CVE-2025-21881

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> uprobes: Reject the shared zeropage in uprobe_write_opcode()<br /> <br /> We triggered the following crash in syzkaller tests:<br /> <br /> BUG: Bad page state in process syz.7.38 pfn:1eff3<br /> page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1eff3<br /> flags: 0x3fffff00004004(referenced|reserved|node=0|zone=1|lastcpupid=0x1fffff)<br /> raw: 003fffff00004004 ffffe6c6c07bfcc8 ffffe6c6c07bfcc8 0000000000000000<br /> raw: 0000000000000000 0000000000000000 00000000fffffffe 0000000000000000<br /> page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set<br /> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1ubuntu1.1 04/01/2014<br /> Call Trace:<br /> <br /> dump_stack_lvl+0x32/0x50<br /> bad_page+0x69/0xf0<br /> free_unref_page_prepare+0x401/0x500<br /> free_unref_page+0x6d/0x1b0<br /> uprobe_write_opcode+0x460/0x8e0<br /> install_breakpoint.part.0+0x51/0x80<br /> register_for_each_vma+0x1d9/0x2b0<br /> __uprobe_register+0x245/0x300<br /> bpf_uprobe_multi_link_attach+0x29b/0x4f0<br /> link_create+0x1e2/0x280<br /> __sys_bpf+0x75f/0xac0<br /> __x64_sys_bpf+0x1a/0x30<br /> do_syscall_64+0x56/0x100<br /> entry_SYSCALL_64_after_hwframe+0x78/0xe2<br /> <br /> BUG: Bad rss-counter state mm:00000000452453e0 type:MM_FILEPAGES val:-1<br /> <br /> The following syzkaller test case can be used to reproduce:<br /> <br /> r2 = creat(&amp;(0x7f0000000000)=&amp;#39;./file0\x00&amp;#39;, 0x8)<br /> write$nbd(r2, &amp;(0x7f0000000580)=ANY=[], 0x10)<br /> r4 = openat(0xffffffffffffff9c, &amp;(0x7f0000000040)=&amp;#39;./file0\x00&amp;#39;, 0x42, 0x0)<br /> mmap$IORING_OFF_SQ_RING(&amp;(0x7f0000ffd000/0x3000)=nil, 0x3000, 0x0, 0x12, r4, 0x0)<br /> r5 = userfaultfd(0x80801)<br /> ioctl$UFFDIO_API(r5, 0xc018aa3f, &amp;(0x7f0000000040)={0xaa, 0x20})<br /> r6 = userfaultfd(0x80801)<br /> ioctl$UFFDIO_API(r6, 0xc018aa3f, &amp;(0x7f0000000140))<br /> ioctl$UFFDIO_REGISTER(r6, 0xc020aa00, &amp;(0x7f0000000100)={{&amp;(0x7f0000ffc000/0x4000)=nil, 0x4000}, 0x2})<br /> ioctl$UFFDIO_ZEROPAGE(r5, 0xc020aa04, &amp;(0x7f0000000000)={{&amp;(0x7f0000ffd000/0x1000)=nil, 0x1000}})<br /> r7 = bpf$PROG_LOAD(0x5, &amp;(0x7f0000000140)={0x2, 0x3, &amp;(0x7f0000000200)=ANY=[@ANYBLOB="1800000000120000000000000000000095"], &amp;(0x7f0000000000)=&amp;#39;GPL\x00&amp;#39;, 0x7, 0x0, 0x0, 0x0, 0x0, &amp;#39;\x00&amp;#39;, 0x0, @fallback=0x30, 0xffffffffffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x10, 0x0, @void, @value}, 0x94)<br /> bpf$BPF_LINK_CREATE_XDP(0x1c, &amp;(0x7f0000000040)={r7, 0x0, 0x30, 0x1e, @val=@uprobe_multi={&amp;(0x7f0000000080)=&amp;#39;./file0\x00&amp;#39;, &amp;(0x7f0000000100)=[0x2], 0x0, 0x0, 0x1}}, 0x40)<br /> <br /> The cause is that zero pfn is set to the PTE without increasing the RSS<br /> count in mfill_atomic_pte_zeropage() and the refcount of zero folio does<br /> not increase accordingly. Then, the operation on the same pfn is performed<br /> in uprobe_write_opcode()-&gt;__replace_page() to unconditional decrease the<br /> RSS count and old_folio&amp;#39;s refcount.<br /> <br /> Therefore, two bugs are introduced:<br /> <br /> 1. The RSS count is incorrect, when process exit, the check_mm() report<br /> error "Bad rss-count".<br /> <br /> 2. The reserved folio (zero folio) is freed when folio-&gt;refcount is zero,<br /> then free_pages_prepare-&gt;free_page_is_bad() report error<br /> "Bad page state".<br /> <br /> There is more, the following warning could also theoretically be triggered:<br /> <br /> __replace_page()<br /> -&gt; ...<br /> -&gt; folio_remove_rmap_pte()<br /> -&gt; VM_WARN_ON_FOLIO(is_zero_folio(folio), folio)<br /> <br /> Considering that uprobe hit on the zero folio is a very rare case, just<br /> reject zero old folio immediately after get_user_page_vma_remote().<br /> <br /> [ mingo: Cleaned up the changelog ]