]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
19 years agoThis commit was generated by cvs2svn to compensate for changes in r147341,
sam [Mon, 13 Jun 2005 17:00:32 +0000 (17:00 +0000)]
This commit was generated by cvs2svn to compensate for changes in r147341,
which included commits to RCS files with non-trunk default branches.

19 years agostripped down import of hostapd 0.3.9
sam [Mon, 13 Jun 2005 17:00:32 +0000 (17:00 +0000)]
stripped down import of hostapd 0.3.9

Approved by: re (dwhite)

19 years agoresolve merge conflicts
sam [Mon, 13 Jun 2005 16:54:21 +0000 (16:54 +0000)]
resolve merge conflicts

19 years agoThis commit was generated by cvs2svn to compensate for changes in r147338,
sam [Mon, 13 Jun 2005 16:43:14 +0000 (16:43 +0000)]
This commit was generated by cvs2svn to compensate for changes in r147338,
which included commits to RCS files with non-trunk default branches.

19 years agostripped down import of wpa_supplicant v0.3.9
sam [Mon, 13 Jun 2005 16:43:14 +0000 (16:43 +0000)]
stripped down import of wpa_supplicant v0.3.9

Approved by: re (dwhite)

19 years agoIt's spelled 'preceded' of course.
brueffer [Mon, 13 Jun 2005 15:49:59 +0000 (15:49 +0000)]
It's spelled 'preceded' of course.

Submitted by: David Adam <zanchey@ucc.gu.uwa.edu.au>
Pointy hat to: brueffer

19 years agoHex strings are _pre_ceeded by 0x, not _pro_ceeded.
brueffer [Mon, 13 Jun 2005 07:55:15 +0000 (07:55 +0000)]
Hex strings are _pre_ceeded by 0x, not _pro_ceeded.

PR: 82187
Submitted by: Anthony Rogers <anthony.j.rogers@gmail.com>
Approved by: re (blanket)
MFC after: 3 days

19 years ago - Don't make vgonel() globally visible, we want to change its prototype
jeff [Mon, 13 Jun 2005 06:26:55 +0000 (06:26 +0000)]
 - Don't make vgonel() globally visible, we want to change its prototype
   anyway and it's not used outside of vfs_subr.c.
 - Change vgonel() to accept a parameter which determines whether or not
   we'll put the vnode on the free list when we're done.
 - Use the new vgonel() parameter rather than VI_DOOMED to signal our
   intentions in vtryrecycle().
 - In vgonel() return if VI_DOOMED is already set, this vnode has already
   been reclaimed.

Sponsored by: Isilon Systems, Inc.

19 years ago - Clear v_dd in cache_zap() instead of cache_purge() as cache_purge() may
jeff [Mon, 13 Jun 2005 05:59:59 +0000 (05:59 +0000)]
 - Clear v_dd in cache_zap() instead of cache_purge() as cache_purge() may
   not be called in all cases where we free the cnp.

Sponsored by: Isilon Systems, Inc.

19 years ago - We should never unlock a buf before we've cleared B_REMFREE. I believe
jeff [Mon, 13 Jun 2005 00:48:58 +0000 (00:48 +0000)]
 - We should never unlock a buf before we've cleared B_REMFREE.  I believe
   this is happening at the moment and sometimes causing panics later on the
   package cluster when we bremfree() a buf whose delayed bremfree() did not
   previously happen.

Sponsored by: Isilon Systems, Inc.

19 years ago - It has long been my suspicion that we don't actually need a loop in
jeff [Mon, 13 Jun 2005 00:47:29 +0000 (00:47 +0000)]
 - It has long been my suspicion that we don't actually need a loop in
   vn_lock().  Add an assert that will help me gain more confidence that this
   is correct.

Sponsored by: Isilon Systems, Inc.

19 years ago - Add KTR_VFS events to vdestroy, vtruncbuf, vinvalbuf, vfreehead.
jeff [Mon, 13 Jun 2005 00:46:37 +0000 (00:46 +0000)]
 - Add KTR_VFS events to vdestroy, vtruncbuf, vinvalbuf, vfreehead.

Sponsored by: Isilon Systems, Inc.

19 years ago - Add KTR_VFS messages for various name cache related events.
jeff [Mon, 13 Jun 2005 00:46:03 +0000 (00:46 +0000)]
 - Add KTR_VFS messages for various name cache related events.

Sponsored by: Isilon Systems, Inc.

19 years ago - Split one KASSERT in bremfree() into two to aid in debugging.
jeff [Mon, 13 Jun 2005 00:45:05 +0000 (00:45 +0000)]
 - Split one KASSERT in bremfree() into two to aid in debugging.

Sponsored by: Isilon Systems, Inc.

19 years ago - switch_point is now unused. This doesn't break module binary compatability
jeff [Sun, 12 Jun 2005 22:33:06 +0000 (22:33 +0000)]
 - switch_point is now unused.  This doesn't break module binary compatability
   since the structure is shrinking, not growing.

19 years ago - Dramatically simplify bioqdisksort(). We no longer do ordered bios so
jeff [Sun, 12 Jun 2005 22:32:29 +0000 (22:32 +0000)]
 - Dramatically simplify bioqdisksort().  We no longer do ordered bios so
   most of the code to deal with them has been dead for sometime.  Simplify
   the code by doing an insert sort hinted by the current head position.

Met with apathy by: arch@

19 years agoDefine IPI_PREEMPT. Update a nearby comment while I'm here.
marcel [Sun, 12 Jun 2005 19:03:01 +0000 (19:03 +0000)]
Define IPI_PREEMPT. Update a nearby comment while I'm here.

19 years agoMark pf callouts as NET_MPSAFE.
mlaier [Sun, 12 Jun 2005 16:46:20 +0000 (16:46 +0000)]
Mark pf callouts as NET_MPSAFE.

Requested by: yongari (serveral times)
Approved by: re (blanket)
MFC after: 1 week

19 years agoWhen doing matching based on dst_ip/src_ip make sure we are really looking
mlaier [Sun, 12 Jun 2005 16:27:10 +0000 (16:27 +0000)]
When doing matching based on dst_ip/src_ip make sure we are really looking
on an IPv4 packet as these variables are uninitialized if not.  This used to
allow arbitrary IPv6 packets depending on the value in the uninitialized
variables.

Some opcodes (most noteably O_REJECT) do not support IPv6 at all right now.

Reviewed by: brooks, glebius
Security: IPFW might pass IPv6 packets depending on stack contents.
Approved by: re (blanket)

19 years agoBuild on pc98.
brooks [Sun, 12 Jun 2005 16:21:44 +0000 (16:21 +0000)]
Build on pc98.

19 years agoStop the interface before detaching and freeing it, rather than after.
brooks [Sun, 12 Jun 2005 15:33:45 +0000 (15:33 +0000)]
Stop the interface before detaching and freeing it, rather than after.

Reported by: marius

19 years agochange 'dev' to 'sc->sc_dev' to fix the build.
scottl [Sun, 12 Jun 2005 15:25:19 +0000 (15:25 +0000)]
change 'dev' to 'sc->sc_dev' to fix the build.

Approved: re (implicit)

19 years agoNew release notes:
hrs [Sun, 12 Jun 2005 08:55:59 +0000 (08:55 +0000)]
New release notes:
security.jail.getfsstatroot_only
--> security.jail.enforce_statfs renamed,
struct ifnet change,
acpi_ibm(4) improved,
atkbdc(4) cleanup,
arkbdc(4), syscons(4), and so on, enabled by default on sparc64,
hwpmc(4) MI support,
ng_tcpmss(4) added, and
rexecd(8) removed.

Approved by: re (implicitly)

19 years agoDo not allocate memory while holding a mutex.
pjd [Sun, 12 Jun 2005 07:03:23 +0000 (07:03 +0000)]
Do not allocate memory while holding a mutex.
I introduce a very small race here (some file system can be mounted or
unmounted between 'count' calculation and file systems list creation),
but it is harmless.

Found by: FreeBSD Kernel Stress Test Suite: http://www.holm.cc/stress/
Reported by: Peter Holm <peter@holm.cc>

19 years agoReturn NULL instead of a bogus pointer from if_alloc when if_com_alloc
brooks [Sun, 12 Jun 2005 00:53:03 +0000 (00:53 +0000)]
Return NULL instead of a bogus pointer from if_alloc when if_com_alloc
fails.

Move detaching the ifnet from the ifindex_table into if_free so we can
both keep the sanity checks and actually delete the ifnets. [0]

Reported by: gallatin [0]
Approved by: re (blanket)

19 years agoRefactor the NETSMBCRYPTO option so that it does the same on all
marcel [Sun, 12 Jun 2005 00:47:21 +0000 (00:47 +0000)]
Refactor the NETSMBCRYPTO option so that it does the same on all
platforms. ARM is excluded as it doesn't yet have any crypto
sources.

Approved by: re (dwhite)
MFC after: 1 day

19 years agoFix IPv6 neighbor discovery by using IF_LLADDR to get the mac address
brooks [Sun, 12 Jun 2005 00:45:24 +0000 (00:45 +0000)]
Fix IPv6 neighbor discovery by using IF_LLADDR to get the mac address
instead of a particularly ugly cast + pointer math hack.

Reported by: kuriyama, kris

19 years agoDocument SA-05:10, SA-05:11, and SA-05:12.
hrs [Sat, 11 Jun 2005 17:27:14 +0000 (17:27 +0000)]
Document SA-05:10, SA-05:11, and SA-05:12.

Approved by: re (implicitly)

19 years agos/sc_enaddr/enaddr/ to fix pc98 build.
brooks [Sat, 11 Jun 2005 16:30:43 +0000 (16:30 +0000)]
s/sc_enaddr/enaddr/ to fix pc98 build.

Approved by: re (blanket)

19 years agoAdd a stub libthr(3) man page to document what it is and why, as well as
rwatson [Sat, 11 Jun 2005 15:27:56 +0000 (15:27 +0000)]
Add a stub libthr(3) man page to document what it is and why, as well as
to point at libmap.conf(5).  This will help answer questions about what
and why it is, although not in great detail.

Approved by: re (scottl)
MFC after: 1 week
MFC note: When MFC'd, don't MFC mention of work not yet MFC'd.

19 years agoDo not allocate memory based on not-checked argument from userland.
pjd [Sat, 11 Jun 2005 14:58:20 +0000 (14:58 +0000)]
Do not allocate memory based on not-checked argument from userland.
It can be used to panic the kernel by giving too big value.
Fix it by moving allocation and size verification into kern_getfsstat().
This even simplifies kern_getfsstat() consumers, but destroys symmetry -
memory is allocated inside kern_getfsstat(), but has to be freed by the
caller.

Found by: FreeBSD Kernel Stress Test Suite: http://www.holm.cc/stress/
Reported by: Peter Holm <peter@holm.cc>

19 years agoo setsockopt(2) cannot remove accept filter. [1]
maxim [Sat, 11 Jun 2005 11:59:48 +0000 (11:59 +0000)]
o setsockopt(2) cannot remove accept filter. [1]
o getsockopt(SO_ACCEPTFILTER) always returns success on listen socket
  even we didn't install accept filter on the socket.
o Fix these bugs and add regression tests for them.

Submitted by: Igor Sysoev [1]
Reviewed by: alfred
MFC after: 2 weeks

19 years agoFix copy&paste bug.
pjd [Sat, 11 Jun 2005 11:46:32 +0000 (11:46 +0000)]
Fix copy&paste bug.

19 years agoUnbreak the PowerPC GENERIC build.
jkoshy [Sat, 11 Jun 2005 10:59:02 +0000 (10:59 +0000)]
Unbreak the PowerPC GENERIC build.

Reviewed by: delphij

19 years ago - Assert that we're not in the name cache anymore in vdestroy().
jeff [Sat, 11 Jun 2005 08:48:09 +0000 (08:48 +0000)]
 - Assert that we're not in the name cache anymore in vdestroy().

Sponsored by: Isilon Systems, Inc.

19 years ago - Assert that we're not adding a doomed vnode to the name cache.
jeff [Sat, 11 Jun 2005 08:47:30 +0000 (08:47 +0000)]
 - Assert that we're not adding a doomed vnode to the name cache.

Sponsored by: Isilon Systems, Inc.

19 years agoAvoid GCC optimizations from injecting a call to memset(?) in order
marcel [Sat, 11 Jun 2005 03:21:20 +0000 (03:21 +0000)]
Avoid GCC optimizations from injecting a call to memset(?) in order
to initialize the buffer array in ata_raid_attach() by removing the
initializer. There's no memset(?) in the kernel. Instead, assign
'\0' to the first element. The buffer array holds strings only, so
this is functionally equivalent.

Applies to: ia64
Tripped over by: tinderbox

19 years agoMove if_alloc() up so it's before mii_phy_probe().
brooks [Sat, 11 Jun 2005 01:37:46 +0000 (01:37 +0000)]
Move if_alloc() up so it's before mii_phy_probe().

19 years ago - Add KTR_VFS tracing to track the life of vnodes. Eventually KTR_VFS
jeff [Sat, 11 Jun 2005 01:16:46 +0000 (01:16 +0000)]
 - Add KTR_VFS tracing to track the life of vnodes.  Eventually KTR_VFS
   events could be added to cover other interesting details.
 - Add some VNASSERTs to discover places where we access vnodes after
   they have been uma_zfree'd before we try to free them again.
 - Add a few more VNASSERTs to vdestroy() to be certain that the vnode is
   really unused.

Sponsored by: Isilon Systems, Inc.

19 years agoMove if_alloc() up in fxp_attach() so there's an ifp before
brooks [Sat, 11 Jun 2005 00:47:34 +0000 (00:47 +0000)]
Move if_alloc() up in fxp_attach() so there's an ifp before
mii_phy_probe() is called.

Committed via: fxp0

19 years ago - Replace the unused KTR_NFS with KTR_VFS. KTR_VFS differs from KTR_VOP
jeff [Sat, 11 Jun 2005 00:40:27 +0000 (00:40 +0000)]
 - Replace the unused KTR_NFS with KTR_VFS.  KTR_VFS differs from KTR_VOP
   in that KTR_VFS will be hand placed, while KTR_VOP traces the individual
   vnode operations and is generated by vnode_if.awk.
 - Add a comment describing KTR_VOP.

19 years agoAdd an evil hack to work around a cast from the softc to the ifnet that
brooks [Sat, 11 Jun 2005 00:20:38 +0000 (00:20 +0000)]
Add an evil hack to work around a cast from the softc to the ifnet that
I missed.  Since I did no rearrange any softcs, casting the result of
device_get_softc() to (struct ifnet **) and derefrencing it yeilds a
pointer to the ifp.  This makes at least vr(4) nics work.

19 years agoAcquire Giant before making some CAM calls that need Giant to be held.
vkashyap [Sat, 11 Jun 2005 00:14:23 +0000 (00:14 +0000)]
Acquire Giant before making some CAM calls that need Giant to be held.

19 years agoThe new contigmalloc(9) has a bad degenerate case where there were
green [Sat, 11 Jun 2005 00:05:16 +0000 (00:05 +0000)]
The new contigmalloc(9) has a bad degenerate case where there were
many regions checked again and again despite knowing the pages
contained were not usable and only satisfied the alignment constraints
This case was compounded, especially for large allocations, by the
practice of looping from the top of memory so as to keep out of the
important low-memory regions.  While the old contigmalloc(9) has the
same problem, it is not as noticeable due to looping from the low
memory to high.

This degenerate case is fixed, as well as reversing the sense of the
rest of the loops within it, to provide a tremendous speed increase.
This makes the best case O(n * VM overhead) much more likely than the
worst case O(4 * VM overhead).  For comparison, the worst case for old
contigmalloc would be O(5 * VM overhead) in addition to its strategy
of turning used memory into free being highly pessimal.

Also, fix a bug that in practice most likely couldn't have been triggered,
int the new contigmalloc(9): it walked backwards from the end of memory
without accounting for how many pages it needed.  Potentially, nonexistant
pages could have been mapped.  This hasn't occurred because the kernel
generally requests as its first contigmalloc(9) a single page.

Reported by: Nicolas Dehaine <nicko@stbernard.com>, wes
MFC After: 1 month
More testing by: Nicolas Dehaine <nicko@stbernard.com>, wes

19 years agoAvoid deadlock in fxp driver when system runs out of mbufs.
wes [Fri, 10 Jun 2005 23:54:52 +0000 (23:54 +0000)]
Avoid deadlock in fxp driver when system runs out of mbufs.

MFC after: 1 week
Provided by: Ernie Smallis <esmallis@stbernard.com>

19 years agoCatch up with the struct ifnet changes and use if_alloc().
thompsa [Fri, 10 Jun 2005 23:52:01 +0000 (23:52 +0000)]
Catch up with the struct ifnet changes and use if_alloc().

Reviewed by: brooks
Approved by: mlaier (mentor)

19 years agoFix a serious deadlock with the NFS client. Given a large enough
green [Fri, 10 Jun 2005 23:50:41 +0000 (23:50 +0000)]
Fix a serious deadlock with the NFS client.  Given a large enough
atomic write request, it can fill the buffer cache with the entirety
of that write in order to handle retries.  However, it never drops
the vnode lock, or else it wouldn't be atomic, so it ends up waiting
indefinitely for more buf memory that cannot be gotten as it has it
all, and it waits in an uncancellable state.

To fix this, hibufspace is exported and scaled to a reasonable
fraction.  This is used as the limit of how much of an atomic write
request by the NFS client will be handled asynchronously.  If the
request is larger than this, it will be turned into a synchronous
request which won't deadlock the system.  It's possible this value is
far off from what is required by some, so it shall be tunable as soon
as mount_nfs(8) learns of the new field.

The slowdown between an asynchronous and a synchronous write on NFS
appears to be on the order of 2x-4x.

General nod by: gad
MFC after: 2 weeks
More testing: wes
PR: kern/79208

19 years ago - I broke binary compat. update KTR_VERSION.
jeff [Fri, 10 Jun 2005 23:27:01 +0000 (23:27 +0000)]
 - I broke binary compat. update KTR_VERSION.

Spotted by: jhb

19 years ago - Add curthread to the state that ktr is saving. The extra information is
jeff [Fri, 10 Jun 2005 23:21:29 +0000 (23:21 +0000)]
 - Add curthread to the state that ktr is saving.  The extra information is
   well worth the bloat.
 - Change the formatting of 'show ktr' slightly to accommodate the
   additional field.  Remove a tab from the verbose output and place the
   actual trace data after a : so it is more easy to understand which
   part is the event and which is part of the record.

19 years agoTurn on atkbdc(4), atkbd(4), creator(4), machfb(4), syscons(4), ohci(4),
marius [Fri, 10 Jun 2005 23:11:16 +0000 (23:11 +0000)]
Turn on atkbdc(4), atkbd(4), creator(4), machfb(4), syscons(4), ohci(4),
psm(4), ukbd(4), ums(4) and usb(4) on by default. Modulo some nits with
the most annoying one probably being USB keyboards no longer working at
the OFW boot prompt after halting FreeBSD these drivers work fine on
sparc64 including X and there's nothing left that I'd consider a show-
stopper. I.e. graphical consoles on sun4u machines should either work
out of the box or by plugging in a card that is supported by either
creator(4) or machfb(4). The exception obviously are SBus-only machines
without UPA slots like some Ultra 1 (but which also still lack support
in other areas) and certain Exx0 (but which probably are mainly used
with serial consoles anyway). I'll try to add a cgsix(4) for these later
as Sun CG6 cards are probably the most common SBus framebuffer cards in
sun4u machines. I however don't see much sense in adding drivers for the
dozen of SBus framebuffers that were destined for sparc v8 machines.
The rest of the USB drivers aren't enabled as I'm only aware of ukbd(4)
and ums(4) as well as ohci(4) working with the on-board ALI M5237 and
Sun PCIO-2 controllers. Aue(4) definitely doesn't work on sparc64, yet.

Thanks to:
- Jake for the initial work on syscons(4) on sparc64 and creator(4).
- Marcel for uart(4) and especially for its support for the SCCs which
  are only used on sparc64 so far. In various regards it wouldn't have
  been possible to enable syscons(4) by default on sparc64, yet, without
  uart(4).
- All that tested patches.

Ok'ed by: scottl (RE hat), tmm

19 years ago- In preparation to turning syscons(4) etc. on by default in the sparc64
marius [Fri, 10 Jun 2005 23:06:14 +0000 (23:06 +0000)]
- In preparation to turning syscons(4) etc. on by default in the sparc64
  GENERIC comment in ttyN.
- Add the name of the device driver creating the device nodes above the
  respectives blocks so it's easier for user to find the right entry to
  shut up warnings from getty(8). Replace 'Requires device 'uart' be
  enabled.' with just 'uart(4)' as the former referred to a sparc64
  GENERIC back when uart(4) wasn't enabled by default, yet.
- Turn off the getty(8) on screen as screen is created by ofw_console(4)
  which is no longer enabled in the sparc64 GENERIC (and also only is a
  last resort) to shut up warnings from getty(8) with the current GENERIC.

19 years agoWrap the calls to the ISA DMA specific sndbuf_dma*() functions of
marius [Fri, 10 Jun 2005 21:33:14 +0000 (21:33 +0000)]
Wrap the calls to the ISA DMA specific sndbuf_dma*() functions of
sys/dev/sound/isa/sndbuf_dma.c (compilation depending on device isa)
in #ifdef DEV_ISA so sound(4) can be compiled without isa(4).

MFC after: 1 month

19 years agodon't look at the wme ie in a beacon unless we negotiated use
sam [Fri, 10 Jun 2005 21:30:29 +0000 (21:30 +0000)]
don't look at the wme ie in a beacon unless we negotiated use

19 years ago- Hook up atkbdc(4), atkbd(4) and psm(4) to the sparc64 build, not
marius [Fri, 10 Jun 2005 20:58:59 +0000 (20:58 +0000)]
- Hook up atkbdc(4), atkbd(4) and psm(4) to the sparc64 build, not
  enabled in GENERIC by default, yet.
- While here remove the exclusion of ukbd(4) from the sparc64 NOTES
  as ukbd(4) compiles and works on sparc64.

19 years ago- Hook up the new locations of the atkbdc(4), atkbd(4) and psm(4) source
marius [Fri, 10 Jun 2005 20:56:38 +0000 (20:56 +0000)]
- Hook up the new locations of the atkbdc(4), atkbd(4) and psm(4) source
  files after they were repo-copied to sys/dev/atkbdc. The sources of
  atkbdc(4) and its children were moved to the new location in preparation
  for adding an EBus front-end to atkbdc(4) for use on sparc64; i.e. in
  order to not further scatter them over the whole tree which would have
  been the result of adding atkbdc_ebus.c in e.g. sys/sparc64/ebus. Another
  reason for the repo-copies was that some of the sources were misfiled,
  e.g. sys/isa/atkbd_isa.c wasn't ISA-specific at all but for hanging
  atkbd(4) off of atkbdc(4) and was renamed to atkbd_atkbdc.c accordingly.
  Most of sys/isa/psm.c, i.e. expect for its PSMC PNP part, also isn't
  ISA-specific.
- Separate the parts of atkbdc_isa.c which aren't actually ISA-specific
  but are shareable between different atkbdc(4) bus front-ends into
  atkbdc_subr.c (repo-copied from atkbdc_isa.c). While here use
  bus_generic_rl_alloc_resource() and bus_generic_rl_release_resource()
  respectively in atkbdc_isa.c instead of rolling own versions.
- Add sparc64 MD bits to atkbdc(4) and atkbd(4) and an EBus front-end for
  atkbdc(4). PS/2 controllers and input devices are used on a couple of
  Sun OEM boards and occur on either the EBus or the ISA bus. Depending on
  the board it's either the only on-board mean to connect a keyboard and
  mouse or an alternative to either RS232 or USB devices.
- Wrap the PSMC PNP part of psm.c in #ifdef DEV_ISA so it can be compiled
  without isa(4) (e.g. for EBus-only machines). This ISA-specific part
  isn't separated into its own source file, yet, as it requires more work
  than was feasible for 6.0 in order to do it in a clean way. Actually
  philip@ is working on a rewrite of psm(4) so a more comprehensive
  clean-up and separation of hardware dependent and independent parts is
  expected to happen after 6.0.

Tested on: i386, sparc64 (AX1105, AXe and AXi boards)
Reviewed by: philip

19 years agoRemove rexecd(8), a server that implements a particularly insecure
nectar [Fri, 10 Jun 2005 20:52:36 +0000 (20:52 +0000)]
Remove rexecd(8), a server that implements a particularly insecure
method of executing commands remotely.  There are no rexec clients in
the FreeBSD tree, and the client function rexec(3) is present only in
libcompat.  It has been documented as "obsolete" since 4.3BSD, and its
use has been discouraged in the man page for over 10 years.

19 years agoRevert the unnecessicary addition of some braces in fxp_attach(). Don't
brooks [Fri, 10 Jun 2005 20:42:02 +0000 (20:42 +0000)]
Revert the unnecessicary addition of some braces in fxp_attach().  Don't
explicitly free the ifp in fxp_detach(), the call to fxp_release() takes
care of it.

19 years agoBack out "rw" locale addition for reason unknown to me (forced by portmgr)
ache [Fri, 10 Jun 2005 20:14:38 +0000 (20:14 +0000)]
Back out "rw" locale addition for reason unknown to me (forced by portmgr)

19 years agoMention hwpmc(4) ABI/API changes in the 20050609 entry.
jkoshy [Fri, 10 Jun 2005 19:59:26 +0000 (19:59 +0000)]
Mention hwpmc(4) ABI/API changes in the 20050609 entry.

Discussed with: imp

19 years agoFix typo.
jkoshy [Fri, 10 Jun 2005 18:06:59 +0000 (18:06 +0000)]
Fix typo.

Reviewed by: rwatson, sam

19 years agoDelete a file that was meant to be renamed while repo-copying it but
marius [Fri, 10 Jun 2005 17:36:34 +0000 (17:36 +0000)]
Delete a file that was meant to be renamed while repo-copying it but
wasn't and now is superfluous.

19 years agoForced commit to denote that the following repo-copies have taken place:
marius [Fri, 10 Jun 2005 17:34:41 +0000 (17:34 +0000)]
Forced commit to denote that the following repo-copies have taken place:

sys/dev/kbd/atkbd.c -> sys/dev/atkbdc/atkbd.c
sys/dev/kbd/atkbdc.c -> sys/dev/atkbdc/atkbdc.c
sys/dev/kbd/atkbdcreg.h -> sys/dev/atkbdc/atkbdcreg.h
sys/dev/kbd/atkbdreg.h -> sys/dev/atkbdc/atkbdreg.h
sys/isa/atkbd_isa.c -> sys/dev/atkbdc/atkbd_atkbdc.c
sys/isa/atkbdc_isa.c -> sys/dev/atkbdc/atkbdc_isa.c
sys/isa/atkbdc_isa.c -> sys/dev/atkbdc/atkbdc_subr.c
sys/isa/psm.c -> sys/dev/atkbdc/psm.c

Repo-copies done by: markm

19 years agoAdd a comment to the effect that fictitious pages do not require the
alc [Fri, 10 Jun 2005 17:27:54 +0000 (17:27 +0000)]
Add a comment to the effect that fictitious pages do not require the
initialization of their machine-dependent fields.

19 years agoDefer ip_output of pfsync updates to an independent callout thread instead
mlaier [Fri, 10 Jun 2005 17:23:49 +0000 (17:23 +0000)]
Defer ip_output of pfsync updates to an independent callout thread instead
of just dropping the lock around the ip_output call.  This used to cause
corrupted state tree walks for some call-paths.

In a second stage all callouts will be marked MPSAFE according to the
setting of mpsafenet.

Reported and tested by: Matthew Grooms <mgrooms at seton dot org>
MFC after: 3 days
X-MFC after: Marking callouts MPSAFE + 1 week

19 years agoUpdate prototype for function pmap_init(), it no longer has the arguments
hmp [Fri, 10 Jun 2005 17:19:27 +0000 (17:19 +0000)]
Update prototype for function pmap_init(), it no longer has the arguments
phys_start and phys_end.

Remove a stale documentation not about dis/uncontiguous memory.

Update manual page date while I am around these ends.

Reviewed by: alc

19 years agoAdd manual page for newly added pmap_page_init(9) function, part of the
hmp [Fri, 10 Jun 2005 17:16:16 +0000 (17:16 +0000)]
Add manual page for newly added pmap_page_init(9) function, part of the
PMAP KPI.

Reviewed-by: alc

19 years agoFix the wording in this man page so that it
rodrigc [Fri, 10 Jun 2005 16:54:17 +0000 (16:54 +0000)]
Fix the wording in this man page so that it
reflects the actual behavior of the API
for listing extended attributes.

PR: docs/79261
Submitted by: rodrigc
Reviewed by: rwatson, kan
Approved by: das (mentor)

19 years agoMention interface API changes.
brooks [Fri, 10 Jun 2005 16:51:49 +0000 (16:51 +0000)]
Mention interface API changes.

19 years agoStop embedding struct ifnet at the top of driver softcs. Instead the
brooks [Fri, 10 Jun 2005 16:49:24 +0000 (16:49 +0000)]
Stop embedding struct ifnet at the top of driver softcs. Instead the
struct ifnet or the layer 2 common structure it was embedded in have
been replaced with a struct ifnet pointer to be filled by a call to the
new function, if_alloc(). The layer 2 common structure is also allocated
via if_alloc() based on the interface type. It is hung off the new
struct ifnet member, if_l2com.

This change removes the size of these structures from the kernel ABI and
will allow us to better manage them as interfaces come and go.

Other changes of note:
 - Struct arpcom is no longer referenced in normal interface code.
   Instead the Ethernet address is accessed via the IFP2ENADDR() macro.
   To enforce this ac_enaddr has been renamed to _ac_enaddr.
 - The second argument to ether_ifattach is now always the mac address
   from driver private storage rather than sometimes being ac_enaddr.

Reviewed by: sobomax, sam

19 years agovalidate the bssid for non-data frames too when operating in
sam [Fri, 10 Jun 2005 16:14:48 +0000 (16:14 +0000)]
validate the bssid for non-data frames too when operating in
adhoc/ahdemo/hostap modes

19 years agoMake ichsmb(4) child device handling properly.
takawata [Fri, 10 Jun 2005 16:12:43 +0000 (16:12 +0000)]
Make ichsmb(4) child device handling properly.

19 years agoo fix wpa w/ wme: don't strip the QoS header on recv as tkip requires
sam [Fri, 10 Jun 2005 16:11:24 +0000 (16:11 +0000)]
o fix wpa w/ wme: don't strip the QoS header on recv as tkip requires
  it; instead pass the space occupied by the header down into the
  crypto modules (except in the demic case which needs it only when
  doing int in s/w)
o while here fix defrag to strip the header from 2nd and later frames
o teach decap code how to handle 4-address frames

19 years agoAdd missing {} in last commit.
mlaier [Fri, 10 Jun 2005 15:53:21 +0000 (15:53 +0000)]
Add missing {} in last commit.

19 years agoRemove the last use of pmap_initialized.
cognet [Fri, 10 Jun 2005 13:31:30 +0000 (13:31 +0000)]
Remove the last use of pmap_initialized.

19 years ago- Remove redundant parenthesis in M_CHECK macro.
glebius [Fri, 10 Jun 2005 12:44:21 +0000 (12:44 +0000)]
- Remove redundant parenthesis in M_CHECK macro.
- Do not edit pullup_len outside M_CHECK macro.
- Do not reimplement NG_FWD_NEW_DATA().
- Remove redundant check for item being not NULL.

Submitted by: ru

19 years agoModify send_pkt() to return the generated packet and have the caller
green [Fri, 10 Jun 2005 12:28:17 +0000 (12:28 +0000)]
Modify send_pkt() to return the generated packet and have the caller
do the subsequent ip_output() in IPFW.  In ipfw_tick(), the keep-alive
packets must be generated from the data that resides under the
stateful lock, but they must not be sent at that time, as this would
cause a lock order reversal with the normal ordering (interface's
lock, then locks belonging to the pfil hooks).

In practice, this caused deadlocks when using IPFW and if_bridge(4)
together to do stateful transparent filtering.

MFC after: 1 week

19 years ago- Only create the led(4) interface, if setting the thinklight is possible
markus [Fri, 10 Jun 2005 11:56:18 +0000 (11:56 +0000)]
- Only create the led(4) interface, if setting the thinklight is possible
- Initialize val_ec with the content of the volume EC register
  for ACPI_IBM_METHOD_VOLUME and ACPI_IBM_METHOD_MUTE in acpi_ibm_sysctl_set()
  if there is no CMOS handle present. This fixes setting volume and mute on
  such models.

Submitted by: ru
Approved by: philip

19 years agoMake the default RB_AUGMENT() produce a 'do {} while (0)' instead
harti [Fri, 10 Jun 2005 11:44:57 +0000 (11:44 +0000)]
Make the default RB_AUGMENT() produce a 'do {} while (0)' instead
of nothing. This prevents the compiler from complaining about empty
if statements when compiled with higher WARN levels.

19 years agoAdd locale/rw
ache [Fri, 10 Jun 2005 11:24:30 +0000 (11:24 +0000)]
Add locale/rw

19 years agoUse MOPT_END in favor of MOPT_NULL, which better describes the purporse
delphij [Fri, 10 Jun 2005 09:51:43 +0000 (09:51 +0000)]
Use MOPT_END in favor of MOPT_NULL, which better describes the purporse
of the macro.

19 years agoSnoopy in quality of life.
phk [Fri, 10 Jun 2005 08:58:44 +0000 (08:58 +0000)]
Snoopy in quality of life.

19 years agoManual page for ng_tcpmss.
glebius [Fri, 10 Jun 2005 08:44:19 +0000 (08:44 +0000)]
Manual page for ng_tcpmss.

19 years agoIdentify the Intel ICH4 EHCI controller.
benno [Fri, 10 Jun 2005 08:28:22 +0000 (08:28 +0000)]
Identify the Intel ICH4 EHCI controller.

19 years agoAttach ng_tcpmss to the build.
glebius [Fri, 10 Jun 2005 08:05:13 +0000 (08:05 +0000)]
Attach ng_tcpmss to the build.

19 years agoAdd ng_tcpmss node, which alters MSS options of TCP packets. Useful to
glebius [Fri, 10 Jun 2005 08:02:34 +0000 (08:02 +0000)]
Add ng_tcpmss node, which alters MSS options of TCP packets. Useful to
hack MSS of packets outgoing via interface with small MTU, to workaround
path MTU discovery problems.

Written by Alexey Popov, with some cleanups from me. There are also plans
to improve mpd port, so that it uses this node, instead of doing MSS
hacking in userland, when 'enable tcpmssfix' option is on.

Submitted by:   Alexey Popov <lollypop@flexuser.ru>

19 years agoEnable SATA hotplug support.
sos [Fri, 10 Jun 2005 07:43:10 +0000 (07:43 +0000)]
Enable SATA hotplug support.

Submitted by: Olivier Houchard

19 years agoClean up after the last change -- remove now stale comment.
ru [Fri, 10 Jun 2005 07:06:07 +0000 (07:06 +0000)]
Clean up after the last change -- remove now stale comment.

19 years agoDon't free(sock) before it's even allocated.
des [Fri, 10 Jun 2005 06:36:03 +0000 (06:36 +0000)]
Don't free(sock) before it's even allocated.

Submitted by: Joerg Sonnenberger <joerg@britannica.bec.de>
MFC after: 1 week

19 years agoUse appropriate error codes for each facility instead of just PAM_AUTH_ERR.
des [Fri, 10 Jun 2005 06:16:13 +0000 (06:16 +0000)]
Use appropriate error codes for each facility instead of just PAM_AUTH_ERR.

Noticed by: pjd

19 years agoRemove kludges intended to support src trees with partial obj trees.
des [Fri, 10 Jun 2005 06:12:53 +0000 (06:12 +0000)]
Remove kludges intended to support src trees with partial obj trees.

Discussed with: ru

19 years agoOnly set mode when the incoming ioctl is IOC_VOID, until we have better
delphij [Fri, 10 Jun 2005 05:20:33 +0000 (05:20 +0000)]
Only set mode when the incoming ioctl is IOC_VOID, until we have better
solution against the ioctl collisions.

Submitted by: Antoine Brodin <antoine.brodin at laposte net>
PR: kern/81867

19 years agomark stations authorized during recv processing instead of doing it
sam [Fri, 10 Jun 2005 05:04:42 +0000 (05:04 +0000)]
mark stations authorized during recv processing instead of doing it
as a side effect of sending an auth success frame; sending mgmt
frames should not have side effects

19 years agoCorrect grammar error in comment
csjp [Fri, 10 Jun 2005 04:44:38 +0000 (04:44 +0000)]
Correct grammar error in comment

MFC after: 3 days

19 years agomove AID implementation defines from the protocol definitions to
sam [Fri, 10 Jun 2005 04:42:34 +0000 (04:42 +0000)]
move AID implementation defines from the protocol definitions to
where they are used

19 years agoaccept diassoc frame in ASSOC state
sam [Fri, 10 Jun 2005 04:37:05 +0000 (04:37 +0000)]
accept diassoc frame in ASSOC state

19 years agoFix tinderbox breakage.
jkoshy [Fri, 10 Jun 2005 03:45:04 +0000 (03:45 +0000)]
Fix tinderbox breakage.

19 years agoAdd support for /etc/dhclient-enter-hooks to match /etc/dhclient-exit-hooks.
brooks [Fri, 10 Jun 2005 03:41:18 +0000 (03:41 +0000)]
Add support for /etc/dhclient-enter-hooks to match /etc/dhclient-exit-hooks.

19 years agoIntroduce a procedure, pmap_page_init(), that initializes the
alc [Fri, 10 Jun 2005 03:33:36 +0000 (03:33 +0000)]
Introduce a procedure, pmap_page_init(), that initializes the
vm_page's machine-dependent fields.  Use this function in
vm_pageq_add_new_page() so that the vm_page's machine-dependent and
machine-independent fields are initialized at the same time.

Remove code from pmap_init() for initializing the vm_page's
machine-dependent fields.

Remove stale comments from pmap_init().

Eliminate the Boolean variable pmap_initialized from the alpha, amd64,
i386, and ia64 pmap implementations.  Its use is no longer required
because of the above changes and earlier changes that result in physical
memory that is being mapped at initialization time being mapped without
pv entries.

Tested by: cognet, kensmith, marcel

19 years agoRestore preemption of idle threads.
ups [Fri, 10 Jun 2005 03:00:29 +0000 (03:00 +0000)]
Restore preemption of idle threads.

Submitted by: jhb

19 years agopull some debug msgs up so they're seen more often
sam [Fri, 10 Jun 2005 01:48:57 +0000 (01:48 +0000)]
pull some debug msgs up so they're seen more often