Hi Zhongjin, On Sat, Dec 03, 2022 at 05:45:27PM +0800, Chen Zhongjin wrote: > syzkaller reported a memleak: > https://syzkaller.appspot.com/bug?id=62f37ff612f0021641eda5b17f056f1668aa9aed > > unreferenced object 0xffff88811009c7f8 (size 136): > ... > backtrace: > [<ffffffff821db19b>] z_erofs_do_read_page+0x99b/0x1740 > [<ffffffff821dee9e>] z_erofs_readahead+0x24e/0x580 > [<ffffffff814bc0d6>] read_pages+0x86/0x3d0 > ... > > syzkaller constructed a case: in z_erofs_register_pcluster(), > ztailpacking = false and map->m_pa = zero. This makes pcl->obj.index > become zero although pcl is not an inline pcluster.
Thanks for the patch! We should just fail out if map->m_pa / EROFS_BLKSIZ == 0. > > Then following path adds refcount for grp, but the it won't be put > because pcl is inline, which makes pcl not released when shrink. > > z_erofs_readahead() > z_erofs_do_read_page() # for another page > z_erofs_collector_begin() > erofs_find_workgroup() > erofs_workgroup_get() > > To fix this, add an attribute in z_erofs_pcluster to mark the inline > state which not depends on index of grp. I think the main reason is "inline pcluster _always_ did memory leak before since I don't find any chance to these free inline pclusters in the current codebase. Actually I submitted a patch for this, could you check/review this if possible? https://lore.kernel.org/r/[email protected] Thanks, Gao Xiang > > Fixes: cecf864d3d76 ("erofs: support inline data decompression") > Reported-by: [email protected] > Signed-off-by: Chen Zhongjin <[email protected]> > --- > fs/erofs/zdata.c | 2 +- > fs/erofs/zdata.h | 5 ++++- > 2 files changed, 5 insertions(+), 2 deletions(-) > > diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c > index b792d424d774..fef2624d19e3 100644 > --- a/fs/erofs/zdata.c > +++ b/fs/erofs/zdata.c > @@ -517,7 +517,7 @@ static int z_erofs_register_pcluster(struct > z_erofs_decompress_frontend *fe) > DBG_BUGON(!mutex_trylock(&pcl->lock)); > > if (ztailpacking) { > - pcl->obj.index = 0; /* which indicates ztailpacking */ > + pcl->is_inline = true; /* which indicates ztailpacking */ > pcl->pageofs_in = erofs_blkoff(map->m_pa); > pcl->tailpacking_size = map->m_plen; > } else { > diff --git a/fs/erofs/zdata.h b/fs/erofs/zdata.h > index d98c95212985..35051ad27521 100644 > --- a/fs/erofs/zdata.h > +++ b/fs/erofs/zdata.h > @@ -78,6 +78,9 @@ struct z_erofs_pcluster { > unsigned short tailpacking_size; > }; > > + /* I: whether it is inline or not */ > + bool is_inline; > + > /* I: compression algorithm format */ > unsigned char algorithmformat; > > @@ -115,7 +118,7 @@ struct z_erofs_decompressqueue { > > static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) > { > - return !pcl->obj.index; > + return pcl->is_inline; > } > > static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster > *pcl) > -- > 2.17.1 >
