Difference between revisions of "Download/kernel/rhel5/028stab031.1/changes"

From OpenVZ Virtuozzo Containers Wiki
Jump to: navigation, search
(created)
 
(-4)
Line 1: Line 1:
−
== Changes ==
+
The fact that teachers need better   training to carry out deliberate instruction in reading, spelling, and writing should prompt action rather than criticism. ,
−
* Mainstream security fix in netlink.
 
−
* Fixes in checkpointing, CPU fair scheduler, VE I/O scheduling, beancounters, VE disk quotas, etc.
 
−
 
 
−
=== Config changes ===
 
−
Added:
 
−
* +<code>CONFIG_IKCONFIG=y</code>
 
−
* +<code>CONFIG_IKCONFIG_PROC=y</code>
 
−
* +<code>CONFIG_KEXEC=y</code> (i686-PAE only)
 
−
* +<code>CONFIG_CRASH_DUMP=y</code> (i686-PAE only)
 
−
* +<code>CONFIG_PROC_VMCORE=y</code> (i686-PAE only)
 
−
Removed:
 
−
* -<code>CONFIG_4KSTACKS=y</code>
 
−
<includeonly>[[{{PAGENAME}}/changes#Patches|{{Long changelog message}}]]</includeonly><noinclude>
 
−
=== Patches ===
 
−
 
 
−
==== diff-arch-4gb-20070409 ====
 
−
<div class="change">
 
−
Patch from Kirill Korotaev &lt;dev@openvz.org&gt;:<br/>
 
−
[4G/4G] i686: 4GB split patch
 
−
 
 
−
4G split patch based on linux-2.6.0-4g4g.patch from RHEL4 from Ingo Molnar.
 
−
 
 
−
Changes:
 
−
* fixed reserved pages handling
 
−
* fixed PI futexes direct user access atomic operations
 
−
* fixed lockdep
 
−
* fixed entry.S to handle multiple irets and 16bit stacks, NMI support
 
−
* reworked hw breakpoints support as in my patch in 2.4.x
 
−
* fixed huge TSS and io bitmaps
 
−
* added PTL locks support
 
−
* fixed kprobes
 
−
* 4-level page tables
 
−
* lots of fixes merged from OVZ 2.6.9 (see below)
 
−
 
 
−
Merged:
 
−
; linux-2.6.9-4g4g-hugemem-warning.patch
 
−
:
 
−
; linux-2.6.9-4g4g-noncachable.patch
 
−
:
 
−
; diff-arch-4gb-swsuspd
 
−
: Fixed broken software suspend
 
−
; diff-arch-4gb-fixaddr2
 
−
: Fixed VSYSCALL_BASE/FIXADDR_TOP interrelations (8k stacks)
 
−
; diff-arch-4gb-stksize
 
−
: Fixed virtual stack mappings for 8k stacks
 
−
; diff-arch-4gb-copyusr
 
−
: Optimization in filemap.c on copy_XXX_user
 
−
; diff-arch-4gb-vmalloc2
 
−
: vmalloc area extending up to 256Mb
 
−
; diff-arch-4gb-tasksize
 
−
: Limiting user address space to 3GB
 
−
; diff-arch-4gb-tssmaps
 
−
: Fixed TSS mapings for online_cpus &gt; 2 and big TSS
 
−
; diff-arch-4gb-ldtleak
 
−
: Fixed leak of LDT pages on error path
 
−
; diff-arch-4gb-gcc296
 
−
: Fixed get_user() compilation bug on GCC 2.96
 
−
; diff-arch-4gb-amd-prefetch
 
−
: Fixed prefetch code detection on AMD
 
−
; diff-arch-4gb-mce-20060824
 
−
: Fixed MCE handling when 4GB split is on
 
−
 
 
−
Not needed (already in):
 
−
* linux-2.6.9-net-b44-4g4g.patch
 
−
* linux-2.6.9-4g4g-maxtasksize.patch
 
−
* diff-arch-4gb-pgdctor
 
−
* diff-arch-4gb-getname
 
−
</div>
 
−
 
 
−
==== diff-ms-lockdep-neighbour-table-class-20070409 ====
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
 
 
−
[LOCKDEP] Fix wrong deadlock report in neigh table
 
−
 
 
−
Lockdep detects a fake deadlock in the calltrace:
 
−
<pre>
 
−
neigh_proxy_process()
 
−
  `- lock(neigh_table-&gt;proxy_queue.lock);
 
−
arp_process (tbl-&gt;proxy_redo)
 
−
neigh_event_ns
 
−
neigh_update
 
−
skb_purge_queue
 
−
  `- lock(neighbour-&gt;arp_queue.lock);
 
−
</pre>
 
−
 
 
−
Actually there is no deadlock as the first lock and the second
 
−
one are different skb_buff_head's locks, but they are initialized
 
−
both in skb_queue_head_init() and thus have on lockdep-class.
 
−
 
 
−
This is a mainstream "BUG".
 
−
Fixed by adding another class to neigh_table's proxy_queue lock.
 
−
 
 
−
Bug #78837.
 
−
 
 
−
(Tested with node bootup, VE start and vzt-prep vzt-ss).
 
−
</div>
 
−
 
 
−
==== diff-ubc-page-uncharge-bug-first-20070406 ====
 
−
<div class="change">
 
−
 
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[BC] Check correct user_beancounter passed first in ub_page_uncharge()
 
−
 
 
−
If page accidentally has a not-removed page_beancounter kernel will
 
−
oops dereferencing ub-&gt;ub_percpu(). Move the BUG_ON upper to be sure
 
−
we work with user_beancounter.
 
−
 
 
−
This might be the lost in bug #78461.
 
−
</div>
 
−
 
 
−
==== diff-ubc-refcount-leak-dupmm-20070409 ====
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:<br/>
 
−
[BC] refcount leak in dup_mm() on error path
 
−
 
 
−
Fix simple beancounter refcount leak on error path
 
−
in dup_mm().
 
−
 
 
−
Bug #77231.
 
−
</div>
 
−
 
 
−
==== diff-ubc-vmguar-enough-null-mm-20070406 ====
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[BC] vmguar_enough_memory() oopses if called form kernel thread
 
−
 
 
−
If vmguar_enough_memory() function is called by kernel thread, it oopses
 
−
due to task_struct-&gt;mm equals NULL. Such situation was encountered when
 
−
aufs was over ramfs.
 
−
 
 
−
</div>
 
−
 
 
−
==== diff-cpt-make-zombie-20070413 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[CPT] alternative way to migrate zombie processes
 
−
 
 
−
In older 2.6.8 kernels do_exit() was very simple, essentially
 
−
it disposed m etc, which is done automatically while checkpointing,
 
−
and did some work on notifying parent. So that it was natural
 
−
to move restored process to zombie state by hands.
 
−
 
 
−
In 2.6.18 do_exit makes _lots_ of work.
 
−
 
 
−
Seems, it is easier to invert logic. We introduce new flag
 
−
PF_RESTART_EXIT, which suppresses the work which was already done,
 
−
when process at source hardware node moved to zombie state
 
−
(mostly, sending signals) and use do_exit() when restoring zombie processes.
 
−
 
 
−
Also, the same patch adds checks for a few of new things, which
 
−
cannot be migrated, it is related because the list of those things
 
−
obtained from do_exit().
 
−
</div>
 
−
 
 
−
==== diff-cpt-off-lockdep-on-sockets-20070413 ====
 
−
<div class="change">
 
−
 
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[CPT] Fix lockdep warning on socket dump
 
−
 
 
−
CPT locks all the sockets it finds for dumping.
 
−
This is OK, but lockdep thinks as if it were a circular locking.
 
−
 
 
−
It happens each time we migrate a VE with more than
 
−
one socked aboard.
 
−
 
 
−
Bug #79164.
 
−
 
 
−
</div>
 
−
 
 
−
==== diff-cpt-robust-list-20070413 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[CPT] checkpointing robust lists
 
−
 
 
−
Otherwise we are going to have problems with migration
 
−
of newer glibcs using robust lists when this is possible.
 
−
 
 
−
</div>
 
−
 
 
−
==== diff-fairsched-best-vcpu-20070413 ====
 
−
<div class="change">
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[SCHED] Reduce starvation of some VCPUs in case of cpu limits
 
−
 
 
−
Change logic of choosing best_vcpu to schedule to.
 
−
There are two potential problems:
 
−
 
−
a) if a vcpu is hot, and last used physical CPU of this vcpu is equal to
 
−
smp_processor_id() it will be always chosen. This is not a good
 
−
decision, because there is no guarantee, that _all_ physical CPU's must
 
−
take vcpu's from a vsched. For example, if cpulimit for a vsched is
 
−
small, this vsched can be run only on one physical CPU forever.
 
−
 
 
−
b) Also now newer 'cold' vcpu's are chosen first,
 
−
because we scan active_list in direct way,
 
−
i.e. from older vcpus to newer vcpus, and a newer one will be chosen finally.
 
−
In this case old vcpu's can starve for a long time
 
−
 
 
−
Bug #79015.
 
−
</div>
 
−
 
 
−
==== diff-fairsched-find-idle-vcpu-20070403 ====
 
−
<div class="change">
 
−
 
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[SCHED] find_idle_vcpu() mask check fix
 
−
 
 
−
In find_idle_vcpu() we skip VCPU's with ID's that is not
 
−
set in physical '*cpus' mask. It's incorrect.
 
−
We must skip VCPU's that has appropriate VCPU-&gt;last_pcpu
 
−
</div>
 
−
 
 
−
==== diff-ms-bridge-unaligned-access-20070411 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Evgeny Kravtsunov &lt;emkravts@openvz.org&gt;:<br/>
 
−
[BRIDGE] Unaligned access on IA64 when compare ether addr
 
−
 
 
−
Patch fixes unaligned access that takes place on ia64 in compare_ether_addr()
 
−
compare_ether_addr() requires address to be aligned on 2-byte boundary,
 
−
while addresses declared in bridges are aligned on 1-byte.
 
−
 
 
−
Bug #79001.<br/>
 
−
Bug #78983.
 
−
</div>
 
−
 
 
−
==== diff-ms-fatal-signal-20070413 ====
 
−
<div class="change">
 
−
Patch from Denis Lunev &lt;den@openvz.org&gt;:<br/>
 
−
[PATCH] Fatal signal processing logic
 
−
 
 
−
This patch changes fatal signal processing logic.
 
−
The SIGKILL should be raised only for all threads in the groups *EXCEPT* recipient one.
 
−
 
 
−
{{Bug|533}}.
 
−
</div>
 
−
 
 
−
==== diff-ms-stopmachine-msleep-20070411 ====
 
−
<div class="change">
 
−
Patch from Konstantin Khorenko &lt;khorenko@openvz.org&gt;:<br/>
 
−
problem found by Vasily (vvs@) &amp; Kirill (dev@):
 
−
 
 
−
a possible situation in stop_machine:
 
−
 
 
−
* stopmachine_state == STOPMACHINE_WAIT;
 
−
* STOPPER (stop_machine()) is in state SM_STOPPER_WAITING, calling yield() in a loop;
 
−
* SLAVES (stopmachine()) also call yield() in a loop.
 
−
 
 
−
This leads to the fairsched_lock suffering on all CPUs and in case of unfair
 
−
getting lock rules (for example on NUMA node), some CPUs can wait for the lock
 
−
forever/for a long time, causing a hang of the node.
 
−
This patch replaces yield() by msleep(10).
 
−
 
 
−
Mainstream kernel is affected as well, though it is harder to trigger:
 
−
One CPU does yield() taking/releasing rq-&gt;lock,
 
−
while another CPU should try to take that rq-&gt;lock (e.g. for balancing)
 
−
and will livelock forever.
 
−
 
 
−
Bug #78975.
 
−
</div>
 
−
 
 
−
==== diff-rh-dlm-misc-device-fix-20070412 ====
 
−
<div class="change">
 
−
Patch from Andrey Mirkin &lt;major@openvz.org&gt;, backported from -nmw git tree:
 
−
 
 
−
<pre>
 
−
From: Patrick Caulfield &lt;pcaulfie@redhat.com&gt;
 
−
Date: Wed, 21 Mar 2007 09:23:53 +0000 (+0000)
 
−
Subject: [DLM] Don't delete misc device if lockspace removal fails
 
−
X-Git-Url: http://git.kernel.org/?p=linux%2Fkernel%2Fgit%2Fsteve%2Fgfs2-2.6-nmw.git;a=commitdiff_plain;h=2ebea8b4cc0b8859a99d2006ec085c1da2d0758a;hp=ead2e9aa2a555bb1d94ff2e5b4c974d0e7cc6518
 
−
 
 
−
[DLM] Don't delete misc device if lockspace removal fails
 
−
 
 
−
Currently if the lockspace removal fails the misc device associated with a
 
−
lockspace is left deleted. After that there is no way to access the orphaned
 
−
lockspace from userland.
 
−
 
 
−
This patch recreates the misc device if th dlm_release_lockspace fails. I
 
−
believe this is better than attempting to remove the lockspace first because
 
−
that leaves an unattached device lying around. The potential gap in which there
 
−
is no access to the lockspace between removing the misc device and recreating it
 
−
is acceptable ... after all the application is trying to remove it, and only new
 
−
users of the lockspace will be affected.
 
−
 
 
−
Signed-Off-By: Patrick Caulfield &lt;pcaulfie@redhat.com&gt;
 
−
Signed-off-by: Steven Whitehouse &lt;swhiteho@redhat.com&gt;
 
−
</pre>
 
−
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=232878
 
−
 
 
−
Bug #79107.
 
−
</div>
 
−
 
 
−
==== diff-ubc-dcache-uncharge-root-20070413 ====
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:<br/>
 
−
[BC] uncharge fs root (/) from dcachesize
 
−
 
 
−
"/" dentry was charged in d_alloc_root(), then charged and uncharged
 
−
during filesystem activity. But at umount time that first charge was
 
−
forgotten.
 
−
 
 
−
So uncharge "/" by hand.
 
−
 
 
−
Bug #77771.
 
−
</div>
 
−
 
 
−
==== diff-ubc-ioacct-context-handle-20070413 ====
 
−
<div class="change">
 
−
 
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[IOACCT] Fix ioacct race
 
−
 
 
−
When page becomes dirty there's no time to store a context
 
−
on it - page may become clean immediately.
 
−
 
 
−
Thus we had a race in accounting when a page became clean
 
−
before we set a context on it and this context got lost and
 
−
not freed.
 
−
 
 
−
Handle the context the other way - in case we're going to
 
−
set a new context on a page that already has one - free it
 
−
and account written bytes in case the page became clean.
 
−
When removing a context from a page - handle the case when
 
−
a page does not have one due to the race in question. In
 
−
any case dirty page will have a context set, and a clean
 
−
one will have not.
 
−
 
−
Bug #79008.
 
−
</div>
 
−
 
 
−
==== diff-ubc-ioprio-clean-active-ub-20070411 ====
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[IOPRIO] cleaning active beancounter
 
−
 
 
−
After beancounter disappears, it still can be active.
 
−
Clean it up.
 
−
 
 
−
Bug #78875.
 
−
</div>
 
−
 
 
−
==== diff-ubc-percpu-sign-expansion-20070410 ====
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[BC] Percpu counters discrepancy on 64bit arches
 
−
 
 
−
Operation
 
−
long += -(unsigned int);
 
−
leads to wrong result on 64bit due to no sign extension.
 
−
 
 
−
Bug #78998.
 
−
</div>
 
−
 
 
−
==== diff-ve-futex-vpid-fix-20070413 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[PATCH] pthead_mutex_lock deadlock inside VE
 
−
 
 
−
This patch replaces diff-ve-futex-EDEADLK-bypass-20061225
 
−
and previous small patch reparing user-level deadlock in pthread_mutex_lock()
 
−
happening because value 0 was not "virtual".
 
−
 
 
−
It undoes unnatural tests for virtuality of pid supplied by user.
 
−
Naively, this can result in kernel warning if user mangles
 
−
pid doing something like:
 
−
<source lang="c"> 
 
−
syscall(__NR_futex, &amp;l, FUTEX_LOCK_PI, 0, 0);
 
−
l = getpid() | 1024;
 
−
syscall(__NR_futex, &amp;l, FUTEX_LOCK_PI, 0, 0);
 
−
</source>
 
−
 
 
−
This happens due to pid aliasing introduced with vpids:
 
−
test pid == virt_pid(current) is not enough to ensure that
 
−
pid does not correspond to current.
 
−
 
 
−
I find more natural to test for this condition directly.
 
−
</div>
 
−
 
 
−
==== diff-ve-kconfig-security-deps-20070411 ====
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[PATCH] kconfig: security depends on !ve
 
−
 
 
−
Many people have CONFIG_SECURITY enabled in their configs.
 
−
When they try to do `make oldconfig` for OpenVZ kernels with such
 
−
configs, no questions appear concerning CONFIG_VE and friends, and
 
−
people have OpenVZ kernels with virtualization features disabled.
 
−
Fix it. Reverse the dependency of VE/SECURITY.
 
−
</div>
 
−
 
 
−
==== diff-ve-venet-stoprace-20070412 ====
 
−
<div class="change">
 
−
Patch from Denis Lunev &lt;den@openvz.org&gt;:<br/>
 
−
[VENET] stop IP management before freeing venet
 
−
 
 
−
The device is freed before the VE&lt;-&gt;IP mapping is cleaned.
 
−
 
 
−
Bug #75502.
 
−
</div>
 
−
 
 
−
==== diff-vzdq-sleep-under-inode-lock-20070412 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:
 
−
 
 
−
The calltrace:
 
−
<source lang="c">
 
−
vzdq_aquotq_lookup
 
−
iget5_locked()
 
−
get_new_inode()
 
−
  `- spin_lock(&amp;inode_lock);
 
−
find_inode()
 
−
->set() /* == vzdq_aquotq_lookset */
 
−
vdq_aquot_lookset()
 
−
user_get_super()
 
−
  `- down_read(...)
 
−
</source>
 
−
 
 
−
So it may sleep with inode_lock taken.
 
−
Move all the sleeping operations out of the lock.
 
−
 
 
−
Bug #79124.
 
−
</div>
 
−
 
 
−
==== diff-i2o-timeout-20071013 ====
 
−
<div class="change">
 
−
Patch from Kostja:
 
−
 
 
−
Adds missed error check while trying to get message slot in I2OPASSTHRU ioctl
 
−
handler. This error caused kernel crash if one uses "raideng" (from raidutils)
 
−
when controller timeouts.
 
−
 
 
−
Bug #79279.
 
−
 
 
−
</div>
 
−
 
 
−
==== diff-ubc-io-release-debug-20070416 ====
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[IOACCT] Debug on page release
 
−
 
 
−
When releasing an IO beancounter from the page that
 
−
is not supposed to IO pb print a warning.
 
−
 
 
−
might help debug bug #79427.
 
−
</div>
 
−
 
 
−
==== diff-ubc-put-beancounters-in-fork-20070416 ====
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:<br/>
 
−
[BC] UB put on error path in fork()
 
−
 
 
−
If fork() fails after ub_task_charge(),
 
−
nobody is putting three beancounters getted there.
 
−
 
 
−
Bug #77231.
 
−
</div>
 
−
 
 
−
==== diff-arch-x86-ioremap-guard-page-fix-20070419 ====
 
−
<div class="change">
 
−
Patch from Dmitry Monakhov &lt;dmonakhov@openvz.org&gt;:<br/>
 
−
[4GB] change_page_attr() BUG's on strange pages
 
−
 
 
−
It is really long story...
 
−
 
 
−
this patch restore iounmap code as it was after:
 
−
<pre>
 
−
commit 2c692eefe4aff109eab9384b6d8c7e1a8f094dad
 
−
Author: ak &lt;ak&gt;
 
−
Date:  Sun Jan 23 18:29:11 2005 +0000
 
−
 
 
−
later this code was removed by this patch:
 
−
commit a7dd5a5f2b5db975bdf1dcaa3f3da6c289630076
 
−
Author: andrea &lt;andrea&gt;
 
−
 
 
−
Date:  Tue Mar 8 18:49:39 2005 +0000
 
−
 
 
−
this code was broken again after:
 
−
commit bf5421c309bb89e5106452bc840983b1b4754d61
 
−
Author: Andi Kleen &lt;ak@suse.de&gt;
 
−
Date:  Mon Dec 12 22:17:09 2005 -0800
 
−
</pre>
 
−
 
 
−
The problem is actually that get_vm_area() allocates
 
−
one more page then was requested for guard page.
 
−
But change_page_attr() doesn't actually takes this into account.
 
−
 
 
−
Bug #79617.
 
−
</div>
 
−
 
 
−
==== diff-cpt-active-callback-eagain-20070423 ====
 
−
<div class="change">
 
−
Patch from Andrey Mirkin &lt;major@openvz.org&gt;:<br/>
 
−
[CPT] retry checkpointing if VE has active netlink callback
 
−
 
 
−
Return -EAGAIN instead of -EBUSY if netlink socket has active callback.
 
−
In this case we will try to freeze VE 3 times.
 
−
</div>
 
−
 
 
−
==== diff-cpt-make-zombie-b-20070420 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Thorsten Schifferdecker:<br/>
 
−
[CPT] tux is missing on vanilla kernels
 
−
 
 
−
Compilation bug was introduced by
 
−
[http://git.openvz.org/?p=linux-2.6.18-openvz;a=commitdiff;h=f90c2c318467829ecde43919ad38f326527f533b http://git.openvz.org/?p=linux-2.6.18-openvz;a=commitdiff;h=f90c2c318467829ecde43919ad38f326527f533b]
 
−
 
 
−
Fixes {{bug|545}}.
 
−
</div>
 
−
 
 
−
==== diff-cpt-restore-netlink-socket-20070419 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[CPT] restore rcv queue on netlink sockets and unbound netlink sockets
 
−
 
 
−
Code restoring queues was forgotten. This fixes bug #79723.
 
−
 
 
−
Unbound sockets were restored incorrectly, they were outbound
 
−
to some post, which prevented subsequent bind by application.
 
−
This fixes bug #79724.
 
−
 
−
The patch overrides previous patch with subj:
 
−
"[CPT] restore rcv queue on netlink sockets", which fixed only bug #79723.
 
−
</div>
 
−
 
 
−
==== diff-cpt-setup-pagein-fix-20070418 ====
 
−
<div class="change">
 
−
Patch from Andrey Mirkin &lt;major@openvz.org&gt;:<br/>
 
−
[CPT] Fix lazy migration
 
−
 
 
−
Accidentally during porting checkpointing on 2.6.18 kernel rst_setup_pagein()
 
−
function was added to code twice. This breaks lazy migration.
 
−
 
 
−
Bug #77921.
 
−
 
 
−
Creation procedure of pgin block device was quite hairy, so almost every time
 
−
during lazy migration we can see annoying message in kernel log:
 
−
register_blkdev: cannot get major 254 for pgin
 
−
 
 
−
In this patch creation of pgin block device is fixed.
 
−
</div>
 
−
 
 
−
==== diff-cpt-ubc-off-20070424 ====
 
−
<div class="change">
 
−
 
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[BC] compilation fix with CONFIG_USER_RESOURCE=n
 
−
 
 
−
Compilation fix with CONFIG_USER_RESOURCE=n
 
−
</div>
 
−
 
 
−
==== diff-cpt-zombie-threads-20070416 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[CPT] thread groups with exited leader did not migrate
 
−
 
 
−
The bug is simple and stupid, it is very strange nobody saw this.
 
−
 
 
−
When thread group leader exits, its mm/files/fs/namespace are released
 
−
but zombie process remain frozen until all the threads exit.
 
−
Restore was not able to restore such configuration.
 
−
 
 
−
Solution is simple: when chackpointing save not real (NULL)
 
−
mm/files/fs/namespace, but mm/files/fs/namespace of this thread group.
 
−
</div>
 
−
 
 
−
==== diff-fairsched-best-vcpu-b-20070423 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[SCHED] Select some vcpu instead of idle even if all vcpus are hot
 
−
 
 
−
We have to use oldest vcpu if all vcpu's are hot.
 
−
In current kernel an idle_vcpu is used and CPU can idle instead of
 
−
doing some job.
 
−
 
 
−
Bug #79676.
 
−
 
 
−
</div>
 
−
==== diff-fairsched-unused-rq-20070424 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[SCHED] cleanup: removed unused variable
 
−
 
 
−
struct rq *rq was used before fairsched patch to compare
 
−
tasks with rq-&gt;idle. With fairsched idle task is bound to
 
−
pcpu, not vcpu and thus struct rq *rq is simply not needed.
 
−
</div>
 
−
 
 
−
==== diff-ms-cfq-allow-merge-b-20070424 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[PATCH] modification of allow merge policy in cfq (mainstream)
 
−
 
 
−
Jens Axboe rewrited allow merge policy one more time after
 
−
we reported the problem and fixed the problem we face curently when some
 
−
tasks experience I/O starvations.
 
−
 
 
−
This is an incremental patch to the previous patch
 
−
diff-ms-cfq-allow-merge-20070117 ported to OpenVZ.
 
−
 
 
−
This patch is cumulative of
 
−
 
 
−
<ul>
 
−
* [http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=719d34027e1a186e46a3952e8a24bf91ecc33837 719d34027e1a186e46a3952e8a24bf91ecc33837]
 
−
 
 
−
* [http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=ec8acb6904fabb8e741f741ec99bb1c18f2b3dee ec8acb6904fabb8e741f741ec99bb1c18f2b3dee]
 
−
</ul>
 
−
 
 
−
Bug #79594.
 
−
 
 
−
</div>
 
−
==== diff-ms-cpufreq-centrino-20070426 ====
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:<br/>
 
−
 
 
−
[PATCH] Make speedstep centrino cpufreq driver use wr/rdmsr_on_cpu()
 
−
 
 
−
speedstep-centrino cpufreq driver was using set_cpus_allowed() and
 
−
checks for smp_processor_id() to confine itself to given CPU.
 
−
 
 
−
Switch to rdmsr_on_cpu/wrmsr_on_cpu() infrastructure.
 
−
 
 
−
Closes http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=420708
 
−
</div>
 
−
 
 
−
==== diff-ms-elf-retval-20070420 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[PATCH] Invalid return value of execve() resulting in oopses (mainstream)
 
−
 
 
−
When elf loader fails to map executable (due to memory shortage
 
−
or because binary is malformed), it can return 0. Normally, this is invisible
 
−
because process is killed with SIGKILL and it never returns to user space.
 
−
 
 
−
But if exec() is called from kernel thread (hotplug, whatever) consequences
 
−
are more interesting and vary depending on architecture.
 
−
 
 
−
i386. Nothing especially interesting, execve() just returns with "success"  :-)
 
−
 
 
−
x86_64. Fake zero frame is used on way to caller, RSP/RIP are loaded
 
−
with zeros, ergo... double fault.
 
−
 
 
−
ia64. Similar to i386, but r32...r95 are corrupted. Sometimes it oopses
 
−
due to return to zero PC, sometimes it sees NaT in rXX and oopses
 
−
due to NaT consumption.
 
−
 
 
−
This fix solves bugs #68582 (i386), #73753 (x86_64) and #79847 (ia64).
 
−
</div>
 
−
 
 
−
==== diff-ms-fib-netlink-lookup-recursion-20070425 ====
 
−
<div class="change">
 
−
Patch from Alexey Kuznetsov &lt;alexey@openvz.org&gt;:<br/>
 
−
[PATCH] stack overflow in netlink (mainstream)
 
−
 
 
−
Reply to NETLINK_FIB_LOOKUP messages were misrouted back to kernel,
 
−
which resulted in infinite recursion and stack overflow.
 
−
 
 
−
The bug is present in all kernel versions since the feature appeared.
 
−
 
 
−
(linux 2.6.13, Jun 20th, 2005,<br/>
 
−
commit 246955fe4c38bd706ae30e37c64892c94213775d,<br/>
 
−
[NETLINK]: fib_lookup() via netlink)
 
−
 
 
−
The patch also makes some minimal cleanup:
 
−
# Return something consistent (-ENOENT) when fib table is missing
 
−
# Do not crash when queue is empty (does not happen, but yet)
 
−
# Put result of lookup
 
−
 
 
−
Franlky, I would delete this thing instead of fixing. It looks ugly
 
−
and was used only for debugging LC-trie.
 
−
 
 
−
Signed-off-by: Alexey Kuznetsov &lt;kuznet@ms2.inr.ac.ru&gt;<br/>
 
−
Acked-by: Dave Miller &lt;davem@davemloft.net&gt;
 
−
</div>
 
−
 
 
−
==== diff-ms-fib-netlink-lookup-recursion-b-20070425 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Sergey Vlasov &lt;vsu@altlinux.ru&gt;:<br/>
 
−
[NETLINK] Fix for Alexey's netlink lookup recursion fix
 
−
 
 
−
When CONFIG_IP_MULTIPLE_TABLES is enabled, the code in nl_fib_lookup()
 
−
needs to initialize the res.r field before fib_res_put(&amp;res) - unlike
 
−
fib_lookup(), a direct call to -&gt;tb_lookup does not set this field.
 
−
</div>
 
−
 
 
−
==== diff-ms-loop-dont-complete-lo-bh-done-20070418 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:<br/>
 
−
[PATCH] loopback: oops on loopback mount/umount (mainstream)
 
−
 
 
−
After LOOP_SET_FD/LOOP_CLR_FD combo loop device's queue gets request
 
−
handler which is persistent.
 
−
 
 
−
After, say
 
−
mount -t iso9660 /dev/loop0 /mnt # sic
 
−
this request handler is called directly with<br/>
 
−
a) -&gt;lo_state being Lo_unbound<br/>
 
−
b) -&gt;lo_pending being zero
 
−
 
 
−
Error path in loop_make_request() completes -&gt;lo_bh_done completion
 
−
which is persistent as well.
 
−
 
 
−
Now, let's start worker thread as usual. It'll set -&gt;lo_pending to 1,
 
−
don't wait for completion because it was already completed (brokenly),
 
−
and will not get out of infinite loop because of -&gt;lo_pending. Loop
 
−
device doesn't have bios at this point and triggers BUG_ON.
 
−
 
 
−
So, don't complete -&gt;lo_bh_done when loop device isn't setup fully.
 
−
In mainstream it was accidentaly fixed when converting to kthreads.
 
−
 
 
−
Bug #79521.
 
−
</div>
 
−
 
 
−
==== diff-ubc-ioprio-force-disp-off-20070424 ====
 
−
 
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[IOPRIO] forced dispatching when CONFIG_UBC_IO_PRIO off
 
−
 
 
−
If CONFIG_UBC_IO_PRIO is off, then no beancounters are in active list,
 
−
consequently we have a bug in forced dispatching case.
 
−
 
 
−
{{Bug|528}}.
 
−
</div>
 
−
 
 
−
==== diff-ubc-ioprio-oops-on-virt-off-20070423 ====
 
−
<div class="change">
 
−
Patch from Vasily Tarasov &lt;vtaras@openvz.org&gt;:<br/>
 
−
[IOPRIO] Oops on IO-prioritization disabling
 
−
 
 
−
If io-prioritization is suddenly turned off via
 
−
/sys/block/&lt;dev&gt;/queue/iosched/virt_mode cfqq owner BC does not equal
 
−
current io context. It is right to get beancounter obtained from queue,
 
−
but not from current IO context.
 
−
</div>
 
−
 
 
−
==== diff-ubc-off-20070424 ====
 
−
<div class="change">
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[BC] compilation fix with CONFIG_USER_RESOURCE=n
 
−
 
 
−
Compilation fix with CONFIG_USER_RESOURCE=n
 
−
</div>
 
−
 
 
−
==== diff-ubc-unusedprivvm-in-zeromap-20070424 ====
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[BC] Leak of privvmpages on zero page maps
 
−
 
 
−
This was obviously forgotten.
 
−
Each mmap of /dev/zero causes this leak.
 
−
 
 
−
Bug #80246.
 
−
</div>
 
−
 
 
−
==== diff-ve-meminfo-b-20070330 ====
 
−
<div class="change">
 
−
Patch from Konstantin Khorenko &lt;khorenko@openvz.org&gt;:<br/>
 
−
[MEMINFO] sysctl for selecting UsedMem source
 
−
 
 
−
Adds sysctl to choose base ubc parameter for memory usage inside a VE.
 
−
Sets PRIVVMPAGES beancounter to be used by default instead of OOMGUARPAGES.
 
−
 
 
−
Bug #78088.
 
−
</div>
 
−
 
 
−
==== diff-ve-netlink-veprintk-20070426 ====
 
−
<div class="change">
 
−
Patch from Pavel Emelianov &lt;xemul@openvz.org&gt;:<br/>
 
−
[NETLINK] VE netlink message should go into VE log
 
−
 
 
−
When parsing netlink arguments kernel may printk that
 
−
some bytes left unparsed. Make this info appear in VE log,
 
−
instead of global one.
 
−
</div>
 
−
 
 
−
==== diff-ve-tun-persist-20070424 ====
 
−
<div class="change">
 
−
Patch from Vasily Averin &lt;vvs@openvz.org&gt;:<br/>
 
−
[TUN] prohibit tun persistent mode inside VE
 
−
 
 
−
Prohibit tun persistent mode inside VE, until resolved via ve hooks.
 
−
 
 
−
Bug #79612.
 
−
</div>
 
−
 
 
−
==== diff-ve-xen-netback-20070426 ====
 
−
<div class="change">
 
−
Patch from Sergey Korshunoff &lt;seyko2@&gt;:
 
−
 
 
−
Fix Xen netback driver, since loopback_dev is defined as a macro
 
−
in OVZ and is substituted.
 
−
</div>
 
−
 
 
−
==== diff-vzdq-ugbad-20070428 ====
 
−
<div class="change">
 
−
Patch from Konstantin Khorenko &lt;khorenko@openvz.org&gt;:<br/>
 
−
[VZDQ] prohibit chown of a file if owner doesn't have ugid struct
 
−
 
 
−
Prohibit chown a file if its owner does not have
 
−
ugid record. This might happen if we somehow exceeded
 
−
the UID/GID (e.g. set ugidlimit less than number of users).
 
−
 
 
−
Bug #79553.
 
−
</div>
 
−
 
 
−
==== diff-vzstat-numa-fixes-b-20070416 ====
 
−
<div class="change">
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
[VZSTAT] Sum up per node stats for pgdat's
 
−
 
 
−
cat /proc/vz/stats in 2.6.18 shows information about zones
 
−
for each node, but vzstat utility parse information about one (last)
 
−
node only, and shows incorrect memory info on hosts with several nodes
 
−
(numa hosts for instance). New code scans all nodes and summarize
 
−
statistics for identical zones per each node.
 
−
 
 
−
Bug #77994.
 
−
</div>
 
−
 
 
−
==== diff-vzstat-zone-id-20070416 ====
 
−
<div class="change">
 
−
Patch from Alexandr Andreev &lt;aandreev@openvz.org&gt;:<br/>
 
−
 
 
−
[VZSTAT] Fix vzstat when DMA32 zone on x8664 (index 1) is empty
 
−
 
 
−
Fix vzstat when DMA32 zone on x8664 (index 1) is empty.
 
−
For this, show ordinal number of a zone instead of it's real
 
−
index in kernel 'zones' array.
 
−
So the output will look like:
 
−
<pre>
 
−
0 DMA
 
−
skipped DMA32 zone
 
−
1 Normal
 
−
</pre>
 
−
 
 
−
Bug #77336.
 
−
</div>
 
−
 
 
−
==== io-accounting-menuconfig.patch ====
 
−
<div class="change">
 
−
Patch from Alexey Dobriyan &lt;adobriyan@openvz.org&gt;:
 
−
 
 
−
1. Move TASK_IO_ACCOUNTING out of EMBEDDED menu cf placement in mainline.
 
−
 
 
−
2. As side effect, EMBEDDED menu will be shown on same level as designed,
 
−
not returning to top level after SYSCTL option.
 
−
 
 
−
{{Bug|550}}.
 
−
</div>
 
−
 
 
−
==== namespaces-utsname-xen.patch ====
 
−
<div class="change">
 
−
Patch from Sergey Korshunoff &lt;seyko2@&gt;:
 
−
 
 
−
Fix utsname handling in process-xen.c, which is a copy of process.c
 
−
</div>
 
−
 
 
−
==== diff-i2o-msgleak-10070423 ====
 
−
<div class="change">
 
−
Patch from Vasiliy:
 
−
 
 
−
This patch fixes i2o message leak.
 
−
We need to free msg itself and i2o message in hw in case
 
−
of error.
 
−
</div>
 
−
 
 
−
==== diff-i2o-msgget-errh-10070423 ====
 
−
<div class="change">
 
−
Patch from Vasiliy:
 
−
 
 
−
This patch fixes access to memory that has not been allocated:
 
−
i2o_msg_get_wait() can returns errors different from I2O_QUEUE_EMPTY.
 
−
But the result is checked only against this code.
 
−
if it is not I2O_QUEUE_EMPTY then we dereference the error code as the pointer later.
 
−
</div>
 
−
 
 
−
==== diff-i2o-cfg-passthru-20070423 ====
 
−
<div class="change">
 
−
Patch from Vasiliy:
 
−
 
 
−
This patch fixes a number of issues in i2o_cfg_passthru{,32}:
 
−
 
 
−
* memory leaks (including i2o_message leak fixed by khorenko@sw.ru)
 
−
* infinite loop to sg_list_cleanup in passthru32
 
−
* bad error paths
 
−
</div>
 
−
 
 
−
==== diff-i2o-proc-perms-20060304 ====
 
−
<div class="change">
 
−
Patch from Vasiliy:
 
−
 
 
−
Reading from some i2o related proc files can lead to the controller hang due
 
−
unknown reasons. As a workaround this patch changes the permission of these
 
−
files to root-only accessible.
 
−
</div>
 
−
 
 
−
==== linux-2.6.18-drbd-8.0.0-8.0.2.patch ====
 
−
<div class="change">
 
−
Drbd update to v8.0.2
 
−
</div>
 
−
 
 
−
</noinclude>
 

Revision as of 15:57, 22 October 2009

The fact that teachers need better training to carry out deliberate instruction in reading, spelling, and writing should prompt action rather than criticism. ,