CVE-2024-57884

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> mm: vmscan: account for free pages to prevent infinite Loop in throttle_direct_reclaim()<br /> <br /> The task sometimes continues looping in throttle_direct_reclaim() because<br /> allow_direct_reclaim(pgdat) keeps returning false. <br /> <br /> #0 [ffff80002cb6f8d0] __switch_to at ffff8000080095ac<br /> #1 [ffff80002cb6f900] __schedule at ffff800008abbd1c<br /> #2 [ffff80002cb6f990] schedule at ffff800008abc50c<br /> #3 [ffff80002cb6f9b0] throttle_direct_reclaim at ffff800008273550<br /> #4 [ffff80002cb6fa20] try_to_free_pages at ffff800008277b68<br /> #5 [ffff80002cb6fae0] __alloc_pages_nodemask at ffff8000082c4660<br /> #6 [ffff80002cb6fc50] alloc_pages_vma at ffff8000082e4a98<br /> #7 [ffff80002cb6fca0] do_anonymous_page at ffff80000829f5a8<br /> #8 [ffff80002cb6fce0] __handle_mm_fault at ffff8000082a5974<br /> #9 [ffff80002cb6fd90] handle_mm_fault at ffff8000082a5bd4<br /> <br /> At this point, the pgdat contains the following two zones:<br /> <br /> NODE: 4 ZONE: 0 ADDR: ffff00817fffe540 NAME: "DMA32"<br /> SIZE: 20480 MIN/LOW/HIGH: 11/28/45<br /> VM_STAT:<br /> NR_FREE_PAGES: 359<br /> NR_ZONE_INACTIVE_ANON: 18813<br /> NR_ZONE_ACTIVE_ANON: 0<br /> NR_ZONE_INACTIVE_FILE: 50<br /> NR_ZONE_ACTIVE_FILE: 0<br /> NR_ZONE_UNEVICTABLE: 0<br /> NR_ZONE_WRITE_PENDING: 0<br /> NR_MLOCK: 0<br /> NR_BOUNCE: 0<br /> NR_ZSPAGES: 0<br /> NR_FREE_CMA_PAGES: 0<br /> <br /> NODE: 4 ZONE: 1 ADDR: ffff00817fffec00 NAME: "Normal"<br /> SIZE: 8454144 PRESENT: 98304 MIN/LOW/HIGH: 68/166/264<br /> VM_STAT:<br /> NR_FREE_PAGES: 146<br /> NR_ZONE_INACTIVE_ANON: 94668<br /> NR_ZONE_ACTIVE_ANON: 3<br /> NR_ZONE_INACTIVE_FILE: 735<br /> NR_ZONE_ACTIVE_FILE: 78<br /> NR_ZONE_UNEVICTABLE: 0<br /> NR_ZONE_WRITE_PENDING: 0<br /> NR_MLOCK: 0<br /> NR_BOUNCE: 0<br /> NR_ZSPAGES: 0<br /> NR_FREE_CMA_PAGES: 0<br /> <br /> In allow_direct_reclaim(), while processing ZONE_DMA32, the sum of<br /> inactive/active file-backed pages calculated in zone_reclaimable_pages()<br /> based on the result of zone_page_state_snapshot() is zero. <br /> <br /> Additionally, since this system lacks swap, the calculation of inactive/<br /> active anonymous pages is skipped.<br /> <br /> crash&gt; p nr_swap_pages<br /> nr_swap_pages = $1937 = {<br /> counter = 0<br /> }<br /> <br /> As a result, ZONE_DMA32 is deemed unreclaimable and skipped, moving on to<br /> the processing of the next zone, ZONE_NORMAL, despite ZONE_DMA32 having<br /> free pages significantly exceeding the high watermark.<br /> <br /> The problem is that the pgdat-&gt;kswapd_failures hasn&amp;#39;t been incremented.<br /> <br /> crash&gt; px ((struct pglist_data *) 0xffff00817fffe540)-&gt;kswapd_failures<br /> $1935 = 0x0<br /> <br /> This is because the node deemed balanced. The node balancing logic in<br /> balance_pgdat() evaluates all zones collectively. If one or more zones<br /> (e.g., ZONE_DMA32) have enough free pages to meet their watermarks, the<br /> entire node is deemed balanced. This causes balance_pgdat() to exit early<br /> before incrementing the kswapd_failures, as it considers the overall<br /> memory state acceptable, even though some zones (like ZONE_NORMAL) remain<br /> under significant pressure.<br /> <br /> <br /> The patch ensures that zone_reclaimable_pages() includes free pages<br /> (NR_FREE_PAGES) in its calculation when no other reclaimable pages are<br /> available (e.g., file-backed or anonymous pages). This change prevents<br /> zones like ZONE_DMA32, which have sufficient free pages, from being<br /> mistakenly deemed unreclaimable. By doing so, the patch ensures proper<br /> node balancing, avoids masking pressure on other zones like ZONE_NORMAL,<br /> and prevents infinite loops in throttle_direct_reclaim() caused by<br /> allow_direct_reclaim(pgdat) repeatedly returning false.<br /> <br /> <br /> The kernel hangs due to a task stuck in throttle_direct_reclaim(), caused<br /> by a node being incorrectly deemed balanced despite pressure in certain<br /> zones, such as ZONE_NORMAL. This issue arises from<br /> zone_reclaimable_pages<br /> ---truncated---