]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
15 years ago[patch] fortune(6): George Bernard Shaw quote fix
edwin [Thu, 4 Jun 2009 22:01:50 +0000 (22:01 +0000)]
[patch] fortune(6): George Bernard Shaw quote fix

From the original PR:

    s/milestones/millstones/
    and less important..
    s/man/Man/
    Not every source I've seen capitalizes 'Man', but it seems
    right. Uncapitalized 'man' would usually be preceded by
    an 'a'. But I haven't seen any reference cite the orignal
    source yet, so I can't say for sure.

http://quotationsbook.com/quote/31568/

PR: conf/131469
Submitted by: John Hein <jhein@timing.com>
MFC after: 2 days

15 years agoRemove duplicate variable setting.
thompsa [Thu, 4 Jun 2009 22:00:48 +0000 (22:00 +0000)]
Remove duplicate variable setting.

Spotted by: Sylvestre Gallon

15 years agorevert r162516. We only support 1 or 2 channels per stream
thompsa [Thu, 4 Jun 2009 21:59:28 +0000 (21:59 +0000)]
revert r162516. We only support 1 or 2 channels per stream
which reflects mono and stereo.

Submitted by: Hans Petter Selasky

15 years ago[patch] fortune(6): file not Y2.01K compliant
edwin [Thu, 4 Jun 2009 21:55:07 +0000 (21:55 +0000)]
[patch] fortune(6): file not Y2.01K compliant

Update the time in the fortune to make the joke a little bit more
realistic again: Bump year from 2009 to 2039.

PR: conf/129860
Submitted by: Alan Amesbury <amesbury@umn.edu>
MFC after: 2 days

15 years ago[patch] calendar.music: Chuck Berry was born in St. Louis, Missouri, not California
edwin [Thu, 4 Jun 2009 21:48:04 +0000 (21:48 +0000)]
[patch] calendar.music: Chuck Berry was born in St. Louis, Missouri, not California

    The /usr/bin/calendar program reports that Chuck Berry was born
    in San Jose California but he was not born in California.

    Chuck Berry was born in St. Louis, Missouri in 1926 on October 18.

    http://www.chuckberry.com/about/bio.htm
    http://www.khaldea.com/charts/chuckberry.shtml
    http://en.wikipedia.org/wiki/Chuck_Berry

PR: conf/128215
Submitted by: comet--berkeley (aka Pablo Picasso) <comet@transbay.net>
MFC after: 2 days

15 years agotrack rename of CSA ie
sam [Thu, 4 Jun 2009 18:22:21 +0000 (18:22 +0000)]
track rename of CSA ie

Submitted by: wxs

15 years agoWhen checking for shared writes, use the struct mount returned from
ps [Thu, 4 Jun 2009 16:50:03 +0000 (16:50 +0000)]
When checking for shared writes, use the struct mount returned from
vn_start_write.

Reviewed by: jhb

15 years agoSupport shared vnode locks for write operations when the offset is
ps [Thu, 4 Jun 2009 16:18:07 +0000 (16:18 +0000)]
Support shared vnode locks for write operations when the offset is
provided on filesystems that support it.  This really improves mysql
+ innodb performance on ZFS.

Reviewed by: jhb, kmacy, jeffr

15 years agoo station mode channel switch support
sam [Thu, 4 Jun 2009 15:57:38 +0000 (15:57 +0000)]
o station mode channel switch support
o IEEE80211_IOC_CHANSWITCH fixups:
  - restrict to hostap vaps
  - return EOPNOTSUPP instead of EINVAL when applied to !hostap vap
    or to a vap w/o 11h enabled
  - interpret count of 0 to mean cancel the current CSA

Reviewed by: rpaulo, avatar

15 years agoFix upcall races in the client side krpc. For the client side upcall,
rmacklem [Thu, 4 Jun 2009 14:49:27 +0000 (14:49 +0000)]
Fix upcall races in the client side krpc. For the client side upcall,
holding SOCKBUF_LOCK() isn't sufficient to guarantee that there is
no upcall in progress, since SOCKBUF_LOCK() is released/re-acquired
in the upcall. An upcall reference counter was added to the upcall
structure that is incremented at the beginning of the upcall and
decremented at the end of the upcall. As such, a reference count == 0
when holding the SOCKBUF_LOCK() guarantees there is no upcall in
progress. Add a function that is called just after soupcall_clear(),
which waits until the reference count == 0.
Also, move the mtx_destroy() down to after soupcall_clear(), so that
the mutex is not destroyed before upcalls are done.

Reviewed by: dfr, jhb
Tested by: pho
Approved by: kib (mentor)

15 years agoFix two races in the server side krpc w.r.t upcalls:
rmacklem [Thu, 4 Jun 2009 14:13:06 +0000 (14:13 +0000)]
Fix two races in the server side krpc w.r.t upcalls:
  Add a flag so that soupcall_clear() is only called once to cancel
  an upcall.
  Move the test for xprt_registered in the upcall down to after the
  mtx_lock() of the pool mutex, to catch the case where it is
  unregistered while the upcall is waiting for the mutex.
Also, move the mtx_destroy() of the pool mutex to after SVC_RELEASE(),
so that it isn't destroyed before the upcalls are disabled.

Reviewed by: dfr, jhb
Tested by: pho
Approved by: kib (mentor)

15 years agofix a bug introduced in rev.190865 related to the signedness
luigi [Thu, 4 Jun 2009 12:27:57 +0000 (12:27 +0000)]
fix a bug introduced in rev.190865 related to the signedness
of the credit of a pipe. On passing, also use explicit
signed/unsigned types for two other fields.
Noticed by Oleg Bulyzhin and Maxim Ignatenko long ago,
i forgot to commit the fix.

Does not affect RELENG_7.

15 years agoCorrect typo; errno => error.
ed [Thu, 4 Jun 2009 11:22:53 +0000 (11:22 +0000)]
Correct typo; errno => error.

15 years agoRe-add opt_mac.h include, which is required in order for MNT_MULTILABEL
rwatson [Thu, 4 Jun 2009 10:30:18 +0000 (10:30 +0000)]
Re-add opt_mac.h include, which is required in order for MNT_MULTILABEL
to be set properly on devfs.  Otherwise, it isn't possible to set labels
on /dev nodes.

Reported by: Sergio Rodriguez <sergiorr at yahoo.com>
MFC after: 3 days

15 years agoDon't panic in nlm_record_lock if we get ENOENT from lf_advlockasync. This
dfr [Thu, 4 Jun 2009 08:13:51 +0000 (08:13 +0000)]
Don't panic in nlm_record_lock if we get ENOENT from lf_advlockasync. This
is likely to be because the file was just removed and in our context this is
harmless.

15 years agoadd two prerequisites; if_zydreg.h and if_zydfw for dependency check.
weongyo [Thu, 4 Jun 2009 03:59:20 +0000 (03:59 +0000)]
add two prerequisites; if_zydreg.h and if_zydfw for dependency check.

15 years agoreimplements RF logic for GCT chipset (as known as UW2453) to support
weongyo [Thu, 4 Jun 2009 02:49:50 +0000 (02:49 +0000)]
reimplements RF logic for GCT chipset (as known as UW2453) to support
ICIDU NI-707503 which is donated by Nick Hibma (great thanks!).  Though
it has a MAXIM RF (0x8) there's some success reports with using GCT RF
(0x9) codes and it worked well for ICIDU NI-707503 too.  So codes for
MAXIM and GCT RFs are integrated.

Before this commit, if I rememeber correctly, MAXIM RF is never tested
that it seems it's a first report working with FreeBSD.

15 years agocleanups the device match list.
weongyo [Thu, 4 Jun 2009 01:55:13 +0000 (01:55 +0000)]
cleanups the device match list.

15 years agoWhen a channel switch is done to a channel with different operating
sam [Wed, 3 Jun 2009 23:33:09 +0000 (23:33 +0000)]
When a channel switch is done to a channel with different operating
characteristics force the stations to re-associate so protocol state
is re-initialized.  Note that for 11h/DFS this is irrelevant as channel
changes are never cross-band.

Reviewed by: ctlaw

15 years agoAfter a channel switch mark associated stations so they will immediately
sam [Wed, 3 Jun 2009 23:30:25 +0000 (23:30 +0000)]
After a channel switch mark associated stations so they will immediately
be probed as inactive; this more quickly weeds out stations that don't
follow to the new channel.

15 years agoRe-format the nfsv4.4 man page so that all sentences start on a
rmacklem [Wed, 3 Jun 2009 21:50:26 +0000 (21:50 +0000)]
Re-format the nfsv4.4 man page so that all sentences start on a
new line.

Approved by: kib (mentor)

15 years agoRemove unused VM includes.
jhb [Wed, 3 Jun 2009 20:25:13 +0000 (20:25 +0000)]
Remove unused VM includes.

15 years agoAdd one further check with mac_policy_count to an mbuf copying case
rwatson [Wed, 3 Jun 2009 19:41:12 +0000 (19:41 +0000)]
Add one further check with mac_policy_count to an mbuf copying case
(limited to netatalk) to avoid MAC label lookup on both mbufs if no
policies are registered.

Obtained from: TrustedBSD Project

15 years agoAdjust the padding of struct pcpu to r193219.
marius [Wed, 3 Jun 2009 19:31:26 +0000 (19:31 +0000)]
Adjust the padding of struct pcpu to r193219.

Submitted by: Eygene Ryabinkin

15 years agoContinue work to optimize performance of "options MAC" when no MAC policy
rwatson [Wed, 3 Jun 2009 18:46:28 +0000 (18:46 +0000)]
Continue work to optimize performance of "options MAC" when no MAC policy
modules are loaded by avoiding mbuf label lookups when policies aren't
loaded, pushing further socket locking into MAC policy modules, and
avoiding locking MAC ifnet locks when no policies are loaded:

- Check mac_policies_count before looking for mbuf MAC label m_tags in MAC
  Framework entry points.  We will still pay label lookup costs if MAC
  policies are present but don't require labels (typically a single mbuf
  header field read, but perhaps further indirection if IPSEC or other
  m_tag consumers are in use).

- Further push socket locking for socket-related access control checks and
  events into MAC policies from the MAC Framework, so that sockets are
  only locked if a policy specifically requires a lock to protect a label.
  This resolves lock order issues during sonewconn() and also in local
  domain socket cross-connect where multiple socket locks could not be
  held at once for the purposes of propagatig MAC labels across multiple
  sockets.  Eliminate mac_policy_count check in some entry points where it
  no longer avoids locking.

- Add mac_policy_count checking in some entry points relating to network
  interfaces that otherwise lock a global MAC ifnet lock used to protect
  ifnet labels.

Obtained from: TrustedBSD Project

15 years ago- Remove unused sparc64-bitops.h file. Our ext2fs code doesn't use
stas [Wed, 3 Jun 2009 17:30:10 +0000 (17:30 +0000)]
- Remove unused sparc64-bitops.h file. Our ext2fs code doesn't use
  sparc64-specific bitops implemetations and relies on generic ones.
  Furthermore, bitops implementations present in sparc64-bitops.h
  are written in C similarly to generic bitops.

15 years agotreat IEEE80211_S_CSA as a "running state"; this fixes
sam [Wed, 3 Jun 2009 17:25:19 +0000 (17:25 +0000)]
treat IEEE80211_S_CSA as a "running state"; this fixes
ap mode 11h channel switch announcements

15 years agoTrim an obsolete comment.
jkoshy [Wed, 3 Jun 2009 17:19:12 +0000 (17:19 +0000)]
Trim an obsolete comment.

Noticed by: lifengkai <lifengkai at huawei dot com>

15 years agoMake GPT style partitiong endian-safe in U-Boot support library.
raj [Wed, 3 Jun 2009 16:28:29 +0000 (16:28 +0000)]
Make GPT style partitiong endian-safe in U-Boot support library.

Submitted by: Piotr Ziecik
Obtained from: Semihalf

15 years ago- Style(9) improvements.
stas [Wed, 3 Jun 2009 14:18:37 +0000 (14:18 +0000)]
- Style(9) improvements.
- Convert all K&R definitions to ANSI equialents.
- Retire bsd_malloc and bsd_free macros and
  use malloc/free directly.
- Drop some unused debugging calls.

This commit brings no functional changes.

15 years ago- Sync our copies of ext2fs Linux headers to current Linux versions.
stas [Wed, 3 Jun 2009 13:25:50 +0000 (13:25 +0000)]
- Sync our copies of ext2fs Linux headers to current Linux versions.
  Minimize differencies between our ext2fs headers and relevant Linux
  versions by using EXT2_SB macro to access the superblock fields. Most
  of the differencies in access to these fields are now hidden inside
  this macro.
- Rename the s_db_per_group field of ext2fs_sb_info to s_gdb_count
  to reflect the similar change in Linux headers. New name also seem
  to be more appropriate for this field.
- Use proper types for s_first_inode and s_inode_size in-core superblock
  fields. Now they reflec types used in the on-disk superblock version.
- Add support for older filesystem revisions that doesn't have proper
  s_first_ino and s_inode_size fields in the on-disk superblock. In these
  cases predefined values for these fields are used.
- Add simple sanity checks for s_first_inode and s_inode_size correctness.

Reviewed by: bde (previous version)
MFC after: 2 weeks

15 years agoAdd vm_lowmem event handler for dirhash. This will cause dirhashes to be
snb [Wed, 3 Jun 2009 09:44:22 +0000 (09:44 +0000)]
Add vm_lowmem event handler for dirhash. This will cause dirhashes to be
deleted when the system is low on memory. This ought to allow an increase to
vfs.ufs.dirhash_maxmem on machines that have lots of memory, without
degrading performance by having too much memory reserved for dirhash when
other things need it. The default value for dirhash_maxmem is being kept at
2MB for now, though.

This work was mostly done during the 2008 Google Summer of Code.

Approved by: dwmalone (mentor), re
MFC after: 3 months

15 years agoWhere if not in examples we should follow style(9)?
pjd [Wed, 3 Jun 2009 09:28:58 +0000 (09:28 +0000)]
Where if not in examples we should follow style(9)?

15 years agolchflags(2) takes int, not u_long like chflags(2) and fchflags(2).
pjd [Wed, 3 Jun 2009 09:24:58 +0000 (09:24 +0000)]
lchflags(2) takes int, not u_long like chflags(2) and fchflags(2).
Strange, isn't it?

Pointed out by: bde

15 years agoCorrect comment.
pjd [Wed, 3 Jun 2009 09:23:31 +0000 (09:23 +0000)]
Correct comment.

15 years agoBy default, label all network interfaces as biba/equal on attach. This
rwatson [Wed, 3 Jun 2009 08:49:44 +0000 (08:49 +0000)]
By default, label all network interfaces as biba/equal on attach.  This
makes it easier for first-time users to configure and work with biba as
remote acess is still allowed.  Effectively, this means that, by default,
only local security properties, not distributed ones, are enforced.

Obtained from: TrustedBSD Project

15 years agoUse ISO C99 style inline semantics in msun.
ed [Wed, 3 Jun 2009 08:16:34 +0000 (08:16 +0000)]
Use ISO C99 style inline semantics in msun.

Because we use ISO C99 nowadays, we can just get rid of enforcing
GNU89-style inlining.

15 years agocalls callout_drain(9) to un-schedule a scan timer to prevent a page
weongyo [Wed, 3 Jun 2009 04:10:22 +0000 (04:10 +0000)]
calls callout_drain(9) to un-schedule a scan timer to prevent a page
fault in softclock.

Submitted by: sam
Reviewed by: jhb, sam (original version), thompsa

15 years agoMark MAC Framework sx and rm locks as NOWITNESS to suppress warnings that
rwatson [Tue, 2 Jun 2009 22:22:09 +0000 (22:22 +0000)]
Mark MAC Framework sx and rm locks as NOWITNESS to suppress warnings that
might arise from WITNESS not understanding its locking protocol, which
should be deadlock-free.  Currently these warnings generally don't occur,
but as object locking is pushed into policies for some object types, they
would otherwise occur more often.

Obtained from: TrustedBSD Project

15 years agoAdd support for the experimental nfs subsystem to the scripts in
rmacklem [Tue, 2 Jun 2009 22:15:47 +0000 (22:15 +0000)]
Add support for the experimental nfs subsystem to the scripts in
/etc/rc.d. They use the following new rc variables:
  nfsv4_server_enable - set to "YES" to run the experimental server
  nfsuserd_enable - set to "YES" to run nfsuserd for NFSv4 client and
    server
  nfsuserd_flags - command line flags for nfsuserd
  nfscbd_enable - set to "YES" to run the experimental nfs client's
    NFSv4 callback daemon
  nfscbd_flags - command line flags for nfscbd

Reviewed by: dougb
Approved by: kib (mentor)

15 years agoimprove raw xmit failure handling
sam [Tue, 2 Jun 2009 21:17:56 +0000 (21:17 +0000)]
improve raw xmit failure handling

15 years agocount frag tx failures as an ifnet error
sam [Tue, 2 Jun 2009 21:13:57 +0000 (21:13 +0000)]
count frag tx failures as an ifnet error

15 years agofix comment
sam [Tue, 2 Jun 2009 21:12:07 +0000 (21:12 +0000)]
fix comment

15 years agorestart tdma beacons after vap destroy
sam [Tue, 2 Jun 2009 21:11:26 +0000 (21:11 +0000)]
restart tdma beacons after vap destroy

15 years agoremove another vestige of the null if_softc on detach hack
sam [Tue, 2 Jun 2009 20:51:59 +0000 (20:51 +0000)]
remove another vestige of the null if_softc on detach hack

15 years agofix typo
sam [Tue, 2 Jun 2009 20:48:12 +0000 (20:48 +0000)]
fix typo

15 years agofix setting of ni_txrate
sam [Tue, 2 Jun 2009 20:32:13 +0000 (20:32 +0000)]
fix setting of ni_txrate

Submitted by: "Paul B. Mahol" <onemda@gmail.com>

15 years agopartially fix mode setting; this no longer returns an error but still
sam [Tue, 2 Jun 2009 20:00:43 +0000 (20:00 +0000)]
partially fix mode setting; this no longer returns an error but still
needs to handle the case where the vap is up+running

Noticed by: "Paul B. Mahol" <onemda@gmail.com>

15 years agoPlace the fifo and ref counting variables on the stack to prevent races.
thompsa [Tue, 2 Jun 2009 19:28:26 +0000 (19:28 +0000)]
Place the fifo and ref counting variables on the stack to prevent races.

Submitted by: Hans Petter Selasky

15 years agomove if_detach to the top of ieee80211_ifdetach to close various races
sam [Tue, 2 Jun 2009 18:55:27 +0000 (18:55 +0000)]
move if_detach to the top of ieee80211_ifdetach to close various races

Reviewed by: jhb

15 years agomove ifq_detach from if_detach to if_free; this permits callers to
sam [Tue, 2 Jun 2009 18:53:21 +0000 (18:53 +0000)]
move ifq_detach from if_detach to if_free; this permits callers to
reference if_snd in the period between detach+free which helps simplify
detach code

Reviewed by: jhb, rwatson

15 years agoRemove MAC kernel config files and add "options MAC" to GENERIC, with the
rwatson [Tue, 2 Jun 2009 18:31:08 +0000 (18:31 +0000)]
Remove MAC kernel config files and add "options MAC" to GENERIC, with the
goal of shipping 8.0 with MAC support in the default kernel.  No policies
will be compiled in or enabled by default, but it will now be possible to
load them at boot or runtime without a kernel recompile.

While the framework is not believed to impose measurable overhead when no
policies are loaded (a result of optimization over the past few months in
HEAD), we'll continue to benchmark and optimize as the release approaches.
Please keep an eye out for performance or functionality regressions that
could be a result of this change.

Approved by: re (kensmith)
Obtained from: TrustedBSD Project

15 years agoInitialize iov and iovlen before use.
pjd [Tue, 2 Jun 2009 18:30:09 +0000 (18:30 +0000)]
Initialize iov and iovlen before use.

Reported by: Lucius Windschuh <lwindschuh@googlemail.com>

15 years agoAdd internal 'mac_policy_count' counter to the MAC Framework, which is a
rwatson [Tue, 2 Jun 2009 18:26:17 +0000 (18:26 +0000)]
Add internal 'mac_policy_count' counter to the MAC Framework, which is a
count of the number of registered policies.

Rather than unconditionally locking sockets before passing them into MAC,
lock them in the MAC entry points only if mac_policy_count is non-zero.

This avoids locking overhead for a number of socket system calls when no
policies are registered, eliminating measurable overhead for the MAC
Framework for the socket subsystem when there are no active policies.

Possibly socket locks should be acquired by policies if they are required
for socket labels, which would further avoid locking overhead when there
are policies but they don't require labeling of sockets, or possibly
don't even implement socket controls.

Obtained from: TrustedBSD Project

15 years ago- Use volatile for signal variables.
lulf [Tue, 2 Jun 2009 17:57:24 +0000 (17:57 +0000)]
- Use volatile for signal variables.

Suggested by: Jaakko Heinonen <jh -at- saunalahti.fi>

15 years agoReorgansise the logic for tranversing the pipe list.
thompsa [Tue, 2 Jun 2009 17:31:59 +0000 (17:31 +0000)]
Reorgansise the logic for tranversing the pipe list.

Submitted by: Hans Petter Selasky

15 years agoFix compile after the removal of bsd_udev.
thompsa [Tue, 2 Jun 2009 17:31:16 +0000 (17:31 +0000)]
Fix compile after the removal of bsd_udev.

Submitted by: Hans Petter Selasky

15 years agoFix multithread issue where the is_uref variable was not set and cleared
thompsa [Tue, 2 Jun 2009 17:30:18 +0000 (17:30 +0000)]
Fix multithread issue where the is_uref variable was not set and cleared
properly in the CDEV private data.

Submitted by: Hans Petter Selasky

15 years agoStaticize ukbd_detach and fix indentation.
thompsa [Tue, 2 Jun 2009 17:29:15 +0000 (17:29 +0000)]
Staticize ukbd_detach and fix indentation.

Submitted by: Sylvestre Gallon

15 years agoRe-enable WARNS=6 after my universe test.
delphij [Tue, 2 Jun 2009 17:27:54 +0000 (17:27 +0000)]
Re-enable WARNS=6 after my universe test.

15 years agoAdd libusb20_tr_get_length to get the transfer length.
thompsa [Tue, 2 Jun 2009 17:27:51 +0000 (17:27 +0000)]
Add libusb20_tr_get_length to get the transfer length.

Submitted by: Hans Petter Selasky

15 years agoRemove hack used to deal with ifnet teardown now that if_detach and the
sam [Tue, 2 Jun 2009 16:57:27 +0000 (16:57 +0000)]
Remove hack used to deal with ifnet teardown now that if_detach and the
bridge do a better job.
o move ether_ifdetach to the top of ieee80211_detach
o do not clear if_softc at the top of ieee80211_detach; we no longer need
  this because we are safeguarded against calls coming back through if_ioctl
o simplify the bpf tracker now that we don't null if_softc

This also fixes an issue where having a bpf consumer active when a vap
is destroyed would cause a crash because bpf referenced free'd memory.

Reviewed by: imp

15 years agoBuf-ring fixes for mxge
gallatin [Tue, 2 Jun 2009 16:52:33 +0000 (16:52 +0000)]
Buf-ring fixes for mxge

- always maintain byte/mcast/drop stats via drbr
- move #define of IFNET_BUF_RING so that its picked
  up by all files in the driver
- conditionalize IFNET_BUF_RING on the FreeBSD_version
  bump just after it appeared in the tree.

Sponsored by: Myricom Inc.

15 years agoIn bwi_newstate, only zero the bssid when we stop a STA. And only
imp [Tue, 2 Jun 2009 16:48:10 +0000 (16:48 +0000)]
In bwi_newstate, only zero the bssid when we stop a STA.  And only
when we've not stopped the card.  It hangs the system when we touch
the CSR after bwistop.

This fixes the hanging on kldunload.

15 years agoRemove unneeded include.
rwatson [Tue, 2 Jun 2009 15:59:46 +0000 (15:59 +0000)]
Remove unneeded include.

MFC after: 3 days

15 years agoRemove window(1) from the base system.
ed [Tue, 2 Jun 2009 13:44:36 +0000 (13:44 +0000)]
Remove window(1) from the base system.

Some time ago Tom Rhodes sent me an email that he was willing to perform
various cleanups to the window(1) source code. After some discussion, we
both decided the best thing to do, was to move window(1) to the ports
tree. The application isn't used a lot nowadays, mainly because it has
been superseeded by screen, tmux, etc.

A couple of hours ago Tom committed window(1) to ports (misc/window), so
I'm removing it from the tree. I don't think people will really miss it,
but I'm describing the change in UPDATING anyway.

Discussed with: trhodes, pav, kib
Approved by: re

15 years agoHandle lock recursion differenty by always checking against LO_RECURSABLE
attilio [Tue, 2 Jun 2009 13:03:35 +0000 (13:03 +0000)]
Handle lock recursion differenty by always checking against LO_RECURSABLE
instead the lock own flag itself.

Tested by: pho

15 years agoInclude <machine/stdarg.h> for va_*(). I'm not sure how this compiled
jhb [Tue, 2 Jun 2009 12:35:04 +0000 (12:35 +0000)]
Include <machine/stdarg.h> for va_*().  I'm not sure how this compiled
on amd64 without this.

15 years agoAdd support for the four PUC serial interfaces found on IBM SurePOS 300
rwatson [Tue, 2 Jun 2009 09:58:17 +0000 (09:58 +0000)]
Add support for the four PUC serial interfaces found on IBM SurePOS 300
series POS terminals.

MFC after: 3 days
Submitted by: Marc Balmer <marc at msys.ch>

15 years agoCorrect a boundary case error in the management of a page's dirty bits by
alc [Tue, 2 Jun 2009 08:02:27 +0000 (08:02 +0000)]
Correct a boundary case error in the management of a page's dirty bits by
shm_dotruncate() and vnode_pager_setsize().  Specifically, if the length of
a shared memory object or a file is truncated such that the length modulo
the page size is between 1 and 511, then all of the page's dirty bits were
cleared.  Now, a dirty bit is cleared only if the corresponding block is
truncated in its entirety.

15 years agoRather than using both -prune (which requires directory-first tree traversal)
brian [Tue, 2 Jun 2009 07:35:51 +0000 (07:35 +0000)]
Rather than using both -prune (which requires directory-first tree traversal)
and -delete (which implies depth-first traversal), avoid using -delete in
favour of -execdir.

This has a side-effect of not removing directories that contain files,
even if we delete all of those files, but IMHO that's a better option
than specifying all possible local filesystem types in this script.

PR: 122811
MFC after: 3 weeks

15 years ago - Use an acquire barrier to increment f_count in fget_unlocked and
jeff [Tue, 2 Jun 2009 06:55:32 +0000 (06:55 +0000)]
 - Use an acquire barrier to increment f_count in fget_unlocked and
   remove the volatile cast.  Describe the reason in detail in a comment.

Discussed with: bde, jhb

15 years agoAdd 88E8071, 88E8072 to the supported hardware list. While I'm here
yongari [Tue, 2 Jun 2009 05:13:02 +0000 (05:13 +0000)]
Add 88E807188E8072 to the supported hardware list. While I'm here
correct description of 88E8070. It's Yukon Extreme and have gigabit
PHY.

15 years agoAdd Yukon Extreme device ids, 88E8071 and 88E8072.
yongari [Tue, 2 Jun 2009 05:08:57 +0000 (05:08 +0000)]
Add Yukon Extreme device ids, 88E8071 and 88E8072.
While I'm here correct description of 88E807088E8070 is Yukon
Extreme and have gigabit PHY.

15 years agoAdd Rx checksum offloading support for Yukon FE+ and Yukon Extreme.
yongari [Tue, 2 Jun 2009 04:59:29 +0000 (04:59 +0000)]
Add Rx checksum offloading support for Yukon FE+ and Yukon Extreme.
These controllers use newer descriptor format and the new descriptor
format uses status LE to indicate the status of checksum. Rx
checksummed value used in previous controllers were very cryptic
and I failed to understand how to use them. In addition most
controllers in previous generations had Rx checksum offloading bug.

While I'm here introduce a MSK_FLAG_NORX_CSUM flag to bypass
checking Rx checksum offloading as Yukon FE+ A0 has status LE bug.

15 years agoCatch up with the times: "mozilla" -> "firefox".
jkoshy [Tue, 2 Jun 2009 04:45:56 +0000 (04:45 +0000)]
Catch up with the times: "mozilla" -> "firefox".

15 years agoAdd frame parser capability of Yukon FE+ and Yukon Extreme. With
yongari [Tue, 2 Jun 2009 04:35:44 +0000 (04:35 +0000)]
Add frame parser capability of Yukon FE+ and Yukon Extreme. With
this feature hardware automatically computes TCP/UDP payload
offset. Introduce MSK_FLAG_AUTOTX_CSUM to mark the capability.
Yukon Extreme B0 revision is known to have a silicon for the
feature so disable it. Yukon Extreme B0 still can do Tx checksum
offloading but CPU have to compute TCP/UDP payload offset. To
enable traditional checksum offloading, disable automatic Tx
checksum calculation capability.
Yukon Extreme A0 revision could not use store-and-forward mode for
jumbo frames(silicon bug) so disable Tx checksum offloading for
jumbo frames.

I believe controllers that have MSK_FLAG_AUTOTX_CSUM capability or
new descriptor format do not have Tx checksum offload bug so
disable checksum offloading workaround for for short frames.

Tested by: jhb, Warren Block ( wblock <> wonkity dot com )

15 years agoAdd preliminary Yukon Extreme support and register definitions.
yongari [Tue, 2 Jun 2009 04:00:17 +0000 (04:00 +0000)]
Add preliminary Yukon Extreme support and register definitions.
Yukon Extreme uses new descriptor format for TSO and has Tx frame
parser which greatly reduces CPU cycles spent in computing TCP/UDP
payload offset calculation in Tx checksum offloading path. The new
descriptor format also removed TCP/UDP payload computation for TSO
which in turn results in better TSO performance. It seems Yukon
Extreme has a lot of new (unknown) features but only basic
offloading is supported at this time. So far there are two known
issues.
 o Sometimes Rx overrun errors happen when pulling data over
   gigabit link. Running over 100Mbps seem to ok.
 o Ethernet hardware address shows all-zeroed value on 88E8070.
   Assigning ethernet address with ifconfig is necessary to make it
   work.
Support for Yukon Extreme is not perfect but it would be better
than having a non-working device. Special thanks to jbh who fixed
several bugs of initial patch.

Tested by: jhb, Warren Block ( wblock <> wonkity dot com )

15 years agoFix monitor mode vaps to work as intended:
sam [Tue, 2 Jun 2009 00:33:28 +0000 (00:33 +0000)]
Fix monitor mode vaps to work as intended:
o track # bpf taps on monitor mode vaps instead of # monitor mode vaps
o spam monitor mode taps on tx/rx
o fix ieee80211_radiotap_rx_all to dispatch frames only if the vap is up
o while here print radiotap (and superg) state in show com

15 years agoProgram LED registers for 88E1116/88E1149 PHYs. These PHYs are
yongari [Tue, 2 Jun 2009 00:30:30 +0000 (00:30 +0000)]
Program LED registers for 88E1116/88E1149 PHYs. These PHYs are
found on Marvell Yukon Ultra, Marvell Yukon Extreme controllers.
While I'm here explicitly issue 'powerup' command for 88E1149 PHY.

Tested by: jhb, Warren Block ( wblock <> wonkity dot com )

15 years agoDon't assume page register value is 0 and restore previous page
yongari [Tue, 2 Jun 2009 00:21:30 +0000 (00:21 +0000)]
Don't assume page register value is 0 and restore previous page
register after issuing 'powerup'.

15 years agodon't dispatch frames to vap's not running
sam [Tue, 2 Jun 2009 00:06:39 +0000 (00:06 +0000)]
don't dispatch frames to vap's not running

15 years agocount packets
sam [Tue, 2 Jun 2009 00:04:10 +0000 (00:04 +0000)]
count packets

15 years agoAdd a note about the change to rcorder for pf and ipfw.
dougb [Mon, 1 Jun 2009 22:47:59 +0000 (22:47 +0000)]
Add a note about the change to rcorder for pf and ipfw.

15 years agoCommit the updates to this file for the new BIND options
dougb [Mon, 1 Jun 2009 22:14:45 +0000 (22:14 +0000)]
Commit the updates to this file for the new BIND options

15 years agoAdd support for the build options that are currently in the port:
dougb [Mon, 1 Jun 2009 21:58:59 +0000 (21:58 +0000)]
Add support for the build options that are currently in the port:
WITH_BIND_IDN
WITH_BIND_LARGE_FILE
WITH_BIND_SIGCHASE
WITH_BIND_XML

15 years ago- Add Hungarian catalog
gabor [Mon, 1 Jun 2009 21:54:59 +0000 (21:54 +0000)]
- Add Hungarian catalog

15 years agoRemove another d_thread_t use that crept in.
jhb [Mon, 1 Jun 2009 21:54:22 +0000 (21:54 +0000)]
Remove another d_thread_t use that crept in.

15 years agoMFp4.
mav [Mon, 1 Jun 2009 21:42:26 +0000 (21:42 +0000)]
MFp4.

Log supported AHCI controller capabilities.

15 years agoAdd an extension to the character device interface that allows character
jhb [Mon, 1 Jun 2009 21:32:52 +0000 (21:32 +0000)]
Add an extension to the character device interface that allows character
device drivers to use arbitrary VM objects to satisfy individual mmap()
requests.
- A new d_mmap_single(cdev, &foff, objsize, &object, prot) callback is
  added to cdevsw.  This function is called for each mmap() request.
  If it returns ENODEV, then the mmap() request will fall back to using
  the device's device pager object and d_mmap().  Otherwise, the method
  can return a VM object to satisfy this entire mmap() request via
  *object.  It can also modify the starting offset into this object via
  *foff.  This allows device drivers to use the file offset as a cookie
  to identify specific VM objects.
- vm_mmap_vnode() has been changed to call vm_mmap_cdev() directly when
  mapping V_CHR vnodes.  This avoids duplicating all the cdev mmap
  handling code and simplifies some of vm_mmap_vnode().
- D_VERSION has been bumped to D_VERSION_02.  Older device drivers
  using D_VERSION_01 are still supported.

MFC after: 1 month

15 years agoV_loif is not an array but a pure pointer, so treat it as such.
zec [Mon, 1 Jun 2009 21:29:54 +0000 (21:29 +0000)]
V_loif is not an array but a pure pointer, so treat it as such.

Reviewed by: bz
Approved by: julian (mentor)

15 years agoFix segfault when giving invalid long option to pkg_info.
jilles [Mon, 1 Jun 2009 21:26:52 +0000 (21:26 +0000)]
Fix segfault when giving invalid long option to pkg_info.

PR: bin/133473
Submitted by: Rafal Grodzinski
Approved by: ed (mentor)
MFC after: 1 week

15 years agoRework socket upcalls to close some races with setup/teardown of upcalls.
jhb [Mon, 1 Jun 2009 21:17:03 +0000 (21:17 +0000)]
Rework socket upcalls to close some races with setup/teardown of upcalls.
- Each socket upcall is now invoked with the appropriate socket buffer
  locked.  It is not permissible to call soisconnected() with this lock
  held; however, so socket upcalls now return an integer value.  The two
  possible values are SU_OK and SU_ISCONNECTED.  If an upcall returns
  SU_ISCONNECTED, then the soisconnected() will be invoked on the
  socket after the socket buffer lock is dropped.
- A new API is provided for setting and clearing socket upcalls.  The
  API consists of soupcall_set() and soupcall_clear().
- To simplify locking, each socket buffer now has a separate upcall.
- When a socket upcall returns SU_ISCONNECTED, the upcall is cleared from
  the receive socket buffer automatically.  Note that a SO_SND upcall
  should never return SU_ISCONNECTED.
- All this means that accept filters should now return SU_ISCONNECTED
  instead of calling soisconnected() directly.  They also no longer need
  to explicitly clear the upcall on the new socket.
- The HTTP accept filter still uses soupcall_set() to manage its internal
  state machine, but other accept filters no longer have any explicit
  knowlege of socket upcall internals aside from their return value.
- The various RPC client upcalls currently drop the socket buffer lock
  while invoking soreceive() as a temporary band-aid.  The plan for
  the future is to add a new flag to allow soreceive() to be called with
  the socket buffer locked.
- The AIO callback for socket I/O is now also invoked with the socket
  buffer locked.  Previously sowakeup() would drop the socket buffer
  lock only to call aio_swake() which immediately re-acquired the socket
  buffer lock for the duration of the function call.

Discussed with: rwatson, rmacklem

15 years agoUpdate VNET base pointer setting macro to use a correct source of
zec [Mon, 1 Jun 2009 21:10:23 +0000 (21:10 +0000)]
Update VNET base pointer setting macro to use a correct source of
vnet context.

Approved by: julian (mentor)

15 years agoInitialize the match structure. This is unnecessary but gcc insists to
delphij [Mon, 1 Jun 2009 21:07:54 +0000 (21:07 +0000)]
Initialize the match structure.  This is unnecessary but gcc insists to
complain about it when we raise the WARNS level.

15 years agoRemove an #undef MIN that slipped under the radar and led me to
zec [Mon, 1 Jun 2009 20:59:40 +0000 (20:59 +0000)]
Remove an #undef MIN that slipped under the radar and led me to
hastily introduce an #define MIN() a few lines below in r191816.

Approved by: julian (mentor)
Discussed with: bz

15 years agoAdd forgotten in previous commit flags argument.
dchagin [Mon, 1 Jun 2009 20:54:41 +0000 (20:54 +0000)]
Add forgotten in previous commit flags argument.

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

15 years agoImplement accept4 syscall.
dchagin [Mon, 1 Jun 2009 20:48:39 +0000 (20:48 +0000)]
Implement accept4 syscall.

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

15 years agoImplement a variation of the accept_common() which takes
dchagin [Mon, 1 Jun 2009 20:44:58 +0000 (20:44 +0000)]
Implement a variation of the accept_common() which takes
a flags argument.

Do not preserve td_retval before kern_fcntl(F_SETFL) as it does not
changed.

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

15 years agoSplit linux_accept() syscall onto linux_accept_common() which should
dchagin [Mon, 1 Jun 2009 20:42:27 +0000 (20:42 +0000)]
Split linux_accept() syscall onto linux_accept_common() which should
be used by linuxulator and linux_accept() itself.

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