On Mon, Jan 12, 2015 at 03:09:52PM -0800, Andrew Morton wrote:
> On Fri, 24 Oct 2014 17:20:40 -0400 Matthew Wilcox <matthew.r.wilcox@xxxxxxxxx> wrote:
>
> > Instead of calling aops->get_xip_mem from the fault handler, the
> > filesystem passes a get_block_t that is used to find the appropriate
> > blocks.
> >
> > ...
> >
> > +static int copy_user_bh(struct page *to, struct buffer_head *bh,
> > + unsigned blkbits, unsigned long vaddr)
> > +{
> > + void *vfrom, *vto;
> > + if (dax_get_addr(bh, &vfrom, blkbits) < 0)
> > + return -EIO;
> > + vto = kmap_atomic(to);
> > + copy_user_page(vto, vfrom, vaddr, to);
> > + kunmap_atomic(vto);
>
> Again, please check the cache-flush aspects. copy_user_page() appears
> to be reponsible for handling coherency issues on the destination
> vaddr, but what about *vto?
vto is a new kernel address ... if there's any dirty data for that
address, it should have been flushed by the prior kunmap_atomic(), right?
> > + mutex_lock(&mapping->i_mmap_mutex);
> > +
> > + /*
> > + * Check truncate didn't happen while we were allocating a block.
> > + * If it did, this block may or may not be still allocated to the
> > + * file. We can't tell the filesystem to free it because we can't
> > + * take i_mutex here.
>
> (what's preventing us from taking i_mutex?)
We're in a page fault handler, and we may already be holding i_mutex.
We're definitely holding mmap_sem, and to quote from mm/rmap.c:
/*
* Lock ordering in mm:
*
* inode->i_mutex (while writing or truncating, not reading or faulting)
* mm->mmap_sem
> > In the worst case, the file still has blocks
> > + * allocated past the end of the file.
> > + */
> > + size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
> > + if (unlikely(vmf->pgoff >= size)) {
> > + error = -EIO;
> > + goto out;
> > + }
>
> How does this play with holepunching? Checking i_size won't work there?
It doesn't. But the same problem exists with non-DAX files too, and
when I pointed it out, it was met with a shrug from the crowd. I saw a
patch series just recently that fixes it for XFS, but as far as I know,
btrfs and ext4 still don't play well with pagefault vs hole-punch races.
> > + memset(&bh, 0, sizeof(bh));
> > + block = (sector_t)vmf->pgoff << (PAGE_SHIFT - blkbits);
> > + bh.b_size = PAGE_SIZE;
>
> ah, there.
>
> PAGE_SIZE varies a lot between architectures. What are the
> implications of this>?
At the moment, you can only do DAX for blocksizes that are equal to
PAGE_SIZE. That's a restriction that existed for the previous XIP code,
and I haven't fixed it all for DAX yet. I'd like to, but it's not high on
my list of things to fix. Since these are in-mmeory filesystems, there's
not likely to be high demand to move the filesystem between machines.
> > + repeat:
> > + page = find_get_page(mapping, vmf->pgoff);
> > + if (page) {
> > + if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
> > + page_cache_release(page);
> > + return VM_FAULT_RETRY;
> > + }
> > + if (unlikely(page->mapping != mapping)) {
> > + unlock_page(page);
> > + page_cache_release(page);
> > + goto repeat;
> > + }
> > + size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
> > + if (unlikely(vmf->pgoff >= size)) {
> > + error = -EIO;
>
> What happened when this happens?
This case is where we have a struct page covering a hole in the file from
a read fault and we've raced with a truncate. It's basically the same code
that's in filemap_fault().
> > + goto unlock_page;
> > + }
> > + }
> > +
> > + error = get_block(inode, block, &bh, 0);
> > + if (!error && (bh.b_size < PAGE_SIZE))
> > + error = -EIO;
>
> How could this happen?
The only way I can think of is if the filesystem was corrupted. But it's
worth programming defensively, no?
--
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>