Skip to contentSkip to conversation

remove follow_pfn

PATCH In review Christoph Hellwig 3 participants 10 messages Mar 24, 2024 – Mar 26, 2024

Subsystems (6)
Memory management x86 MM core x86 MM +2 more subsystems
lore
Share & tools
lore
#0 lore raw

Hi all,

this series open codes follow_pfn in the only remaining caller, although the code there remains questionable. It then also moves follow_phys into the only user and simplifies it a bit.

Diffstat:

 arch/x86/mm/pat/memtype.c |   22 ++++++++++++++-
 drivers/virt/acrn/mm.c    |   10 +++++--
 include/linux/mm.h        |    4 --
 mm/memory.c               |   64 +---------------------------------------------
 mm/nommu.c                |   21 ---------------
 5 files changed, 30 insertions(+), 91 deletions(-)
#1 lore raw

PATCH 1/3 virt: acrn: stop using follow_pfn

Switch from follow_pfn to follow_pte so that we can get rid of follow_pfn. Note that this doesn't fix any of the pre-existing raciness and lack of permission checking in the code.

---
 drivers/virt/acrn/mm.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/virt/acrn/mm.c +8 -2
index fa5d9ca6be5706..69c3f619f88196 100644
@@ -171,18 +171,24 @@ int acrn_vm_ram_map(struct acrn_vm *vm, struct acrn_vm_memmap *memmap)
 	mmap_read_lock(current->mm);
 	vma = vma_lookup(current->mm, memmap->vma_base);
 	if (vma && ((vma->vm_flags & VM_PFNMAP) != 0)) {
+		spinlock_t *ptl;
+		pte_t *ptep;
+
 		if ((memmap->vma_base + memmap->len) > vma->vm_end) {
 			mmap_read_unlock(current->mm);
 			return -EINVAL;
 		}
 
-		ret = follow_pfn(vma, memmap->vma_base, &pfn);
-		mmap_read_unlock(current->mm);
+		ret = follow_pte(vma->vm_mm, memmap->vma_base, &ptep, &ptl);
 		if (ret < 0) {
+			mmap_read_unlock(current->mm);
 			dev_dbg(acrn_dev.this_device,
 				"Failed to lookup PFN at VMA:%pK.\n", (void *)memmap->vma_base);
 			return ret;
 		}
+		pfn = pte_pfn(ptep_get(ptep));
+		pte_unmap_unlock(ptep, ptl);
+		mmap_read_unlock(current->mm);
 
 		return acrn_mm_region_add(vm, memmap->user_vm_pa,
 			 PFN_PHYS(pfn), memmap->len,
Signature · 1 line
2.39.2
Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] virt: acrn: stop using follow_pfn

Quoted 30 lines · Christoph Hellwig

Switch from follow_pfn to follow_pte so that we can get rid of follow_pfn. Note that this doesn't fix any of the pre-existing raciness and lack of permission checking in the code.

---
  drivers/virt/acrn/mm.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)
drivers/virt/acrn/mm.c +5 -2
index fa5d9ca6be5706..69c3f619f88196 100644
@@ -171,18 +171,24 @@ int acrn_vm_ram_map(struct acrn_vm *vm, struct acrn_vm_memmap *memmap)
  	mmap_read_lock(current->mm);
  	vma = vma_lookup(current->mm, memmap->vma_base);
  	if (vma && ((vma->vm_flags & VM_PFNMAP) != 0)) {
+		spinlock_t *ptl;
+		pte_t *ptep;
+
  		if ((memmap->vma_base + memmap->len) > vma->vm_end) {
  			mmap_read_unlock(current->mm);
  			return -EINVAL;
  		}
 
-		ret = follow_pfn(vma, memmap->vma_base, &pfn);
-		mmap_read_unlock(current->mm);
+		ret = follow_pte(vma->vm_mm, memmap->vma_base, &ptep, &ptl);
  		if (ret < 0) {
+			mmap_read_unlock(current->mm);

drivers/virt/acrn/mm.c int acrn_vm_ram_map(struct acrn_vm *vm, struct acrn_vm_memmap *memmap) · line 194

Quoted 10 lines
drivers/virt/acrn/mm.c +3 -0
  			dev_dbg(acrn_dev.this_device,
  				"Failed to lookup PFN at VMA:%pK.\n", (void *)memmap->vma_base);
  			return ret;
  		}
+		pfn = pte_pfn(ptep_get(ptep));
+		pte_unmap_unlock(ptep, ptl);
+		mmap_read_unlock(current->mm);
 
  		return acrn_mm_region_add(vm, memmap->user_vm_pa,
  			 PFN_PHYS(pfn), memmap->len,

... I have similar patches lying around here (see bwlow). I added some actual access permission checks.

(I also realized, that if we get an anon folio in a COW mapping via follow_pte() here, I suspect one might be able to do some nasty things. Just imagine if we munmap(), free the anon folio, and then it gets used in other context ... At least KVM/vfio handle that using references+MMU notifiers.)

commit 812e577dea97327bcc68d34504e7387dff2ffd8f

Author: David Hildenbrand <[email protected]>
Date:   Fri Mar 8 13:53:04 2024 +0100

virt/acrn/mm: use follow_pte() instead of follow_pfn()

follow_pfn() should not be used. Instead, use follow_pte() and do some best-guess PTE permission checks.

Should we simply always require pte_write()? Maybe. Performing no checks clearly looks wrong, and pin_user_pages_fast() is unconditionally called with FOLL_WRITE.

drivers/virt/acrn/mm.c +11 -1
index fa5d9ca6be57..563c545adb2c 100644
@@ -171,12 +171,22 @@ int acrn_vm_ram_map(struct acrn_vm *vm, struct acrn_vm_memmap *memmap)
        mmap_read_lock(current->mm);
        vma = vma_lookup(current->mm, memmap->vma_base);
        if (vma && ((vma->vm_flags & VM_PFNMAP) != 0)) {
+               spinlock_t *ptl;
+               pte_t *ptep;
+
                if ((memmap->vma_base + memmap->len) > vma->vm_end) {
                        mmap_read_unlock(current->mm);
                        return -EINVAL;
                }
 
-               ret = follow_pfn(vma, memmap->vma_base, &pfn);
+               ret = follow_pte(vma, memmap->vma_base, &ptep, &ptl);
+               if (!ret) {
+                       pfn = pte_pfn(ptep_get(ptep));
+                       if (!pte_write(ptep_get(ptep)) &&
+                           (memmap->attr & ACRN_MEM_ACCESS_WRITE))
+                               ret = -EFAULT;
+                       pte_unmap_unlock(ptep, ptl);
+               }
                mmap_read_unlock(current->mm);
                if (ret < 0) {
 
Signature · 3 lines
Cheers,

David / dhildenb
#3 lore raw

Re: [PATCH 1/3] virt: acrn: stop using follow_pfn

Quoted 7 lines · David Hildenbrand

... I have similar patches lying around here (see bwlow). I added some actual access permission checks.

(I also realized, that if we get an anon folio in a COW mapping via follow_pte() here, I suspect one might be able to do some nasty things. Just imagine if we munmap(), free the anon folio, and then it gets used in other context ... At least KVM/vfio handle that using references+MMU notifiers.)

How about you just send out your series that seems to go further and I retract mine?

#4 lore raw

Re: [PATCH 1/3] virt: acrn: stop using follow_pfn

Quoted 11 lines · Christoph Hellwig
Quoted 7 lines · David Hildenbrand

... I have similar patches lying around here (see bwlow). I added some actual access permission checks.

(I also realized, that if we get an anon folio in a COW mapping via follow_pte() here, I suspect one might be able to do some nasty things. Just imagine if we munmap(), free the anon folio, and then it gets used in other context ... At least KVM/vfio handle that using references+MMU notifiers.)

How about you just send out your series that seems to go further and I retract mine?

Let's go with yours first and I'll rebase.

Regarding above issue, I still have not made up my mind: likely we should reject any PFN in acrn that has a valid "struct page", and that page does not have PG_reserved set. That's what VFIO effectively does IIRC.

Signature · 3 lines
Cheers,

David / dhildenb
#5 lore raw

PATCH 2/3 mm: remove follow_pfn

In reply to Christoph Hellwig

Remove follow_pfn now that the last user is gone.

---
 include/linux/mm.h |  2 --
 mm/memory.c        | 36 ++----------------------------------
 mm/nommu.c         | 21 ---------------------
 3 files changed, 2 insertions(+), 57 deletions(-)
include/linux/mm.h +0 -2
index 0436b919f1c7fc..9cd2c69f913601 100644
@@ -2393,8 +2393,6 @@ int
 copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
 int follow_pte(struct mm_struct *mm, unsigned long address,
 	       pte_t **ptepp, spinlock_t **ptlp);
-int follow_pfn(struct vm_area_struct *vma, unsigned long address,
-	unsigned long *pfn);
 int follow_phys(struct vm_area_struct *vma, unsigned long address,
 		unsigned int flags, unsigned long *prot, resource_size_t *phys);
 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
mm/memory.c +2 -34
index f2bc6dd15eb830..d7f09d5aae6e53 100644
@@ -5880,8 +5880,8 @@ int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
  * Only IO mappings and raw PFN mappings are allowed.  The mmap semaphore
  * should be taken for read.
  *
- * KVM uses this function.  While it is arguably less bad than ``follow_pfn``,
- * it is not a good general-purpose API.
+ * KVM uses this function.  While it is arguably less bad than the historic
+ * ``follow_pfn``, it is not a good general-purpose API.
  *
  * Return: zero on success, -ve otherwise.
  */
@@ -5923,38 +5923,6 @@ int follow_pte(struct mm_struct *mm, unsigned long address,
 }
 EXPORT_SYMBOL_GPL(follow_pte);
 
-/**
- * follow_pfn - look up PFN at a user virtual address
- * @vma: memory mapping
- * @address: user virtual address
- * @pfn: location to store found PFN
- *
- * Only IO mappings and raw PFN mappings are allowed.
- *
- * This function does not allow the caller to read the permissions
- * of the PTE.  Do not use it.
- *
- * Return: zero and the pfn at @pfn on success, -ve otherwise.
- */
-int follow_pfn(struct vm_area_struct *vma, unsigned long address,
-	unsigned long *pfn)
-{
-	int ret = -EINVAL;
-	spinlock_t *ptl;
-	pte_t *ptep;
-
-	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
-		return ret;
-
-	ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
-	if (ret)
-		return ret;
-	*pfn = pte_pfn(ptep_get(ptep));
-	pte_unmap_unlock(ptep, ptl);
-	return 0;
-}
-EXPORT_SYMBOL(follow_pfn);
-
 #ifdef CONFIG_HAVE_IOREMAP_PROT
 int follow_phys(struct vm_area_struct *vma,
 		unsigned long address, unsigned int flags,
mm/nommu.c +0 -21
index 5ec8f44e7ce976..38d3ecc30c3ba9 100644
@@ -110,27 +110,6 @@ unsigned int kobjsize(const void *objp)
 	return page_size(page);
 }
 
-/**
- * follow_pfn - look up PFN at a user virtual address
- * @vma: memory mapping
- * @address: user virtual address
- * @pfn: location to store found PFN
- *
- * Only IO mappings and raw PFN mappings are allowed.
- *
- * Returns zero and the pfn at @pfn on success, -ve otherwise.
- */
-int follow_pfn(struct vm_area_struct *vma, unsigned long address,
-	unsigned long *pfn)
-{
-	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
-		return -EINVAL;
-
-	*pfn = address >> PAGE_SHIFT;
-	return 0;
-}
-EXPORT_SYMBOL(follow_pfn);
-
 void vfree(const void *addr)
 {
 	kfree(addr);
Signature · 1 line
2.39.2
#7 lore raw

PATCH 3/3 mm: move follow_phys to arch/x86/mm/pat/memtype.c

In reply to Christoph Hellwig

follow_phys is only used by two allers in arch/x86/mm/pat/memtype.c. Move it there and hardcode the two arguments that get the same values passed by both caller.

---
 arch/x86/mm/pat/memtype.c | 22 ++++++++++++++++++++--
 include/linux/mm.h        |  2 --
 mm/memory.c               | 28 ----------------------------
 3 files changed, 20 insertions(+), 32 deletions(-)
arch/x86/mm/pat/memtype.c +20 -2
index 0d72183b5dd028..bad99eb5c95b0d 100644
@@ -947,6 +947,24 @@ static void free_pfn_range(u64 paddr, unsigned long size)
 		memtype_free(paddr, paddr + size);
 }
 
+static int follow_phys(struct vm_area_struct *vma, unsigned long *prot,
+		resource_size_t *phys)
+{
+	pte_t *ptep, pte;
+	spinlock_t *ptl;
+
+	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
+		return -EINVAL;
+
+	if (follow_pte(vma->vm_mm, vma->vm_start, &ptep, &ptl))
+		return -EINVAL;
+	pte = ptep_get(ptep);
+	*prot = pgprot_val(pte_pgprot(pte));
+	*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
+	pte_unmap_unlock(ptep, ptl);
+	return 0;
+}
+
 /*
  * track_pfn_copy is called when vma that is covering the pfnmap gets
  * copied through copy_page_range().
@@ -966,7 +984,7 @@ int track_pfn_copy(struct vm_area_struct *vma)
 		 * reserve the whole chunk covered by vma. We need the
 		 * starting address and protection from pte.
 		 */
-		if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
+		if (follow_phys(vma, &prot, &paddr)) {
 			WARN_ON_ONCE(1);
 			return -EINVAL;
 		}
@@ -1053,7 +1071,7 @@ void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 	/* free the chunk starting from pfn or the whole chunk */
 	paddr = (resource_size_t)pfn << PAGE_SHIFT;
 	if (!paddr && !size) {
-		if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
+		if (follow_phys(vma, &prot, &paddr)) {
 			WARN_ON_ONCE(1);
 			return;
 		}
include/linux/mm.h +0 -2
index 9cd2c69f913601..51cfc8267da755 100644
@@ -2393,8 +2393,6 @@ int
 copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
 int follow_pte(struct mm_struct *mm, unsigned long address,
 	       pte_t **ptepp, spinlock_t **ptlp);
-int follow_phys(struct vm_area_struct *vma, unsigned long address,
-		unsigned int flags, unsigned long *prot, resource_size_t *phys);
 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
 			void *buf, int len, int write);
 
mm/memory.c +0 -28
index d7f09d5aae6e53..d18f6d15f56e6c 100644
@@ -5924,34 +5924,6 @@ int follow_pte(struct mm_struct *mm, unsigned long address,
 EXPORT_SYMBOL_GPL(follow_pte);
 
 #ifdef CONFIG_HAVE_IOREMAP_PROT
-int follow_phys(struct vm_area_struct *vma,
-		unsigned long address, unsigned int flags,
-		unsigned long *prot, resource_size_t *phys)
-{
-	int ret = -EINVAL;
-	pte_t *ptep, pte;
-	spinlock_t *ptl;
-
-	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
-		goto out;
-
-	if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
-		goto out;
-	pte = ptep_get(ptep);
-
-	if ((flags & FOLL_WRITE) && !pte_write(pte))
-		goto unlock;
-
-	*prot = pgprot_val(pte_pgprot(pte));
-	*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
-
-	ret = 0;
-unlock:
-	pte_unmap_unlock(ptep, ptl);
-out:
-	return ret;
-}
-
 /**
  * generic_access_phys - generic implementation for iomem mmap access
  * @vma: the vma to access
Signature · 1 line
2.39.2
#8 lore raw

Re: [PATCH 3/3] mm: move follow_phys to arch/x86/mm/pat/memtype.c

Quoted 3 lines · Christoph Hellwig

follow_phys is only used by two allers in arch/x86/mm/pat/memtype.c. Move it there and hardcode the two arguments that get the same values passed by both caller.

s/aller
/caller

s/both caller.
/both callers.

Quoted 23 lines
---
 arch/x86/mm/pat/memtype.c | 22 ++++++++++++++++++++--
 include/linux/mm.h        |  2 --
 mm/memory.c               | 28 ----------------------------
 3 files changed, 20 insertions(+), 32 deletions(-)
arch/x86/mm/pat/memtype.c +7 -0
index 0d72183b5dd028..bad99eb5c95b0d 100644
@@ -947,6 +947,24 @@ static void free_pfn_range(u64 paddr, unsigned long size)
 		memtype_free(paddr, paddr + size);
 }
 
+static int follow_phys(struct vm_area_struct *vma, unsigned long *prot,
+		resource_size_t *phys)
+{
+	pte_t *ptep, pte;
+	spinlock_t *ptl;
+
+	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))

arch/x86/mm/pat/memtype.c static void free_pfn_range(u64 paddr, unsigned long size) · line 966

Quoted 10 lines
arch/x86/mm/pat/memtype.c +10 -0
+		return -EINVAL;
+
+	if (follow_pte(vma->vm_mm, vma->vm_start, &ptep, &ptl))
+		return -EINVAL;
+	pte = ptep_get(ptep);
+	*prot = pgprot_val(pte_pgprot(pte));
+	*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
+	pte_unmap_unlock(ptep, ptl);
+	return 0;
+}

Please keep the readability newlines as the original had, ie.:

+ if (follow_pte(vma->vm_mm, vma->vm_start, &ptep, &ptl))
+ return -EINVAL;
+
+ pte = ptep_get(ptep);
+ *prot = pgprot_val(pte_pgprot(pte));
+ *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
+ pte_unmap_unlock(ptep, ptl);
+
+ return 0;

Thanks,

Ingo

#9 lore raw

Re: [PATCH 3/3] mm: move follow_phys to arch/x86/mm/pat/memtype.c

In reply to Christoph Hellwig
Quoted 20 lines · Christoph Hellwig

follow_phys is only used by two allers in arch/x86/mm/pat/memtype.c. Move it there and hardcode the two arguments that get the same values passed by both caller.

---
  arch/x86/mm/pat/memtype.c | 22 ++++++++++++++++++++--
  include/linux/mm.h        |  2 --
  mm/memory.c               | 28 ----------------------------
  3 files changed, 20 insertions(+), 32 deletions(-)
arch/x86/mm/pat/memtype.c +1 -0
index 0d72183b5dd028..bad99eb5c95b0d 100644
@@ -947,6 +947,24 @@ static void free_pfn_range(u64 paddr, unsigned long size)
  		memtype_free(paddr, paddr + size);
  }
 
+static int follow_phys(struct vm_area_struct *vma, unsigned long *prot,

arch/x86/mm/pat/memtype.c static void free_pfn_range(u64 paddr, unsigned long size) · line 960

Quoted 10 lines
arch/x86/mm/pat/memtype.c +10 -0
+		resource_size_t *phys)
+{
+	pte_t *ptep, pte;
+	spinlock_t *ptl;
+
+	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
+		return -EINVAL;
+
+	if (follow_pte(vma->vm_mm, vma->vm_start, &ptep, &ptl))
+		return -EINVAL;

I have a patch here that (a) makes follow_pte() consume a VMA instead of a MM, and (b) moves the VMA check into follow_pte().

While you're at it, and if you're resending this, you might want to include similar changes; otherwise I'll can send this myself later on top.

Signature · 3 lines
Cheers,

David / dhildenb