diff mbox series

[v5.5,09/10] mmap locking API: add mmap_assert_locked() and mmap_assert_write_locked()

Message ID 20200424013858.GB158937@google.com (mailing list archive)
State New, archived
Headers show
Series None | expand

Commit Message

Michel Lespinasse April 24, 2020, 1:38 a.m. UTC 
Add new APIs to assert that mmap_sem is held.

Using this instead of rwsem_is_locked and lockdep_assert_held[_write]
makes the assertions more tolerant of future changes to the lock type.

Signed-off-by: Michel Lespinasse <walken@google.com>
---
 arch/x86/events/core.c    |  2 +-
 fs/userfaultfd.c          |  6 +++---
 include/linux/mmap_lock.h | 14 ++++++++++++++
 mm/gup.c                  |  2 +-
 mm/hmm.c                  |  2 +-
 mm/memory.c               |  2 +-
 mm/mmu_notifier.c         |  6 +++---
 mm/pagewalk.c             |  6 +++---
 mm/util.c                 |  2 +-
 9 files changed, 28 insertions(+), 14 deletions(-)

Comments

Vlastimil Babka May 18, 2020, 11:01 a.m. UTC | #1 
On 4/24/20 3:38 AM, Michel Lespinasse wrote:
> Add new APIs to assert that mmap_sem is held.
> 
> Using this instead of rwsem_is_locked and lockdep_assert_held[_write]
> makes the assertions more tolerant of future changes to the lock type.
> 
> Signed-off-by: Michel Lespinasse <walken@google.com>
> ---
>  arch/x86/events/core.c    |  2 +-
>  fs/userfaultfd.c          |  6 +++---
>  include/linux/mmap_lock.h | 14 ++++++++++++++
>  mm/gup.c                  |  2 +-
>  mm/hmm.c                  |  2 +-
>  mm/memory.c               |  2 +-
>  mm/mmu_notifier.c         |  6 +++---
>  mm/pagewalk.c             |  6 +++---
>  mm/util.c                 |  2 +-
>  9 files changed, 28 insertions(+), 14 deletions(-)
> 

...

> @@ -73,4 +75,16 @@ static inline void mmap_read_unlock_non_owner(struct mm_struct *mm)
>  	up_read_non_owner(&mm->mmap_sem);
>  }
>  
> +static inline void mmap_assert_locked(struct mm_struct *mm)
> +{
> +	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, -1), mm);
> +	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> +}
> +
> +static inline void mmap_assert_write_locked(struct mm_struct *mm)
> +{
> +	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, 0), mm);
> +	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> +}

I would remove VM_BUG_ON_MM() from the lockdep part. If kernel has lockdep
enabled, it's already in heavy debugging mode enough so let's just use it and
not depend on DEBUG_VM. Many sites you convert don't require DEBUG_VM for the
lockdep checks.

With that you can also use the standard lockdep_assert_held() and
lockdep_assert_held_write() wrappers.

If user has both lockdep and DEBUG_VM enabled, should we run both variants?
Perhaps lockdep is enough as it's more comprehensive? Your initial v5 version
was doing that.
Michel Lespinasse May 19, 2020, 1:06 p.m. UTC | #2 
On Mon, May 18, 2020 at 01:01:33PM +0200, Vlastimil Babka wrote:
> On 4/24/20 3:38 AM, Michel Lespinasse wrote:
> > +static inline void mmap_assert_locked(struct mm_struct *mm)
> > +{
> > +	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, -1), mm);
> > +	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> > +}
> > +
> > +static inline void mmap_assert_write_locked(struct mm_struct *mm)
> > +{
> > +	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, 0), mm);
> > +	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> > +}
> 
> I would remove VM_BUG_ON_MM() from the lockdep part. If kernel has lockdep
> enabled, it's already in heavy debugging mode enough so let's just use it and
> not depend on DEBUG_VM. Many sites you convert don't require DEBUG_VM for the
> lockdep checks.
> 
> With that you can also use the standard lockdep_assert_held() and
> lockdep_assert_held_write() wrappers.
> 
> If user has both lockdep and DEBUG_VM enabled, should we run both variants?
> Perhaps lockdep is enough as it's more comprehensive? Your initial v5 version
> was doing that.

Thanks, changed these to lockdep_assert_held() /
lockdep_assert_held_write() as suggested. This misses dumping out the
mm, but I think there is only limited value in that. I did keep the
rwsem_is_locked fallback as people had commented earlier about
getting assertions in the non-lockdep case. If both are enabled...
then we'll get somewhat redundant assertions, but I think that is fine
(better to keep the code simple than try to work around that).
diff mbox series

Patch

diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index a619763e96e1..66559ac4f89e 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -2182,7 +2182,7 @@  static void x86_pmu_event_mapped(struct perf_event *event, struct mm_struct *mm)
 	 * For now, this can't happen because all callers hold mmap_sem
 	 * for write.  If this changes, we'll need a different solution.
 	 */
-	lockdep_assert_held_write(&mm->mmap_sem);
+	mmap_assert_write_locked(mm);
 
 	if (atomic_inc_return(&mm->context.perf_rdpmc_allowed) == 1)
 		on_each_cpu_mask(mm_cpumask(mm), refresh_pce, NULL, 1);
diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 9c645eee1a59..12b492409040 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -234,7 +234,7 @@  static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx,
 	pte_t *ptep, pte;
 	bool ret = true;
 
-	VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
+	mmap_assert_locked(mm);
 
 	ptep = huge_pte_offset(mm, address, vma_mmu_pagesize(vma));
 
@@ -286,7 +286,7 @@  static inline bool userfaultfd_must_wait(struct userfaultfd_ctx *ctx,
 	pte_t *pte;
 	bool ret = true;
 
-	VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
+	mmap_assert_locked(mm);
 
 	pgd = pgd_offset(mm, address);
 	if (!pgd_present(*pgd))
@@ -405,7 +405,7 @@  vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
 	 * Coredumping runs without mmap_sem so we can only check that
 	 * the mmap_sem is held, if PF_DUMPCORE was not set.
 	 */
-	WARN_ON_ONCE(!rwsem_is_locked(&mm->mmap_sem));
+	mmap_assert_locked(mm);
 
 	ctx = vmf->vma->vm_userfaultfd_ctx.ctx;
 	if (!ctx)
diff --git a/include/linux/mmap_lock.h b/include/linux/mmap_lock.h
index 9e104835a0d1..5bf7cee5d93b 100644
--- a/include/linux/mmap_lock.h
+++ b/include/linux/mmap_lock.h
@@ -1,6 +1,8 @@ 
 #ifndef _LINUX_MMAP_LOCK_H
 #define _LINUX_MMAP_LOCK_H
 
+#include <linux/mmdebug.h>
+
 #define MMAP_LOCK_INITIALIZER(name) \
 	.mmap_sem = __RWSEM_INITIALIZER(name.mmap_sem),
 
@@ -73,4 +75,16 @@  static inline void mmap_read_unlock_non_owner(struct mm_struct *mm)
 	up_read_non_owner(&mm->mmap_sem);
 }
 
+static inline void mmap_assert_locked(struct mm_struct *mm)
+{
+	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, -1), mm);
+	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
+}
+
+static inline void mmap_assert_write_locked(struct mm_struct *mm)
+{
+	VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, 0), mm);
+	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
+}
+
 #endif /* _LINUX_MMAP_LOCK_H */
diff --git a/mm/gup.c b/mm/gup.c
index 0404e52513b2..e12993ceb711 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -1403,7 +1403,7 @@  long populate_vma_page_range(struct vm_area_struct *vma,
 	VM_BUG_ON(end   & ~PAGE_MASK);
 	VM_BUG_ON_VMA(start < vma->vm_start, vma);
 	VM_BUG_ON_VMA(end   > vma->vm_end, vma);
-	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
+	mmap_assert_locked(mm);
 
 	gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK;
 	if (vma->vm_flags & VM_LOCKONFAULT)
diff --git a/mm/hmm.c b/mm/hmm.c
index 280585833adf..660a4bcf932a 100644
--- a/mm/hmm.c
+++ b/mm/hmm.c
@@ -581,7 +581,7 @@  long hmm_range_fault(struct hmm_range *range)
 	struct mm_struct *mm = range->notifier->mm;
 	int ret;
 
-	lockdep_assert_held(&mm->mmap_sem);
+	mmap_assert_locked(mm);
 
 	do {
 		/* If range is no longer valid force retry. */
diff --git a/mm/memory.c b/mm/memory.c
index e6dd3309c5a3..20f98ea8968e 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1214,7 +1214,7 @@  static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
 		next = pud_addr_end(addr, end);
 		if (pud_trans_huge(*pud) || pud_devmap(*pud)) {
 			if (next - addr != HPAGE_PUD_SIZE) {
-				VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma);
+				mmap_assert_locked(tlb->mm);
 				split_huge_pud(vma, pud, addr);
 			} else if (zap_huge_pud(tlb, vma, pud, addr))
 				goto next;
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
index cfd0a03bf5cc..24eb9d1ed0a7 100644
--- a/mm/mmu_notifier.c
+++ b/mm/mmu_notifier.c
@@ -609,7 +609,7 @@  int __mmu_notifier_register(struct mmu_notifier *subscription,
 	struct mmu_notifier_subscriptions *subscriptions = NULL;
 	int ret;
 
-	lockdep_assert_held_write(&mm->mmap_sem);
+	mmap_assert_write_locked(mm);
 	BUG_ON(atomic_read(&mm->mm_users) <= 0);
 
 	if (IS_ENABLED(CONFIG_LOCKDEP)) {
@@ -761,7 +761,7 @@  struct mmu_notifier *mmu_notifier_get_locked(const struct mmu_notifier_ops *ops,
 	struct mmu_notifier *subscription;
 	int ret;
 
-	lockdep_assert_held_write(&mm->mmap_sem);
+	mmap_assert_write_locked(mm);
 
 	if (mm->notifier_subscriptions) {
 		subscription = find_get_mmu_notifier(mm, ops);
@@ -1006,7 +1006,7 @@  int mmu_interval_notifier_insert_locked(
 		mm->notifier_subscriptions;
 	int ret;
 
-	lockdep_assert_held_write(&mm->mmap_sem);
+	mmap_assert_write_locked(mm);
 
 	if (!subscriptions || !subscriptions->has_itree) {
 		ret = __mmu_notifier_register(NULL, mm);
diff --git a/mm/pagewalk.c b/mm/pagewalk.c
index 928df1638c30..3b452aa05cd0 100644
--- a/mm/pagewalk.c
+++ b/mm/pagewalk.c
@@ -395,7 +395,7 @@  int walk_page_range(struct mm_struct *mm, unsigned long start,
 	if (!walk.mm)
 		return -EINVAL;
 
-	lockdep_assert_held(&walk.mm->mmap_sem);
+	mmap_assert_locked(walk.mm);
 
 	vma = find_vma(walk.mm, start);
 	do {
@@ -453,7 +453,7 @@  int walk_page_range_novma(struct mm_struct *mm, unsigned long start,
 	if (start >= end || !walk.mm)
 		return -EINVAL;
 
-	lockdep_assert_held(&walk.mm->mmap_sem);
+	mmap_assert_locked(walk.mm);
 
 	return __walk_page_range(start, end, &walk);
 }
@@ -472,7 +472,7 @@  int walk_page_vma(struct vm_area_struct *vma, const struct mm_walk_ops *ops,
 	if (!walk.mm)
 		return -EINVAL;
 
-	lockdep_assert_held(&walk.mm->mmap_sem);
+	mmap_assert_locked(walk.mm);
 
 	err = walk_page_test(vma->vm_start, vma->vm_end, &walk);
 	if (err > 0)
diff --git a/mm/util.c b/mm/util.c
index ea2e15b21446..e7cf9c815f5a 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -437,7 +437,7 @@  int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
 	unsigned long locked_vm, limit;
 	int ret = 0;
 
-	lockdep_assert_held_write(&mm->mmap_sem);
+	mmap_assert_write_locked(mm);
 
 	locked_vm = mm->locked_vm;
 	if (inc) {