mp [Mon, 15 Oct 2007 15:23:07 +0000 (15:23 +0000)]
Import two vendor fixes from tcsh-6.15.01 for MFC to 7.0. The fixes are:
- Fix pty detection for autologout setting
- kill `foo` got stuck because sigchld was disabled too soon
daichi [Sun, 14 Oct 2007 13:55:38 +0000 (13:55 +0000)]
Added whiteout behavior option. ``-o whiteout=always'' is default mode
(it is established practice) and ``-o whiteout=whenneeded'' is less
disk-space using mode especially for resource restricted environments
like embedded environments. (Contributed by Ed Schouten. Thanks)
daichi [Sun, 14 Oct 2007 13:53:38 +0000 (13:53 +0000)]
Default copy mode has been changed from traditional-mode to transparent-mode.
Some folks who have reported some issues have solved with transparent mode.
We guess it is time to change the default copy mode. The transparent-mode is
the best in most situations.
daichi [Sun, 14 Oct 2007 13:49:30 +0000 (13:49 +0000)]
- It has been become MPSAFE.
- Fixed lock panic issue under MPSAFE.
- Fixed panic issue whenever it locks vnode with reclaim.
- Fixed lock implementations not conforming to vnode_if.src style.
daichi [Sun, 14 Oct 2007 13:46:11 +0000 (13:46 +0000)]
- Added support for vfs_cache on unionfs. As a result, you can use
applications that use procfs on unionfs.
- Removed unionfs internal cache mechanism because it has
vfs_cache support instead. As a result, it just simplified code of
unionfs.
- Fixed kern/111262 issue.
This commit includes the following core components:
* sample configuration file for sensorsd
* rc(8) script and glue code for sensorsd(8)
* sysctl(3) doc fixes for CTL_HW tree
* sysctl(3) documentation for hardware sensors
* sysctl(8) documentation for hardware sensors
* support for the sensor structure for sysctl(8)
* rc.conf(5) documentation for starting sensorsd(8)
* sensor_attach(9) et al documentation
* /sys/kern/kern_sensors.c
o sensor_attach(9) API for drivers to register ksensors
o sensor_task_register(9) API for the update task
o sysctl(3) glue code
o hw.sensors shadow tree for sysctl(8) internal magic
* <sys/sensors.h>
* HW_SENSORS definition for <sys/sysctl.h>
* sensors display for systat(1), including documentation
* sensorsd(8) and all applicable documentation
The userland part of the framework is entirely source-code
compatible with OpenBSD 4.1, 4.2 and -current as of today.
All sensor readings can be viewed with `sysctl hw.sensors`,
monitored in semi-realtime with `systat -sensors` and also
logged with `sensorsd`.
Submitted by: Constantine A. Murenin <cnst@FreeBSD.org>
Sponsored by: Google Summer of Code 2007 (GSoC2007/cnst-sensors)
Mentored by: syrinx
Tested by: many
OKed by: kensmith
Obtained from: OpenBSD (parts)
csjp [Sun, 14 Oct 2007 00:52:30 +0000 (00:52 +0000)]
Fix id -A when the subject has an extended subject token associated with
them (for example when they have logged in from an ip6 source).
- Stick with the initial call to getaudit(2), if it returns E2BIG, use
getaudit_addr(2) instead and set the "extended" flag to indicate that
we the calling credential has an extended subject state.
- Additionally, add the printing of the machine/at_addr (the ip/ip6
addresses)
sam [Sat, 13 Oct 2007 22:30:41 +0000 (22:30 +0000)]
revert 1.18: the negotiated rate set may not match the hal
rate tables, so using the hal's rateCodeToIndex array
will produce wrong indices for the negotiated rate set
ache [Sat, 13 Oct 2007 16:28:22 +0000 (16:28 +0000)]
The problem is: currently our single byte ctype(3) functions are broken
for wide characters locales in the argument range >= 0x80 - they may
return false positives.
Example 1: for UTF-8 locale we currently have:
iswspace(0xA0)==1 and isspace(0xA0)==1
(because iswspace() and isspace() are the same code)
but must have
iswspace(0xA0)==1 and isspace(0xA0)==0
(because there is no such character and all others in the range
0x80..0xff for the UTF-8 locale, it keeps ASCII only in the single byte
range because our internal wchar_t representation for UTF-8 is UCS-4).
Example 2: for all wide character locales isalpha(arg) when arg > 0xFF may
return false positives (must be 0).
(because iswalpha() and isalpha() are the same code)
This change address this issue separating single byte and wide ctype
and also fix iswascii() (currently iswascii() is broken for
arguments > 0xFF).
This change is 100% binary compatible with old binaries.
jhb [Fri, 12 Oct 2007 19:47:42 +0000 (19:47 +0000)]
The ukbd driver has some questionable "magic" to for a default keyboard
which is ukbd0. Specifically, the keyboard driver structures for ukbd0
are not allocated/freed but are statically allocated via a persistent
global variable. There is some additional magic for the ukbd0 such that
if the keyboard is marked as probed in this global variable, then we
don't check to see if the device_t we are probing has an interface.
This causes a problem if an attach of ukbd0 fails without fulling clearing
the state in the global variable. Specifically, if the keyboard fails to
initialize in init_keyboard() or kbd_register(), then the keyboard will
still be marked as probed. The USB layer will then try to offer the
"generic" version of the USB keyboard device (as opposed to the
per-interface sub-devices) and the ukbd(4) driver will see that the
keyboard is marked probe and will skip the "is this a per-interface device"
check. Later in ukbd_attach() it panics because it tries to dereference
the interface pointer which is NULL.
The fix is to clear the flags in the persistent keyboard data for ukbd0
when init_keyboard() or kbd_register() fail.
mohans [Fri, 12 Oct 2007 19:12:21 +0000 (19:12 +0000)]
NFS MP scaling changes.
- Eliminate the hideous nfs_sndlock that serialized NFS/TCP request senders
thru the sndlock.
- Institute a new nfs_connectlock that serializes NFS/TCP reconnects. Add
logic to wait for pending request senders to finish sending before
reconnecting. Dial down the sb_timeo for NFS/TCP sockets to 1 sec.
- Break out the nfs xid manipulation under a new nfs xid lock, rather than
over loading the nfs request lock for this purpose.
- Fix some of the locking in nfs_request.
Many thanks to Kris Kennaway for his help with this and for initiating the
MP scaling analysis and work. Kris also tested this patch thorougly.
Approved by: re@ (Ken Smith)
emax [Fri, 12 Oct 2007 16:35:36 +0000 (16:35 +0000)]
Teach /etc/rc.d/ppp to start multiple instances of ppp.
ppp_profile variable can now contain multiple profiles.
Overrides for ppp mode and nat can go into ppp_$profile_mode
and ppp_$profile_nat variables respectively. If those are
not specified, defaults from ppp_mode and ppp_nat are used.
Submitted by: Yuri Kurenkov < y dot kurenkov at init dot ru >
Reviewed by: mtm
MFC after: 1 week
csjp [Fri, 12 Oct 2007 15:09:02 +0000 (15:09 +0000)]
- Change the wakeup logic associated with having multiple sleepers
on multiple different audit pipes. The old method used cv_signal()
which would result in only one thread being woken up after we
appended a record to it's queue. This resulted in un-timely wake-ups
when processing audit records real-time.
- Assign PSOCK priority to threads that have been sleeping on a read(2).
This is the same priority threads are woken up with when they select(2)
or poll(2). This yields fairness between various forms of sleep on
the audit pipes.
csjp [Fri, 12 Oct 2007 14:58:34 +0000 (14:58 +0000)]
Make sure that we refresh the PID on read(2) and write(2) operations.
This fixes the process portion of the bpf(4) stats if the peer forks
into the background after it's opened the descriptor. This bug
results in the following behavior for netstat -B:
# netstat -B
Pid Netif Flags Recv Drop Match Sblen Hblen Command
netstat: kern.proc.pid failed: No such process
78023 em0 p--s-- 2237404 43119 2237404 13986 0 ??????
csjp [Fri, 12 Oct 2007 14:57:39 +0000 (14:57 +0000)]
Add a signal handler for SIGINT to make sure that the PID file
gets cleaned up upon receiving SIGINT. This un-breaks subsequent
executions of ipfwpcap and helps when debugging network/divert
issues like this:
csjp [Fri, 12 Oct 2007 14:56:52 +0000 (14:56 +0000)]
Revision 1.12 of lockf.c fixed a "thundering herd" scenario when the
lock experienced contention a number of processes would race to acquire
lock when it was released. This problem resulted in a lot of CPU
load as well as locks being picked up out of order.
Unfortunately, a regression snuck in which allowed multiple threads
to pickup the same lock when -k was not used. This could occur when
multiple processes open a file descriptor to inode X (one process
will be blocked) and the file is unlinked on unlock (thereby removing
the directory entry allow another process to create a new directory
entry for the same file name and lock it).
This changes restores the old algorithm of: wait for the lock, then
acquire lock when we want to unlink the file on exit (specifically
when -k is not used) and keeps the new algorithm for when -k is used,
which yields fairness and improved performance.
Also, update the man page to inform users that if lockf(1) is being
used to facilitate concurrency between a number of processes, it
is recommended that -k be used to reduce CPU load and yeld
fairness with regard to lock ordering.
csjp [Fri, 12 Oct 2007 14:55:41 +0000 (14:55 +0000)]
Add pts/pty to the un-hidden devices for logins. This un-breaks
logins to jailed environments when the system is using PTS style
ptys (kern.pts.enable=1).
kib [Fri, 12 Oct 2007 10:38:05 +0000 (10:38 +0000)]
When pidfile is already locked and has zero length, do not return
success and zero pid from pidfile_read(). Return EAGAIN instead. Sleep
up to three times for 5 ms while waiting for pidfile to be written.
mount(8) does the kill(mountpid, SIGHUP). If mountd pidfile is truncated,
that would result in the SIGHUP delivered to the mount' process group
instead of the mountd.
Found and analyzed by: Peter Holm
Tested by: Peter Holm, kris
Reviewed by: pjd
MFC after: 1 week
ru [Fri, 12 Oct 2007 08:03:51 +0000 (08:03 +0000)]
Though it was possible to configure our BIND to build even when
libpthread support isn't present, our maintainer felt it's an
overkill, so instead enforce the BIND dependency on libpthread.
thompsa [Fri, 12 Oct 2007 05:23:00 +0000 (05:23 +0000)]
Update ipw to work with the new net80211 stack, plus other driver improvements.
- Add proper scanning support rather than letting the firmware grab the first
access point
- Overhaul state changes
- Use macros for locking and provide _locked() versions of some functions
- Increase debugging output
- Use a callout rather than the old watchdog interface
- Improve style, function names and defines
- Add WPA (TKIP) support
Based heavily on a patchset provided by Sam Leffler.
kientzle [Fri, 12 Oct 2007 04:11:31 +0000 (04:11 +0000)]
Correct the cpio writers to not accept data for non-regular files.
In particular, the previous code led to archives that had
non-empty bodies following directory entries. Not a fatal
problem, as bsdtar and GNU cpio are both happy to just skip
this bogus data, but it still shouldn't be there.
kientzle [Fri, 12 Oct 2007 04:08:28 +0000 (04:08 +0000)]
Correct the return values of the final zero-length block at EOF.
Return EOF immediately if an entry in a ZIP archive has no body.
In particular, the latter issue was causing bsdtar to emit spurious
warnings when extracting directory entries from ZIP archives.
mohans [Fri, 12 Oct 2007 03:56:27 +0000 (03:56 +0000)]
Set the NFS server sockbuf high watermarks to the system defaults
(up form 32KB). The low highwatermark setting caused UDP fullsock
request drops, throttling thruput greatly.
Reported by: Kris Kennaway
Approved by: re@ (Ken Smith)
yongari [Fri, 12 Oct 2007 03:32:55 +0000 (03:32 +0000)]
Not all VIA Rhine chips support 256 register space. So touching
VR_STICKHW register would result in unexpected results on these
hardwares. wpaul said the following for the issue.
The vr_attach() routine unconditionally does this for all supported
chips:
/*
* Windows may put the chip in suspend mode when it
* shuts down. Be sure to kick it in the head to wake it
* up again.
*/
VR_CLRBIT(sc, VR_STICKHW, (VR_STICKHW_DS0|VR_STICKHW_DS1));
The problem is, the VR_STICKHW register is not valid on all Rhine
devices. The VT86C100A chip, which is present on the D-Link DFE-530TX
boards, doesn't support power management, and its register space is
only 128 bytes wide. The VR_STICKHW register offset falls outside this
range. This may go unnoticed in most scenarios, but if you happen to have
another PCI device in your system which is assigned the register
space immediately after that of the Rhine, the vr(4) driver will
incorrectly stomp it. In my case, the BIOS on my test board decided
to put the register space for my PRO/100 ethernet board right next
to the Rhine, and the Rhine driver ended up clobbering the IMR register
of the PRO/100 device. (Long story short: the board kept locking up on
boot. Took me the better part of the morning suss out why.)
The strictly correct thing to do would be to check the PCI config space
to make sure the device supports the power management capability and only
write to the VR_STICKHW register if it does.
Instead of inspecting chip revision numbers for the availability of
VR_STICKHW register, check the existence of power management capability
of the hardware as wpaul suggested.
thompsa [Fri, 12 Oct 2007 03:03:16 +0000 (03:03 +0000)]
Fix two panics in lagg.
1. The locking was changed to shared but roundrobin mode still updated a
pointer in the softc with the next tx interface to use. This will panic
under high load. Change this to an atomically incremented sequence number in
order to choose the tx port in round robin.
2. IFQ_HANDOFF will free the mbuf if the queue is full, this will then be freed
again by lagg_start() and panic. Reorganised the error handling and freeing
to fix this.