From: Benjamin Berg <benjamin.b...@intel.com> Hi,
using mincore() to check whether a page is owned by UML is not correct as it returns whether the page is resident in memory and not whether something has been mapped at the address. This means that UML could get spurious failures in *_nofault functions like copy_from_kernel_nofault). If this happens, it can create hard to debug issues. For us the problem showed up because hostfs was randomly throwing ENOENT errors when loading libraries (including ld.so). This can be traced back to dentry_path_raw failing as it uses an optimization that requires the use of copy_from_kernel_nofault internally (see the documentation of prepend_name and prepend_copy). Pull in the existing RODATA patch from Johannes (with light editing) and then remove the call to mincore() as it is incorrect. Benjamin Benjamin Berg (1): um: remove copy_from_kernel_nofault_allowed Johannes Berg (1): um: mark rodata read-only and implement _nofault accesses arch/um/Kconfig | 1 + arch/um/include/asm/processor-generic.h | 2 + arch/um/include/asm/uaccess.h | 20 +++++++--- arch/um/include/shared/arch.h | 2 + arch/um/include/shared/as-layout.h | 2 +- arch/um/include/shared/irq_user.h | 3 +- arch/um/include/shared/kern_util.h | 12 ++++-- arch/um/include/shared/os.h | 1 - arch/um/kernel/Makefile | 2 +- arch/um/kernel/irq.c | 3 +- arch/um/kernel/maccess.c | 19 --------- arch/um/kernel/mem.c | 10 +++++ arch/um/kernel/trap.c | 28 ++++++++++--- arch/um/os-Linux/process.c | 51 ------------------------ arch/um/os-Linux/signal.c | 4 +- arch/um/os-Linux/skas/process.c | 8 ++-- arch/x86/um/os-Linux/mcontext.c | 12 ++++++ arch/x86/um/shared/sysdep/faultinfo_32.h | 12 ++++++ arch/x86/um/shared/sysdep/faultinfo_64.h | 12 ++++++ 19 files changed, 109 insertions(+), 95 deletions(-) delete mode 100644 arch/um/kernel/maccess.c -- 2.48.1
