On Sat, 2007-06-16 at 12:14 +0400, Dmitriy Monakhov wrote:
> On 16:16 Птн 15 Июн , Mingming Cao wrote:
> > I hit almost the same issue today also, but with different error #, and
> > one more kernel oops, when run fsstress on x86_64.
> >
> > EXT4-fs: writeback error = -2
> > EXT4-fs: writeback error = -2
> This error never happens in writeback in my case, only ENOSPC.
I see why, error -2 is ENOENT, your kernel already patched with your
previus patch to fix the return error from ext4_reserve_global() from
ENOENT to ENOSPC.
> > > > I've digged this a litle bit with folowig results:
> > > >
> > > > int block_read_full_page(struct page *page, get_block_t *get_block)
> > > > {
> > > > ...
> > > > 1914: if (!page_has_buffers(page)) <<< page_has_buffers(page) == true
> > > > create_empty_buffers(page, blocksize, 0);
> > > > head = page_buffers(page); <<<< page_buffers(page) == NULL
> > > > <<<i've add debug info here:
> > > > <<< page->flags == 100000000000821
> > > > <<< PagePrivate(page) == 1, (page)->private == NULL
> > > > <<< So we have private page without buffers, it is WRONG.
> > > >
Right, I think the problem is the PagePrivate() is being set without a
valid page->private pointer.
The PagePrivate flag is being set by delayed allocation in the
ext4_wb_commit_write().
Looking at ext4-delayed-allocation.patch
+int ext4_wb_commit_write(struct file *file, struct page *page,
+ unsigned from, unsigned to)
+{
+ loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
+ struct inode *inode = page->mapping->host;
+ int err = 0;
+
+ wb_debug("commit page %lu (%u-%u) for inode %lu\n",
+ page->index, from, to, inode->i_ino);
+
+ /* mark page private so that we get
+ * called to invalidate/release page */
+ SetPagePrivate(page);
+
+ if (!PageBooked(page) && !PageMappedToDisk(page)) {
+ /* ->prepare_write() observed that block for this
+ * page hasn't been allocated yet. there fore it
+ * asked to reserve block for later allocation */
+ BUG_ON(page->private == 0);
+ page->private = 0;
+ err = ext4_wb_reserve_space_page(page, 1);
+ if (err)
+ return err;
+ }
It sets the PagePrivate flag at the commit_write() time. And the page-
>private is cleared to 0. page->private is being set to 1 at
ext4_wb_prepare_write():
+int ext4_wb_prepare_write(struct file *file, struct page *page,
+ unsigned from, unsigned to)
+{
+ struct inode *inode = page->mapping->host;
+ struct buffer_head bh, *bhw = &bh;
+ int err = 0;
+
+ wb_debug("prepare page %lu (%u-%u) for inode %lu\n",
+ page->index, from, to, page->mapping->host->i_ino);
+
+ /* if page is uptodate this means that ->prepare_write() has
+ * been called on page before and page is mapped to disk or
+ * we did reservation. page is protected and nobody can
+ * access it. hence, it safe to use page->private to pass
+ * flag that ->commit_write() has to reserve blocks. because
+ * an error may occur after ->prepare_write() we should not
+ * reserve block here. it's better to do in ->commit_write()
+ * when we're sure page is to be written */
+ page->private = 0;
+ if (!PageUptodate(page)) {
+ /* first write to this page */
+ bh.b_state = 0;
+ err = ext4_get_block(inode, page->index, bhw, 0);
+ if (err)
+ return err;
+ if (!buffer_mapped(bhw)) {
+ /* this block isn't allocated yet, reserve space */
+ wb_debug("reserve space for new block\n");
+ page->private = 1;
+ ext4_wb_clear_page(page, from, to);
+ ClearPageMappedToDisk(page);
>From the comments it says the page->private is set to 1 to letting
commit_write know that it needs block reservation, but I don't see the
page->private value being checked in ext4_wb_commit_write(). Instead,
the PageMappedToDisk(page) flag is being checked.
Alex, can you clarify the use of page->private and PagePrivate flag
here? Do we still need the page->private for delayed allocation, with
PageBooked flag and PageMappedToDisk page flag?
Thanks,
Mingming
-
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
