Skip to contentSkip to conversation

mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables

PATCH In review Suren Baghdasaryan 2 participants 6 messages Mar 3, 2023 – Mar 4, 2023

Subsystems (3)
lore
Share & tools
lore

Suren Baghdasaryan

author reviewer
#0 lore raw

PATCH 1/1 mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables

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.

Fixes 44a83f2083bd ("mm/khugepaged: write-lock VMA while collapsing a huge page")
Reported-by [email protected]
---
 include/linux/mm.h | 39 ++++++++++++++++++++++++++++-----------
 mm/khugepaged.c    |  5 ++++-
 2 files changed, 32 insertions(+), 12 deletions(-)
include/linux/mm.h +28 -11
index 1b9be34a24fb..5f16263d176d 100644
@@ -676,18 +676,23 @@ static inline void vma_end_read(struct vm_area_struct *vma)
 	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->vm_lock->lock);
@@ -695,14 +700,26 @@ static inline void vma_start_write(struct vm_area_struct *vma)
 	up_write(&vma->vm_lock->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);
 }
 
 static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
mm/khugepaged.c +4 -1
index c64e01f03f27..408fed42c9f5 100644
@@ -1795,6 +1795,10 @@ static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,
 		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 address_space *mapping, pgoff_t pgoff,
 				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);
Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
#1 lore raw
I love your patch! Yet something to improve:

Hi Suren,

I love your patch! Yet something to improve:

[auto build test ERROR on akpm-mm/mm-everything]

url:    https://github.com/intel-lab-lkp/linux/commits/Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401
base:   https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-everything
patch link:    https://lore.kernel.org/r/20230303213250.3555716-1-surenb%40google.com
patch subject: [PATCH 1/1] mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
config: s390-randconfig-r024-20230302 (https://download.01.org/0day-ci/archive/20230304/[email protected]/config)
compiler: clang version 17.0.0 (https://github.com/llvm/llvm-project 67409911353323ca5edf2049ef0df54132fa1ca7)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# install s390 cross compiling tool for clang build
# apt-get install binutils-s390x-linux-gnu
# https://github.com/intel-lab-lkp/linux/commit/338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401 git checkout 338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=s390 olddefconfig COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=s390 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

>> mm/khugepaged.c:1799:9: error: call to undeclared function 'vma_try_start_write'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
                           if (!vma_try_start_write(vma))
                                ^
   mm/khugepaged.c:1799:9: note: did you mean 'vma_start_write'?
   include/linux/mm.h:742:20: note: 'vma_start_write' declared here
   static inline void vma_start_write(struct vm_area_struct *vma) {}
                      ^

1 error generated.

vim +/vma_try_start_write +1799 mm/khugepaged.c

1736
1737 static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,

  1738				       struct mm_struct *target_mm,
  1739				       unsigned long target_addr, struct page *hpage,
  1740				       struct collapse_control *cc)
  1741	{
  1742		struct vm_area_struct *vma;
  1743		int target_result = SCAN_FAIL;
  1744	
  1745		i_mmap_lock_write(mapping);
  1746		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
  1747			int result = SCAN_FAIL;
  1748			struct mm_struct *mm = NULL;
  1749			unsigned long addr = 0;
  1750			pmd_t *pmd;
  1751			bool is_target = false;
  1752	
  1753			/*
  1754			 * Check vma->anon_vma to exclude MAP_PRIVATE mappings that
  1755			 * got written to. These VMAs are likely not worth investing
  1756			 * mmap_write_lock(mm) as PMD-mapping is likely to be split
  1757			 * later.
  1758			 *
  1759			 * Note that vma->anon_vma check is racy: it can be set up after
  1760			 * the check but before we took mmap_lock by the fault path.
  1761			 * But page lock would prevent establishing any new ptes of the
  1762			 * page, so we are safe.
  1763			 *
  1764			 * An alternative would be drop the check, but check that page
  1765			 * table is clear before calling pmdp_collapse_flush() under
  1766			 * ptl. It has higher chance to recover THP for the VMA, but
  1767			 * has higher cost too. It would also probably require locking
  1768			 * the anon_vma.
  1769			 */
  1770			if (READ_ONCE(vma->anon_vma)) {
  1771				result = SCAN_PAGE_ANON;
  1772				goto next;
  1773			}
  1774			addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
  1775			if (addr & ~HPAGE_PMD_MASK ||
  1776			    vma->vm_end < addr + HPAGE_PMD_SIZE) {
  1777				result = SCAN_VMA_CHECK;
  1778				goto next;
  1779			}
  1780			mm = vma->vm_mm;
  1781			is_target = mm == target_mm && addr == target_addr;
  1782			result = find_pmd_or_thp_or_none(mm, addr, &pmd);
  1783			if (result != SCAN_SUCCEED)
  1784				goto next;
  1785			/*
  1786			 * We need exclusive mmap_lock to retract page table.
  1787			 *
  1788			 * We use trylock due to lock inversion: we need to acquire
  1789			 * mmap_lock while holding page lock. Fault path does it in
  1790			 * reverse order. Trylock is a way to avoid deadlock.
  1791			 *
  1792			 * Also, it's not MADV_COLLAPSE's job to collapse other
  1793			 * mappings - let khugepaged take care of them later.
  1794			 */
  1795			result = SCAN_PTE_MAPPED_HUGEPAGE;
  1796			if ((cc->is_khugepaged || is_target) &&
  1797			    mmap_write_trylock(mm)) {
  1798				/* trylock for the same lock inversion as above */
> 1799				if (!vma_try_start_write(vma))

1800 goto unlock_next;
1801
1802 /*
1803 * Re-check whether we have an ->anon_vma, because
1804 * collapse_and_free_pmd() requires that either no
1805 * ->anon_vma exists or the anon_vma is locked.
1806 * We already checked ->anon_vma above, but that check
1807 * is racy because ->anon_vma can be populated under the
1808 * mmap lock in read mode.
1809 */
1810 if (vma->anon_vma) {
1811 result = SCAN_PAGE_ANON;
1812 goto unlock_next;
1813 }
1814 /*
1815 * When a vma is registered with uffd-wp, we can't
1816 * recycle the pmd pgtable because there can be pte
1817 * markers installed. Skip it only, so the rest mm/vma
1818 * can still have the same file mapped hugely, however
1819 * it'll always mapped in small page size for uffd-wp
1820 * registered ranges.
1821 */
1822 if (hpage_collapse_test_exit(mm)) {
1823 result = SCAN_ANY_PROCESS;
1824 goto unlock_next;
1825 }
1826 if (userfaultfd_wp(vma)) {
1827 result = SCAN_PTE_UFFD_WP;
1828 goto unlock_next;
1829 }
1830 collapse_and_free_pmd(mm, vma, addr, pmd);
1831 if (!cc->is_khugepaged && is_target)
1832 result = set_huge_pmd(vma, addr, pmd, hpage);
1833 else
1834 result = SCAN_SUCCEED;
1835
1836 unlock_next:
1837 mmap_write_unlock(mm);
1838 goto next;
1839 }
1840 /*
1841 * Calling context will handle target mm/addr. Otherwise, let
1842 * khugepaged try again later.
1843 */
1844 if (!is_target) {
1845 khugepaged_add_pte_mapped_thp(mm, addr);
1846 continue;
1847 }
1848 next:
1849 if (is_target)
1850 target_result = result;
1851 }
1852 i_mmap_unlock_write(mapping);
1853 return target_result;
1854 }
1855

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests
#2 lore raw
In reply to Suren Baghdasaryan
I love your patch! Yet something to improve:

Hi Suren,

I love your patch! Yet something to improve:

[auto build test ERROR on akpm-mm/mm-everything]

url:    https://github.com/intel-lab-lkp/linux/commits/Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401
base:   https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-everything
patch link:    https://lore.kernel.org/r/20230303213250.3555716-1-surenb%40google.com
patch subject: [PATCH 1/1] mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
config: x86_64-randconfig-a013 (https://download.01.org/0day-ci/archive/20230304/[email protected]/config)
compiler: gcc-11 (Debian 11.3.0-8) 11.3.0
reproduce (this is a W=1 build):
        # https://github.com/intel-lab-lkp/linux/commit/338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401
        git checkout 338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        make W=1 O=build_dir ARCH=x86_64 olddefconfig
        make W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

mm/khugepaged.c: In function 'retract_page_tables':

>> mm/khugepaged.c:1799:30: error: implicit declaration of function 'vma_try_start_write'; did you mean 'vma_start_write'? [-Werror=implicit-function-declaration]
    1799 |                         if (!vma_try_start_write(vma))
         |                              ^~~~~~~~~~~~~~~~~~~
         |                              vma_start_write

cc1: some warnings being treated as errors

vim +1799 mm/khugepaged.c

1736
1737 static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,

  1738				       struct mm_struct *target_mm,
  1739				       unsigned long target_addr, struct page *hpage,
  1740				       struct collapse_control *cc)
  1741	{
  1742		struct vm_area_struct *vma;
  1743		int target_result = SCAN_FAIL;
  1744	
  1745		i_mmap_lock_write(mapping);
  1746		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
  1747			int result = SCAN_FAIL;
  1748			struct mm_struct *mm = NULL;
  1749			unsigned long addr = 0;
  1750			pmd_t *pmd;
  1751			bool is_target = false;
  1752	
  1753			/*
  1754			 * Check vma->anon_vma to exclude MAP_PRIVATE mappings that
  1755			 * got written to. These VMAs are likely not worth investing
  1756			 * mmap_write_lock(mm) as PMD-mapping is likely to be split
  1757			 * later.
  1758			 *
  1759			 * Note that vma->anon_vma check is racy: it can be set up after
  1760			 * the check but before we took mmap_lock by the fault path.
  1761			 * But page lock would prevent establishing any new ptes of the
  1762			 * page, so we are safe.
  1763			 *
  1764			 * An alternative would be drop the check, but check that page
  1765			 * table is clear before calling pmdp_collapse_flush() under
  1766			 * ptl. It has higher chance to recover THP for the VMA, but
  1767			 * has higher cost too. It would also probably require locking
  1768			 * the anon_vma.
  1769			 */
  1770			if (READ_ONCE(vma->anon_vma)) {
  1771				result = SCAN_PAGE_ANON;
  1772				goto next;
  1773			}
  1774			addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
  1775			if (addr & ~HPAGE_PMD_MASK ||
  1776			    vma->vm_end < addr + HPAGE_PMD_SIZE) {
  1777				result = SCAN_VMA_CHECK;
  1778				goto next;
  1779			}
  1780			mm = vma->vm_mm;
  1781			is_target = mm == target_mm && addr == target_addr;
  1782			result = find_pmd_or_thp_or_none(mm, addr, &pmd);
  1783			if (result != SCAN_SUCCEED)
  1784				goto next;
  1785			/*
  1786			 * We need exclusive mmap_lock to retract page table.
  1787			 *
  1788			 * We use trylock due to lock inversion: we need to acquire
  1789			 * mmap_lock while holding page lock. Fault path does it in
  1790			 * reverse order. Trylock is a way to avoid deadlock.
  1791			 *
  1792			 * Also, it's not MADV_COLLAPSE's job to collapse other
  1793			 * mappings - let khugepaged take care of them later.
  1794			 */
  1795			result = SCAN_PTE_MAPPED_HUGEPAGE;
  1796			if ((cc->is_khugepaged || is_target) &&
  1797			    mmap_write_trylock(mm)) {
  1798				/* trylock for the same lock inversion as above */
> 1799				if (!vma_try_start_write(vma))

1800 goto unlock_next;
1801
1802 /*
1803 * Re-check whether we have an ->anon_vma, because
1804 * collapse_and_free_pmd() requires that either no
1805 * ->anon_vma exists or the anon_vma is locked.
1806 * We already checked ->anon_vma above, but that check
1807 * is racy because ->anon_vma can be populated under the
1808 * mmap lock in read mode.
1809 */
1810 if (vma->anon_vma) {
1811 result = SCAN_PAGE_ANON;
1812 goto unlock_next;
1813 }
1814 /*
1815 * When a vma is registered with uffd-wp, we can't
1816 * recycle the pmd pgtable because there can be pte
1817 * markers installed. Skip it only, so the rest mm/vma
1818 * can still have the same file mapped hugely, however
1819 * it'll always mapped in small page size for uffd-wp
1820 * registered ranges.
1821 */
1822 if (hpage_collapse_test_exit(mm)) {
1823 result = SCAN_ANY_PROCESS;
1824 goto unlock_next;
1825 }
1826 if (userfaultfd_wp(vma)) {
1827 result = SCAN_PTE_UFFD_WP;
1828 goto unlock_next;
1829 }
1830 collapse_and_free_pmd(mm, vma, addr, pmd);
1831 if (!cc->is_khugepaged && is_target)
1832 result = set_huge_pmd(vma, addr, pmd, hpage);
1833 else
1834 result = SCAN_SUCCEED;
1835
1836 unlock_next:
1837 mmap_write_unlock(mm);
1838 goto next;
1839 }
1840 /*
1841 * Calling context will handle target mm/addr. Otherwise, let
1842 * khugepaged try again later.
1843 */
1844 if (!is_target) {
1845 khugepaged_add_pte_mapped_thp(mm, addr);
1846 continue;
1847 }
1848 next:
1849 if (is_target)
1850 target_result = result;
1851 }
1852 i_mmap_unlock_write(mapping);
1853 return target_result;
1854 }
1855

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests
#3 lore raw
In reply to Suren Baghdasaryan
I love your patch! Yet something to improve:

Hi Suren,

I love your patch! Yet something to improve:

[auto build test ERROR on akpm-mm/mm-everything]

url:    https://github.com/intel-lab-lkp/linux/commits/Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401
base:   https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-everything
patch link:    https://lore.kernel.org/r/20230303213250.3555716-1-surenb%40google.com
patch subject: [PATCH 1/1] mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
config: i386-randconfig-a013 (https://download.01.org/0day-ci/archive/20230304/[email protected]/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project f28c006a5895fc0e329fe15fead81e37457cb1d1)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# https://github.com/intel-lab-lkp/linux/commit/338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401 git checkout 338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 olddefconfig COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

>> mm/khugepaged.c:1799:9: error: implicit declaration of function 'vma_try_start_write' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
                           if (!vma_try_start_write(vma))
                                ^
   mm/khugepaged.c:1799:9: note: did you mean 'vma_start_write'?
   include/linux/mm.h:742:20: note: 'vma_start_write' declared here
   static inline void vma_start_write(struct vm_area_struct *vma) {}
                      ^

1 error generated.

vim +/vma_try_start_write +1799 mm/khugepaged.c

1736
1737 static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,

  1738				       struct mm_struct *target_mm,
  1739				       unsigned long target_addr, struct page *hpage,
  1740				       struct collapse_control *cc)
  1741	{
  1742		struct vm_area_struct *vma;
  1743		int target_result = SCAN_FAIL;
  1744	
  1745		i_mmap_lock_write(mapping);
  1746		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
  1747			int result = SCAN_FAIL;
  1748			struct mm_struct *mm = NULL;
  1749			unsigned long addr = 0;
  1750			pmd_t *pmd;
  1751			bool is_target = false;
  1752	
  1753			/*
  1754			 * Check vma->anon_vma to exclude MAP_PRIVATE mappings that
  1755			 * got written to. These VMAs are likely not worth investing
  1756			 * mmap_write_lock(mm) as PMD-mapping is likely to be split
  1757			 * later.
  1758			 *
  1759			 * Note that vma->anon_vma check is racy: it can be set up after
  1760			 * the check but before we took mmap_lock by the fault path.
  1761			 * But page lock would prevent establishing any new ptes of the
  1762			 * page, so we are safe.
  1763			 *
  1764			 * An alternative would be drop the check, but check that page
  1765			 * table is clear before calling pmdp_collapse_flush() under
  1766			 * ptl. It has higher chance to recover THP for the VMA, but
  1767			 * has higher cost too. It would also probably require locking
  1768			 * the anon_vma.
  1769			 */
  1770			if (READ_ONCE(vma->anon_vma)) {
  1771				result = SCAN_PAGE_ANON;
  1772				goto next;
  1773			}
  1774			addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
  1775			if (addr & ~HPAGE_PMD_MASK ||
  1776			    vma->vm_end < addr + HPAGE_PMD_SIZE) {
  1777				result = SCAN_VMA_CHECK;
  1778				goto next;
  1779			}
  1780			mm = vma->vm_mm;
  1781			is_target = mm == target_mm && addr == target_addr;
  1782			result = find_pmd_or_thp_or_none(mm, addr, &pmd);
  1783			if (result != SCAN_SUCCEED)
  1784				goto next;
  1785			/*
  1786			 * We need exclusive mmap_lock to retract page table.
  1787			 *
  1788			 * We use trylock due to lock inversion: we need to acquire
  1789			 * mmap_lock while holding page lock. Fault path does it in
  1790			 * reverse order. Trylock is a way to avoid deadlock.
  1791			 *
  1792			 * Also, it's not MADV_COLLAPSE's job to collapse other
  1793			 * mappings - let khugepaged take care of them later.
  1794			 */
  1795			result = SCAN_PTE_MAPPED_HUGEPAGE;
  1796			if ((cc->is_khugepaged || is_target) &&
  1797			    mmap_write_trylock(mm)) {
  1798				/* trylock for the same lock inversion as above */
> 1799				if (!vma_try_start_write(vma))

1800 goto unlock_next;
1801
1802 /*
1803 * Re-check whether we have an ->anon_vma, because
1804 * collapse_and_free_pmd() requires that either no
1805 * ->anon_vma exists or the anon_vma is locked.
1806 * We already checked ->anon_vma above, but that check
1807 * is racy because ->anon_vma can be populated under the
1808 * mmap lock in read mode.
1809 */
1810 if (vma->anon_vma) {
1811 result = SCAN_PAGE_ANON;
1812 goto unlock_next;
1813 }
1814 /*
1815 * When a vma is registered with uffd-wp, we can't
1816 * recycle the pmd pgtable because there can be pte
1817 * markers installed. Skip it only, so the rest mm/vma
1818 * can still have the same file mapped hugely, however
1819 * it'll always mapped in small page size for uffd-wp
1820 * registered ranges.
1821 */
1822 if (hpage_collapse_test_exit(mm)) {
1823 result = SCAN_ANY_PROCESS;
1824 goto unlock_next;
1825 }
1826 if (userfaultfd_wp(vma)) {
1827 result = SCAN_PTE_UFFD_WP;
1828 goto unlock_next;
1829 }
1830 collapse_and_free_pmd(mm, vma, addr, pmd);
1831 if (!cc->is_khugepaged && is_target)
1832 result = set_huge_pmd(vma, addr, pmd, hpage);
1833 else
1834 result = SCAN_SUCCEED;
1835
1836 unlock_next:
1837 mmap_write_unlock(mm);
1838 goto next;
1839 }
1840 /*
1841 * Calling context will handle target mm/addr. Otherwise, let
1842 * khugepaged try again later.
1843 */
1844 if (!is_target) {
1845 khugepaged_add_pte_mapped_thp(mm, addr);
1846 continue;
1847 }
1848 next:
1849 if (is_target)
1850 target_result = result;
1851 }
1852 i_mmap_unlock_write(mapping);
1853 return target_result;
1854 }
1855

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests
author reviewer
#4 lore raw

Hi Andrew,
I missed vma_try_start_write() definition for CONFIG_PER_VMA_LOCK=n configuration. Could you please patch this with the following?:

include/linux/mm.h +0 -0
@@ -740,6 +740,8 @@ static inline bool vma_start_read(struct

vm_area_struct *vma)
{ return false; }
static inline void vma_end_read(struct vm_area_struct *vma) {} static inline void vma_start_write(struct vm_area_struct *vma) {}
+static inline bool vma_try_start_write(struct vm_area_struct *vma) { return true; }
static inline void vma_assert_write_locked(struct vm_area_struct *vma) {}

or should I send a separate patch?
Thanks,
Suren.

Quoted 167 lines · kernel test robot

Hi Suren,

I love your patch! Yet something to improve:

[auto build test ERROR on akpm-mm/mm-everything]

url:    https://github.com/intel-lab-lkp/linux/commits/Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401
base:   https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-everything
patch link:    https://lore.kernel.org/r/20230303213250.3555716-1-surenb%40google.com
patch subject: [PATCH 1/1] mm/khugepaged: fix vm_lock/i_mmap_rwsem inversion in retract_page_tables
config: i386-randconfig-a013 (https://download.01.org/0day-ci/archive/20230304/[email protected]/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project f28c006a5895fc0e329fe15fead81e37457cb1d1)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# https://github.com/intel-lab-lkp/linux/commit/338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Suren-Baghdasaryan/mm-khugepaged-fix-vm_lock-i_mmap_rwsem-inversion-in-retract_page_tables/20230304-053401 git checkout 338d2ee433a0559b3c6c3966b8d4ffe55e8dc6c9
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 olddefconfig COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

mm/khugepaged.c:1799:9: error: implicit declaration of function 'vma_try_start_write' is invalid in C99 [-Werror,-Wimplicit-function-declaration]

                           if (!vma_try_start_write(vma))
                                ^
   mm/khugepaged.c:1799:9: note: did you mean 'vma_start_write'?
   include/linux/mm.h:742:20: note: 'vma_start_write' declared here
   static inline void vma_start_write(struct vm_area_struct *vma) {}
                      ^

1 error generated.

vim +/vma_try_start_write +1799 mm/khugepaged.c

1736
1737 static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,

  1738                                 struct mm_struct *target_mm,
  1739                                 unsigned long target_addr, struct page *hpage,
  1740                                 struct collapse_control *cc)
  1741  {
  1742          struct vm_area_struct *vma;
  1743          int target_result = SCAN_FAIL;
  1744
  1745          i_mmap_lock_write(mapping);
  1746          vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
  1747                  int result = SCAN_FAIL;
  1748                  struct mm_struct *mm = NULL;
  1749                  unsigned long addr = 0;
  1750                  pmd_t *pmd;
  1751                  bool is_target = false;
  1752
  1753                  /*
  1754                   * Check vma->anon_vma to exclude MAP_PRIVATE mappings that
  1755                   * got written to. These VMAs are likely not worth investing
  1756                   * mmap_write_lock(mm) as PMD-mapping is likely to be split
  1757                   * later.
  1758                   *
  1759                   * Note that vma->anon_vma check is racy: it can be set up after
  1760                   * the check but before we took mmap_lock by the fault path.
  1761                   * But page lock would prevent establishing any new ptes of the
  1762                   * page, so we are safe.
  1763                   *
  1764                   * An alternative would be drop the check, but check that page
  1765                   * table is clear before calling pmdp_collapse_flush() under
  1766                   * ptl. It has higher chance to recover THP for the VMA, but
  1767                   * has higher cost too. It would also probably require locking
  1768                   * the anon_vma.
  1769                   */
  1770                  if (READ_ONCE(vma->anon_vma)) {
  1771                          result = SCAN_PAGE_ANON;
  1772                          goto next;
  1773                  }
  1774                  addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
  1775                  if (addr & ~HPAGE_PMD_MASK ||
  1776                      vma->vm_end < addr + HPAGE_PMD_SIZE) {
  1777                          result = SCAN_VMA_CHECK;
  1778                          goto next;
  1779                  }
  1780                  mm = vma->vm_mm;
  1781                  is_target = mm == target_mm && addr == target_addr;
  1782                  result = find_pmd_or_thp_or_none(mm, addr, &pmd);
  1783                  if (result != SCAN_SUCCEED)
  1784                          goto next;
  1785                  /*
  1786                   * We need exclusive mmap_lock to retract page table.
  1787                   *
  1788                   * We use trylock due to lock inversion: we need to acquire
  1789                   * mmap_lock while holding page lock. Fault path does it in
  1790                   * reverse order. Trylock is a way to avoid deadlock.
  1791                   *
  1792                   * Also, it's not MADV_COLLAPSE's job to collapse other
  1793                   * mappings - let khugepaged take care of them later.
  1794                   */

1795 result = SCAN_PTE_MAPPED_HUGEPAGE;
1796 if ((cc->is_khugepaged || is_target) &&
1797 mmap_write_trylock(mm)) {
1798 /* trylock for the same lock inversion as above */

Quoted 1 line

1799 if (!vma_try_start_write(vma))

1800 goto unlock_next;
1801
1802 /*

  1803                           * Re-check whether we have an ->anon_vma, because
  1804                           * collapse_and_free_pmd() requires that either no
  1805                           * ->anon_vma exists or the anon_vma is locked.
  1806                           * We already checked ->anon_vma above, but that check
  1807                           * is racy because ->anon_vma can be populated under the
  1808                           * mmap lock in read mode.
  1809                           */
  1810                          if (vma->anon_vma) {
  1811                                  result = SCAN_PAGE_ANON;
  1812                                  goto unlock_next;
  1813                          }
  1814                          /*
  1815                           * When a vma is registered with uffd-wp, we can't
  1816                           * recycle the pmd pgtable because there can be pte
  1817                           * markers installed.  Skip it only, so the rest mm/vma
  1818                           * can still have the same file mapped hugely, however
  1819                           * it'll always mapped in small page size for uffd-wp
  1820                           * registered ranges.
  1821                           */
  1822                          if (hpage_collapse_test_exit(mm)) {
  1823                                  result = SCAN_ANY_PROCESS;
  1824                                  goto unlock_next;
  1825                          }
  1826                          if (userfaultfd_wp(vma)) {
  1827                                  result = SCAN_PTE_UFFD_WP;
  1828                                  goto unlock_next;
  1829                          }
  1830                          collapse_and_free_pmd(mm, vma, addr, pmd);
  1831                          if (!cc->is_khugepaged && is_target)
  1832                                  result = set_huge_pmd(vma, addr, pmd, hpage);
  1833                          else
  1834                                  result = SCAN_SUCCEED;
  1835
  1836  unlock_next:
  1837                          mmap_write_unlock(mm);
  1838                          goto next;
  1839                  }
  1840                  /*
  1841                   * Calling context will handle target mm/addr. Otherwise, let
  1842                   * khugepaged try again later.
  1843                   */
  1844                  if (!is_target) {
  1845                          khugepaged_add_pte_mapped_thp(mm, addr);
  1846                          continue;
  1847                  }
  1848  next:
  1849                  if (is_target)
  1850                          target_result = result;
  1851          }
  1852          i_mmap_unlock_write(mapping);
  1853          return target_result;

1854 }
1855

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests

author reviewer
#5 lore raw
In reply to Suren Baghdasaryan
Quoted 85 lines · Suren Baghdasaryan

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.

Fixes 44a83f2083bd ("mm/khugepaged: write-lock VMA while collapsing a huge page")
Reported-by [email protected]
---
 include/linux/mm.h | 39 ++++++++++++++++++++++++++++-----------
 mm/khugepaged.c    |  5 ++++-
 2 files changed, 32 insertions(+), 12 deletions(-)
include/linux/mm.h +25 -5
index 1b9be34a24fb..5f16263d176d 100644
@@ -676,18 +676,23 @@ static inline void vma_end_read(struct vm_area_struct *vma)
        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->vm_lock->lock);
@@ -695,14 +700,26 @@ static inline void vma_start_write(struct vm_area_struct *vma)
        up_write(&vma->vm_lock->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)
 {

include/linux/mm.h static inline void vma_start_write(struct vm_area_struct *vma) · line 723

Quoted 10 lines
include/linux/mm.h +3 -6
-       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);
 }

Hi Andrew,
I missed vma_try_start_write() definition for CONFIG_PER_VMA_LOCK=n configuration. Could you please patch this with the following?:

include/linux/mm.h +0 -0
@@ -740,6 +740,8 @@ static inline bool vma_start_read(struct

vm_area_struct *vma)
{ return false; }
static inline void vma_end_read(struct vm_area_struct *vma) {} static inline void vma_start_write(struct vm_area_struct *vma) {}
+static inline bool vma_try_start_write(struct vm_area_struct *vma) { return true; }
static inline void vma_assert_write_locked(struct vm_area_struct *vma) {}

or should I send a separate patch?
Thanks,
Suren.

Quoted 28 lines
include/linux/mm.h +0 -0
 static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
mm/khugepaged.c +4 -2
index c64e01f03f27..408fed42c9f5 100644
@@ -1795,6 +1795,10 @@ static int retract_page_tables(struct address_space *mapping, pgoff_t pgoff,
                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 address_space *mapping, pgoff_t pgoff,
                                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);
--

2.40.0.rc0.216.gc4246ad0f0-goog