Currently we fill as much as posible data in a socket and then use
half of these data to make a flow in the connection.
These data is send and recv buffers. When a connection is dumped and
restored, it needs time to go back to the normal work. In this time we
can reliably fill only send buffers. So at the result the sockets may
contain less data than required for heartbeating and the test will
block.
The test with this patch will monitor actuall size of data in a test
socket.
Cc: Konstantin Neumoin <kneumoin@parallels.com>
Reported-by: Konstantin Neumoin <kneumoin@parallels.com>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
@action_names is rather a const array, so make
sure we never access some data outside of it
defining its size.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
In addition it checks that criu closes all its descriptors
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
strstr is a really heavy one, lets use already defined
and filled @file_path variable instead.
Reported-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Andrew Vagin <avagin@parallels.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
In criu we carry SysV memory via @VMA_AREA_SYSVIPC
attribute but to the test must have a live reference
to real memory area to check that shmem address was
restored correctly. Add it.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Andrew Vagin <avagin@parallels.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
This would allow to restore parameters with user namespaces
enabled.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Need to compare the /pts/ part in the
former path, otherwise it will produce
bogus /dev/pts/pts/index form.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We need this only once -- while calling the mmap from remote
context -- so it's enough to have on-stack variable.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Andrew Vagin <avagin@parallels.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Because directories are opened via direct mkdir
call with name taken from ghost path don't postfix
it with cr.%x.ghost, otherwise that's the name
directory will have after restore complete causing
cwd01 test to fail.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Users don't like errors even if criu returns zero.
So I suggest to start looking at them.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
In this case we have another chance to not skip errors.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We don't support posix mqueues at the moment
but in case if this fs is simply mounted and
not used lets proceed without errors.
In case if someone is using it we detect it
because fs won't be empty and refuse to dump.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
This timeout is too small for tcp test cases.
This timeout should be bigger than TCP_RTO_MAX 120 seconds
Signed-off-by: Konstantin Neumoin <kneumoin@parallels.com>
Acked-by: Andrew Vagin <avagin@parallels.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
It's bad idea to have only one target object.
In addition I add a shared region to the second process,
which is not mapped in the first.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
/proc/PID/map_files are protected by the global CAP_SYS_ADMIN, so we
need to avoid using them to support user namespaces.
We are going to use memfd_create() to get the first file descriptor and
then all others processes will able to open it via /proc/PID/fd/X.
In this patch memfd_create() is used to get a file descriptor, which is
associated with a shared memory region.
If memfd_create() isn't supported, the old scheme will be used.
v2: don't set a name for memfd. It isn't required.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
/proc/PID/map_files are protected by the global CAP_SYS_ADMIN, so we
need to avoid using them to support user namespaces.
We are going to use memfd_create() to get the first file descriptor and
then all others processes will able to open it via /proc/PID/fd/X.
This patch reworks slave processes to not use map_files.
v2: add more comments
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
v2: Follow the kerndat style that "features" are described
just by global boolean variables.
v3: give NULL as a name to get EFAULT if memfd_create is supported
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Currently we're using predefined format for master/slave pty peers:
masters are always /dev/ptmx, while slaves are /dev/pts/$index,
where $index is the peer number.
While fitting most of distros this is not always correct and slave peers
might be mounted to an arbitrary place, so that we need to somehow
carry paths with ourself in image.
Instead of bloating current tty image lets use regular file engine instead
and on checkpoint stage save a path to the link in regfiles set, then on
restore simply fetch it from the image.
Such approach will help in future when we need to support multiple
instances of devpts filesystem.
To support backward compatibility with images where no regfile
records are present we generate new one on the fly in
pty_alloc_reg() helper.
Because of the need to restore dead slave peers and restore of
the controlling terminal we need to generate that named "fake
inverted" pty_alloc_fake_reg() helper: in particular if
we need to open dead slave peer we generate fake master peer,
open it and the close out. Almost the same situation in
restoring contolling terminal -- we get master peer, generate
appropriate fake slave object, open it, manipulate, then
close it out and free.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
The idea is to be able to lookup for special id
which might be not present and we should not
yield the error.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We will use reg-files engine for tty c/r so lets prepare scaffolds
here preventing the engine to generate ghost files for PTY peers
(mostly because one can't create slave peers without opening
master peers first which is a sole part of tty engine itself).
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Otherwise this postfix will be accumulated on
subsequent checkpoints.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We will need it for tty restore.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
To use it in tty code even when file
descriptor is not added into files chain.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We will use this path in reg-files engine anyway
so simply switch to this ability now.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We will have to support more tty types in future so
make calls depending on type of ttys.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Instead of calling case() with majors all over the places lets
introduce own enum for tty types and use it instead.
Because we're using not @major numbers now but taking @minors
into account as well, this brings more strict check of which
kind of terminals we can dump now thus it's potentially should
fix the cases when we're trying to c/r terminals which we don't
understand yet (in particular /dev/console [5:1]).
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We should not have a chance to exit with a wrong code on error paths.
Now dump_one_task() returs zero, if allocation of dfds failed:
ret = collect_mappings(pid, &vmas);
if (ret) {
pr_err("Collect mappings (pid: %d) failed with %d\n", pid, ret);
goto err;
}
if (!shared_fdtable(item)) {
dfds = xmalloc(sizeof(*dfds));
if (!dfds)
goto err;
...
err:
return -1;
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
See the comment below for an explanation of what is going on. We will
ultimately need to handle dumping the netlink data, but I think it is good to
prevent injecting events into the stream during a dump. So we pre-load the
modules, even though it isn't very pretty.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
I found this solution in the LXC code. We can open the old root, call
pivot_root(".", "."), call fchdir to the old root and call umount(".").
Now restore will not fail, if the root is read-only.
In addition it's a bit faster.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
The main problem with the
test that it's never 'fail'.
Test can only 'pass' or hangs.
Signed-off-by: Konstantin Neumoin <kneumoin@parallels.com>
Acked-by: Andrew Vagin <avagin@parallels.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
TASK_HELPERs are created with CLONE_FILES, so if we always close the cg yard
here, it will close it for the other helpers and cause problems. Instead, we
close it much later, in code only called by alive tasks, to ensure that there
is no conflict.
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Actually these @locked and @packet_mode are
valid for unix98 ptys only so since moving
them into @tty_pty_entry is impossible at
least lets comment this.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
- parse_index -> parse_pty_index since it's PTY specific
- pty_is_master -> tty_is_master because it will test
not only PTY peers but other types as well
- pty_is_hung -> tty_is_hung for same reason
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
TASK_SIZE for AArch64 may be any of several options between (1 << 39)
and (1 << 48), inclusive, depending on kernel configuration. Go back
to just checking the most significant bit, as was done before commit
3f12d688ae563b91d8a738d68902ea2e23f2b59c was made to accomodate 32-bit
ARM (before mmap_seized got architecture-specific implementations).
This fixes the following error for AArch64 kernels with
CONFIG_ARM64_64K_PAGES=y.
Error (parasite-syscall.c:1105): Can't allocate memory for parasite blob (pid: 104)
Signed-off-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
