CVE-2025-68356

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> gfs2: Prevent recursive memory reclaim<br /> <br /> Function new_inode() returns a new inode with inode-&gt;i_mapping-&gt;gfp_mask<br /> set to GFP_HIGHUSER_MOVABLE. This value includes the __GFP_FS flag, so<br /> allocations in that address space can recurse into filesystem memory<br /> reclaim. We don&amp;#39;t want that to happen because it can consume a<br /> significant amount of stack memory.<br /> <br /> Worse than that is that it can also deadlock: for example, in several<br /> places, gfs2_unstuff_dinode() is called inside filesystem transactions.<br /> This calls filemap_grab_folio(), which can allocate a new folio, which<br /> can trigger memory reclaim. If memory reclaim recurses into the<br /> filesystem and starts another transaction, a deadlock will ensue.<br /> <br /> To fix these kinds of problems, prevent memory reclaim from recursing<br /> into filesystem code by making sure that the gfp_mask of inode address<br /> spaces doesn&amp;#39;t include __GFP_FS.<br /> <br /> The "meta" and resource group address spaces were already using GFP_NOFS<br /> as their gfp_mask (which doesn&amp;#39;t include __GFP_FS). The default value<br /> of GFP_HIGHUSER_MOVABLE is less restrictive than GFP_NOFS, though. To<br /> avoid being overly limiting, use the default value and only knock off<br /> the __GFP_FS flag. I&amp;#39;m not sure if this will actually make a<br /> difference, but it also shouldn&amp;#39;t hurt.<br /> <br /> This patch is loosely based on commit ad22c7a043c2 ("xfs: prevent stack<br /> overflows from page cache allocation").<br /> <br /> Fixes xfstest generic/273.