userfaultfd: mcopy_atomic|mfill_zeropage: UFFDIO_COPY|UFFDIO_ZEROPAGE preparation
authorAndrea Arcangeli <aarcange@redhat.com>
Fri, 4 Sep 2015 22:47:04 +0000 (15:47 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 4 Sep 2015 23:54:41 +0000 (16:54 -0700)
This implements mcopy_atomic and mfill_zeropage that are the lowlevel
VM methods that are invoked respectively by the UFFDIO_COPY and
UFFDIO_ZEROPAGE userfaultfd commands.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Sanidhya Kashyap <sanidhya.gatech@gmail.com>
Cc: zhang.zhanghailiang@huawei.com
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Feiner <pfeiner@google.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Huangpeng (Peter)" <peter.huangpeng@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/userfaultfd_k.h
mm/Makefile
mm/userfaultfd.c [new file with mode: 0644]

index e1e43609a179a3f425b2f8d70d86d7b348816e22..587480ad41b7b10c8dc9e71d921665ffddfb2c39 100644 (file)
 extern int handle_userfault(struct vm_area_struct *vma, unsigned long address,
                            unsigned int flags, unsigned long reason);
 
+extern ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start,
+                           unsigned long src_start, unsigned long len);
+extern ssize_t mfill_zeropage(struct mm_struct *dst_mm,
+                             unsigned long dst_start,
+                             unsigned long len);
+
 /* mm helpers */
 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
                                        struct vm_userfaultfd_ctx vm_ctx)
index 98c4eaeabdcb4592d0b15fd71f3f520aa1bd3d35..b424d5e5b6ff5b1dec8f95fdd089451dba4dd19c 100644 (file)
@@ -78,3 +78,4 @@ obj-$(CONFIG_CMA)     += cma.o
 obj-$(CONFIG_MEMORY_BALLOON) += balloon_compaction.o
 obj-$(CONFIG_PAGE_EXTENSION) += page_ext.o
 obj-$(CONFIG_CMA_DEBUGFS) += cma_debug.o
+obj-$(CONFIG_USERFAULTFD) += userfaultfd.o
diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
new file mode 100644 (file)
index 0000000..c54c761
--- /dev/null
@@ -0,0 +1,269 @@
+/*
+ *  mm/userfaultfd.c
+ *
+ *  Copyright (C) 2015  Red Hat, Inc.
+ *
+ *  This work is licensed under the terms of the GNU GPL, version 2. See
+ *  the COPYING file in the top-level directory.
+ */
+
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/rmap.h>
+#include <linux/swap.h>
+#include <linux/swapops.h>
+#include <linux/userfaultfd_k.h>
+#include <linux/mmu_notifier.h>
+#include <asm/tlbflush.h>
+#include "internal.h"
+
+static int mcopy_atomic_pte(struct mm_struct *dst_mm,
+                           pmd_t *dst_pmd,
+                           struct vm_area_struct *dst_vma,
+                           unsigned long dst_addr,
+                           unsigned long src_addr)
+{
+       struct mem_cgroup *memcg;
+       pte_t _dst_pte, *dst_pte;
+       spinlock_t *ptl;
+       struct page *page;
+       void *page_kaddr;
+       int ret;
+
+       ret = -ENOMEM;
+       page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, dst_vma, dst_addr);
+       if (!page)
+               goto out;
+
+       page_kaddr = kmap(page);
+       ret = -EFAULT;
+       if (copy_from_user(page_kaddr, (const void __user *) src_addr,
+                          PAGE_SIZE))
+               goto out_kunmap_release;
+       kunmap(page);
+
+       /*
+        * The memory barrier inside __SetPageUptodate makes sure that
+        * preceeding stores to the page contents become visible before
+        * the set_pte_at() write.
+        */
+       __SetPageUptodate(page);
+
+       ret = -ENOMEM;
+       if (mem_cgroup_try_charge(page, dst_mm, GFP_KERNEL, &memcg))
+               goto out_release;
+
+       _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
+       if (dst_vma->vm_flags & VM_WRITE)
+               _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
+
+       ret = -EEXIST;
+       dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
+       if (!pte_none(*dst_pte))
+               goto out_release_uncharge_unlock;
+
+       inc_mm_counter(dst_mm, MM_ANONPAGES);
+       page_add_new_anon_rmap(page, dst_vma, dst_addr);
+       mem_cgroup_commit_charge(page, memcg, false);
+       lru_cache_add_active_or_unevictable(page, dst_vma);
+
+       set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+
+       /* No need to invalidate - it was non-present before */
+       update_mmu_cache(dst_vma, dst_addr, dst_pte);
+
+       pte_unmap_unlock(dst_pte, ptl);
+       ret = 0;
+out:
+       return ret;
+out_release_uncharge_unlock:
+       pte_unmap_unlock(dst_pte, ptl);
+       mem_cgroup_cancel_charge(page, memcg);
+out_release:
+       page_cache_release(page);
+       goto out;
+out_kunmap_release:
+       kunmap(page);
+       goto out_release;
+}
+
+static int mfill_zeropage_pte(struct mm_struct *dst_mm,
+                             pmd_t *dst_pmd,
+                             struct vm_area_struct *dst_vma,
+                             unsigned long dst_addr)
+{
+       pte_t _dst_pte, *dst_pte;
+       spinlock_t *ptl;
+       int ret;
+
+       _dst_pte = pte_mkspecial(pfn_pte(my_zero_pfn(dst_addr),
+                                        dst_vma->vm_page_prot));
+       ret = -EEXIST;
+       dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
+       if (!pte_none(*dst_pte))
+               goto out_unlock;
+       set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+       /* No need to invalidate - it was non-present before */
+       update_mmu_cache(dst_vma, dst_addr, dst_pte);
+       ret = 0;
+out_unlock:
+       pte_unmap_unlock(dst_pte, ptl);
+       return ret;
+}
+
+static pmd_t *mm_alloc_pmd(struct mm_struct *mm, unsigned long address)
+{
+       pgd_t *pgd;
+       pud_t *pud;
+       pmd_t *pmd = NULL;
+
+       pgd = pgd_offset(mm, address);
+       pud = pud_alloc(mm, pgd, address);
+       if (pud)
+               /*
+                * Note that we didn't run this because the pmd was
+                * missing, the *pmd may be already established and in
+                * turn it may also be a trans_huge_pmd.
+                */
+               pmd = pmd_alloc(mm, pud, address);
+       return pmd;
+}
+
+static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,
+                                             unsigned long dst_start,
+                                             unsigned long src_start,
+                                             unsigned long len,
+                                             bool zeropage)
+{
+       struct vm_area_struct *dst_vma;
+       ssize_t err;
+       pmd_t *dst_pmd;
+       unsigned long src_addr, dst_addr;
+       long copied = 0;
+
+       /*
+        * Sanitize the command parameters:
+        */
+       BUG_ON(dst_start & ~PAGE_MASK);
+       BUG_ON(len & ~PAGE_MASK);
+
+       /* Does the address range wrap, or is the span zero-sized? */
+       BUG_ON(src_start + len <= src_start);
+       BUG_ON(dst_start + len <= dst_start);
+
+       down_read(&dst_mm->mmap_sem);
+
+       /*
+        * Make sure the vma is not shared, that the dst range is
+        * both valid and fully within a single existing vma.
+        */
+       err = -EINVAL;
+       dst_vma = find_vma(dst_mm, dst_start);
+       if (!dst_vma || (dst_vma->vm_flags & VM_SHARED))
+               goto out;
+       if (dst_start < dst_vma->vm_start ||
+           dst_start + len > dst_vma->vm_end)
+               goto out;
+
+       /*
+        * Be strict and only allow __mcopy_atomic on userfaultfd
+        * registered ranges to prevent userland errors going
+        * unnoticed. As far as the VM consistency is concerned, it
+        * would be perfectly safe to remove this check, but there's
+        * no useful usage for __mcopy_atomic ouside of userfaultfd
+        * registered ranges. This is after all why these are ioctls
+        * belonging to the userfaultfd and not syscalls.
+        */
+       if (!dst_vma->vm_userfaultfd_ctx.ctx)
+               goto out;
+
+       /*
+        * FIXME: only allow copying on anonymous vmas, tmpfs should
+        * be added.
+        */
+       if (dst_vma->vm_ops)
+               goto out;
+
+       /*
+        * Ensure the dst_vma has a anon_vma or this page
+        * would get a NULL anon_vma when moved in the
+        * dst_vma.
+        */
+       err = -ENOMEM;
+       if (unlikely(anon_vma_prepare(dst_vma)))
+               goto out;
+
+       for (src_addr = src_start, dst_addr = dst_start;
+            src_addr < src_start + len; ) {
+               pmd_t dst_pmdval;
+               BUG_ON(dst_addr >= dst_start + len);
+               dst_pmd = mm_alloc_pmd(dst_mm, dst_addr);
+               if (unlikely(!dst_pmd)) {
+                       err = -ENOMEM;
+                       break;
+               }
+
+               dst_pmdval = pmd_read_atomic(dst_pmd);
+               /*
+                * If the dst_pmd is mapped as THP don't
+                * override it and just be strict.
+                */
+               if (unlikely(pmd_trans_huge(dst_pmdval))) {
+                       err = -EEXIST;
+                       break;
+               }
+               if (unlikely(pmd_none(dst_pmdval)) &&
+                   unlikely(__pte_alloc(dst_mm, dst_vma, dst_pmd,
+                                        dst_addr))) {
+                       err = -ENOMEM;
+                       break;
+               }
+               /* If an huge pmd materialized from under us fail */
+               if (unlikely(pmd_trans_huge(*dst_pmd))) {
+                       err = -EFAULT;
+                       break;
+               }
+
+               BUG_ON(pmd_none(*dst_pmd));
+               BUG_ON(pmd_trans_huge(*dst_pmd));
+
+               if (!zeropage)
+                       err = mcopy_atomic_pte(dst_mm, dst_pmd, dst_vma,
+                                              dst_addr, src_addr);
+               else
+                       err = mfill_zeropage_pte(dst_mm, dst_pmd, dst_vma,
+                                                dst_addr);
+
+               cond_resched();
+
+               if (!err) {
+                       dst_addr += PAGE_SIZE;
+                       src_addr += PAGE_SIZE;
+                       copied += PAGE_SIZE;
+
+                       if (fatal_signal_pending(current))
+                               err = -EINTR;
+               }
+               if (err)
+                       break;
+       }
+
+out:
+       up_read(&dst_mm->mmap_sem);
+       BUG_ON(copied < 0);
+       BUG_ON(err > 0);
+       BUG_ON(!copied && !err);
+       return copied ? copied : err;
+}
+
+ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start,
+                    unsigned long src_start, unsigned long len)
+{
+       return __mcopy_atomic(dst_mm, dst_start, src_start, len, false);
+}
+
+ssize_t mfill_zeropage(struct mm_struct *dst_mm, unsigned long start,
+                      unsigned long len)
+{
+       return __mcopy_atomic(dst_mm, start, 0, len, true);
+}