From: Zhihao Cheng <chengzhihao1@huawei.com> hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/IAWTIW CVE: NA -------------------------------- When an ext4 image is mounted with '-obuffered_iomap', the readahead and defragmentation on the same file may let readahead process miss restoring 'PF_MEMALLOC_NOFS' flag, which could fail subsequent memory allocations in the same process. Specifically: ioctl(fd, EXT4_IOC_MOVE_EXT) read(fd) page_cache_ra_order nofs = memalloc_nofs_save() // set 'PF_MEMALLOC_NOFS' for current ext4_move_extents ext4_disable_buffered_iomap_aops mapping_clear_large_folios if (unlikely(!mapping_large_folio_support(mapping))) goto fallback // Forget clearing 'PF_MEMALLOC_NOFS' Fix it by adding 'memalloc_nofs_restore' in 'fallback' branch. Fixes: 1325aef8d896 ("mm: use memalloc_nofs_save() in page_cache_ra_order()") Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> --- mm/readahead.c | 1 + 1 file changed, 1 insertion(+) diff --git a/mm/readahead.c b/mm/readahead.c index 12ae30a966ef..08a7490ef793 100644 --- a/mm/readahead.c +++ b/mm/readahead.c @@ -546,6 +546,7 @@ void page_cache_ra_order(struct readahead_control *ractl, if (unlikely(!mapping_large_folio_support(mapping))) { filemap_invalidate_unlock_shared(mapping); + memalloc_nofs_restore(nofs); goto fallback; } -- 2.39.2