From: David Hildenbrand <david@redhat.com> mainline inclusion from mainline-v6.17-rc3 commit 772e5b4a5e8360743645b9a466842d16092c4f94 category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/784 CVE: CVE-2025-21696 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i... ------------------ Registering userfaultd on a VMA that spans at least one PMD and then mremap()'ing that VMA can trigger a WARN when recovering from a failed page table move due to a page table allocation error. The code ends up doing the right thing (recurse, avoiding moving actual page tables), but triggering that WARN is unpleasant: WARNING: CPU: 2 PID: 6133 at mm/mremap.c:357 move_normal_pmd mm/mremap.c:357 [inline] WARNING: CPU: 2 PID: 6133 at mm/mremap.c:357 move_pgt_entry mm/mremap.c:595 [inline] WARNING: CPU: 2 PID: 6133 at mm/mremap.c:357 move_page_tables+0x3832/0x44a0 mm/mremap.c:852 Modules linked in: CPU: 2 UID: 0 PID: 6133 Comm: syz.0.19 Not tainted 6.17.0-rc1-syzkaller-00004-g53e760d89498 #0 PREEMPT(full) Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014 RIP: 0010:move_normal_pmd mm/mremap.c:357 [inline] RIP: 0010:move_pgt_entry mm/mremap.c:595 [inline] RIP: 0010:move_page_tables+0x3832/0x44a0 mm/mremap.c:852 Code: ... RSP: 0018:ffffc900037a76d8 EFLAGS: 00010293 RAX: 0000000000000000 RBX: 0000000032930007 RCX: ffffffff820c6645 RDX: ffff88802e56a440 RSI: ffffffff820c7201 RDI: 0000000000000007 RBP: ffff888037728fc0 R08: 0000000000000007 R09: 0000000000000000 R10: 0000000032930007 R11: 0000000000000000 R12: 0000000000000000 R13: ffffc900037a79a8 R14: 0000000000000001 R15: dffffc0000000000 FS: 000055556316a500(0000) GS:ffff8880d68bc000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000001b30863fff CR3: 0000000050171000 CR4: 0000000000352ef0 Call Trace: <TASK> copy_vma_and_data+0x468/0x790 mm/mremap.c:1215 move_vma+0x548/0x1780 mm/mremap.c:1282 mremap_to+0x1b7/0x450 mm/mremap.c:1406 do_mremap+0xfad/0x1f80 mm/mremap.c:1921 __do_sys_mremap+0x119/0x170 mm/mremap.c:1977 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline] do_syscall_64+0xcd/0x4c0 arch/x86/entry/syscall_64.c:94 entry_SYSCALL_64_after_hwframe+0x77/0x7f RIP: 0033:0x7f00d0b8ebe9 Code: ... RSP: 002b:00007ffe5ea5ee98 EFLAGS: 00000246 ORIG_RAX: 0000000000000019 RAX: ffffffffffffffda RBX: 00007f00d0db5fa0 RCX: 00007f00d0b8ebe9 RDX: 0000000000400000 RSI: 0000000000c00000 RDI: 0000200000000000 RBP: 00007ffe5ea5eef0 R08: 0000200000c00000 R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000246 R12: 0000000000000002 R13: 00007f00d0db5fa0 R14: 00007f00d0db5fa0 R15: 0000000000000005 </TASK> The underlying issue is that we recurse during the original page table move, but not during the recovery move. Fix it by checking for both VMAs and performing the check before the pmd_none() sanity check. Add a new helper where we perform+document that check for the PMD and PUD level. Thanks to Harry for bisecting. Link: https://lkml.kernel.org/r/20250818175358.1184757-1-david@redhat.com Fixes: 0cef0bb836e3 ("mm: clear uffd-wp PTE/PMD state on mremap()") Signed-off-by: David Hildenbrand <david@redhat.com> Reported-by: syzbot+4d9a13f0797c46a29e42@syzkaller.appspotmail.com Closes: https://lkml.kernel.org/r/689bb893.050a0220.7f033.013a.GAE@google.com Tested-by: Harry Yoo <harry.yoo@oracle.com> Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com> Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Jann Horn <jannh@google.com> Cc: Pedro Falcato <pfalcato@suse.de> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Conflicts: mm/mremap.c [The mainline patch is based on pagetable_move_control, introducing this structure that relies on numerous pre-patches. Here, we directly adapt according to the code logic.] Signed-off-by: Tong Tiangen <tongtiangen@huawei.com> Signed-off-by: Ze Zuo <zuoze1@huawei.com> --- mm/mremap.c | 77 ++++++++++++++++++++++++++++------------------------- 1 file changed, 41 insertions(+), 36 deletions(-) diff --git a/mm/mremap.c b/mm/mremap.c index 98674ed21b62..36ead7184d8e 100644 --- a/mm/mremap.c +++ b/mm/mremap.c @@ -276,9 +276,29 @@ static inline bool arch_supports_page_table_move(void) } #endif +static inline bool uffd_supports_page_table_move(struct vm_area_struct *vma, + struct vm_area_struct *new_vma) +{ + /* + * If we are moving a VMA that has uffd-wp registered but with + * remap events disabled (new VMA will not be registered with uffd), we + * need to ensure that the uffd-wp state is cleared from all pgtables. + * This means recursing into lower page tables in move_page_tables(). + * + * We might get called with VMAs reversed when recovering from a + * failed page table move. In that case, the + * "old"-but-actually-"originally new" VMA during recovery will not have + * a uffd context. Recursing into lower page tables during the original + * move but not during the recovery move will cause trouble, because we + * run into already-existing page tables. So check both VMAs. + */ + return !vma_has_uffd_without_event_remap(vma) && + !vma_has_uffd_without_event_remap(new_vma); +} + #ifdef CONFIG_HAVE_MOVE_PMD -static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, - unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd) +static bool move_normal_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma, + unsigned long old_addr, unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd) { spinlock_t *old_ptl, *new_ptl; struct mm_struct *mm = vma->vm_mm; @@ -287,6 +307,8 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, if (!arch_supports_page_table_move()) return false; + if (!uffd_supports_page_table_move(vma, new_vma)) + return false; /* * The destination pmd shouldn't be established, free_pgtables() * should have released it. @@ -313,15 +335,6 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, if (WARN_ON_ONCE(!pmd_none(*new_pmd))) return false; - /* If this pmd belongs to a uffd vma with remap events disabled, we need - * to ensure that the uffd-wp state is cleared from all pgtables. This - * means recursing into lower page tables in move_page_tables(), and we - * can reuse the existing code if we simply treat the entry as "not - * moved". - */ - if (vma_has_uffd_without_event_remap(vma)) - return false; - /* * We don't have to worry about the ordering of src and dst * ptlocks because exclusive mmap_lock prevents deadlock. @@ -352,17 +365,16 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, return res; } #else -static inline bool move_normal_pmd(struct vm_area_struct *vma, - unsigned long old_addr, unsigned long new_addr, pmd_t *old_pmd, - pmd_t *new_pmd) +static inline bool move_normal_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma, + unsigned long old_addr, unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd) { return false; } #endif #if CONFIG_PGTABLE_LEVELS > 2 && defined(CONFIG_HAVE_MOVE_PUD) -static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr, - unsigned long new_addr, pud_t *old_pud, pud_t *new_pud) +static bool move_normal_pud(struct vm_area_struct *vma, struct vm_area_struct *new_vma, + unsigned long old_addr, unsigned long new_addr, pud_t *old_pud, pud_t *new_pud) { spinlock_t *old_ptl, *new_ptl; struct mm_struct *mm = vma->vm_mm; @@ -370,6 +382,8 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr, if (!arch_supports_page_table_move()) return false; + if (!uffd_supports_page_table_move(vma, new_vma)) + return false; /* * The destination pud shouldn't be established, free_pgtables() * should have released it. @@ -377,15 +391,6 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr, if (WARN_ON_ONCE(!pud_none(*new_pud))) return false; - /* If this pud belongs to a uffd vma with remap events disabled, we need - * to ensure that the uffd-wp state is cleared from all pgtables. This - * means recursing into lower page tables in move_page_tables(), and we - * can reuse the existing code if we simply treat the entry as "not - * moved". - */ - if (vma_has_uffd_without_event_remap(vma)) - return false; - /* * We don't have to worry about the ordering of src and dst * ptlocks because exclusive mmap_lock prevents deadlock. @@ -410,9 +415,8 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr, return true; } #else -static inline bool move_normal_pud(struct vm_area_struct *vma, - unsigned long old_addr, unsigned long new_addr, pud_t *old_pud, - pud_t *new_pud) +static inline bool move_normal_pud(struct vm_area_struct *vma, struct vm_area_struct *new_vma, + unsigned long old_addr, unsigned long new_addr, pud_t *old_pud, pud_t *new_pud) { return false; } @@ -518,8 +522,9 @@ static __always_inline unsigned long get_extent(enum pgt_entry entry, * pgt_entry. Returns true if the move was successful, else false. */ static bool move_pgt_entry(enum pgt_entry entry, struct vm_area_struct *vma, - unsigned long old_addr, unsigned long new_addr, - void *old_entry, void *new_entry, bool need_rmap_locks) + struct vm_area_struct *new_vma, unsigned long old_addr, + unsigned long new_addr, void *old_entry, void *new_entry, + bool need_rmap_locks) { bool moved = false; @@ -529,11 +534,11 @@ static bool move_pgt_entry(enum pgt_entry entry, struct vm_area_struct *vma, switch (entry) { case NORMAL_PMD: - moved = move_normal_pmd(vma, old_addr, new_addr, old_entry, + moved = move_normal_pmd(vma, new_vma, old_addr, new_addr, old_entry, new_entry); break; case NORMAL_PUD: - moved = move_normal_pud(vma, old_addr, new_addr, old_entry, + moved = move_normal_pud(vma, new_vma, old_addr, new_addr, old_entry, new_entry); break; case HPAGE_PMD: @@ -598,14 +603,14 @@ unsigned long move_page_tables(struct vm_area_struct *vma, break; if (pud_trans_huge(*old_pud) || pud_devmap(*old_pud)) { if (extent == HPAGE_PUD_SIZE) { - move_pgt_entry(HPAGE_PUD, vma, old_addr, new_addr, + move_pgt_entry(HPAGE_PUD, vma, new_vma, old_addr, new_addr, old_pud, new_pud, need_rmap_locks); /* We ignore and continue on error? */ continue; } } else if (IS_ENABLED(CONFIG_HAVE_MOVE_PUD) && extent == PUD_SIZE) { - if (move_pgt_entry(NORMAL_PUD, vma, old_addr, new_addr, + if (move_pgt_entry(NORMAL_PUD, vma, new_vma, old_addr, new_addr, old_pud, new_pud, true)) continue; } @@ -621,7 +626,7 @@ unsigned long move_page_tables(struct vm_area_struct *vma, if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd) || pmd_devmap(*old_pmd)) { if (extent == HPAGE_PMD_SIZE && - move_pgt_entry(HPAGE_PMD, vma, old_addr, new_addr, + move_pgt_entry(HPAGE_PMD, vma, new_vma, old_addr, new_addr, old_pmd, new_pmd, need_rmap_locks)) continue; split_huge_pmd(vma, old_pmd, old_addr); @@ -631,7 +636,7 @@ unsigned long move_page_tables(struct vm_area_struct *vma, * If the extent is PMD-sized, try to speed the move by * moving at the PMD level if possible. */ - if (move_pgt_entry(NORMAL_PMD, vma, old_addr, new_addr, + if (move_pgt_entry(NORMAL_PMD, vma, new_vma, old_addr, new_addr, old_pmd, new_pmd, true)) continue; } -- 2.25.1