Arch-dependend way to restore extended registers set.
Use it straight-away to restore per-thread registers.
Signed-off-by: Dmitry Safonov <dima@arista.com>
Extended registers set for task is restored with rt_sigreturn() through
prepared sigframe. For threads it's currently lost.
Preserve it inside thread context to restore on thread curing.
Signed-off-by: Dmitry Safonov <dima@arista.com>
CRIU dumps main thread and sub-threads differently, so there needed
a test to check if fpu is preserved across C/R in sub-threads.
Signed-off-by: Dmitry Safonov <dima@arista.com>
With pseudo-random garbage, the seed is printed with pr_err().
get_task_regs() is called during seizing the task and also for each
thread.
At this moment only for x86.
Signed-off-by: Dmitry Safonov <dima@arista.com>
If snprintf was truncated we should probably know about it instead of
continuing to increase off, as snprintf returns number of characters
which would have been written and not the number which was actually
written.
Normally we check snprintf only for overflow not for error, some modern
compilers print warnings if truncation was not checked.
Probably it was the case why we implemented [1], the truncation happened
and on the next iteration of for loop we've hit negative size for
snprintf and got -1.
Fixes: 90f043dea ("namespaces: handle errors of snprintf") [1]
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
To minimize things done in parasite, PTRACE_GET_THREAD_AREA can be
used to get remote tls. That also removes an additional compat stack
(de)allocation in the parasite (also asm-coded syscall).
In order to use PTRACE_GET_THREAD_AREA, the dumpee should be stopped.
So, let's move this from criu to compel to non-seized state and put tls
into thread info on x86.
Signed-off-by: Dmitry Safonov <dima@arista.com>
It seems the Fedora rawhide /tmp is no longer 1777 but 755.
Change it back to 1777 to make our CI runs successful again.
Signed-off-by: Adrian Reber <areber@redhat.com>
New compilators print warnings if snprintf return value is not checked
for truncation. Let's make them happy.
Fixes: #1372
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
This will surpress false gcc warnings like this:
criu/stats.c:85:10: error: array subscript 4 is above array bounds
of 'struct timing[2]' [-Werror=array-bounds]
85 | return &rstats->timings[t];
| ^~~~~~~~~~~~~~~~~~~
criu/stats.c:25:16: note: while referencing 'timings'
25 | struct timing timings[RESTORE_TIME_NS_STATS];
| ^~~~~~~
cc1: all warnings being treated as errors
Signed-off-by: Andrei Vagin <avagin@gmail.com>
The latest glibc has redefined SIGSTKSZ as 'sysconf (_SC_SIGSTKSZ)' and
this breaks a static char[] definition.
Hardcoding TESTSIGSTKSZ to 16384 in the test. This fixes:
sigaltstack.c:17:13: error: variably modified 'stack_thread' at file scope
17 | static char stack_thread[SIGSTKSZ + TEST_MSG_BUFFER_SIZE] __stack_aligned__;
| ^~~~~~~~~~~~
sigaltstack.c:18:13: error: variably modified 'stack_main' at file scope
18 | static char stack_main[SIGSTKSZ + TEST_MSG_BUFFER_SIZE] __stack_aligned__;
| ^~~~~~~~~~
Signed-off-by: Adrian Reber <areber@redhat.com>
The latest podman pulls in crun instead of runc. Unfortunately crun is
not built against libcriu and does not support checkpoint/restore.
Switch back to runc.
Signed-off-by: Adrian Reber <areber@redhat.com>
In parse_pid_status(), it is assumed that the seccomp field can be
missing from /proc/pid/status. When the field is missing, it is not
properly initialized, leading to bad behavior.
We initialize seccomp_mode to SECCOMP_MODE_DISABLED by default,
similarly to what is done in compel/src/lib/infect.c:parse_pid_status.
Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
This differs from the previous one by
1. using relative path instead of absolute
2. chdir() after setup
Reviewed-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Reviewed-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
Create unix socket and unlink it. Make fifo in this place.
Reviewed-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Reviewed-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
Use SIOCUNIXFILE ioctl approach to get socket fd opened with O_PATH. Utilise it
for detecting deletion and resolving relative name. Preserve old method as
fallback if this new IOCTL fails.
Also remove overmount_sock crfail in zdtm. With the unix_resolve_name
reworked to use SIOCUNIXFILE criu can now pass this test.
Reviewed-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Reviewed-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
If there are any file in place of ghost unix socket, criu will delete it at
restore while trying to recreate ghost one.
Reviewed-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Reviewed-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
Moved strip_deleted to util and reworked it so other parts of criu can use
it without the need of files-reg.h
Reviewed-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Reviewed-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
Otherwise we gonna accumulated "(deleted)" postfix generated by
kernel on every c/r iteration eventually overflowing PATH_MAX
which will make container undumpable.
Signed-off-by: Cyrill Gorcunov <gorcunov@virtuozzo.com>
If criu finds a file lock and the --file-locks option isn't set, it
stops dumping processes, resumes them and exits with an error.
Signed-off-by: Andrei Vagin <avagin@gmail.com>
If criu finds a file lock and the --file-locks option isn't set, it
stops dumping processes, resumes them and exits with an error.
Signed-off-by: Andrei Vagin <avagin@gmail.com>
When I compile criu with "make DEBUG=1" and run it to restore my
program, it produces a segmentation fault.
In aarch64, with compile flag "-O0", when criu executes the code in pie,
it is unable to visit the content of ARCH_VDSO_SYMBOLS. So I put these
variables into the stack.
Signed-off-by: anatasluo <luolongjuna@gmail.com>
This moves Fedora Rawhide based tests away from Travis. To Github
Actions for x86_64 and to Drone for aarch64.
Signed-off-by: Adrian Reber <areber@redhat.com>
To run Fedora Rawhide based aarch64 containers on Drone CI our current
Dockerfile setup does not work.
This moves the package installation out of the Dockerfile into
scripts/ci/prepare-for-fedora-rawhide.sh to be usable in the Dockerfile
environment and in the Drone CI environment.
Signed-off-by: Adrian Reber <areber@redhat.com>
The updates to the latest Vagrant version and from Fedora 32 to 33.
Also using --no-tty instead of > /dev/null for vagrant up.
Also run 'dnf upgrade -y' in out vagrant VM to get the latest kernel.
Signed-off-by: Adrian Reber <areber@redhat.com>
Running in an environment with clang and without gcc even installed
does not work as compel-host-bin uses HOSTCC which defaults to gcc.
If CLANG=1 is set this also sets HOSTCC to clang to actually build
compel-host-bin with clang and not with gcc.
Signed-off-by: Adrian Reber <areber@redhat.com>
Besides Travis CI Drone CI seems to be only service providing ARM based
builds. This switches the aarch64 and arm32 builds to drone.io.
Because Drone CI is running in a Docker container we cannot use 'setarch
linux32' as it requires the blocked syscall 'personality(2)'.
But Drone CI provides an 'arch: arm' which gives the same architecture
as 'setarch linux32' on Travis aarch64: armv8l
Signed-off-by: Adrian Reber <areber@redhat.com>
1) Let's do test_init earlier so that max_nr test_msg is now visible in
thread-bomb.out.
2) Also lets check errors from malloc and pthread_... functions, print
messages about their errors and do proper deallocation at least.
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
With Travis dramatically reducing the minutes available for CI, CRIU
needs a new place to run tests. This moves the Vagrant based Fedora 32
no VDSO test cases to Cirrus CI. Cirrus CI seems to be one of the very
few free CI services allowing access to /dev/kvm.
Signed-off-by: Adrian Reber <areber@redhat.com>
open_handle and first part of alloc_openable do the same work. Both these
function are called from check_open_handle. Rework check_open_handle to call
only alloc_openable.
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
mnt_is_dir is used when looking up for suitable mount point. In some cases
that function may fail several times. Error level seems to strict for this
cases.
Added error message to lookup_mnt_sdev in case all mnt_is_dir failed.
As for open_handle and alloc_openable which are calling mnt_is_dir, they are
used in check_open_handle, which will call error afterwards.
Adjusted log level for __open_mountpoint result in open_handle since it is
allowed to fail several times. open_handle caller get_mark_path expect
possible failure and will print error in case.
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko@virtuozzo.com>
