mm/ksm: unmerge and clear VM_MERGEABLE when setting PR_SET_MEMORY_MERGE=0

Patch series "mm/ksm: improve PR_SET_MEMORY_MERGE=0 handling and cleanup
disabling KSM", v2.

(1) Make PR_SET_MEMORY_MERGE=0 unmerge pages like setting MADV_UNMERGEABLE
does, (2) add a selftest for it and (3) factor out disabling of KSM from
s390/gmap code.


This patch (of 3):

Let's unmerge any KSM pages when setting PR_SET_MEMORY_MERGE=0, and clear
the VM_MERGEABLE flag from all VMAs -- just like KSM would.  Of course,
only do that if we previously set PR_SET_MEMORY_MERGE=1.

Link: https://lkml.kernel.org/r/20230422205420.30372-1-david@redhat.com
Link: https://lkml.kernel.org/r/20230422205420.30372-2-david@redhat.comSigned-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Stefan Roesch <shr@devkernel.io>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Claudio Imbrenda <imbrenda@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Janosch Frank <frankja@linux.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

include/linux/ksm.h

@@ -21,6 +21,7 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
 
 void ksm_add_vma(struct vm_area_struct *vma);
 int ksm_enable_merge_any(struct mm_struct *mm);
+int ksm_disable_merge_any(struct mm_struct *mm);
 
 int __ksm_enter(struct mm_struct *mm);
 void __ksm_exit(struct mm_struct *mm);

kernel/sys.c

@@ -2695,16 +2695,10 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 		if (mmap_write_lock_killable(me->mm))
 			return -EINTR;
 
-		if (arg2) {
+		if (arg2)
 			error = ksm_enable_merge_any(me->mm);
-		} else {
-			/*
-			 * TODO: we might want disable KSM on all VMAs and
-			 * trigger unsharing to completely disable KSM.
-			 */
-			clear_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
-			error = 0;
-		}
+		else
+			error = ksm_disable_merge_any(me->mm);
 		mmap_write_unlock(me->mm);
 		break;
 	case PR_GET_MEMORY_MERGE:

mm/ksm.c

@@ -2520,6 +2520,22 @@ static void __ksm_add_vma(struct vm_area_struct *vma)
 		vm_flags_set(vma, VM_MERGEABLE);
 }
 
+static int __ksm_del_vma(struct vm_area_struct *vma)
+{
+	int err;
+
+	if (!(vma->vm_flags & VM_MERGEABLE))
+		return 0;
+
+	if (vma->anon_vma) {
+		err = unmerge_ksm_pages(vma, vma->vm_start, vma->vm_end);
+		if (err)
+			return err;
+	}
+
+	vm_flags_clear(vma, VM_MERGEABLE);
+	return 0;
+}
 /**
  * ksm_add_vma - Mark vma as mergeable if compatible
  *
@@ -2542,6 +2558,20 @@ static void ksm_add_vmas(struct mm_struct *mm)
 		__ksm_add_vma(vma);
 }
 
+static int ksm_del_vmas(struct mm_struct *mm)
+{
+	struct vm_area_struct *vma;
+	int err;
+
+	VMA_ITERATOR(vmi, mm, 0);
+	for_each_vma(vmi, vma) {
+		err = __ksm_del_vma(vma);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
 /**
  * ksm_enable_merge_any - Add mm to mm ksm list and enable merging on all
  *                        compatible VMA's
@@ -2569,6 +2599,35 @@ int ksm_enable_merge_any(struct mm_struct *mm)
 	return 0;
 }
 
+/**
+ * ksm_disable_merge_any - Disable merging on all compatible VMA's of the mm,
+ *			   previously enabled via ksm_enable_merge_any().
+ *
+ * Disabling merging implies unmerging any merged pages, like setting
+ * MADV_UNMERGEABLE would. If unmerging fails, the whole operation fails and
+ * merging on all compatible VMA's remains enabled.
+ *
+ * @mm: Pointer to mm
+ *
+ * Returns 0 on success, otherwise error code
+ */
+int ksm_disable_merge_any(struct mm_struct *mm)
+{
+	int err;
+
+	if (!test_bit(MMF_VM_MERGE_ANY, &mm->flags))
+		return 0;
+
+	err = ksm_del_vmas(mm);
+	if (err) {
+		ksm_add_vmas(mm);
+		return err;
+	}
+
+	clear_bit(MMF_VM_MERGE_ANY, &mm->flags);
+	return 0;
+}
+
 int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
 		unsigned long end, int advice, unsigned long *vm_flags)
 {