| Age | Commit message (Collapse) | Author |
|---|
|
The atomic replace allows to create cumulative patches. They are useful when
you maintain many livepatches and want to remove one that is lower on the
stack. In addition it is very useful when more patches touch the same function
and there are dependencies between them.
It's also a feature some of the distros are using already to distribute
their patches.
|
|
There are already macros to iterate over struct klp_func and klp_object.
Add also klp_for_each_patch(). But make it internal because also
klp_patches list is internal.
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Joe Lawrence <joe.lawrence@redhat.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
|
|
The fake signal is send automatically now. We can rely on it completely
and remove the sysfs attribute.
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
An administrator may send a fake signal to all remaining blocking tasks
of a running transition by writing to
/sys/kernel/livepatch/<patch>/signal attribute. Let's do it
automatically after 15 seconds. The timeout is chosen deliberately. It
gives the tasks enough time to transition themselves.
Theoretically, sending it once should be more than enough. However,
every task must get outside of a patched function to be successfully
transitioned. It could prove not to be simple and resending could be
helpful in that case.
A new workqueue job could be a cleaner solution to achieve it, but it
could also introduce deadlocks and cause more headaches with
synchronization and cancelling.
[jkosina@suse.cz: removed added newline]
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Replaced patches are removed from the stack when the transition is
finished. It means that Nop structures will never be needed again
and can be removed. Why should we care?
+ Nop structures give the impression that the function is patched
even though the ftrace handler has no effect.
+ Ftrace handlers do not come for free. They cause slowdown that might
be visible in some workloads. The ftrace-related slowdown might
actually be the reason why the function is no longer patched in
the new cumulative patch. One would expect that cumulative patch
would help solve these problems as well.
+ Cumulative patches are supposed to replace any earlier version of
the patch. The amount of NOPs depends on which version was replaced.
This multiplies the amount of scenarios that might happen.
One might say that NOPs are innocent. But there are even optimized
NOP instructions for different processors, for example, see
arch/x86/kernel/alternative.c. And klp_ftrace_handler() is much
more complicated.
+ It sounds natural to clean up a mess that is no longer needed.
It could only be worse if we do not do it.
This patch allows to unpatch and free the dynamic structures independently
when the transition finishes.
The free part is a bit tricky because kobject free callbacks are called
asynchronously. We could not wait for them easily. Fortunately, we do
not have to. Any further access can be avoided by removing them from
the dynamic lists.
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Sometimes we would like to revert a particular fix. Currently, this
is not easy because we want to keep all other fixes active and we
could revert only the last applied patch.
One solution would be to apply new patch that implemented all
the reverted functions like in the original code. It would work
as expected but there will be unnecessary redirections. In addition,
it would also require knowing which functions need to be reverted at
build time.
Another problem is when there are many patches that touch the same
functions. There might be dependencies between patches that are
not enforced on the kernel side. Also it might be pretty hard to
actually prepare the patch and ensure compatibility with the other
patches.
Atomic replace && cumulative patches:
A better solution would be to create cumulative patch and say that
it replaces all older ones.
This patch adds a new "replace" flag to struct klp_patch. When it is
enabled, a set of 'nop' klp_func will be dynamically created for all
functions that are already being patched but that will no longer be
modified by the new patch. They are used as a new target during
the patch transition.
The idea is to handle Nops' structures like the static ones. When
the dynamic structures are allocated, we initialize all values that
are normally statically defined.
The only exception is "new_func" in struct klp_func. It has to point
to the original function and the address is known only when the object
(module) is loaded. Note that we really need to set it. The address is
used, for example, in klp_check_stack_func().
Nevertheless we still need to distinguish the dynamically allocated
structures in some operations. For this, we add "nop" flag into
struct klp_func and "dynamic" flag into struct klp_object. They
need special handling in the following situations:
+ The structures are added into the lists of objects and functions
immediately. In fact, the lists were created for this purpose.
+ The address of the original function is known only when the patched
object (module) is loaded. Therefore it is copied later in
klp_init_object_loaded().
+ The ftrace handler must not set PC to func->new_func. It would cause
infinite loop because the address points back to the beginning of
the original function.
+ The various free() functions must free the structure itself.
Note that other ways to detect the dynamic structures are not considered
safe. For example, even the statically defined struct klp_object might
include empty funcs array. It might be there just to run some callbacks.
Also note that the safe iterator must be used in the free() functions.
Otherwise already freed structures might get accessed.
Special callbacks handling:
The callbacks from the replaced patches are _not_ called by intention.
It would be pretty hard to define a reasonable semantic and implement it.
It might even be counter-productive. The new patch is cumulative. It is
supposed to include most of the changes from older patches. In most cases,
it will not want to call pre_unpatch() post_unpatch() callbacks from
the replaced patches. It would disable/break things for no good reasons.
Also it should be easier to handle various scenarios in a single script
in the new patch than think about interactions caused by running many
scripts from older patches. Not to say that the old scripts even would
not expect to be called in this situation.
Removing replaced patches:
One nice effect of the cumulative patches is that the code from the
older patches is no longer used. Therefore the replaced patches can
be removed. It has several advantages:
+ Nops' structs will no longer be necessary and might be removed.
This would save memory, restore performance (no ftrace handler),
allow clear view on what is really patched.
+ Disabling the patch will cause using the original code everywhere.
Therefore the livepatch callbacks could handle only one scenario.
Note that the complication is already complex enough when the patch
gets enabled. It is currently solved by calling callbacks only from
the new cumulative patch.
+ The state is clean in both the sysfs interface and lsmod. The modules
with the replaced livepatches might even get removed from the system.
Some people actually expected this behavior from the beginning. After all
a cumulative patch is supposed to "completely" replace an existing one.
It is like when a new version of an application replaces an older one.
This patch does the first step. It removes the replaced patches from
the list of patches. It is safe. The consistency model ensures that
they are no longer used. By other words, each process works only with
the structures from klp_transition_patch.
The removal is done by a special function. It combines actions done by
__disable_patch() and klp_complete_transition(). But it is a fast
track without all the transaction-related stuff.
Signed-off-by: Jason Baron <jbaron@akamai.com>
[pmladek@suse.com: Split, reuse existing code, simplified]
Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Jessica Yu <jeyu@kernel.org>
Cc: Jiri Kosina <jikos@kernel.org>
Cc: Miroslav Benes <mbenes@suse.cz>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
The possibility to re-enable a registered patch was useful for immediate
patches where the livepatch module had to stay until the system reboot.
The improved consistency model allows to achieve the same result by
unloading and loading the livepatch module again.
Also we are going to add a feature called atomic replace. It will allow
to create a patch that would replace all already registered patches.
The aim is to handle dependent patches more securely. It will obsolete
the stack of patches that helped to handle the dependencies so far.
Then it might be unclear when a cumulative patch re-enabling is safe.
It would be complicated to support the many modes. Instead we could
actually make the API and code easier to understand.
Therefore, remove the two step public API. All the checks and init calls
are moved from klp_register_patch() to klp_enabled_patch(). Also the patch
is automatically freed, including the sysfs interface when the transition
to the disabled state is completed.
As a result, there is never a disabled patch on the top of the stack.
Therefore we do not need to check the stack in __klp_enable_patch().
And we could simplify the check in __klp_disable_patch().
Also the API and logic is much easier. It is enough to call
klp_enable_patch() in module_init() call. The patch can be disabled
by writing '0' into /sys/kernel/livepatch/<patch>/enabled. Then the module
can be removed once the transition finishes and sysfs interface is freed.
The only problem is how to free the structures and kobjects safely.
The operation is triggered from the sysfs interface. We could not put
the related kobject from there because it would cause lock inversion
between klp_mutex and kernfs locks, see kn->count lockdep map.
Therefore, offload the free task to a workqueue. It is perfectly fine:
+ The patch can no longer be used in the livepatch operations.
+ The module could not be removed until the free operation finishes
and module_put() is called.
+ The operation is asynchronous already when the first
klp_try_complete_transition() fails and another call
is queued with a delay.
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
module_put() is currently never called in klp_complete_transition() when
klp_force is set. As a result, we might keep the reference count even when
klp_enable_patch() fails and klp_cancel_transition() is called.
This might give the impression that a module might get blocked in some
strange init state. Fortunately, it is not the case. The reference count
is ignored when mod->init fails and erroneous modules are always removed.
Anyway, this might be confusing. Instead, this patch moves
the global klp_forced flag into struct klp_patch. As a result,
we block only modules that might still be in use after a forced
transition. Newly loaded livepatches might be eventually completely
removed later.
It is not a big deal. But the code is at least consistent with
the reality.
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
The address of the to be patched function and new function is stored
in struct klp_func as:
void *new_func;
unsigned long old_addr;
The different naming scheme and type are derived from the way
the addresses are set. @old_addr is assigned at runtime using
kallsyms-based search. @new_func is statically initialized,
for example:
static struct klp_func funcs[] = {
{
.old_name = "cmdline_proc_show",
.new_func = livepatch_cmdline_proc_show,
}, { }
};
This patch changes unsigned long old_addr -> void *old_func. It removes
some confusion when these address are later used in the code. It is
motivated by a followup patch that adds special NOP struct klp_func
where we want to assign func->new_func = func->old_addr respectively
func->new_func = func->old_func.
This patch does not modify the existing behavior.
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Alice Ferrazzi <alice.ferrazzi@gmail.com>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Now that synchronize_rcu() waits for preempt-disable regions of code
as well as RCU read-side critical sections, synchronize_sched() can be
replaced by synchronize_rcu(). This commit therefore makes this change,
even though it is but a comment.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
|
|
Support for immediate flag was removed by commit d0807da78e11
("livepatch: Remove immediate feature"). We bail out during
patch registration for architectures, those don't support
reliable stack trace. Remove the check in klp_try_switch_task(),
as its not required.
Signed-off-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Immediate flag has been used to disable per-task consistency and patch
all tasks immediately. It could be useful if the patch doesn't change any
function or data semantics.
However, it causes problems on its own. The consistency problem is
currently broken with respect to immediate patches.
func a
patches 1i
2i
3
When the patch 3 is applied, only 2i function is checked (by stack
checking facility). There might be a task sleeping in 1i though. Such
task is migrated to 3, because we do not check 1i in
klp_check_stack_func() at all.
Coming atomic replace feature would be easier to implement and more
reliable without immediate.
Thus, remove immediate feature completely and save us from the problems.
Note that force feature has the similar problem. However it is
considered as a last resort. If used, administrator should not apply any
new live patches and should plan for reboot into an updated kernel.
The architectures would now need to provide HAVE_RELIABLE_STACKTRACE to
fully support livepatch.
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
If a task sleeps in a set of patched functions uninterruptedly, it could
block the whole transition indefinitely. Thus it may be useful to clear
its TIF_PATCH_PENDING to allow the process to finish.
Admin can do that now by writing to force sysfs attribute in livepatch
sysfs directory. TIF_PATCH_PENDING is then cleared for all tasks and the
transition can finish successfully.
Important note! Administrator should not use this feature without a
clearance from a patch distributor. It must be checked that by doing so
the consistency model guarantees are not violated. Removal (rmmod) of
patch modules is permanently disabled when the feature is used. It
cannot be guaranteed there is no task sleeping in such module.
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Live patching consistency model is of LEAVE_PATCHED_SET and
SWITCH_THREAD. This means that all tasks in the system have to be marked
one by one as safe to call a new patched function. Safe means when a
task is not (sleeping) in a set of patched functions. That is, no
patched function is on the task's stack. Another clearly safe place is
the boundary between kernel and userspace. The patching waits for all
tasks to get outside of the patched set or to cross the boundary. The
transition is completed afterwards.
The problem is that a task can block the transition for quite a long
time, if not forever. It could sleep in a set of patched functions, for
example. Luckily we can force the task to leave the set by sending it a
fake signal, that is a signal with no data in signal pending structures
(no handler, no sign of proper signal delivered). Suspend/freezer use
this to freeze the tasks as well. The task gets TIF_SIGPENDING set and
is woken up (if it has been sleeping in the kernel before) or kicked by
rescheduling IPI (if it was running on other CPU). This causes the task
to go to kernel/userspace boundary where the signal would be handled and
the task would be marked as safe in terms of live patching.
There are tasks which are not affected by this technique though. The
fake signal is not sent to kthreads. They should be handled differently.
They can be woken up so they leave the patched set and their
TIF_PATCH_PENDING can be cleared thanks to stack checking.
For the sake of completeness, if the task is in TASK_RUNNING state but
not currently running on some CPU it doesn't get the IPI, but it would
eventually handle the signal anyway. Second, if the task runs in the
kernel (in TASK_RUNNING state) it gets the IPI, but the signal is not
handled on return from the interrupt. It would be handled on return to
the userspace in the future when the fake signal is sent again. Stack
checking deals with these cases in a better way.
If the task was sleeping in a syscall it would be woken by our fake
signal, it would check if TIF_SIGPENDING is set (by calling
signal_pending() predicate) and return ERESTART* or EINTR. Syscalls with
ERESTART* return values are restarted in case of the fake signal (see
do_signal()). EINTR is propagated back to the userspace program. This
could disturb the program, but...
* each process dealing with signals should react accordingly to EINTR
return values.
* syscalls returning EINTR happen to be quite common situation in the
system even if no fake signal is sent.
* freezer sends the fake signal and does not deal with EINTR anyhow.
Thus EINTR values are returned when the system is resumed.
The very safe marking is done in architectures' "entry" on syscall and
interrupt/exception exit paths, and in a stack checking functions of
livepatch. TIF_PATCH_PENDING is cleared and the next
recalc_sigpending() drops TIF_SIGPENDING. In connection with this, also
call klp_update_patch_state() before do_signal(), so that
recalc_sigpending() in dequeue_signal() can clear TIF_PATCH_PENDING
immediately and thus prevent a double call of do_signal().
Note that the fake signal is not sent to stopped/traced tasks. Such task
prevents the patching to finish till it continues again (is not traced
anymore).
Last, sending the fake signal is not automatic. It is done only when
admin requests it by writing 1 to signal sysfs attribute in livepatch
sysfs directory.
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: x86@kernel.org
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Log a few kernel debug messages at the beginning of the following livepatch
transition functions:
klp_complete_transition()
klp_cancel_transition()
klp_init_transition()
klp_reverse_transition()
Also update the log notice message in klp_start_transition() for similar
verbiage as the above messages.
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
klp_complete_transition() performs a bit of housework before a
transition to KLP_PATCHED or KLP_UNPATCHED is actually completed
(including post-(un)patch callbacks). To be consistent, move the
transition "complete" kernel log notice out of
klp_try_complete_transition() and into klp_complete_transition().
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Provide livepatch modules a klp_object (un)patching notification
mechanism. Pre and post-(un)patch callbacks allow livepatch modules to
setup or synchronize changes that would be difficult to support in only
patched-or-unpatched code contexts.
Callbacks can be registered for target module or vmlinux klp_objects,
but each implementation is klp_object specific.
- Pre-(un)patch callbacks run before any (un)patching transition
starts.
- Post-(un)patch callbacks run once an object has been (un)patched and
the klp_patch fully transitioned to its target state.
Example use cases include modification of global data and registration
of newly available services/handlers.
See Documentation/livepatch/callbacks.txt for details and
samples/livepatch/ for examples.
Signed-off-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
rcu_read_(un)lock(), list_*_rcu(), and synchronize_rcu() are used for a secure
access and manipulation of the list of patches that modify the same function.
In particular, it is the variable func_stack that is accessible from the ftrace
handler via struct ftrace_ops and klp_ops.
Of course, it synchronizes also some states of the patch on the top of the
stack, e.g. func->transition in klp_ftrace_handler.
At the same time, this mechanism guards also the manipulation of
task->patch_state. It is modified according to the state of the transition and
the state of the process.
Now, all this works well as long as RCU works well. Sadly livepatching might
get into some corner cases when this is not true. For example, RCU is not
watching when rcu_read_lock() is taken in idle threads. It is because they
might sleep and prevent reaching the grace period for too long.
There are ways how to make RCU watching even in idle threads, see
rcu_irq_enter(). But there is a small location inside RCU infrastructure when
even this does not work.
This small problematic location can be detected either before calling
rcu_irq_enter() by rcu_irq_enter_disabled() or later by rcu_is_watching().
Sadly, there is no safe way how to handle it. Once we detect that RCU was not
watching, we might see inconsistent state of the function stack and the related
variables in klp_ftrace_handler(). Then we could do a wrong decision, use an
incompatible implementation of the function and break the consistency of the
system. We could warn but we could not avoid the damage.
Fortunately, ftrace has similar problems and they seem to be solved well there.
It uses a heavy weight implementation of some RCU operations. In particular, it
replaces:
+ rcu_read_lock() with preempt_disable_notrace()
+ rcu_read_unlock() with preempt_enable_notrace()
+ synchronize_rcu() with schedule_on_each_cpu(sync_work)
My understanding is that this is RCU implementation from a stone age. It meets
the core RCU requirements but it is rather ineffective. Especially, it does not
allow to batch or speed up the synchronize calls.
On the other hand, it is very trivial. It allows to safely trace and/or
livepatch even the RCU core infrastructure. And the effectiveness is a not a
big issue because using ftrace or livepatches on productive systems is a rare
operation. The safety is much more important than a negligible extra load.
Note that the alternative implementation follows the RCU principles. Therefore,
we could and actually must use list_*_rcu() variants when manipulating the
func_stack. These functions allow to access the pointers in the right
order and with the right barriers. But they do not use any other
information that would be set only by rcu_read_lock().
Also note that there are actually two problems solved in ftrace:
First, it cares about the consistency of RCU read sections. It is being solved
the way as described and used in this patch.
Second, ftrace needs to make sure that nobody is inside the dynamic trampoline
when it is being freed. For this, it also calls synchronize_rcu_tasks() in
preemptive kernel in ftrace_shutdown().
Livepatch has similar problem but it is solved by ftrace for free.
klp_ftrace_handler() is a good guy and never sleeps. In addition, it is
registered with FTRACE_OPS_FL_DYNAMIC. It causes that
unregister_ftrace_function() calls:
* schedule_on_each_cpu(ftrace_sync) - always
* synchronize_rcu_tasks() - in preemptive kernel
The effect is that nobody is neither inside the dynamic trampoline nor inside
the ftrace handler after unregister_ftrace_function() returns.
[jkosina@suse.cz: reformat changelog, fix comment]
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
klp_init_transition() does not set func->transition for immediate patches.
Then klp_ftrace_handler() could use the new code immediately. As a result,
it is not safe to put the livepatch module in klp_cancel_transition().
This patch reverts most of the last minute changes klp_cancel_transition().
It keeps the warning about a misuse because it still makes sense.
Fixes: 3ec24776bfd0 ("livepatch: allow removal of a disabled patch")
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
klp_mutex is shared between core.c and transition.c, and as such would
rather be properly located in a header so that we don't have to play
'extern' games from .c sources.
This also silences sparse warning (wrongly) suggesting that klp_mutex
should be defined static.
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Currently we do not allow patch module to unload since there is no
method to determine if a task is still running in the patched code.
The consistency model gives us the way because when the unpatching
finishes we know that all tasks were marked as safe to call an original
function. Thus every new call to the function calls the original code
and at the same time no task can be somewhere in the patched code,
because it had to leave that code to be marked as safe.
We can safely let the patch module go after that.
Completion is used for synchronization between module removal and sysfs
infrastructure in a similar way to commit 942e443127e9 ("module: Fix
mod->mkobj.kobj potentially freed too early").
Note that we still do not allow the removal for immediate model, that is
no consistency model. The module refcount may increase in this case if
somebody disables and enables the patch several times. This should not
cause any harm.
With this change a call to try_module_get() is moved to
__klp_enable_patch from klp_register_patch to make module reference
counting symmetric (module_put() is in a patch disable path) and to
allow to take a new reference to a disabled module when being enabled.
Finally, we need to be very careful about possible races between
klp_unregister_patch(), kobject_put() functions and operations
on the related sysfs files.
kobject_put(&patch->kobj) must be called without klp_mutex. Otherwise,
it might be blocked by enabled_store() that needs the mutex as well.
In addition, enabled_store() must check if the patch was not
unregisted in the meantime.
There is no need to do the same for other kobject_put() callsites
at the moment. Their sysfs operations neither take the lock nor
they access any data that might be freed in the meantime.
There was an attempt to use kobjects the right way and prevent these
races by design. But it made the patch definition more complicated
and opened another can of worms. See
https://lkml.kernel.org/r/1464018848-4303-1-git-send-email-pmladek@suse.com
[Thanks to Petr Mladek for improving the commit message.]
Signed-off-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
Change livepatch to use a basic per-task consistency model. This is the
foundation which will eventually enable us to patch those ~10% of
security patches which change function or data semantics. This is the
biggest remaining piece needed to make livepatch more generally useful.
This code stems from the design proposal made by Vojtech [1] in November
2014. It's a hybrid of kGraft and kpatch: it uses kGraft's per-task
consistency and syscall barrier switching combined with kpatch's stack
trace switching. There are also a number of fallback options which make
it quite flexible.
Patches are applied on a per-task basis, when the task is deemed safe to
switch over. When a patch is enabled, livepatch enters into a
transition state where tasks are converging to the patched state.
Usually this transition state can complete in a few seconds. The same
sequence occurs when a patch is disabled, except the tasks converge from
the patched state to the unpatched state.
An interrupt handler inherits the patched state of the task it
interrupts. The same is true for forked tasks: the child inherits the
patched state of the parent.
Livepatch uses several complementary approaches to determine when it's
safe to patch tasks:
1. The first and most effective approach is stack checking of sleeping
tasks. If no affected functions are on the stack of a given task,
the task is patched. In most cases this will patch most or all of
the tasks on the first try. Otherwise it'll keep trying
periodically. This option is only available if the architecture has
reliable stacks (HAVE_RELIABLE_STACKTRACE).
2. The second approach, if needed, is kernel exit switching. A
task is switched when it returns to user space from a system call, a
user space IRQ, or a signal. It's useful in the following cases:
a) Patching I/O-bound user tasks which are sleeping on an affected
function. In this case you have to send SIGSTOP and SIGCONT to
force it to exit the kernel and be patched.
b) Patching CPU-bound user tasks. If the task is highly CPU-bound
then it will get patched the next time it gets interrupted by an
IRQ.
c) In the future it could be useful for applying patches for
architectures which don't yet have HAVE_RELIABLE_STACKTRACE. In
this case you would have to signal most of the tasks on the
system. However this isn't supported yet because there's
currently no way to patch kthreads without
HAVE_RELIABLE_STACKTRACE.
3. For idle "swapper" tasks, since they don't ever exit the kernel, they
instead have a klp_update_patch_state() call in the idle loop which
allows them to be patched before the CPU enters the idle state.
(Note there's not yet such an approach for kthreads.)
All the above approaches may be skipped by setting the 'immediate' flag
in the 'klp_patch' struct, which will disable per-task consistency and
patch all tasks immediately. This can be useful if the patch doesn't
change any function or data semantics. Note that, even with this flag
set, it's possible that some tasks may still be running with an old
version of the function, until that function returns.
There's also an 'immediate' flag in the 'klp_func' struct which allows
you to specify that certain functions in the patch can be applied
without per-task consistency. This might be useful if you want to patch
a common function like schedule(), and the function change doesn't need
consistency but the rest of the patch does.
For architectures which don't have HAVE_RELIABLE_STACKTRACE, the user
must set patch->immediate which causes all tasks to be patched
immediately. This option should be used with care, only when the patch
doesn't change any function or data semantics.
In the future, architectures which don't have HAVE_RELIABLE_STACKTRACE
may be allowed to use per-task consistency if we can come up with
another way to patch kthreads.
The /sys/kernel/livepatch/<patch>/transition file shows whether a patch
is in transition. Only a single patch (the topmost patch on the stack)
can be in transition at a given time. A patch can remain in transition
indefinitely, if any of the tasks are stuck in the initial patch state.
A transition can be reversed and effectively canceled by writing the
opposite value to the /sys/kernel/livepatch/<patch>/enabled file while
the transition is in progress. Then all the tasks will attempt to
converge back to the original patch state.
[1] https://lkml.kernel.org/r/20141107140458.GA21774@suse.cz
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Ingo Molnar <mingo@kernel.org> # for the scheduler changes
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
