[bug report] mpage: Convert do_mpage_readpage() to use a folio

[Dan Carpenter <dan.carpenter@oracle.com>]

Hello Matthew Wilcox (Oracle),

The patch 4c27dc762d7b: "mpage: Convert do_mpage_readpage() to use a
folio" from Jun 10, 2022, leads to the following Smatch static
checker warning:

	fs/mpage.c:102 map_buffer_to_folio()
	error: we previously assumed 'page_bh' could be null (see line 99)

fs/mpage.c
    78 static void map_buffer_to_folio(struct folio *folio, struct buffer_head *bh,
    79                 int page_block)
    80 {
    81         struct inode *inode = folio->mapping->host;
    82         struct buffer_head *page_bh, *head;
    83         int block = 0;
    84 
    85         head = folio_buffers(folio);
    86         if (!head) {
    87                 /*
    88                  * don't make any buffers if there is only one buffer on
    89                  * the folio and the folio just needs to be set up to date
    90                  */
    91                 if (inode->i_blkbits == PAGE_SHIFT &&
    92                     buffer_uptodate(bh)) {
    93                         folio_mark_uptodate(folio);
    94                         return;
    95                 }
    96                 create_empty_buffers(&folio->page, i_blocksize(inode), 0);
    97         }

Originally there was a "head = page_buffers(page);" here but now head
is left as NULL.

    98 
    99         page_bh = head;
    100         do {
    101                 if (block == page_block) {
--> 102                         page_bh->b_state = bh->b_state;
    103                         page_bh->b_bdev = bh->b_bdev;
    104                         page_bh->b_blocknr = bh->b_blocknr;
    105                         break;
    106                 }
    107                 page_bh = page_bh->b_this_page;
                                  ^^^^^^^^^
Which will crash.

    108                 block++;
    109         } while (page_bh != head);
    110 }

regards,
dan carpenter