[BC] Allocate percpu stats via alloc_percpuThis reduces the amount of memory used by UB on strong SMP systems
like IA64 where NUM_CPUS is set by default to 1024
[CPT][IA64] pause() could hang foreverIA64 does not have sys_pause(), but simulates it with sys_sigsuspend()
with non-sensial sigmask == current sigmask. We did not make special
workaround preventing false restart for this case, hence pause()
could hang forever.
It would not be so bad idea to add the same workaround for sigsuspend()
on all the arch's even if pause() is present. It will be done after testing.
The fact is that sigs...
[CPT] sigsuspend could hang forever after restoreDo not restart syscalls with TIF_RESTORE_SIGMASK in cpt.
It was severe bug. First, we do not need to restart such syscalls,
they are restarted by core on exit from syscall. Second, it was wrong
to restart syscall but do not clear TIF_RESTORE_SIGMASK and do not
restore mask. If some signal happens here, it will be delivered,
but syscall is restarted and sigsusend() will not exit hanging forever...
igmp: handle GFP_ATOMIC allocation failuresadd_grhead() in IGMP code does GFP_ATOMIC allocations. At least two
places, doesn't check for NULL skb and happily feed it into skb_put
for dereference. With help from David Stevens.
Assign predefined priorities to ve start/stop hooksThis at least solves the problem found by Denis:
nfs must be started after and stopped before venet.
Simplify lutime(2)sys_lutime() can be made mush simpler by reusing code in
do_utimes(). This patch mirrors lutimesat(2) series sent
to mainline. After and if lutimesat(2) will be merged,
lutime can de dropped and lutimesat(2) will be used.
Fix /proc/<pid>/stat's VE and BC info.All numbers dumped should be in fixed positions, but
2.6.18 kernels have one extra field vs 2.6.9 - delayacct.
So shift VZ info one field left.
[CPT] sync migration to new MM conventions in 2.6.18The changes are about:
1. mm->page_table_lock has been split (most of the patch);
2. mmlist requires special handling, it is updated only when mm
gets some swap pages;
3. use vm_normal_page() instead of raw macros;
4. use new macros to walk page tables.
(5) (pseudo fix) walking task list is wrong. It will compile
and most likely work, but we remember it is broken
[NETFILTER] iptables sockopt checksIf ip{6}_tables is loaded after VE start or prohibitted for this VE, than
VE user may oops kernel by any iptables command. This is due to
nonvirtualized sockopts, which are registered on the module load and
available for all VEs.
Added additional checks in order to avoid this.
Core-dumping unreadable binaries via PT_INTERP (CVE-2004-1073)Proposed patch to fix #5 in
http://www.isec.pl/vulnerabilities/isec-0017-binfmt_elf.txt
aka
http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2004-1073
To reproduce, do
* grab poc at the end of advisory.
* add line "eph.p_memsz = 4096;" after "eph.p_filesz = 4096;"
where first "4096" is something equal to or greater than 4096.
* ./poc /usr/bin/sudo && ls -l
Here I get:
-rw------- 1 ad a...
Fix virt pids hash manipulationsvpids are inserted and removed from vpid_hash under different locks.
This patch:
- fixes locking
- relaxes locking on vpid<->pid conversions
- optimizes find_task_by_pid_ve
- fixes types (int -> pid_t)
[BC] Oomkiller fixes merged from stable branch - small API cleanups in OOM
- 5 sec waiting in out_of_memory removed (done in ub_oom_lock)
- deadlock on cpusets mutex fixed
[BC] Save and restore interrupts in delayed_put_beancounter()Otherwise the following can happen:
/* in process context */
...
kmem_cache_free()
`- local_irq_save()
ub_slab_uncharge()
put_beancounter()
release_beancounter()
if (!in_interrupt())
/* true since we're in process context
* though with IRQs disabled
*/
delayed_release_beancounter()
`- local_irq_disable()
...
local_irq_enable()
/* BUG! We've enabled interrupts
...
Venet module unload cleanupNew venet refcounting scheme:
- veip attached to a VE holds the module
- venet device in VE0 holds the module
- veip_stop/veip_start is safe to call module_get/put, as they are
called from hook with ownership specified
[PATCH] rewrite lchmod(2) via lchmodat2.6.16 got new system call which was almost exactly was needed for
implementing lchown(2):
sys_fchmodat(int dfd, const char __user *filename, mode_t mode)
Unfortunately, Ulrich haven't supplied one more argument "int flags", so
it'd be possible to pass AT_SYMLINK_NOFOLLOW and get desired behaviour.
lchmod(2) could be just
sys_fchmodat(AT_FDCWD, filename, mode, AT_SYMLINK_NOFOLLOW)
Patch ...
ext[23]-fs warns when acl is not supportedWhe the kernel is build without ACL support incorrect permissions
management takes place on partition with already setted acl flag.
Make ext[23] to check if acl flag is set by default and print warning
in case when kernel booted was build without acl support.
http://bugzilla.openvz.org/show_bug.cgi?id=439
[SIMFS] Fix statfs behaviour over reiserfs and no-quota case1. Update for sim_statfs: when DISK_QUOTA is switched off
in /etc/vz/vz.config , sim_statfs takes kstatfs values using statfs function
from super operations. If reiserfs partition is used for /vz we obtain a bug.
Because reiserfs do not initialize f_ffree (free inodes) and f_files(number
of current allocated inodes) fields of kstatfs. Patch solves the problem by
initializing str...
[BC] Handle exec_ub == NULL case in IO acct.Some code may set exec_ub to indicate that resource
charging is not required, but IO should be accounted.
Since this may (should) happen in task context only
it's safe to use task_ub here.
[VZQUOTA] add virtinfo notifier call to disable vzquota on an inodeComparing to previous version:
1. vzquota_dentry_off() is replaced with vzquota_inode_off()
2. all the operations with S_NOQUOTA are moved under inode_qmblk_lock().
A few of exceptions (in standard dquot.h header) rely on the fact,
that S_NOQUOTA is never cleared.
3. Two patches are merged together, because S_NOQUOTA handling
essentially trivialized.
This allows to deaccount some ino...
OpenVZ team
