- 20 Sep, 2017 21 commits
-
-
Dmitry Safonov authored
No functional change expected. Signed-off-by:
Dmitry Safonov <dsafonov@virtuozzo.com> Signed-off-by:
Andrei Vagin <avagin@virtuozzo.com>
-
Dmitry Safonov authored
This is just to split this oversized outgrowed plumped up parasite_fixup_vdso() and separate it logically. Signed-off-by:
-
Dmitry Safonov authored
Generalize addr-to-pfn conversion in one function. Signed-off-by:
-
Dmitry Safonov authored
Expand vaddr_to_pfn() so it can read and translate to PFN from already opened fd. This can be used as optimization if one need to translate a couple of addresses and also to read other task's PFN. I'll use it in the next patch for reading other tasks's vdso's PFN. Signed-off-by:
-
Dmitry Safonov authored
Let's hide those kernel details in enum. Further in this patches set I'll add kdat test for presence of "[vdso]" hint after mremap(), so we will skip any checking on kernels > v3.16 and do not init vdso_pfn also (skipping parsing of self-pagemap). Signed-off-by:
-
Michael Holzheu authored
On s390 we don't use vvar. Therefore vvar_size in not been initialized and the value remains VVAR_BAD_SIZE. Fix the BUG() statement to also cover this case. Signed-off-by:
Michael Holzheu <holzheu@linux.vnet.ibm.com> Reviewed-by:
-
Andrei Vagin authored
In file included from criu/include/util-vdso.h:23:0, from criu/include/kerndat.h:8, from criu/lsm.c:8: criu/arch/aarch64/include/asm/vdso.h:17:35: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'aarch_vdso_symbol1' static const char* __maybe_unused aarch_vdso_symbol1 = "__kernel_clock_getres"; Signed-off-by: -
Dmitry Safonov authored
Don't need to parse vdso symtable each restore - it's boot-persistent, so move it into criu.kdat file on tmpfs. That will also remove syscalls made for filling compat vdso symtable by compat vdso helper. Reviewed-by:
Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by:
-
Dmitry Safonov authored
On dump we only need vdso_pfn to be filled, on restore we need filled symtables. That means, that we can omit filling symtables on dump, which also means no need in calling compat_vdso_helper process with fork(), pipe(), read(), write() and so on syscalls. Reviewed-by:
-
Dmitry Safonov authored
As ASLR randomizes {vdso,vvar}_start between criu launches, vdso_parse_maps() should be called each launch: - on restore to know {vdso,vvar}_start position for later parking in restorer's save zone - on checkpointing to get vdso's pfn for pre-v3.16 kernels which lose "[vdso]" hint in maps file. But vdso_fill_symtable() call may be omitted if symtable is inside kdat file. Reviewed-by: -
Dmitry Safonov authored
Preparation for saving vdso_symtable in kdat, which will allow skip parsing of native and compat symtables - that at least will save from running compat helper each criu launch. As {vvar,vdso}_start are randomized with ASLR, there is no point in saving them into kdat. We'll still need to reread them from /proc/self/maps for awhile. Reviewed-by: -
Dmitry Safonov authored
I plan to keep boot-persistent vdso_symtable inside kdat, moving {vvar,vdso}_start addresses out into new structure. As order of vdso/vvar is preserved across one booting store it inside of vdso_symtable. Reviewed-by: -
Dmitry Safonov authored
Make a function which decides if we need to insert jump trampolines, or if the blobs are equal and just a remap if enough. Reviewed-by:
-
Dmitry Safonov authored
Ordering check for vvar/vdso blobs was introduced with the commit b00bdb2d ("vdso: x86 -- Test VMAs order in vdso_proxify") Let's simplify it to more readable version. As above we've compared that vvar/vdso's blob size from dumpee matches sizes of rt-vvar/rt-vdso, kill that xor here. Reviewed-by:
-
Dmitry Safonov authored
Check that task without vvar & vdso blobs is restored without them. Reviewed-by:
-
Dmitry Safonov authored
Let's just unmap criu's vdso & vvar if restoree doesn't have them. This could be fired e.g. by migrating task from vdso-disabled kernel to vdso-enabled one. Reviewed-by:
-
Dmitry Safonov authored
The plan is to keep boot-persistent vdso properties in symtable, to omit parsing it in each invocation of criu. As sizes of vdso/vvar are being stable on the same kernel, move them into symtable, substituting end addresses. Begin/end addresses are randomized by ASLR so there is no point in storing them in kdat. Reviewed-by:
-
Dmitry Safonov authored
Just slightly prettify code, no functional change. Reviewed-by:
-
Kirill Tkhai authored
The right order for all of our 4 archs is: SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, int __user *, parent_tidptr, unsigned long, tls, int __user *, child_tidptr) See Linux kernel for the details. Note, this is just a fix, and it's not connected with the second patch. Signed-off-by:Kirill Tkhai <ktkhai@virtuozzo.com> Reviewed-by:
-
Kir Kolyshkin authored
It makes sense to show errno to a user. Signed-off-by:
Kir Kolyshkin <kir@openvz.org> Signed-off-by:
-
Adrian Reber authored
The check_features RPC function required that both known fields are present. Without those fields it exited with an error. If RPC users where not specifying all parameters it would fail. It should, however, be possible to only check for a subset of options. Each supported option is checked separately anyway in the forked criu which does the actual check. Removing the check also enables RPC clients with older protobuf definitions to use the feature check. Signed-off-by:
Adrian Reber <areber@redhat.com> Signed-off-by:
-
- 18 Sep, 2017 1 commit
-
-
Mike Rapoport authored
The recent changes to user address space limits in the Linux kernel break the assumption that TASK_SIZE is 128TB. For now, the maximal task size on ppc64le is 512TB and we need to detect it in runtime for compatibility with older kernels. Signed-off-by:
Mike Rapoport <rppt@linux.vnet.ibm.com> Acked-by:
Laurent Dufour <ldufour@linux.vnet.ibm.com> Signed-off-by:
-
- 17 Sep, 2017 18 commits
-
-
Mike Rapoport authored
Even with 2 parallel jobs maps04 takes too much time with --remote-lazy-pages. Let's skip it for now. Signed-off-by:
-
Mike Rapoport authored
If there were fork()'s during lazy-restore we create a shallow copy of the parent process' page-read. Since all the copies reference the same open files, we cannot close the page-read until we finish restore of all the processes that share the page-read. Signed-off-by:
-
Mike Rapoport authored
Running zdtm/static/maps04 with --remote-lazy-pages in parallel with 3 other tests takes too much time on the Jenkins builder. Let's try running with --parallel 2. Signed-off-by:
-
Mike Rapoport authored
When there is a stale kdat cache file it's contents is read into the memory and we end up requesting random userfaultfd features. Explicitly set the kdat.uffd_features to zero before querying the kernel resolves the issue. ✓ travis-ci: success for kerndat: set uffd features to 0 before querying kernel Signed-off-by:
Pavel Emelyanov <xemul@virtuozzo.com>
-
Mike Rapoport authored
Without non-cooperative userfaultfd some programs may fail during lazy restore because they perform operations that cannot be handled by the lazy-pages daemon. ✓ travis-ci: success for lazy-pages: update checks for availability of userfaultfd (rev3) Signed-off-by:
-
Mike Rapoport authored
The kerndat_init() is now called before the jump to action handler. This allows us to directly use kdat without calling to the corresponding kerndat_*() methods. ✓ travis-ci: success for lazy-pages: update checks for availability of userfaultfd (rev3) Signed-off-by:
-
Dmitry Safonov authored
After the commit 02c763939c10 ("test/zdtm: unify common code") CFLAGS with -D_GNU_SOURCE defined in the top Makefile are being passed to tests Makefiles. As _GNU_SOURCE is also defined by tests, that resulted in zdtm tests build failures: make[2]: Entering directory `/home/criu/test/zdtm/lib' CC test.o test.c:1:0: error: "_GNU_SOURCE" redefined [-Werror] #define _GNU_SOURCE ^ <command-line>:0:0: note: this is the location of the previous definition cc1: all warnings being treated as errors make[2]: *** [test.o] Error 1 However, we didn't catch this in time by Travis-CI, as zdtm.py doesn't do `make zdtm`, rather it does `make -C test/zdtm/{lib,static,transition}`. By calling middle makefile this way, it doesn't have _GNU_SOURCE in CFLAGS from top-Makefile. I think the right thing to do here - is following CRIU's way: rely on definition of _GNU_SOURCE by Makefiles. This patch is almost fully generated with find test/zdtm/ -name '*.c' -type f \ -exec sed -i '/define _GNU_SOURCE/{n;/^$/d;}' '{}' \; \ -exec sed -i '/define _GNU_SOURCE/d' '{}' \; With an exception for adding -D_GNU_SOURCE in tests Makefile.inc for keeping the same behaviour for zdtm.py. Also changed utsname.c to use utsname::domainname, rather private utsname::__domainname, as now it's uncovered (from sys/utsname.h): > struct utsname > { ... > # ifdef __USE_GNU > char domainname[_UTSNAME_DOMAIN_LENGTH]; > # else > char __domainname[_UTSNAME_DOMAIN_LENGTH]; > # endif Reported-by: -
Mike Rapoport authored
With kdat cache and unified kerndat_init() we can call it very early in crtools and then kdat information will be available for all cr-* actions. Signed-off-by:
-
Mike Rapoport authored
kdat and lazy-pages use nearly the same sequence to open userfault. This code can definitely live in a dedicated function. Signed-off-by:
-
Mike Rapoport authored
Newer kernels will report ESRCH when uffdio_copy is attempted when the process is exiting [1]. We take care of the new errno value and keep ENOSPC for compatibility. [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/fs/userfaultfd.c?id=e86b298bebf7e799e4b7232e9135799f1947552eSigned-off-by:
-
Mike Rapoport authored
Signed-off-by:
-
Mike Rapoport authored
Signed-off-by:
-
Mike Rapoport authored
Signed-off-by:
-
Mike Rapoport authored
Raise an exception for kernels that do not have userfaultfd. For the kernels that have userfaultfd but do not provide non-cooperative events (4.3 - 4.11) just print a warning. Fixes: #363 Signed-off-by:
-
Mike Rapoport authored
For the older kernels the implementation of userfaultfd would not include non-cooperative mode. In such case it is still possible to use uffd and enable lazy-pages, but if the restored process will change its virtual memory layout during restore, we'll get memory corruption. After this change 'criu check --feature uffd' will report success if the kernel supports userfaultfd at all and 'criu check --feature uffd-noncoop' will report success if the kernel supports non-cooperative userfaultfd. Suggested-by:
-
Mike Rapoport authored
Signed-off-by:
-
Mike Rapoport authored
Acked-by:
-
Mike Rapoport authored
Acked-by:
-
