From: Yang Shi shy828301@gmail.com
commit 24d7275ce2791829953ed4e72f68277ceb2571c6 upstream.
The syzbot reported the below BUG:
kernel BUG at include/linux/page-flags.h:785! invalid opcode: 0000 [#1] PREEMPT SMP KASAN CPU: 1 PID: 4392 Comm: syz-executor560 Not tainted 5.16.0-rc6-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:PageDoubleMap include/linux/page-flags.h:785 [inline] RIP: 0010:__page_mapcount+0x2d2/0x350 mm/util.c:744 Call Trace: page_mapcount include/linux/mm.h:837 [inline] smaps_account+0x470/0xb10 fs/proc/task_mmu.c:466 smaps_pte_entry fs/proc/task_mmu.c:538 [inline] smaps_pte_range+0x611/0x1250 fs/proc/task_mmu.c:601 walk_pmd_range mm/pagewalk.c:128 [inline] walk_pud_range mm/pagewalk.c:205 [inline] walk_p4d_range mm/pagewalk.c:240 [inline] walk_pgd_range mm/pagewalk.c:277 [inline] __walk_page_range+0xe23/0x1ea0 mm/pagewalk.c:379 walk_page_vma+0x277/0x350 mm/pagewalk.c:530 smap_gather_stats.part.0+0x148/0x260 fs/proc/task_mmu.c:768 smap_gather_stats fs/proc/task_mmu.c:741 [inline] show_smap+0xc6/0x440 fs/proc/task_mmu.c:822 seq_read_iter+0xbb0/0x1240 fs/seq_file.c:272 seq_read+0x3e0/0x5b0 fs/seq_file.c:162 vfs_read+0x1b5/0x600 fs/read_write.c:479 ksys_read+0x12d/0x250 fs/read_write.c:619 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae
The reproducer was trying to read /proc/$PID/smaps when calling MADV_FREE at the mean time. MADV_FREE may split THPs if it is called for partial THP. It may trigger the below race:
CPU A CPU B ----- ----- smaps walk: MADV_FREE: page_mapcount() PageCompound() split_huge_page() page = compound_head(page) PageDoubleMap(page)
When calling PageDoubleMap() this page is not a tail page of THP anymore so the BUG is triggered.
This could be fixed by elevated refcount of the page before calling mapcount, but that would prevent it from counting migration entries, and it seems overkilling because the race just could happen when PMD is split so all PTE entries of tail pages are actually migration entries, and smaps_account() does treat migration entries as mapcount == 1 as Kirill pointed out.
Add a new parameter for smaps_account() to tell this entry is migration entry then skip calling page_mapcount(). Don't skip getting mapcount for device private entries since they do track references with mapcount.
Pagemap also has the similar issue although it was not reported. Fixed it as well.
[shy828301@gmail.com: v4] Link: https://lkml.kernel.org/r/20220203182641.824731-1-shy828301@gmail.com [nathan@kernel.org: avoid unused variable warning in pagemap_pmd_range()] Link: https://lkml.kernel.org/r/20220207171049.1102239-1-nathan@kernel.org Link: https://lkml.kernel.org/r/20220120202805.3369-1-shy828301@gmail.com Fixes: e9b61f19858a ("thp: reintroduce split_huge_page()") Signed-off-by: Yang Shi shy828301@gmail.com Signed-off-by: Nathan Chancellor nathan@kernel.org Reported-by: syzbot+1f52b3a18d5633fa7f82@syzkaller.appspotmail.com Acked-by: David Hildenbrand david@redhat.com Cc: "Kirill A. Shutemov" kirill.shutemov@linux.intel.com Cc: Jann Horn jannh@google.com Cc: Matthew Wilcox willy@infradead.org Cc: Alexey Dobriyan adobriyan@gmail.com Cc: stable@vger.kernel.org Signed-off-by: Andrew Morton akpm@linux-foundation.org Signed-off-by: Linus Torvalds torvalds@linux-foundation.org Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- fs/proc/task_mmu.c | 40 +++++++++++++++++++++++++++++++--------- 1 file changed, 31 insertions(+), 9 deletions(-)
--- a/fs/proc/task_mmu.c +++ b/fs/proc/task_mmu.c @@ -429,7 +429,8 @@ static void smaps_page_accumulate(struct }
static void smaps_account(struct mem_size_stats *mss, struct page *page, - bool compound, bool young, bool dirty, bool locked) + bool compound, bool young, bool dirty, bool locked, + bool migration) { int i, nr = compound ? compound_nr(page) : 1; unsigned long size = nr * PAGE_SIZE; @@ -456,8 +457,15 @@ static void smaps_account(struct mem_siz * page_count(page) == 1 guarantees the page is mapped exactly once. * If any subpage of the compound page mapped with PTE it would elevate * page_count(). + * + * The page_mapcount() is called to get a snapshot of the mapcount. + * Without holding the page lock this snapshot can be slightly wrong as + * we cannot always read the mapcount atomically. It is not safe to + * call page_mapcount() even with PTL held if the page is not mapped, + * especially for migration entries. Treat regular migration entries + * as mapcount == 1. */ - if (page_count(page) == 1) { + if ((page_count(page) == 1) || migration) { smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty, locked, true); return; @@ -506,6 +514,7 @@ static void smaps_pte_entry(pte_t *pte, struct vm_area_struct *vma = walk->vma; bool locked = !!(vma->vm_flags & VM_LOCKED); struct page *page = NULL; + bool migration = false;
if (pte_present(*pte)) { page = vm_normal_page(vma, addr, *pte); @@ -525,8 +534,11 @@ static void smaps_pte_entry(pte_t *pte, } else { mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; } - } else if (is_pfn_swap_entry(swpent)) + } else if (is_pfn_swap_entry(swpent)) { + if (is_migration_entry(swpent)) + migration = true; page = pfn_swap_entry_to_page(swpent); + } } else { smaps_pte_hole_lookup(addr, walk); return; @@ -535,7 +547,8 @@ static void smaps_pte_entry(pte_t *pte, if (!page) return;
- smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked); + smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), + locked, migration); }
#ifdef CONFIG_TRANSPARENT_HUGEPAGE @@ -546,6 +559,7 @@ static void smaps_pmd_entry(pmd_t *pmd, struct vm_area_struct *vma = walk->vma; bool locked = !!(vma->vm_flags & VM_LOCKED); struct page *page = NULL; + bool migration = false;
if (pmd_present(*pmd)) { /* FOLL_DUMP will return -EFAULT on huge zero page */ @@ -553,8 +567,10 @@ static void smaps_pmd_entry(pmd_t *pmd, } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) { swp_entry_t entry = pmd_to_swp_entry(*pmd);
- if (is_migration_entry(entry)) + if (is_migration_entry(entry)) { + migration = true; page = pfn_swap_entry_to_page(entry); + } } if (IS_ERR_OR_NULL(page)) return; @@ -566,7 +582,9 @@ static void smaps_pmd_entry(pmd_t *pmd, /* pass */; else mss->file_thp += HPAGE_PMD_SIZE; - smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked); + + smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), + locked, migration); } #else static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, @@ -1367,6 +1385,7 @@ static pagemap_entry_t pte_to_pagemap_en { u64 frame = 0, flags = 0; struct page *page = NULL; + bool migration = false;
if (pte_present(pte)) { if (pm->show_pfn) @@ -1388,13 +1407,14 @@ static pagemap_entry_t pte_to_pagemap_en frame = swp_type(entry) | (swp_offset(entry) << MAX_SWAPFILES_SHIFT); flags |= PM_SWAP; + migration = is_migration_entry(entry); if (is_pfn_swap_entry(entry)) page = pfn_swap_entry_to_page(entry); }
if (page && !PageAnon(page)) flags |= PM_FILE; - if (page && page_mapcount(page) == 1) + if (page && !migration && page_mapcount(page) == 1) flags |= PM_MMAP_EXCLUSIVE; if (vma->vm_flags & VM_SOFTDIRTY) flags |= PM_SOFT_DIRTY; @@ -1410,8 +1430,9 @@ static int pagemap_pmd_range(pmd_t *pmdp spinlock_t *ptl; pte_t *pte, *orig_pte; int err = 0; - #ifdef CONFIG_TRANSPARENT_HUGEPAGE + bool migration = false; + ptl = pmd_trans_huge_lock(pmdp, vma); if (ptl) { u64 flags = 0, frame = 0; @@ -1450,11 +1471,12 @@ static int pagemap_pmd_range(pmd_t *pmdp if (pmd_swp_uffd_wp(pmd)) flags |= PM_UFFD_WP; VM_BUG_ON(!is_pmd_migration_entry(pmd)); + migration = is_migration_entry(entry); page = pfn_swap_entry_to_page(entry); } #endif
- if (page && page_mapcount(page) == 1) + if (page && !migration && page_mapcount(page) == 1) flags |= PM_MMAP_EXCLUSIVE;
for (; addr != end; addr += PAGE_SIZE) {