bin/111226 (part of)
Submitted by: Martin Kammerhofer
Approved by: ed (mentor)
MFC after: 3 weeks