On Fri, Jan 26, 2024 at 04:33:40PM +0000, Matthew Wilcox wrote: > > +static inline bool > > +xfile_has_lost_data( > > + struct inode *inode, > > + struct folio *folio) > > +{ > > + struct address_space *mapping = inode->i_mapping; > > + > > + /* This folio itself has been poisoned. */ > > + if (folio_test_hwpoison(folio)) > > + return true; > > + > > + /* A base page under this large folio has been poisoned. */ > > + if (folio_test_large(folio) && folio_test_has_hwpoisoned(folio)) > > + return true; > > + > > + /* Data loss has occurred anywhere in this shmem file. */ > > + if (test_bit(AS_EIO, &mapping->flags)) > > + return true; > > + if (filemap_check_wb_err(mapping, 0)) > > + return true; > > + > > + return false; > > +} > > This is too much. filemap_check_wb_err() will do just fine for your > needs unless you really want to get fine-grained and perhaps try to > reconstruct the contents of the file. As in only call filemap_check_wb_err and do away with all the hwpoisoned checks and the extra AS_EIO check?