* Dmitry Safonov <dsafonov@xxxxxxxxxxxxx> wrote: > On 05/16/2016 01:54 PM, Ingo Molnar wrote: > > > >* Dmitry Safonov <dsafonov@xxxxxxxxxxxxx> wrote: > > > >>On 05/10/2016 04:29 PM, Dmitry Safonov wrote: > >>>Add possibility for userspace 32-bit applications to move > >>>vdso mapping. Previously, when userspace app called > >>>mremap for vdso, in return path it would land on previous > >>>address of vdso page, resulting in segmentation violation. > >>>Now it lands fine and returns to userspace with remapped vdso. > >>>This will also fix context.vdso pointer for 64-bit, which does not > >>>affect the user of vdso after mremap by now, but this may change. > >>> > >>>As suggested by Andy, return EINVAL for mremap that splits vdso image. > >>> > >>>Renamed and moved text_mapping structure declaration inside > >>>map_vdso, as it used only there and now it complement > >>>vvar_mapping variable. > >>> > >>>There is still problem for remapping vdso in glibc applications: > >>>linker relocates addresses for syscalls on vdso page, so > >>>you need to relink with the new addresses. Or the next syscall > >>>through glibc may fail: > >>> Program received signal SIGSEGV, Segmentation fault. > >>> #0 0xf7fd9b80 in __kernel_vsyscall () > >>> #1 0xf7ec8238 in _exit () from /usr/lib32/libc.so.6 > >>> > >>>Signed-off-by: Dmitry Safonov <dsafonov@xxxxxxxxxxxxx> > >>>Acked-by: Andy Lutomirski <luto@xxxxxxxxxx> > >>>--- > >>>v8: add WARN_ON_ONCE on current->mm != new_vma->vm_mm > >>>v7: build fix > >>>v6: moved vdso_image_32 check and fixup code into vdso_fix_landing function > >>> with ifdefs around > >>>v5: as Andy suggested, add a check that new_vma->vm_mm and current->mm are > >>> the same, also check not only in_ia32_syscall() but image == &vdso_image_32 > >>>v4: drop __maybe_unused & use image from mm->context instead vdso_image_32 > >>>v3: as Andy suggested, return EINVAL in case of splitting vdso blob on mremap; > >>> used is_ia32_task instead of ifdefs > >>>v2: added __maybe_unused for pt_regs in vdso_mremap > >> > >>Ping? > > > >There's no 0/2 boilerplate explaining the background of the changes - why do you > >want to mremap() the vDSO? > > Thanks for the answer. > > Well, one could move vdso vma before this patch, but doing fast > syscalls through it will not work because of code relying on > mm->context.vdso pointer. > So all this code is just fixup for that pointer on moving. > (Also adds preventing for splitting vdso vma). > As Andy notted, vDSO mremap for !i386 tasks also worked only by a chance > before this patch. > > I need to move vdso vma in CRIU - on restore we need to choose it's > position: > - if vDSO blob of restoring application is the same as the kernel has, > we need to move it on the same place; > - if it differs, we need to choose place that wasn't tooken by other > vma of restoring application and add jump trampolines to it from the > place of vDSO in restoring application. > And CRIU code now relies on possibility on x86_64 to mremap vDSO. > Without this patch that may be broken in future. > And as I work on C/R of compatible 32-bit applications on x86_64, > I need this to work also for 32-bit vDSO. Which does not work, > because of pointer mentioned above. Ok, this looks useful - please add this information to the changelog (with typos fixed). Thanks, Ingo -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>