There are two goals of this merge. First is to reduce the amount
of image files we generate and scan on restore. The latter is
more importaint, as even if we have no weird stuff like signalfd,
we still try to open this file. So after the merge we try to
open ~15 image files (out of ~30) less %) which is nice.
The 2nd goal is to simplify the C/R support for SCM messages.
This becomes possible with the fact, that all files we have can
be distinguished by their ID only, w/o type. This, in turn,
makes image layout for SCMs much simpler.
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
I'm about to merge all the file images into one big image,
so here's the place that would read it and collect.
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
(Try to) Fix the readme problems found by the feedmereadmes guys:
* Tuned up the tagline
* Added info about what for, history, competitors, etc.
* Put more links for potential contributors
This device is an API endpoint, just skip it. Also, prepare the
ground for sit dump.
Issue #11
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Most of the changes this time are preparation for future
new features and optimizations, that hasn't yet been well
tested and polished.
However, we have several new features. The most important
one is the --tcp-close option to help migration of Docker
containers, that constantly change their IP address.
And, as usually, a set of bugfixes.
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
The struct fd_link link allocated in inner context might be
freed before dump call (depending on compiler), instead
use a safer approach.
Signed-off-by: Cyrill Gorcunov <gorcunov@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The unified hierarchy controller doesn't have any name, so criu hangs
while parsing this sort of strings.
Let's skip those for now, until we implement cgroup2 support.
https://github.com/xemul/criu/issues/252
v2: Code comments
Reported-and-tested-by: Adrian Reber <adrian@lisas.de>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
It looks like this call to open_image was left from the
times when we reworked packets sending and collected
them early.
✓ travis-ci: success for sk-packet: Don't open image when not needed
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Code for tcp-close option was added after boolean-valued options
were modified to accept their negations as well. Since tcp-close
option works only with boolean values and does not accept any
argument, it should be wriiten in the same manner to allow users
to override the value when using configuration files.
Signed-off-by: Veronika Kabatova <vkabatov@redhat.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Only long options are supported in configuration files. "-v" is the only
short option which doesn't have a corresponding long option, therefore
users wouldn't be able to setup verbosity in configuration files without
this patch.
Signed-off-by: Veronika Kabatova <vkabatov@redhat.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The TEST_STRING "Hello World!" has 13 bytes size. Therefore the 10 bytes
used for the buffer size are not sufficient.
Fix this and use sizeof() to be on the safe side.
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The "args" parameter in apply_sb_flags() is unsigned long. Therefore also
use unsigned long for the cast.
The "sflags" variable in do_new_mount() is unsigned long. Therefore use
sizeof(sflags) instead of sizeof(int).
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The htole64() function returns a 64 bit value. This is stored in
a 32 bit integer in wait_fds_event() which results in 0 for big-endian
64 bit platforms. The fuxtex() system call also requires 32 bit values.
Fix this and use htole32() also for 64 bit platforms.
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The remap is searhced by dev:ino pair that is known only when
root task starts (and opend the remap). The collection stage
is about to happen in crtools, so move the resolve in post-prep.
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
The call to unlink() should happen in root task context,
while the collection stage is about to be moved into the
crtools, so keep the stale-unlink piece in post-prep
for a socket.
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Having pprep_head built-in it becomes much easier
to do socket-by-socket peer resolution.
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Having ppreh_heads we can simplify the _once helper
by re-using the .next field.
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Now the post-prep actions are xmalloc()-ed then added to the list.
Let's make it sit right on the structures that need the post-prep,
this would make some things much simpler and nicer soon.
The approach is inspired by kernel rcu_head-s :)
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
This flag now means "image was opened", while the "empty image"
concept makes this always true. Since there's one user of this
flag, which is obviously incorrect, remove the latter and fix
the former to mean "at least one object was read".
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
We don't want to run tests, if a kernel has a problem. And we want
to know when the taint flag is changed.
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Acked-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
With the recent cleanups in fdinfo parsing the memory
allocation policy for entries has changed, thus keeping
a pointer on the entry handle is no longer valid (it
will be freed soon).
Said that -- duplicate the memory for queued irmap work.
https://github.com/xemul/criu/issues/328
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
'tcp-close' is a option that replaces connected tcp-sockets with
'closed' ones during restoring, therefore it have no sense without
restore stage. Thus let's skip related tests if the 'norst' flag was
specified.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
1: Signed underflow
criu/files-reg.c: In function ‘open_path’:
criu/files-reg.c:1299:14: error: iteration 2147483647
invokes undefined behavior [-Werror=aggressive-loop-optimizations]
while (count--) {
~~~~~^~
2: Uninitialized variable
criu/files-reg.c: In function ‘make_parent_dirs_if_need’:
criu/files-reg.c:1354:13: warning: ‘count’ may be used uninitialized in
this function [-Wmaybe-uninitialized]
P.S. arch linux 4.7.5-1. gcc (GCC) 6.3.1 20170109
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
In case of success, we want to be silent when on default log level.
This is a time-honored UNIX tradition, who we are to break it?
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
Surprisingly, but
offsetof(inotify_wd_entry, node) == offsetof(fanotify_mark_entry, node)
even though both structures have node field in the middle and the
fields sitting above one are totally different.
So the code works, but ... we're risking :)
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
If the ghost file is too big, it might make sence to try seeking
for holes in it, thus reducing the image size.
We've seen this once for tmpfs files in issue #230.
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
We don't need gf there, but will need more than just mode.
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
GCC isn't happy if we use %rbp as register for local variable
with -ggdb3 option.
Which resulted in the following build error for `make DEBUG=1`:
> In file included from criu/arch/x86/crtools.c:10:0:
> criu/arch/x86/include/asm/compat.h: In function ‘do_full_int80’:
> criu/arch/x86/include/asm/compat.h:50:1: error: bp cannot be used in asm here
Fix it by saving/restoring %rbp around 32-bit syscall manually.
Just while at it - add a comment about r8-r11 clobbers.
Reported-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
There are 2 test cases:
1. Connected socket should be restored in the closed state
2. Listening socket state should not change after restore
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Eugene Batalov <eabatalov89@gmail.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>
