On 24.02.21 10:19, Michael S. Tsirkin wrote:
On Wed, Feb 24, 2021 at 10:47:31AM +0200, Adrian Catangiu wrote:
- Background and problem
The System Generation ID feature is required in virtualized or
containerized environments by applications that work with local copies
or caches of world-unique data such as random values, uuids,
monotonically increasing counters, etc.
Such applications can be negatively affected by VM or container
snapshotting when the VM or container is either cloned or returned to
an earlier point in time.
Furthermore, simply finding out about a system generation change is
only the starting point of a process to renew internal states of
possibly multiple applications across the system. This process requires
a standard interface that applications can rely on and through which
orchestration can be easily done.
- Solution
The System Generation ID is meant to help in these scenarios by
providing a monotonically increasing u32 counter that changes each time
the VM or container is restored from a snapshot.
The `sysgenid` driver exposes a monotonic incremental System Generation
u32 counter via a char-dev filesystem interface accessible
through `/dev/sysgenid`. It provides synchronous and asynchronous SysGen
counter update notifications, as well as counter retrieval and
confirmation mechanisms.
The counter starts from zero when the driver is initialized and
monotonically increments every time the system generation changes.
Userspace applications or libraries can (a)synchronously consume the
system generation counter through the provided filesystem interface, to
make any necessary internal adjustments following a system generation
update.
The provided filesystem interface operations can be used to build a
system level safe workflow that guest software can follow to protect
itself from negative system snapshot effects.
The `sysgenid` driver exports the `void sysgenid_bump_generation()`
symbol which can be used by backend drivers to drive system generation
changes based on hardware events.
System generation changes can also be driven by userspace software
through a dedicated driver ioctl.
**Please note**, SysGenID alone does not guarantee complete snapshot
safety to applications using it. A certain workflow needs to be
followed at the system level, in order to make the system
snapshot-resilient. Please see the "Snapshot Safety Prerequisites"
section in the included documentation.
Signed-off-by: Adrian Catangiu <acatan@xxxxxxxxxx>
---
Documentation/misc-devices/sysgenid.rst | 229 +++++++++++++++
Documentation/userspace-api/ioctl/ioctl-number.rst | 1 +
MAINTAINERS | 8 +
drivers/misc/Kconfig | 15 +
drivers/misc/Makefile | 1 +
drivers/misc/sysgenid.c | 322 +++++++++++++++++++++
include/uapi/linux/sysgenid.h | 18 ++
7 files changed, 594 insertions(+)
create mode 100644 Documentation/misc-devices/sysgenid.rst
create mode 100644 drivers/misc/sysgenid.c
create mode 100644 include/uapi/linux/sysgenid.h
[...]
+``ioctl()``:
+ The driver also adds support for waiting on open file descriptors
+ that haven't acknowledged a generation counter update, as well as a
+ mechanism for userspace to *trigger* a generation update:
+
+ - SYSGENID_SET_WATCHER_TRACKING: takes a bool argument to set tracking
+ status for current file descriptor. When watcher tracking is
+ enabled, the driver tracks this file descriptor as an independent
+ *watcher*. The driver keeps accounting of how many watchers have
+ confirmed the latest Sys-Gen-Id counter and how many of them are
+ *outdated*; an outdated watcher is a *tracked* open file descriptor
+ that has lived through a Sys-Gen-Id change but has not yet confirmed
+ the new generation counter.
+ Software that wants to be waited on by the system while it adjusts
+ to generation changes, should turn tracking on. The sysgenid driver
+ then keeps track of it and can block system-level adjustment process
+ until the software has finished adjusting and confirmed it through a
+ ``write()``.
+ Tracking is disabled by default and file descriptors need to
+ explicitly opt-in using this IOCTL.
+ - SYSGENID_WAIT_WATCHERS: blocks until there are no more *outdated*
+ tracked watchers or, if a ``timeout`` argument is provided, until
+ the timeout expires.
+ If the current caller is *outdated* or a generation change happens
+ while waiting (thus making current caller *outdated*), the ioctl
+ returns ``-EINTR`` to signal the user to handle event and retry.
+ - SYSGENID_TRIGGER_GEN_UPDATE: triggers a generation counter increment.
+ It takes a ``minimum-generation`` argument which represents the
+ minimum value the generation counter will be set to. For example if
+ current generation is ``5`` and ``SYSGENID_TRIGGER_GEN_UPDATE(8)``
+ is called, the generation counter will increment to ``8``.
And what if it's 9?
Then it becomes 10. The hint only tells you what the smallest version
the system is matching against is.
The only thing I have a slight concern over here is an overflow. What if
my generation id is 0x7fffffff? For starters, it'd probably be better to
treat the counter as ulong so it matches the atomic_t, no?
But then you would still have the same situation, just with a wrap to 0
instead of a wrap to negative. I guess the answer is "users of this API
will not get a guarantee that the counters are monotonically increasing.
They have to check for != instead of < or >".
+ This IOCTL can only be used by processes with CAP_CHECKPOINT_RESTORE
+ or CAP_SYS_ADMIN capabilities.
+
+``mmap()``:
+ The driver supports ``PROT_READ, MAP_SHARED`` mmaps of a single page
+ in size. The first 4 bytes of the mapped page will contain an
+ up-to-date u32 copy of the system generation counter.
+ The mapped memory can be used as a low-latency generation counter
+ probe mechanism in critical sections.
+ The mmap() interface is targeted at libraries or code that needs to
+ check for generation changes in-line, where an event loop is not
+ available or read()/write() syscalls are too expensive.
+ In such cases, logic can be added in-line with the sensitive code to
+ check and trigger on-demand/just-in-time readjustments when changes
+ are detected on the memory mapped generation counter.
+ Users of this interface that plan to lazily adjust should not enable
+ watcher tracking, since waiting on them doesn't make sense.
+
+``close()``:
+ Removes the file descriptor as a system generation counter *watcher*.
+
+Snapshot Safety Prerequisites
+=============================
+
+If VM, container or other system-level snapshots happen asynchronously,
+at arbitrary times during an active workload there is no practical way
+to ensure that in-flight local copies or caches of world-unique data
+such as random values, secrets, UUIDs, etc are properly scrubbed and
+regenerated.
+The challenge stems from the fact that the categorization of data as
+snapshot-sensitive is only known to the software working with it, and
+this software has no logical control over the moment in time when an
+external system snapshot occurs.
+
+Let's take an OpenSSL session token for example. Even if the library
+code is made 100% snapshot-safe, meaning the library guarantees that
+the session token is unique (any snapshot that happened during the
+library call did not duplicate or leak the token), the token is still
+vulnerable to snapshot events while it transits the various layers of
+the library caller, then the various layers of the OS before leaving
+the system.
+
+To catch a secret while it's in-flight, we'd have to validate system
+generation at every layer, every step of the way. Even if that would
+be deemed the right solution, it would be a long road and a whole
+universe to patch before we get there.
+
+Bottom line is we don't have a way to track all of these in-flight
+secrets and dynamically scrub them from existence with snapshot
+events happening arbitrarily.
Above should try harder to explan what are the things that need to be
scrubbed and why. For example, I personally don't really know what is
the OpenSSL session token example and what makes it vulnerable. I guess
snapshots can attack each other?
Here's a simple example of a workflow that submits transactions
to a database and wants to avoid duplicate transactions.
This does not require overseer magic. It does however require
a correct genid from hypervisor, so no mmap tricks work.
int genid, oldgenid;
read(&genid);
start:
oldgenid = genid;
transid = submit transaction
read(&genid);
if (genid != oldgenid) {
revert transaction (transid);
goto start:
}
I'm not sure I fully follow. For starters, if this is a VM local
database, I don't think you'd care about the genid. If it's a remote
database, your connection would get dropped already at the point when
you clone/resume, because TCP and your connection state machine will get
really confused when you suddenly have a different IP address or two
consumers of the same stream :).
But for the sake of the argument, let's assume you can have a
connectionless database connection that maintains its own connection
uniqueness logic. That database connector would need to understand how
to abort the connection (and thus the transaction!) when the generation
changes. And that's logic you would do with the read/write/notify
mechanism. So your main loop would check for reads on the genid fd and
after sending a connection termination, notify the overlord that it's
safe to use the VM now.
The OpenSSL case (with mmap) is for libraries that are stateless and can
not guarantee that they receive a genid notification event timely.
Since you asked, this is mainly important for the PRNG. Imagine an https
server. You create a snapshot. You resume from that snapshot. OpenSSL is
fully initialized with a user space PRNG randomness pool that it
considers safe to consume. However, that means your first connection
after resume will be 100% predictable randomness wise.
The mmap mechanism allows the PRNG to reseed after a genid change.
Because we don't have an event mechanism for this code path, that can
happen minutes after the resume. But that's ok, we "just" have to ensure
that nobody is consuming secret data at the point of the snapshot.
+Simplifyng assumption - safety prerequisite
+-------------------------------------------
+
+**Control the snapshot flow**, disallow snapshots coming at arbitrary
+moments in the workload lifetime.
+
+Use a system-level overseer entity that quiesces the system before
+snapshot, and post-snapshot-resume oversees that software components
+have readjusted to new environment, to the new generation. Only after,
+will the overseer un-quiesce the system and allow active workloads.
+
+Software components can choose whether they want to be tracked and
+waited on by the overseer by using the ``SYSGENID_SET_WATCHER_TRACKING``
+IOCTL.
+
+The sysgenid framework standardizes the API for system software to
+find out about needing to readjust and at the same time provides a
+mechanism for the overseer entity to wait for everyone to be done, the
+system to have readjusted, so it can un-quiesce.
+
+Example snapshot-safe workflow
+------------------------------
+
+1) Before taking a snapshot, quiesce the VM/container/system. Exactly
+ how this is achieved is very workload-specific, but the general
+ description is to get all software to an expected state where their
+ event loops dry up and they are effectively quiesced.
If you have ability to do this by communicating with
all processes e.g. through a unix domain socket,
why do you need the rest of the stuff in the kernel?
Quescing is a harder problem than waking up.
That depends. Think of a typical VM workload. Let's take the web server
example again. You can preboot the full VM and snapshot it as is. As
long as you don't allow any incoming connections, you can guarantee that
the system is "quiesced" well enough for the snapshot.
This is really what this bullet point is about. The point is that you're
not consuming randomness you can't reseed asynchronously (see the above
OpenSSL PRNG example).
Alex
Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss
Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879
