fs/proc/vmcore.c:mmap_vmcore: skip non-ram pages reported by hypervisors
authorVitaly Kuznetsov <vkuznets@redhat.com>
Fri, 8 Aug 2014 21:22:05 +0000 (14:22 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 8 Aug 2014 22:57:23 +0000 (15:57 -0700)
We have a special check in read_vmcore() handler to check if the page was
reported as ram or not by the hypervisor (pfn_is_ram()).  However, when
vmcore is read with mmap() no such check is performed.  That can lead to
unpredictable results, e.g.  when running Xen PVHVM guest memcpy() after
mmap() on /proc/vmcore will hang processing HVMMEM_mmio_dm pages creating
enormous load in both DomU and Dom0.

Fix the issue by mapping each non-ram page to the zero page.  Keep direct
path with remap_oldmem_pfn_range() to avoid looping through all pages on
bare metal.

The issue can also be solved by overriding remap_oldmem_pfn_range() in
xen-specific code, as remap_oldmem_pfn_range() was been designed for.
That, however, would involve non-obvious xen code path for all x86 builds
with CONFIG_XEN_PVHVM=y and would prevent all other hypervisor-specific
code on x86 arch from doing the same override.

[fengguang.wu@intel.com: remap_oldmem_pfn_checked() can be static]
[akpm@linux-foundation.org: clean up layout]
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Cc: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/vmcore.c

index 382aa890e228cdaeda6a9038fad01c9474d20eda..a90d6d3541992552d30d2b4a2bac062d352f576d 100644 (file)
@@ -328,6 +328,82 @@ static inline char *alloc_elfnotes_buf(size_t notes_sz)
  * virtually contiguous user-space in ELF layout.
  */
 #ifdef CONFIG_MMU
+/*
+ * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
+ * reported as not being ram with the zero page.
+ *
+ * @vma: vm_area_struct describing requested mapping
+ * @from: start remapping from
+ * @pfn: page frame number to start remapping to
+ * @size: remapping size
+ * @prot: protection bits
+ *
+ * Returns zero on success, -EAGAIN on failure.
+ */
+static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
+                                   unsigned long from, unsigned long pfn,
+                                   unsigned long size, pgprot_t prot)
+{
+       unsigned long map_size;
+       unsigned long pos_start, pos_end, pos;
+       unsigned long zeropage_pfn = my_zero_pfn(0);
+       size_t len = 0;
+
+       pos_start = pfn;
+       pos_end = pfn + (size >> PAGE_SHIFT);
+
+       for (pos = pos_start; pos < pos_end; ++pos) {
+               if (!pfn_is_ram(pos)) {
+                       /*
+                        * We hit a page which is not ram. Remap the continuous
+                        * region between pos_start and pos-1 and replace
+                        * the non-ram page at pos with the zero page.
+                        */
+                       if (pos > pos_start) {
+                               /* Remap continuous region */
+                               map_size = (pos - pos_start) << PAGE_SHIFT;
+                               if (remap_oldmem_pfn_range(vma, from + len,
+                                                          pos_start, map_size,
+                                                          prot))
+                                       goto fail;
+                               len += map_size;
+                       }
+                       /* Remap the zero page */
+                       if (remap_oldmem_pfn_range(vma, from + len,
+                                                  zeropage_pfn,
+                                                  PAGE_SIZE, prot))
+                               goto fail;
+                       len += PAGE_SIZE;
+                       pos_start = pos + 1;
+               }
+       }
+       if (pos > pos_start) {
+               /* Remap the rest */
+               map_size = (pos - pos_start) << PAGE_SHIFT;
+               if (remap_oldmem_pfn_range(vma, from + len, pos_start,
+                                          map_size, prot))
+                       goto fail;
+       }
+       return 0;
+fail:
+       do_munmap(vma->vm_mm, from, len);
+       return -EAGAIN;
+}
+
+static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
+                           unsigned long from, unsigned long pfn,
+                           unsigned long size, pgprot_t prot)
+{
+       /*
+        * Check if oldmem_pfn_is_ram was registered to avoid
+        * looping over all pages without a reason.
+        */
+       if (oldmem_pfn_is_ram)
+               return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
+       else
+               return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
+}
+
 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 {
        size_t size = vma->vm_end - vma->vm_start;
@@ -387,9 +463,9 @@ static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 
                        tsz = min_t(size_t, m->offset + m->size - start, size);
                        paddr = m->paddr + start - m->offset;
-                       if (remap_oldmem_pfn_range(vma, vma->vm_start + len,
-                                                  paddr >> PAGE_SHIFT, tsz,
-                                                  vma->vm_page_prot))
+                       if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
+                                                   paddr >> PAGE_SHIFT, tsz,
+                                                   vma->vm_page_prot))
                                goto fail;
                        size -= tsz;
                        start += tsz;