dax: add huge page fault support
authorMatthew Wilcox <willy@linux.intel.com>
Tue, 8 Sep 2015 21:58:57 +0000 (14:58 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 8 Sep 2015 22:35:28 +0000 (15:35 -0700)
This is the support code for DAX-enabled filesystems to allow them to
provide huge pages in response to faults.

Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/filesystems/dax.txt
fs/dax.c
include/linux/dax.h

index 7af2851..7bde640 100644 (file)
@@ -60,9 +60,10 @@ Filesystem support consists of
 - implementing the direct_IO address space operation, and calling
   dax_do_io() instead of blockdev_direct_IO() if S_DAX is set
 - implementing an mmap file operation for DAX files which sets the
-  VM_MIXEDMAP flag on the VMA, and setting the vm_ops to include handlers
-  for fault and page_mkwrite (which should probably call dax_fault() and
-  dax_mkwrite(), passing the appropriate get_block() callback)
+  VM_MIXEDMAP and VM_HUGEPAGE flags on the VMA, and setting the vm_ops to
+  include handlers for fault, pmd_fault and page_mkwrite (which should
+  probably call dax_fault(), dax_pmd_fault() and dax_mkwrite(), passing the
+  appropriate get_block() callback)
 - calling dax_truncate_page() instead of block_truncate_page() for DAX files
 - calling dax_zero_page_range() instead of zero_user() for DAX files
 - ensuring that there is sufficient locking between reads, writes,
index a7f77e1..15f8ffc 100644 (file)
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -494,6 +494,158 @@ int dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf,
 }
 EXPORT_SYMBOL_GPL(dax_fault);
 
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+/*
+ * The 'colour' (ie low bits) within a PMD of a page offset.  This comes up
+ * more often than one might expect in the below function.
+ */
+#define PG_PMD_COLOUR  ((PMD_SIZE >> PAGE_SHIFT) - 1)
+
+int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address,
+               pmd_t *pmd, unsigned int flags, get_block_t get_block,
+               dax_iodone_t complete_unwritten)
+{
+       struct file *file = vma->vm_file;
+       struct address_space *mapping = file->f_mapping;
+       struct inode *inode = mapping->host;
+       struct buffer_head bh;
+       unsigned blkbits = inode->i_blkbits;
+       unsigned long pmd_addr = address & PMD_MASK;
+       bool write = flags & FAULT_FLAG_WRITE;
+       long length;
+       void *kaddr;
+       pgoff_t size, pgoff;
+       sector_t block, sector;
+       unsigned long pfn;
+       int result = 0;
+
+       /* Fall back to PTEs if we're going to COW */
+       if (write && !(vma->vm_flags & VM_SHARED))
+               return VM_FAULT_FALLBACK;
+       /* If the PMD would extend outside the VMA */
+       if (pmd_addr < vma->vm_start)
+               return VM_FAULT_FALLBACK;
+       if ((pmd_addr + PMD_SIZE) > vma->vm_end)
+               return VM_FAULT_FALLBACK;
+
+       pgoff = ((pmd_addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
+       size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       if (pgoff >= size)
+               return VM_FAULT_SIGBUS;
+       /* If the PMD would cover blocks out of the file */
+       if ((pgoff | PG_PMD_COLOUR) >= size)
+               return VM_FAULT_FALLBACK;
+
+       memset(&bh, 0, sizeof(bh));
+       block = (sector_t)pgoff << (PAGE_SHIFT - blkbits);
+
+       bh.b_size = PMD_SIZE;
+       length = get_block(inode, block, &bh, write);
+       if (length)
+               return VM_FAULT_SIGBUS;
+       i_mmap_lock_read(mapping);
+
+       /*
+        * If the filesystem isn't willing to tell us the length of a hole,
+        * just fall back to PTEs.  Calling get_block 512 times in a loop
+        * would be silly.
+        */
+       if (!buffer_size_valid(&bh) || bh.b_size < PMD_SIZE)
+               goto fallback;
+
+       /* Guard against a race with truncate */
+       size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       if (pgoff >= size) {
+               result = VM_FAULT_SIGBUS;
+               goto out;
+       }
+       if ((pgoff | PG_PMD_COLOUR) >= size)
+               goto fallback;
+
+       if (is_huge_zero_pmd(*pmd))
+               unmap_mapping_range(mapping, pgoff << PAGE_SHIFT, PMD_SIZE, 0);
+
+       if (!write && !buffer_mapped(&bh) && buffer_uptodate(&bh)) {
+               bool set;
+               spinlock_t *ptl;
+               struct mm_struct *mm = vma->vm_mm;
+               struct page *zero_page = get_huge_zero_page();
+               if (unlikely(!zero_page))
+                       goto fallback;
+
+               ptl = pmd_lock(mm, pmd);
+               set = set_huge_zero_page(NULL, mm, vma, pmd_addr, pmd,
+                                                               zero_page);
+               spin_unlock(ptl);
+               result = VM_FAULT_NOPAGE;
+       } else {
+               sector = bh.b_blocknr << (blkbits - 9);
+               length = bdev_direct_access(bh.b_bdev, sector, &kaddr, &pfn,
+                                               bh.b_size);
+               if (length < 0) {
+                       result = VM_FAULT_SIGBUS;
+                       goto out;
+               }
+               if ((length < PMD_SIZE) || (pfn & PG_PMD_COLOUR))
+                       goto fallback;
+
+               if (buffer_unwritten(&bh) || buffer_new(&bh)) {
+                       int i;
+                       for (i = 0; i < PTRS_PER_PMD; i++)
+                               clear_page(kaddr + i * PAGE_SIZE);
+                       count_vm_event(PGMAJFAULT);
+                       mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
+                       result |= VM_FAULT_MAJOR;
+               }
+
+               result |= vmf_insert_pfn_pmd(vma, address, pmd, pfn, write);
+       }
+
+ out:
+       i_mmap_unlock_read(mapping);
+
+       if (buffer_unwritten(&bh))
+               complete_unwritten(&bh, !(result & VM_FAULT_ERROR));
+
+       return result;
+
+ fallback:
+       count_vm_event(THP_FAULT_FALLBACK);
+       result = VM_FAULT_FALLBACK;
+       goto out;
+}
+EXPORT_SYMBOL_GPL(__dax_pmd_fault);
+
+/**
+ * dax_pmd_fault - handle a PMD fault on a DAX file
+ * @vma: The virtual memory area where the fault occurred
+ * @vmf: The description of the fault
+ * @get_block: The filesystem method used to translate file offsets to blocks
+ *
+ * When a page fault occurs, filesystems may call this helper in their
+ * pmd_fault handler for DAX files.
+ */
+int dax_pmd_fault(struct vm_area_struct *vma, unsigned long address,
+                       pmd_t *pmd, unsigned int flags, get_block_t get_block,
+                       dax_iodone_t complete_unwritten)
+{
+       int result;
+       struct super_block *sb = file_inode(vma->vm_file)->i_sb;
+
+       if (flags & FAULT_FLAG_WRITE) {
+               sb_start_pagefault(sb);
+               file_update_time(vma->vm_file);
+       }
+       result = __dax_pmd_fault(vma, address, pmd, flags, get_block,
+                               complete_unwritten);
+       if (flags & FAULT_FLAG_WRITE)
+               sb_end_pagefault(sb);
+
+       return result;
+}
+EXPORT_SYMBOL_GPL(dax_pmd_fault);
+#endif /* CONFIG_TRANSPARENT_HUGEPAGES */
+
 /**
  * dax_pfn_mkwrite - handle first write to DAX page
  * @vma: The virtual memory area where the fault occurred
index 9b51f9d..b415e52 100644 (file)
@@ -14,6 +14,20 @@ int dax_fault(struct vm_area_struct *, struct vm_fault *, get_block_t,
                dax_iodone_t);
 int __dax_fault(struct vm_area_struct *, struct vm_fault *, get_block_t,
                dax_iodone_t);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+int dax_pmd_fault(struct vm_area_struct *, unsigned long addr, pmd_t *,
+                               unsigned int flags, get_block_t, dax_iodone_t);
+int __dax_pmd_fault(struct vm_area_struct *, unsigned long addr, pmd_t *,
+                               unsigned int flags, get_block_t, dax_iodone_t);
+#else
+static inline int dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
+                               pmd_t *pmd, unsigned int flags, get_block_t gb,
+                               dax_iodone_t di)
+{
+       return VM_FAULT_FALLBACK;
+}
+#define __dax_pmd_fault dax_pmd_fault
+#endif
 int dax_pfn_mkwrite(struct vm_area_struct *, struct vm_fault *);
 #define dax_mkwrite(vma, vmf, gb, iod)         dax_fault(vma, vmf, gb, iod)
 #define __dax_mkwrite(vma, vmf, gb, iod)       __dax_fault(vma, vmf, gb, iod)