The patch titled
Subject: mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
has been added to the -mm mm-unstable branch. Its filename is
mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page-fix.patch
This patch will shortly appear at
https://git.kernel.org/pub/scm/linux/kernel/git/akpm/25-new.git/tree/patches/mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page-fix.patch
This patch will later appear in the mm-unstable branch at
git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/process/submit-checklist.rst when testing your code ***
The -mm tree is included into linux-next via the mm-everything
branch at git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
and is updated there every 2-3 working days
------------------------------------------------------
From: Suren Baghdasaryan <surenb@xxxxxxxxxx>
Subject: mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
Date: Fri, 3 Mar 2023 13:32:50 -0800
Internal syscaller on linux-next reported a lock inversion cause by
vm_lock being taken after i_mmap_rwsem:
======================================================
WARNING: possible circular locking dependency detected
6.2.0-next-20230301-syzkaller #0 Not tainted
------------------------------------------------------
syz-executor115/5084 is trying to acquire lock:
ffff888078307a90 (&vma->vm_lock->lock){++++}-{3:3}, at: vma_start_write include/linux/mm.h:678 [inline]
ffff888078307a90 (&vma->vm_lock->lock){++++}-{3:3}, at: retract_page_tables mm/khugepaged.c:1826 [inline]
ffff888078307a90 (&vma->vm_lock->lock){++++}-{3:3}, at: collapse_file+0x4fa5/0x5980 mm/khugepaged.c:2204
but task is already holding lock:
ffff88801f93efa8 (&mapping->i_mmap_rwsem){++++}-{3:3}, at: i_mmap_lock_write include/linux/fs.h:468 [inline]
ffff88801f93efa8 (&mapping->i_mmap_rwsem){++++}-{3:3}, at: retract_page_tables mm/khugepaged.c:1745 [inline]
ffff88801f93efa8 (&mapping->i_mmap_rwsem){++++}-{3:3}, at: collapse_file+0x3da6/0x5980 mm/khugepaged.c:2204
retract_page_tables takes i_mmap_rwsem before exclusive mmap_lock, which
is inverse to normal order. Deadlock is avoided by try-locking mmap_lock
and skipping on failure to obtain it. Locking the VMA should use the same
locking pattern to avoid this lock inversion.
Link: https://lkml.kernel.org/r/20230303213250.3555716-1-surenb@xxxxxxxxxx
Fixes: 44a83f2083bd ("mm/khugepaged: write-lock VMA while collapsing a huge page")
Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
Reported-by: <syzbot+8955a9646d1a48b8be92@xxxxxxxxxxxxxxxxxxxxxxxxx>
Cc: Arjun Roy <arjunroy@xxxxxxxxxx>
Cc: David Hildenbrand <david@xxxxxxxxxx>
Cc: David Howells <dhowells@xxxxxxxxxx>
Cc: Davidlohr Bueso <dave@xxxxxxxxxxxx>
Cc: Greg Thelen <gthelen@xxxxxxxxxx>
Cc: Hugh Dickins <hughd@xxxxxxxxxx>
Cc: Jann Horn <jannh@xxxxxxxxxx>
Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
Cc: kernel-team@xxxxxxxxxxx
Cc: Liam R. Howlett <Liam.Howlett@xxxxxxxxxx>
Cc: Lorenzo Stoakes <lstoakes@xxxxxxxxx>
Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
Cc: Michel Lespinasse <michel@xxxxxxxxxxxxxx>
Cc: Mike Rapoport <rppt@xxxxxxxxxx>
Cc: Peter Oskolkov <posk@xxxxxxxxxx>
Cc: Peter Xu <peterx@xxxxxxxxxx>
Cc: Punit Agrawal <punit.agrawal@xxxxxxxxxxxxx>
Cc: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
Cc: Shakeel Butt <shakeelb@xxxxxxxxxx>
Cc: Song Liu <songliubraving@xxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
--- a/include/linux/mm.h~mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page-fix
+++ a/include/linux/mm.h
@@ -664,18 +664,23 @@ static inline void vma_end_read(struct v
rcu_read_unlock();
}
-static inline void vma_start_write(struct vm_area_struct *vma)
+static bool __is_vma_write_locked(struct vm_area_struct *vma, int *mm_lock_seq)
{
- int mm_lock_seq;
-
mmap_assert_write_locked(vma->vm_mm);
/*
* current task is holding mmap_write_lock, both vma->vm_lock_seq and
* mm->mm_lock_seq can't be concurrently modified.
*/
- mm_lock_seq = READ_ONCE(vma->vm_mm->mm_lock_seq);
- if (vma->vm_lock_seq == mm_lock_seq)
+ *mm_lock_seq = READ_ONCE(vma->vm_mm->mm_lock_seq);
+ return (vma->vm_lock_seq == *mm_lock_seq);
+}
+
+static inline void vma_start_write(struct vm_area_struct *vma)
+{
+ int mm_lock_seq;
+
+ if (__is_vma_write_locked(vma, &mm_lock_seq))
return;
down_write(&vma->lock);
@@ -683,14 +688,26 @@ static inline void vma_start_write(struc
up_write(&vma->lock);
}
+static inline bool vma_try_start_write(struct vm_area_struct *vma)
+{
+ int mm_lock_seq;
+
+ if (__is_vma_write_locked(vma, &mm_lock_seq))
+ return true;
+
+ if (!down_write_trylock(&vma->vm_lock->lock))
+ return false;
+
+ vma->vm_lock_seq = mm_lock_seq;
+ up_write(&vma->vm_lock->lock);
+ return true;
+}
+
static inline void vma_assert_write_locked(struct vm_area_struct *vma)
{
- mmap_assert_write_locked(vma->vm_mm);
- /*
- * current task is holding mmap_write_lock, both vma->vm_lock_seq and
- * mm->mm_lock_seq can't be concurrently modified.
- */
- VM_BUG_ON_VMA(vma->vm_lock_seq != READ_ONCE(vma->vm_mm->mm_lock_seq), vma);
+ int mm_lock_seq;
+
+ VM_BUG_ON_VMA(!__is_vma_write_locked(vma, &mm_lock_seq), vma);
}
#else /* CONFIG_PER_VMA_LOCK */
--- a/mm/khugepaged.c~mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page-fix
+++ a/mm/khugepaged.c
@@ -1795,6 +1795,10 @@ static int retract_page_tables(struct ad
result = SCAN_PTE_MAPPED_HUGEPAGE;
if ((cc->is_khugepaged || is_target) &&
mmap_write_trylock(mm)) {
+ /* trylock for the same lock inversion as above */
+ if (!vma_try_start_write(vma))
+ goto unlock_next;
+
/*
* Re-check whether we have an ->anon_vma, because
* collapse_and_free_pmd() requires that either no
@@ -1823,7 +1827,6 @@ static int retract_page_tables(struct ad
result = SCAN_PTE_UFFD_WP;
goto unlock_next;
}
- vma_start_write(vma);
collapse_and_free_pmd(mm, vma, addr, pmd);
if (!cc->is_khugepaged && is_target)
result = set_huge_pmd(vma, addr, pmd, hpage);
_
Patches currently in -mm which might be from surenb@xxxxxxxxxx are
mm-introduce-config_per_vma_lock.patch
mm-move-mmap_lock-assert-function-definitions.patch
mm-add-per-vma-lock-and-helper-functions-to-control-it.patch
mm-mark-vma-as-being-written-when-changing-vm_flags.patch
mm-mmap-move-vma_prepare-before-vma_adjust_trans_huge.patch
mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page.patch
mm-khugepaged-write-lock-vma-while-collapsing-a-huge-page-fix.patch
mm-mmap-write-lock-vmas-in-vma_prepare-before-modifying-them.patch
mm-mmap-write-lock-vmas-in-vma_prepare-before-modifying-them-fix.patch
mm-mremap-write-lock-vma-while-remapping-it-to-a-new-address-range.patch
mm-write-lock-vmas-before-removing-them-from-vma-tree.patch
mm-write-lock-vmas-before-removing-them-from-vma-tree-fix.patch
mm-conditionally-write-lock-vma-in-free_pgtables.patch
kernel-fork-assert-no-vma-readers-during-its-destruction.patch
mm-mmap-prevent-pagefault-handler-from-racing-with-mmu_notifier-registration.patch
mm-introduce-vma-detached-flag.patch
mm-introduce-lock_vma_under_rcu-to-be-used-from-arch-specific-code.patch
mm-fall-back-to-mmap_lock-if-vma-anon_vma-is-not-yet-set.patch
mm-add-fault_flag_vma_lock-flag.patch
mm-add-fault_flag_vma_lock-flag-fix.patch
mm-prevent-do_swap_page-from-handling-page-faults-under-vma-lock.patch
mm-prevent-userfaults-to-be-handled-under-per-vma-lock.patch
mm-introduce-per-vma-lock-statistics.patch
x86-mm-try-vma-lock-based-page-fault-handling-first.patch
arm64-mm-try-vma-lock-based-page-fault-handling-first.patch
mm-mmap-free-vm_area_struct-without-call_rcu-in-exit_mmap.patch
mm-separate-vma-lock-from-vm_area_struct.patch
