Matthew Dillon [Tue, 23 Nov 2010 07:24:25 +0000 (23:24 -0800)]
kernel - Fix namecache memory exhaustion edge case
* The namecache has traditionally relied on vnode limits to avoid
blowing up its memory pool. However, creating millions of hardlinks
to the same vnode easily circumvents this assumption.
* Enhance the negative cache hysteresis code with a section for the
positive cache that does the same thing, preventing the namecache
from growing greater than 2x maxvnodes.
This situation will not occur under normal operation, it is strictly
for correctness. The kernel should never panic on normal filesystem
operations even if those operations are unrealistic.
* Tested w/HAMMER and a few million hardlinks to a single file.
Reported-by: Venkatesh Srinivas <me@acm.jhu.edu>
Sascha Wildner [Tue, 25 Jan 2011 01:34:41 +0000 (02:34 +0100)]
Sync zoneinfo database with tzdata2011a from elsie.nci.nih.gov
australasia: 8.20 -> 8.22
northamerica: 8.34 -> 8.39
Theory: 8.5 -> 8.6
* australasia: Simplification and gratuitous optimization of
South Australia rules.
* northamerica: Changes to Hawaiian time history (from an article
by Schmitt and Cox): abandoned local mean time in 1896 rather
than 1900; used DST during World War II.
* Theory: Add section on scope of the database.
Peter Avalos [Sat, 8 Jan 2011 01:36:38 +0000 (15:36 -1000)]
pthread.h: Fix prototypes.
We need to remove the condition of _POSIX_THREAD_PROCESS_SHARED for the
pshared prototypes. Otherwise, programs have to include unistd.h (to
get the definition of _POSIX_THREAD_PROCESS_SHARED) before pthread.h.
This is not the way to do things.
This fixes at least kde4, and I'm sure there's others.
Sascha Wildner [Mon, 3 Jan 2011 06:39:17 +0000 (07:39 +0100)]
<dev/netif/wi/if_wavelan_ieee.h>: Uncomment wi_req.
This, in conjunction with a few patches I submitted to pkgsrc, will
eventually fix building the following packages:
net/gtk_wicontrol
net/wimon
net/wistumbler
net/wistumbler2
net/xfce4-wavelan-plugin (and thereby meta/xfce4-extras)
YONETANI Tomokazu [Mon, 22 Nov 2010 02:35:48 +0000 (11:35 +0900)]
tmpfs - fix chgrp(), or chown() when one of uid/gid is to be unchanged
uid/gid values from function arguments were being used instead of the ones
arranged by vop_helper_chown(), accidentally setting one of them to -1
(
4294967295) instead of leaving it unchanged.
Matthew Dillon [Sun, 28 Nov 2010 06:40:13 +0000 (22:40 -0800)]
HAMMER VFS - Fix HAMMER_INODE_XDIRTY flags / state assertion
* The HAMMER_INODE_XDIRTY flag was not being immediately synchronized
with the state of ip->rec_tree in one case and the delay could create
a window of opportunity where an assertion would get hit.
* Remove the window of opportunity, the assertion no longer gets hit.
Reported-by: YONETANI Tomokazu <qhwt.dfly@les.ath.cx>
Matthew Dillon [Sun, 28 Nov 2010 06:27:10 +0000 (22:27 -0800)]
kernel - Fix MP race in signotify_remote()
* In rare cases it is possible for a process to hop multiple cpus while
a signal is trying to chase the process. Adjust signotify_remote() to
deal with the case.
Matthew Dillon [Sun, 28 Nov 2010 06:20:04 +0000 (22:20 -0800)]
kernel - Fix NFS stall and likely also x86-64 seg-fault issue
* The nfs node hash code was not MPSAFE due to its use of a global
hash table. This could lead to a stall condition due to the
global hash table losing track of its manual lock. For now use a
token to interlock hash table access and replace the manual lock
with a lockmgr lock.
* This appears to have also fixed the long-standing random seg-fault
issue with x86-64, where a buildworld loop would seg-fault every once
in a while for no apparent reason. /usr/src on my test box has always
been NFS mounted but was never considered a possible contributor to
the problem.
If NFS loses track of related nfs nodes or races operations on underlying
vnodes while expecting their storage to remain stable it is possible
for a random vnode to become corrupt. How this could translate to a
completely random seg-fault is not entirely understood but I surmise
the unrelated cc1's binary mmap could become corrupt.
Reported-by: Thomas Nikolajsen <thomas.nikolajsen@mail.dk> (nfs issue)
Matthew Dillon [Sun, 28 Nov 2010 06:28:52 +0000 (22:28 -0800)]
kernel - Hold vm_token() for certain buf/bio/vm_page interactions
* Hold the vm_token when messing around with page valid and dirty bits
in buffer cache operations. Such race conditions are possible but would
be very rare.
* Fix inconsistent use of bio2.bio_offset when swapcache intercepts an
I/O operation. When this case occurs the normal filesystem VOP_STRATEGY
function will not be called and any assumption that bio2.bio_offset
will represent a block device offset will be invalid.
To deal with this case swapcache will re-set bio_offset to NOOFFSET.
Filesystems use this value to determine that no block device offset
has been cached for a logical file block.
Reported-by: Jan Lentfer <Jan.Lentfer@web.de> (bio_offset issue)
Sascha Wildner [Tue, 23 Nov 2010 11:41:36 +0000 (12:41 +0100)]
arcmsr(4): Fix up the device description a bit.
Sascha Wildner [Tue, 23 Nov 2010 09:33:41 +0000 (10:33 +0100)]
printf(1): Fix the printing of floats which I broke back in 2007.
I forgot braces in
f386bcf2683a4edef25828e3958ba5b087d702d2
Peter Avalos [Sat, 20 Nov 2010 19:50:41 +0000 (09:50 -1000)]
OpenSSL: Reduce diffs to vendor source.
Peter Avalos [Sat, 20 Nov 2010 07:41:06 +0000 (21:41 -1000)]
Make OpenSSL thread-safe.
Sascha Wildner [Sun, 21 Nov 2010 13:23:02 +0000 (14:23 +0100)]
kbdmap(1): Fix a segmentation fault with -r.
Reported-by: Mikael Moghadam <binarybusiness@gmail.com>
Dragonfly-bug: <http://bugs.dragonflybsd.org/issue1918>
Matthew Dillon [Sun, 21 Nov 2010 02:09:23 +0000 (18:09 -0800)]
build - Fix IP_DIVERT w/UP build error
Reported-by: Stephane Russell <srussell@prodigeinfo.com>
Sascha Wildner [Sat, 20 Nov 2010 16:36:09 +0000 (17:36 +0100)]
HAMMER: Deactivate dedup in 2.8.
It was MFC'd due to suboptimal commit separation, so disallow access
to the HAMMERIOC_DEDUP ioctl in 2.8.
Also clarify this in the message we print when hammer(8) encounters
a <v5 file system.
Submitted-by: Ilya Dryomov <idryomov@gmail.com>
Sascha Wildner [Mon, 15 Nov 2010 20:08:30 +0000 (21:08 +0100)]
Move net.wlan.force_swcrypto out of IEEE80211_DEBUG.
This fixes kernel builds that don't have IEEE80211_DEBUG in the config
(such as SOEKRIS).
Matthew Dillon [Thu, 18 Nov 2010 17:45:52 +0000 (09:45 -0800)]
dummy commit
Matthew Dillon [Thu, 18 Nov 2010 17:20:23 +0000 (09:20 -0800)]
kernel - Fix global counter used in lockf assertions
* lockf elements can be accessed MP.
* This may fix the 'panic: assertion: lf_global_counter>=0 in lf_destroy_range'
panic.
Reported-by: Peter Avalos <peter@theshell.com>
Alex Hornung [Thu, 11 Nov 2010 16:21:10 +0000 (16:21 +0000)]
Temporarily bump the mbuf cluster size
* This solves issues with wpi. Once there is a different solution this
can be reverted.
Suggested-by: Matthew Dillon (dillon@)
Matthew Dillon [Sun, 14 Nov 2010 16:02:48 +0000 (08:02 -0800)]
kernel - Add sysctl to bypass wifi chipset hardware crypto
* Add sysctl net.wlan.force_swcrypto, default disabled. If set to 1
prior to attaching wlan to the underlying wireless interface this
will force software crypto to be used and bypass the hardware
crypto implemented by the wireless chipset.
* Currently we know there is at least one serious issue with the
Atheros 9280 when used with WEP2 (shows up as RSN in list scan) + WPS,
and possibly other combinations. The wrong transmit key winds up
being used for broadcast packets, including ARP.
* This sysctl works around the problem. If your wpa_supplicant is able
to connect to a network but you are unable to dhclient / use the
network you might be hitting this problem and setting the sysctl to
1 may fix it.
Note that the sysctl must be set prior to attaching the wlan to the
underlying wireless driver, so it is typically best to do it at boot
time or near the top of whatever script you use.
Inspired-by: 2-days of connectivity issues at NYCBSDCON.
Matthew Dillon [Sun, 14 Nov 2010 15:03:41 +0000 (07:03 -0800)]
kernel - add kern.sysbeep_enable feature
* Add a sysctl kern.sysbeep_enable (default on). Setting this to 0
will disable the system beep / bell.
* Useful when messing with a laptop in a conference, those beeps can
get really annoying.
Stathis Kamperis [Sat, 13 Nov 2010 16:40:48 +0000 (18:40 +0200)]
HAMMER - Check if supplied conf file exists
Dragonfly-bug: <http://bugs.dragonflybsd.org/issue1911>
Sascha Wildner [Sat, 13 Nov 2010 12:59:18 +0000 (13:59 +0100)]
hammer.8: Add some words about HAMMER version 5 and about upgrading to it.
Submitted-by: Ilya Dryomov <idryomov@gmail.com>
Matthew Dillon [Thu, 11 Nov 2010 18:15:46 +0000 (10:15 -0800)]
kernel - Make sure invalidated pages are unmapped in rare situation.
* vfs_vmio_release() was ignoring pages busied by other MP users while
releasing VM pages associated with the buffer cache. This could
potentially lead to invalida pages remaining memory mapped.
* Ensure that the released pages found to be in this state are still
unmapped.
Matthew Dillon [Thu, 11 Nov 2010 18:12:59 +0000 (10:12 -0800)]
kernel - Close possible hole in hammer inval code
* Do not use FINDBLK_TEST, instead have findblk() return a locked buffer
cache buffer and deal with it from there. While the original code
should have been ok (it would getblk() the buffer cache in either case),
it depended on certain MP race characteristics that might not hold so
don't take any chances.
* This does not fix any known issues but removes some uncertainty.
Matthew Dillon [Wed, 10 Nov 2010 20:12:45 +0000 (12:12 -0800)]
HAMMER utility - Handle pruning when only snapshot meta-data present
* If only snapshot meta-data is present and no snapshot softlinks are found
hammer prune (and hammer cleanup) previous exited with an error. This case
can occur when creating manual snapshot softlinks and/or trying to prune
a slave PFS.
* In fact this is incorrect. If meta-data is present we can prune based
on that.
Matthew Dillon [Sun, 7 Nov 2010 18:03:03 +0000 (10:03 -0800)]
HAMMER - Add hammer dedup directive and support
* Adjust the utility to exit with an appropriate error if an
attempt is made to dedup a filesystem which has not been upgraded
to version 5.
Matthew Dillon [Sun, 7 Nov 2010 17:46:15 +0000 (09:46 -0800)]
HAMMER - Add hammer dedup directive and support
* Minor adjustments to fix compile errors on 64-bit systems
(printf %lld -> %jd and use intmax_t casts).
Matthew Dillon [Sun, 7 Nov 2010 17:43:08 +0000 (09:43 -0800)]
HAMMER - Add hammer dedup filesystem version dump
* Require a version 5 filesystem for dedup operation. Upgrading can be done
in-place. All this does is make the layer2 bytes_free field a signed
value instead of unsigned, allowing it to go negative.
This allows dedup to use the field to account for multiple references to
the same block of data, causing the field to (likely) go negative.
Matthew Dillon [Sun, 7 Nov 2010 17:29:39 +0000 (09:29 -0800)]
HAMMER - Add hammer dedup directive and support
* Implements all the logic required to dedup a HAMMER filesystem.
* There is one remaining issue and that is the reblocker's propensity to
undo de-dup's hard work in certain cases.
* Code bounty for hammer_dedup
Submitted-by: Ilya Dryomov <idryomov@gmail.com>
Matthew Dillon [Sun, 7 Nov 2010 17:20:11 +0000 (09:20 -0800)]
libutil - Add humanize_unsigned()
* Add a humanize_unsigned() function as a companion to humanize_number()
which does a much better job on large unsigned 64-bit quantities.
Submitted-by: Ilya Dryomov <idryomov@gmail.com>
Peter Avalos [Thu, 18 Nov 2010 05:57:01 +0000 (19:57 -1000)]
Update build for OpenSSL-1.0.0b import.
Peter Avalos [Thu, 18 Nov 2010 06:51:07 +0000 (20:51 -1000)]
Merge branch 'vendor/OPENSSL' into DragonFly_RELEASE_2_8
Peter Avalos [Thu, 18 Nov 2010 05:41:45 +0000 (19:41 -1000)]
Import OpenSSL-1.0.0b.
This primarily fixes CVE-2010-3864 and CVE-2010-2939.
Sascha Wildner [Wed, 3 Nov 2010 20:30:03 +0000 (21:30 +0100)]
encrypted_root.sh: Fix a typo.
Joe Talbott [Wed, 3 Nov 2010 18:29:47 +0000 (14:29 -0400)]
network - Add initializer for ifnet.if_groups.
Prevents a NULL pointer dereference when pfsync is compiled into the
kernel.
Reported-By: mat on Efnet#dragonflybsd
Tested-By: mat on Efnet#dragonflybsd
Matthew Dillon [Tue, 2 Nov 2010 20:58:15 +0000 (13:58 -0700)]
libstand - Fix CD9660 name matching code
* The code inherited from FBsd did not properly deal with matching
path elements after a '.'. For example "kernel.FUBAR_XYZ" would match
"kernel.FUBAR" on the CD.
* Fixed.
Matthew Dillon [Tue, 2 Nov 2010 20:36:46 +0000 (13:36 -0700)]
kernel - Reenable acpi sleep states
* Reenable acpi sleep state handling by default, by setting
machdep.cpu_idle_hlt to 2 by default.
Matthew Dillon [Tue, 2 Nov 2010 19:55:17 +0000 (12:55 -0700)]
kernel - Attempt to fix lost rpc issue with NFS timeout/retry
* An edge case in the timeout/retry code failed to pick up on
async RPCs which are replied to while the timeout code has the
rpc locked for retry.
Reported-by: Thomas Nikolajsen <thomas.nikolajsen@mail.dk>
Matthew Dillon [Tue, 2 Nov 2010 17:12:14 +0000 (10:12 -0700)]
network - Fix missing tailq initializer for if_creategroup() API
* The ifg_head list was missing its initializer, causing calls to
if_creategroup() (e.g. by pf/pfsync) to panic the box.
Matthew Dillon [Tue, 2 Nov 2010 05:08:35 +0000 (22:08 -0700)]
network - Fix netstat output missing entries
* netstat was missing entries due to the kernel pcb listing sysctl
code doing a bogus generation count test which did not take into
account pcb's which move between cpus due to e.g. a bind() operation.
Reported-by: Thomas Nikolajsen <thomas.nikolajsen@mail.dk>
Matthew Dillon [Tue, 2 Nov 2010 05:11:59 +0000 (22:11 -0700)]
Merge branch 'DragonFly_RELEASE_2_8' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into DragonFly_RELEASE_2_8
Alex Hornung [Mon, 1 Nov 2010 20:38:14 +0000 (20:38 +0000)]
devfs - fix some issues with rules
* devfs_rule_alloc was still assuming that the rule_type and rule_cmd are not
bit fields, but they now are.
* return error in a number of places when a null pointer or nil-length string
is passed in.
Dragonfly-bug: http://bugs.dragonflybsd.org/issue1885
Reported-by: 'fanch'
Sascha Wildner [Mon, 1 Nov 2010 16:32:13 +0000 (17:32 +0100)]
Sync zoneinfo database with tzdata2010o from elsie.nci.nih.gov
australasia: 8.19 -> 8.20
* Fiji ends DST in 2011 on March 6 and not on March 27.
Matthew Dillon [Mon, 1 Nov 2010 16:29:31 +0000 (09:29 -0700)]
kernel - Fix access chceck in tmpfs that prevented chroot's into it
* Fix broken code which caused access() calls into tmpfs to assume
write permission was being requested when it was not.
Submitted-by: Rumko <rumcic@gmail.com>
Matthew Dillon [Mon, 1 Nov 2010 02:38:46 +0000 (19:38 -0700)]
kernel - Do not back out critical section for exchange_scr() call
* The critical section may be protecting a spinlock and cannot be safely
backed out when calling exchange_scr().
* Fixes a panic.
Reported-by: rob
Matthew Dillon [Sun, 31 Oct 2010 15:52:00 +0000 (08:52 -0700)]
kernel - Fix serious issue w/ smp_invltlb(), plus other issues (3)
* splz() and doreti on both i386 and x86_64 were leaving interrupts
disabled through calls to e.g. sched_ithd() and IPI function callbacks.
As before this can lead to a deadlock if these functions call
smp_invltlb() or some other cpu calls smp_invltlb() while these
functions are spinning on a full IPIQ.
* sti in appropriate places to fix the problem.
Thomas Nikolajsen [Sat, 30 Oct 2010 07:09:52 +0000 (09:09 +0200)]
loader(8): adjustment & man update
* add/enhance info on if*/else*/endif
* add info on default_kernel variable
* include more variables in kenv_vars, including autoboot_delay,
to make setting them in loader.conf work
* remove a few unused variables
Matthew Dillon [Sat, 30 Oct 2010 05:46:29 +0000 (22:46 -0700)]
kernel - Fix longstanding VM long-duration stall issues (2)
* Refactor the pageout daemon's hysteresis, in particular the
free page queue had no real hysteressis and could cause excessive
pagedaemon wakeups.
Use a calculation that maintains a free page queue about half the
size of the minimum cache queue, giving us pretty good pipelining
when a system is under constant memory pressure.
* Add a sysctl for monitoring ppwakeups (wakeups of processes waiting
for memory).
Matthew Dillon [Sat, 30 Oct 2010 03:58:00 +0000 (20:58 -0700)]
kernel - Fix longstanding VM long-duration stall issues
* The pageout daemon was using the wrong wakeup metrics and essentially
would not start cleaning out pages until after processes began to
enter vmwait/vmrate or otherwise begin to stall due to low free memory.
* Fix the metric and also double-check every 5 seconds when the pagedaemon
statistics wakeup occurs.
* vm.v_free_min + vm.v_cache_min may now be used to tune when the
pagedaemon is woken up.
Matthew Dillon [Fri, 29 Oct 2010 06:27:32 +0000 (23:27 -0700)]
dummy commit so a 2.8 commitid is at the top for tagging.
Matthew Dillon [Fri, 29 Oct 2010 06:23:48 +0000 (23:23 -0700)]
nata - properly teardown interrupts on ENXIO
* When a ata controller cannot be attached (e.g. due to AHCI having
already taken its resources) be sure to teardown and release any
resources that had been obtained.
* This fixes a panic on certain motherboards (e.g. gigabyte / phenomX6)
where the ata controller ENXIO'd but left interrupt handlers setup.
An interrupt would then occur, access stale memory, and blow the machine
up.
Matthew Dillon [Fri, 29 Oct 2010 04:33:15 +0000 (21:33 -0700)]
boot - Multiple changes to try to make cdboot work with more bioses
* Go back to single sector read. This seems to help gigabyte bioses
(phenom x 6). Multi-sector reads tend to cause reads to fail with
a timeout for reasons unknown.
* Enable interrupts on start.
* (non operator) replace jcxz. It's a bit unclear whether this is an
issue or not but some assemblers do not generate the correct prefix
for jcxz.
* Report read timeouts and delay 1 second before retrying. At the
very least this allows timeouts to be reported distinctly.
Antonio Huete Jimenez [Mon, 25 Oct 2010 15:04:37 +0000 (17:04 +0200)]
nullfs - Add 'noexec' flag support.
* Make possible to mount a null-mount with 'noexec' option.
* If the underlying filesystem is mounted as 'noexec' and no options
were given at mount-time, the null mount will automatically
have 'noexec'.
Pointed-out-by: @dillon
Matthew Dillon [Thu, 28 Oct 2010 21:31:29 +0000 (14:31 -0700)]
nrelease gui - add xterm resources
* Add some resources for xterms
Matthew Dillon [Thu, 28 Oct 2010 19:36:32 +0000 (12:36 -0700)]
kernel - Fix serious issue w/ smp_invltlb(), plus other issues (2)
It looks like there are a number of additional cases where kernel threads
can wind up running with interrupts physically disabled, which can create
serious problems for smp_invltlb().
It is not entirely clear how these deadlocks occur since the IPI code does
a forced "STI" if it would otherwise loop, but there are several other
hard loops that do not: lwkt_switch looping on a vm_token, spin locks,
and probably a few other places.
We want interrupts to be enabled in all cases where these sorts of loops
occur in order to be able to service Xinvltlb via smp_invltlb() as well
as to prevent the LAPIC interrupt queue from filling up and livelocking
something that it shouldn't.
* lwkt_preempt() must save, zero, and restore gd->gd_intr_nesting_level
when issuing a preemption. Otherwise it will improperly panic on
an assertion if the preempting interrupt thread code tries to switch out.
It is perfectly acceptable for the preempting thread to block (it just
switches back to the thread that got preempted).
Why the assertion was not occuring before I do not know but it is
probably related to threads getting stuck in 'cli' mode. The additional
changes below appear to significantly increase the number of interrupt
thread preemptions which succeed (lwkt.preempt_{hit,miss} counters).
* STI prior to running ithread_fast_handler() from all IDTVECs related
to device interrupts.
* STI in Xcpustop, Xipiq, and Xtimer. These functions can call more
complex C code and doing so with interrupts disabled may prevent
Xinvltlb (via smp_invltlb()) from being executed, deadlocking the
system.
* Reorder a mfence(). Probably not needed but do it anyway.
Matthew Dillon [Thu, 28 Oct 2010 06:55:58 +0000 (23:55 -0700)]
kernel - Fix serious issue w/ smp_invltlb(), plus other issues.
* smp_invltlb() was running asynchronously when it really needs to run
synchronously. Generally speaking the asynchronous ipi did in fact work
pretty well but it still presents a 1uS window of opportunity which
bypasses normal write ordering safeties.
Run smp_invltlb() synchronously.
* Fixing the above lea to the discovery of an ACPI issue. The ACPI
cpu idle halt code, at least on the gigabyte phenom x 6 I've been
testing with, can cause IPIs to be lost. Not just delayed, straight
out lost. Gone. Poof. It doesn't matter whether the IPI is a
broadcast IPI or a directed IPI, it can still get lost.
This was particularly noticeable when I fixed smp_invltlb() and my
test box started locking up due to a random cpu sometimes not receiving
the Xinvltlb IPI, and it is quite possible that this issue was also
responsible for the random seg-faults we would sometimes get on 64-bit
boxes.
For now the acpi halt code has been disabled. It can be enabled with
sysctl machdep.cpu_idle_hlt=2 if you want to risk it.
* Use doreti_syscall_ret and doreti_iret in several cases that were
previously popping the interrupt frame and iret'ing manually. This
is operationally equivalent.
* Add a missing "sti" in the idle loop. Usually the cpu_idle_hook()
deals with this but there are some alternative paths which might not,
potentially causing interrupts to be delayed unnecessarily.
At worst the idle thread has an extra sti in it.
* Add v_smpinvltlb to struct vmmeter, plus some reserved slots for
future expansion.
* Adjust vmstat -s to report smpinvltlb's.
Matthew Dillon [Thu, 28 Oct 2010 21:16:58 +0000 (14:16 -0700)]
Merge branch 'DragonFly_RELEASE_2_8' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into DragonFly_RELEASE_2_8
Sascha Wildner [Thu, 28 Oct 2010 20:19:12 +0000 (22:19 +0200)]
nrelease: Add two options (dri and inet6) to the default mk.conf we supply.
We use them for building the packages that go to the LiveCD.
However, we don't add PKG_OPTIONS.scmgit=-scmgit-gui (used for the LiveCD's
scmgit to save space). So people who want to continue using git without X11
support will need to tweak mk.conf like before.
Sascha Wildner [Thu, 28 Oct 2010 19:43:27 +0000 (21:43 +0200)]
nrelease/installer: Rename the UP/SMP kernel directories on the LiveCD.
Changed from UP -> kernel.UP and from SMP -> kernel.SMP.
Requested-by: dillon
Sascha Wildner [Wed, 27 Oct 2010 07:08:48 +0000 (09:08 +0200)]
installer: Install the kernel used to install the system to /boot/kernel.
We assume that the kernel the user booted the LiveCD/image with for
installing his system will be the kernel he wants to boot the system
from, too.
Sascha Wildner [Wed, 27 Oct 2010 07:07:13 +0000 (09:07 +0200)]
nrelease: Fix an annoying bug that was preventing the ISOs from booting UP.
It seems our CD9660 support in libstand at least has some problems with
directory names containing dots. This prevented the LiveCD to properly
boot the UP kernel. Don't ask me about how it ended up using the SMP
kernel in this case, but that's what it did, making the UP boot
impossible. My guess is it has something to do with the order in which
stuff had been added to the ISO. But that's pure speculation. I'm not
even sure about the dot part. It might as well be directory name length.
In any case, UP boot from the LiveCD was broken.
To fix all this, install the UP kernel to /boot/UP and the SMP kernel to
/boot/SMP on the LiveCD and images. They will be picked up by the
installer from there by a separate commit.
While here, remove some non-functional code from dloader.menu. We can't
easily build i386/x86_64 dual boot ISOs at the moment (for this we would
need cross-building pkgsrc).
Thanks to sjg for bringing me on the right track.
Sascha Wildner [Wed, 27 Oct 2010 04:56:42 +0000 (06:56 +0200)]
installer: Add __printflike()s and fix resulting format warnings.
Sascha Wildner [Wed, 27 Oct 2010 01:45:43 +0000 (03:45 +0200)]
nrelease: Add a missing semicolon.
Sascha Wildner [Tue, 26 Oct 2010 18:29:18 +0000 (20:29 +0200)]
nrelease: Remove unused variable.
Sascha Wildner [Tue, 26 Oct 2010 17:20:39 +0000 (19:20 +0200)]
dloader: Fix error messages.
Sascha Wildner [Tue, 26 Oct 2010 13:24:31 +0000 (15:24 +0200)]
zic(8): Sync with tzcode2010n from elsie.nci.nih.gov
* Conditionally output extra types with most-recently-used offsets last
(for use by systems with pre-2011 versions of localtime.c, helping to
ensure that globals "altzone" and "timezone" get set correctly).
* Fix generation of POSIX strings for zones with rules using
"weekday<=n" forms of dates (thanks to Lei Liu for finding the
problem). Also, limit output for non-POSIX-specificable zones defined
to follow the same rules every year. (Note that no zones of either of
the above types appear in the distribution; these changes cater to
add-on zones).
Sascha Wildner [Tue, 26 Oct 2010 13:20:01 +0000 (15:20 +0200)]
libc/stdtime: Sync localtime.c with tzcode2010n from elsie.nci.nih.gov
* Change to set timezone and altzone based on time types with greatest
transition times (for the benefit of Asia/Seoul).
* Initialize ttinfo structures filled by tzparse (thanks to Ravindra
for reporting a valgrind warning).
Sascha Wildner [Tue, 26 Oct 2010 13:14:11 +0000 (15:14 +0200)]
Sync zoneinfo database with tzdata2010n from elsie.nci.nih.gov
Theory: 8.4 -> 8.5
asia: 8.61 -> 8.62
australasia: 8.18 -> 8.19
factory: 8.2 -> n/a
zone.tab: 8.37 -> 8.38
* Remove the 'factory' file and references to it. It is no longer needed.
* asia: No DST in Hong Kong in 1977 (thanks to Jesper Norgaard for
finding the problem).
* australasia: Change end of DST in Samoa in 2011 from 2011-04-03 0:00 to
2011-04-03 1:00 (thanks to Raymond Hughes for a reference).
* zone.tab: Remove obsolete association of Vostok Station with South
Magnetic Pole; add association with Lake Vostok (thanks to
Petr Machata for finding the problem).
Matthew Dillon [Mon, 25 Oct 2010 23:33:01 +0000 (16:33 -0700)]
nrelease - gui - Change window manager, cpdup additional directories
* Change to fvwm2
* Autostart two xterms and clocks
* Add simple button bar with two buttons 'xterm' and 'firefox'.
* Installer now also installs /usr/src and /usr/pkgsrc from the stick.
Matthew Dillon [Mon, 25 Oct 2010 22:12:06 +0000 (15:12 -0700)]
nrelease - Provide default /etc/X11, better rc.conf
* Don't make users agonize over /etc/X11, provide one even if we aren't
the gui. Also provide it in /etc.hdd so it gets installed onto the
HD.
* This also fixes the problem where the gui install was failing to install
/etc/X11 onto the HD.
* The normal and gui installs were not installing reasonable rc.conf
defaults. Do so.
Matthew Dillon [Mon, 25 Oct 2010 20:46:13 +0000 (13:46 -0700)]
nrelease - gui build adjustments
* Remove packages that failed to build. Unfortunately rxvt was one of them.
Note that I did some manual build hacks to other packages to build
which aren't committed to pkgsrc.
* Fix size calculation for usb disk img.
* Adjust fluxbox menu firefox3->firefox and urxvt->xterm.
Matthew Dillon [Mon, 25 Oct 2010 18:35:44 +0000 (11:35 -0700)]
build - define struct in6_addr in arpa/inet.h
* Use the same mechanic used for struct in_addr to define in6_addr in
arpa/inet.h, conditionalizing the define in in6.h.
* This helps w/pkgsrc builds.
Matthew Dillon [Mon, 25 Oct 2010 02:37:25 +0000 (19:37 -0700)]
dummy commit so a 2.8 commitid is at the top for tagging.
Matthew Dillon [Mon, 25 Oct 2010 02:35:16 +0000 (19:35 -0700)]
nrelease - installer - Use different emergency interrupt mechanic
* Use a tuneable instead of a sysctl.
YONETANI Tomokazu [Mon, 25 Oct 2010 01:51:35 +0000 (10:51 +0900)]
ehci - add missing dependency on usb driver
Matthew Dillon [Mon, 25 Oct 2010 01:23:45 +0000 (18:23 -0700)]
docs - update tuning(7)
* Get rid of a ton of old cruft that no longer applies and replace with
more applicable information.
Antonio Huete Jimenez [Sun, 24 Oct 2010 22:59:45 +0000 (00:59 +0200)]
Fix typo in argv handling.
Matthew Dillon [Sun, 24 Oct 2010 22:29:00 +0000 (15:29 -0700)]
kernel - Temporarily back-out smp_invltlb() changes
* These changes are deadlocking i386 kernels, and it is too close to
the release so back them out for the release. The issue will be
fixed another way post-release.
Matthew Dillon [Sun, 24 Oct 2010 22:27:14 +0000 (15:27 -0700)]
kernel - Fix bug in the emergency interrupt polling thread
* This thread is only active when kern.emergency_intr_enable is non-zero
* Get the next intr handler before processing the current one, since the
act of processing the current handler may remove it.
* In the emergency interrupt polling thread do not stall if we cannot get
the serializer for the interrupt. just skip the handler and we'll try
again next time.
Matthew Dillon [Sun, 24 Oct 2010 18:29:22 +0000 (11:29 -0700)]
sysctl.conf - Enable emergency interrupt polling at a low rate by default
* Set kern.emergency_intr_enable=1 in the ISO/IMG sysctl.conf for
newly installed kernels. The idea is so people can get things working
more quickly and worry about performance afterwords.
* This commit actually got into the previous commit accidently so I am
posting a following commit so we get a proper comment associated with it.
Matthew Dillon [Sun, 24 Oct 2010 18:01:41 +0000 (11:01 -0700)]
hammer utility - Add force support to cleanup
* Allow hammer -f cleanup to override the normal timestamp checking and
force the cleanup to run.
Matthew Dillon [Sun, 24 Oct 2010 16:27:35 +0000 (09:27 -0700)]
nrelease - Add *_GENERIC_SMP for this release
* We will be stuffing both UP and SMP kernels on our release img/iso
for 2.8.
Matthew Dillon [Sun, 24 Oct 2010 07:00:59 +0000 (00:00 -0700)]
nrelease - Release build adjustments
* Add GITHOST override to Makefile.usr
* For now checkout 2010Q3 by default in master, this will be changed
to 'master' in a month or two but for now we want to match the release.
* Add GITHOST feature to the nrelease Makefile, defaulting to
git.dragonflybsd.org. This allows developers who do release builds
to specify a more local clone/mirror/copy of the pkgsrcv2.git and
dragonfly.git repos.
nrelease pulls from these repos so this can save a bunch of time.
* nrelease now installs each kernel into its own kernel.XXX directory
in /boot, with a complete set of modules for each kernel. This is
instead of installing all kernels into /boot/kernel/ and naming them
differently inside /boot/kernel/.
This is to conform to the recent dloader work and very recent new
menu features.
* All kernels+modules are installed with INSTALLSTRIPPEDMODULES so the
ISO comes in at a reasonable size.
* The release ISO now contains both a UP and a SMP kernel, selectable
at boot time.
* Do not try to include pkgsrc in the release ISO/IMG (at 1G+ it is too
big). We continue to include system sources. There isn't enough room
for full sources in the ISO/IMG. Note that GUI builds will include
a full pkgsrc and full sources, including a .git base for them.
* nrelease now supports a IMGSIZE override on the make line (in sectors),
for the usb image.
* nrelease now autosizes the IMG file (when not overriden by IMGSIZE) to
the nearest base-10 gigabyte to ensure that it fits on the USB stick
and provide a bit of extra space for messing around.
Matthew Dillon [Sun, 24 Oct 2010 03:27:30 +0000 (20:27 -0700)]
mtree - no longer create /boot/modules
* /boot/modules is no longer used, do not create it in mtree.
Matthew Dillon [Sun, 24 Oct 2010 02:32:03 +0000 (19:32 -0700)]
boot - Massage dloader.menu a bit more
* Don't present the 64-bit / 64-bit-smp boot options if the current default
is already 64-bit. The generic SMP boot option will still be presented
if the current default is the 64-bit UP kernel.
Matthew Dillon [Sun, 24 Oct 2010 02:03:51 +0000 (19:03 -0700)]
boot - Add a ton of features to the boot loader & adjust default menu
* Add conditionals ifset ifexists else elseifexists and endif
* Conditionalize dloader.menu to only present menu options for which
the related kernels are present in /boot.
* Set the default kernel to one of:
kernel, kernel.GENERIC, kernel.GENERIC_SMP, kernel.X86_64_GENERIC,
or kernel.X86_64_GENERIC_SMP. The first one in the list found becomes
the default.
* If ${default_kernel}_SMP is available supply a menu option to change
the default to that kernel, aka UP->SMP
* If kernel.X86_64_GENERIC[_SMP] is available and the current default is
not the same supply menu options to change the default to the 64-bit
UP or SMP kernel.
Note however that for this to work the related boot kernel directory
had better have a loader.conf.local in it that points the root mount
to a 64-bit root 'cause 64-bit kernels can't run 32 bit binaries yet.
* Adjust the menu item execution code to copy the items before executing
them, allowing recursive menu execution. The recursive menu feature
is a bit of a hack right now however.
* NOTE: the optcd for people ESCaping into the boot prompt was moved
to dloader.rc, when updating you may have to manually reinstall that
file to get the functionality.
Matthew Dillon [Sat, 23 Oct 2010 17:51:06 +0000 (10:51 -0700)]
network - Fix issue with recent unix domain socket gc fixes
* The gc's tracking of the deferral count could get out of sync and wind
up looping forever due to the (now necessary) tsleep between loops.
* Fix by recalculating the count on each loop.
Matthew Dillon [Sat, 23 Oct 2010 16:43:18 +0000 (09:43 -0700)]
kernel - Fix pmap deactivate/reactivation race.
* The LWKT thread switch code clears the cpu mask bit in
proc->p_vmspace->vm_pmap.pm_active, and the switch-in code sets the
mask bit.
This code has a bug because the switch code ALSO optimizes the loading
of %cr3 to avoid reloading it if it hasn't changed, for example when
switching between two user threads associated with the process,
because the other cpu(s) running similar threads may lose track of
the fact that our cpu also needs an IPI for page invalidations in the
pmap for a short period of time.
Because we don't reload %cr3 in this case, our tlb can become invalid.
This can also occur with vfork() sequences.
* Fix by testing that we are switching to the same vmspace and do not
clear the pm_active bit in that case. Retain the %cr3 optimization.
Matthew Dillon [Sat, 23 Oct 2010 16:41:44 +0000 (09:41 -0700)]
kernel - synchronize smp_invltlb()
* With much of the kernel running MPSAFE, smp_invltlb() cannot run
asynchronously.
Matthew Dillon [Sat, 23 Oct 2010 16:35:14 +0000 (09:35 -0700)]
kernel - x86_64 - Add additional checks to lwp_wait() & friends
* lwp_wait() must defer reaping of a lwp that is still running on another
cpu (i.e. in the midst of exiting).
* It shouldn't be possible for this to happen but just incase the thread
gets switched out after TDF_EXITING has been set, also make sure
the threat is no longer on the LWKT run queue.
* Remove old debugging in the LWKT scheduler path.
* Protect gd_freetd (per-cpu td cache) with a critical section. Again
this case should not occur as new threads are not allocated from
interrupts, but protect it anyway. Also assert that the cached free
td is in no way still scheduled.
Note that this cache is required to ensure that the td does not
end up in the MP-accessible objcache before it has been fully
descheduled.
Matthew Dillon [Sat, 23 Oct 2010 16:33:49 +0000 (09:33 -0700)]
kernel - x86_64 - Remove intrframe->if_gs
* The if_gs field doesn't exist in the actual trap frame.
Alex Hornung [Sat, 23 Oct 2010 11:16:22 +0000 (12:16 +0100)]
fix typo in previous commit
Alex Hornung [Sat, 23 Oct 2010 10:59:18 +0000 (11:59 +0100)]
Add comment regarding T_DNA passthrough with db
Requested-by: Aggelos Economopoulos (aggelos@)
Alex Hornung [Sat, 23 Oct 2010 08:18:10 +0000 (09:18 +0100)]
libdevmapper, liblvm - stop inter-lib linking
* Don't link libraries to other libraries, this only causes trouble like
having to do: -lprop -ldevmapper -lprop.
* Remove the double-linking to prop and crypto in cryptdisks.
Suggested-by: Sascha Wildner (swildner@)
Alex Hornung [Sat, 23 Oct 2010 07:34:01 +0000 (08:34 +0100)]
cryptdisks rc.d - remove rcvar
* Remove the rc variable to enable cryptdisks so that it always runs.
It'll just check for the crypttab file, and if it isn't present, it
doesn't do anything.
Alex Hornung [Sat, 23 Oct 2010 07:27:56 +0000 (08:27 +0100)]
crypttab.5 - Add LUKS-only note, remote empty line
Sascha Wildner [Sat, 23 Oct 2010 07:16:26 +0000 (09:16 +0200)]
Fix UP kernel build.
Matthew Dillon [Sat, 23 Oct 2010 02:32:01 +0000 (19:32 -0700)]
libkvm - Add ability to access userspace from kgdb on cores
* Use %cr3 from the dumppcb instead of KPML4phys on x86_64, and similarly
for i386, to access the full page table as of when the panic occured
instead of just the kernel page table.
* minidumps do not dump userspace so userspace will still not be available,
but this gives us the option of sysctl'ing off minidumps when userspace
access is desired, and kgdb will then be able to access the current
userspace context as of the panic, as well.