[PATCH 0/5] uprobes: write_opcode() cleanups

[PATCH 0/5] uprobes: write_opcode() cleanups

[Oleg Nesterov]

Hello,

write_opcode() cleanups I promised before.

Srikar, please review ;)

Oleg.

[PATCH 1/5] uprobes: don't recheck vma/f_mapping in write_opcode()

[Oleg Nesterov]

write_opcode() rechecks valid_vma() and ->f_mapping, this is pointless.
The caller, register_for_each_vma() or uprobe_mmap(), has already done
these checks under mmap_sem.

To clarify, uprobe_mmap() checks valid_vma() only, but we can rely on
build_probe_list(vm_file->f_mapping->host).

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 kernel/events/uprobes.c |   19 +------------------
 1 files changed, 1 insertions(+), 18 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index f935327..a2b32a5 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -206,33 +206,16 @@ static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
 			unsigned long vaddr, uprobe_opcode_t opcode)
 {
 	struct page *old_page, *new_page;
-	struct address_space *mapping;
 	void *vaddr_old, *vaddr_new;
 	struct vm_area_struct *vma;
-	struct uprobe *uprobe;
 	int ret;
+
 retry:
 	/* Read the page with vaddr into memory */
 	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
 	if (ret <= 0)
 		return ret;
 
-	ret = -EINVAL;
-
-	/*
-	 * We are interested in text pages only. Our pages of interest
-	 * should be mapped for read and execute only. We desist from
-	 * adding probes in write mapped pages since the breakpoints
-	 * might end up in the file copy.
-	 */
-	if (!valid_vma(vma, is_swbp_insn(&opcode)))
-		goto put_out;
-
-	uprobe = container_of(auprobe, struct uprobe, arch);
-	mapping = uprobe->inode->i_mapping;
-	if (mapping != vma->vm_file->f_mapping)
-		goto put_out;
-
 	ret = -ENOMEM;
 	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
 	if (!new_page)
-- 
1.5.5.1

[PATCH 2/5] uprobes: __replace_page() should not use page_address_in_vma()

[Oleg Nesterov]

page_address_in_vma(old_page) in __replace_page() is ugly and wrong.
The caller already knows the correct virtual address, this page was
found by get_user_pages(vaddr).

However, page_address_in_vma() can actually fail if page->mapping was
cleared by __delete_from_page_cache() after get_user_pages() returns.
But this means the race with page reclaim, write_opcode() should not
fail, it should retry and read this page again. Not sure this race is
really possible though, page_freeze_refs() logic should prevent it.

We could change __replace_page() to return -EAGAIN in this case, but
it would be better to simply use the caller's vaddr and rely on
page_check_address().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 kernel/events/uprobes.c |   10 +++-------
 1 files changed, 3 insertions(+), 7 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index a2b32a5..5b10705 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -132,17 +132,13 @@ static loff_t vma_address(struct vm_area_struct *vma, loff_t offset)
  *
  * Returns 0 on success, -EFAULT on failure.
  */
-static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage)
+static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
+				struct page *page, struct page *kpage)
 {
 	struct mm_struct *mm = vma->vm_mm;
-	unsigned long addr;
 	spinlock_t *ptl;
 	pte_t *ptep;
 
-	addr = page_address_in_vma(page, vma);
-	if (addr == -EFAULT)
-		return -EFAULT;
-
 	ptep = page_check_address(page, mm, addr, &ptl, 0);
 	if (!ptep)
 		return -EAGAIN;
@@ -243,7 +239,7 @@ retry:
 		goto unlock_out;
 
 	lock_page(new_page);
-	ret = __replace_page(vma, old_page, new_page);
+	ret = __replace_page(vma, vaddr, old_page, new_page);
 	unlock_page(new_page);
 
 unlock_out:
-- 
1.5.5.1

[PATCH 3/5] uprobes: kill write_opcode()->lock_page(new_page)

[Oleg Nesterov]

write_opcode() does lock_page(new_page) for no reason. Nobody can
see this page until __replace_page() exposes it under ptl lock, and
we do nothing with this page after pte_unmap_unlock().

If nothing else, the similar code in do_wp_page() doesn't lock the
new page for page_add_new_anon_rmap/set_pte_at_notify.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 kernel/events/uprobes.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 5b10705..16a3b69 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -238,9 +238,7 @@ retry:
 	if (ret)
 		goto unlock_out;
 
-	lock_page(new_page);
 	ret = __replace_page(vma, vaddr, old_page, new_page);
-	unlock_page(new_page);
 
 unlock_out:
 	unlock_page(old_page);
-- 
1.5.5.1

[PATCH 4/5] uprobes: cleanup and document write_opcode()->lock_page(old_page)

[Oleg Nesterov]

The comment above write_opcode()->lock_page(old_page) tells about
the race with do_wp_page(). I don't really understand which exactly
race it means, but afaics this lock_page() was not enough to close
all races with do_wp_page().

Anyway, since 77fc4af1 this code is always called with ->mmap_sem
hold for writing so we can forget about do_wp_page().

However, we can't simply remove this lock_page(), and the only
(afaics) reason is __replace_page()->try_to_free_swap().

Nothing in write_opcode() needs it, move it into __replace_page()
and fix the comment.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 kernel/events/uprobes.c |   27 ++++++++++++++-------------
 1 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 16a3b69..442064d 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -138,10 +138,15 @@ static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
 	struct mm_struct *mm = vma->vm_mm;
 	spinlock_t *ptl;
 	pte_t *ptep;
+	int err;
 
+	/* freeze PageSwapCache() for try_to_free_swap() below */
+	lock_page(page);
+
+	err = -EAGAIN;
 	ptep = page_check_address(page, mm, addr, &ptl, 0);
 	if (!ptep)
-		return -EAGAIN;
+		goto unlock;
 
 	get_page(kpage);
 	page_add_new_anon_rmap(kpage, vma, addr);
@@ -161,7 +166,10 @@ static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
 	put_page(page);
 	pte_unmap_unlock(ptep, ptl);
 
-	return 0;
+	err = 0;
+ unlock:
+	unlock_page(page);
+	return err;
 }
 
 /**
@@ -215,15 +223,10 @@ retry:
 	ret = -ENOMEM;
 	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
 	if (!new_page)
-		goto put_out;
+		goto put_old;
 
 	__SetPageUptodate(new_page);
 
-	/*
-	 * lock page will serialize against do_wp_page()'s
-	 * PageAnon() handling
-	 */
-	lock_page(old_page);
 	/* copy the page now that we've got it stable */
 	vaddr_old = kmap_atomic(old_page);
 	vaddr_new = kmap_atomic(new_page);
@@ -236,15 +239,13 @@ retry:
 
 	ret = anon_vma_prepare(vma);
 	if (ret)
-		goto unlock_out;
+		goto put_new;
 
 	ret = __replace_page(vma, vaddr, old_page, new_page);
 
-unlock_out:
-	unlock_page(old_page);
+put_new:
 	page_cache_release(new_page);
-
-put_out:
+put_old:
 	put_page(old_page);
 
 	if (unlikely(ret == -EAGAIN))
-- 
1.5.5.1

[PATCH 5/5] uprobes: write_opcode: alloc the new page outside of "retry" loop

[Oleg Nesterov]

It is ugly to free and re-alloc new_page if write_opcode() needs to
retry. Move alloc_page_vma/page_cache_release outside of "retry" loop.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 kernel/events/uprobes.c |   26 +++++++++++---------------
 1 files changed, 11 insertions(+), 15 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 442064d..3fd7bdf 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -214,19 +214,18 @@ static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
 	struct vm_area_struct *vma;
 	int ret;
 
-retry:
-	/* Read the page with vaddr into memory */
-	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
-	if (ret <= 0)
-		return ret;
-
-	ret = -ENOMEM;
 	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
 	if (!new_page)
-		goto put_old;
+		return -ENOMEM;
 
 	__SetPageUptodate(new_page);
 
+ retry:
+	/* Read the page with vaddr into memory */
+	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
+	if (ret <= 0)
+		goto out;
+
 	/* copy the page now that we've got it stable */
 	vaddr_old = kmap_atomic(old_page);
 	vaddr_new = kmap_atomic(new_page);
@@ -238,18 +237,15 @@ retry:
 	kunmap_atomic(vaddr_old);
 
 	ret = anon_vma_prepare(vma);
-	if (ret)
-		goto put_new;
+	if (!ret)
+		ret = __replace_page(vma, vaddr, old_page, new_page);
 
-	ret = __replace_page(vma, vaddr, old_page, new_page);
-
-put_new:
-	page_cache_release(new_page);
-put_old:
 	put_page(old_page);
 
 	if (unlikely(ret == -EAGAIN))
 		goto retry;
+ out:
+	page_cache_release(new_page);
 	return ret;
 }
 
-- 
1.5.5.1

Re: [PATCH 2/5] uprobes: __replace_page() should not use page_address_in_vma()

[Srikar Dronamraju]

* Oleg Nesterov <oleg@redhat.com> [2012-06-19 21:47:12]:

> page_address_in_vma(old_page) in __replace_page() is ugly and wrong.
> The caller already knows the correct virtual address, this page was
> found by get_user_pages(vaddr).
> 
> However, page_address_in_vma() can actually fail if page->mapping was
> cleared by __delete_from_page_cache() after get_user_pages() returns.
> But this means the race with page reclaim, write_opcode() should not
> fail, it should retry and read this page again. Not sure this race is
> really possible though, page_freeze_refs() logic should prevent it.
> 
> We could change __replace_page() to return -EAGAIN in this case, but
> it would be better to simply use the caller's vaddr and rely on
> page_check_address().
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> ---
>  kernel/events/uprobes.c |   10 +++-------
>  1 files changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index a2b32a5..5b10705 100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -132,17 +132,13 @@ static loff_t vma_address(struct vm_area_struct *vma, loff_t offset)
>   *
>   * Returns 0 on success, -EFAULT on failure.
>   */
> -static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage)
> +static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
> +				struct page *page, struct page *kpage)

Could please update the comment above __replace_page to mention that it
now takes addr as a parameter?

Re: [PATCH 5/5] uprobes: write_opcode: alloc the new page outside of "retry" loop

[Anton Arapov]

On Tue, Jun 19, 2012 at 09:47:59PM +0200, Oleg Nesterov wrote:
> It is ugly to free and re-alloc new_page if write_opcode() needs to
> retry. Move alloc_page_vma/page_cache_release outside of "retry" loop.
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> ---
>  kernel/events/uprobes.c |   26 +++++++++++---------------
>  1 files changed, 11 insertions(+), 15 deletions(-)
> 
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index 442064d..3fd7bdf 100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -214,19 +214,18 @@ static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
>  	struct vm_area_struct *vma;
>  	int ret;
>  
> -retry:
> -	/* Read the page with vaddr into memory */
> -	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
> -	if (ret <= 0)
> -		return ret;
> -
> -	ret = -ENOMEM;
>  	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
  Should we want NULL instead of vma in alloc_page_vma()? Or
initialize vma with NULL...

>  	if (!new_page)
> -		goto put_old;
> +		return -ENOMEM;
>  
>  	__SetPageUptodate(new_page);
>  
> + retry:
> +	/* Read the page with vaddr into memory */
> +	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
> +	if (ret <= 0)
> +		goto out;
> +
[cut]

Anton.

Re: [PATCH 2/5] uprobes: __replace_page() should not use page_address_in_vma()

[Oleg Nesterov]

On 06/20, Srikar Dronamraju wrote:
>
> > -static int __replace_page(struct vm_area_struct *vma, struct page *page, struct page *kpage)
> > +static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
> > +				struct page *page, struct page *kpage)
>
> Could please update the comment above __replace_page to mention that it
> now takes addr as a parameter?

Will do, thanks.

Oleg.

Re: [PATCH 5/5] uprobes: write_opcode: alloc the new page outside of "retry" loop

[Oleg Nesterov]

On 06/20, Anton Arapov wrote:
>
> On Tue, Jun 19, 2012 at 09:47:59PM +0200, Oleg Nesterov wrote:
> > @@ -214,19 +214,18 @@ static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
> >  	struct vm_area_struct *vma;
> >  	int ret;
> >
> > -retry:
> > -	/* Read the page with vaddr into memory */
> > -	ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
> > -	if (ret <= 0)
> > -		return ret;
> > -
> > -	ret = -ENOMEM;
> >  	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
>   Should we want NULL instead of vma in alloc_page_vma()? Or
> initialize vma with NULL...

OOPS. Thanks Anton, this should be fixed.

Oleg.