[PATCH 0/2] ceph: Properly clean up after failures in ceph_submit_write()

[PATCH 0/2] ceph: Properly clean up after failures in ceph_submit_write()

[Sam Edwards]

Hi list,

This series contains two patches from my previous series [1] that are
independent and have no outstanding review comments.

The patches fix a crash that can occur if ceph_inc_osd_stopping_blocker()
fails on write, and split out the complex page-array cleanup logic to a new
helper function (per Slava's suggestion). The first patch is intended for
stable.

There is no known reproducer for this race condition; it appears to be
timing-sensitive. I have a fault-injection patch that can trigger it, which I
can provide on request.

Warm regards,
Sam

[1] https://lore.kernel.org/all/20260107210139.40554-1-CFSworks@gmail.com/

Sam Edwards (2):
  ceph: free page array when ceph_submit_write() fails
  ceph: split out page-array discarding to a function

 fs/ceph/addr.c | 59 ++++++++++++++++++++++++++++++++------------------
 1 file changed, 38 insertions(+), 21 deletions(-)

-- 
2.52.0

[PATCH 1/2] ceph: free page array when ceph_submit_write() fails

[Sam Edwards]

If `locked_pages` is zero, the page array must not be allocated:
ceph_process_folio_batch() uses `locked_pages` to decide when to
allocate `pages`, and redundant allocations trigger
ceph_allocate_page_array()'s BUG_ON(), resulting in a worker oops (and
writeback stall) or even a kernel panic. Consequently, the main loop in
ceph_writepages_start() assumes that the lifetime of `pages` is confined
to a single iteration.

The ceph_submit_write() function claims ownership of the page array on
success (it is later freed when the write concludes). But failures only
redirty/unlock the pages and fail to free the array, making the failure
case in ceph_submit_write() fatal.

Free the page array (and reset locked_pages) in ceph_submit_write()'s
error-handling 'if' block so that the caller's invariant (that the array
does not remain in ceph_wbc) is maintained unconditionally, making
failures in ceph_submit_write() recoverable as originally intended.

Fixes: 1551ec61dc55 ("ceph: introduce ceph_submit_write() method")
Cc: stable@vger.kernel.org
Signed-off-by: Sam Edwards <CFSworks@gmail.com>
---
 fs/ceph/addr.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 63b75d214210..c3e0b5b429ea 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -1470,6 +1470,14 @@ int ceph_submit_write(struct address_space *mapping,
 			unlock_page(page);
 		}
 
+		if (ceph_wbc->from_pool) {
+			mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
+			ceph_wbc->from_pool = false;
+		} else
+			kfree(ceph_wbc->pages);
+		ceph_wbc->pages = NULL;
+		ceph_wbc->locked_pages = 0;
+
 		ceph_osdc_put_request(req);
 		return -EIO;
 	}
-- 
2.52.0

Re: [PATCH 1/2] ceph: free page array when ceph_submit_write() fails

[Viacheslav Dubeyko]

On Sun, 2026-01-25 at 18:27 -0800, Sam Edwards wrote:
> If `locked_pages` is zero, the page array must not be allocated:
> ceph_process_folio_batch() uses `locked_pages` to decide when to
> allocate `pages`, and redundant allocations trigger
> ceph_allocate_page_array()'s BUG_ON(), resulting in a worker oops (and
> writeback stall) or even a kernel panic. Consequently, the main loop in
> ceph_writepages_start() assumes that the lifetime of `pages` is confined
> to a single iteration.
> 
> The ceph_submit_write() function claims ownership of the page array on
> success (it is later freed when the write concludes). But failures only
> redirty/unlock the pages and fail to free the array, making the failure
> case in ceph_submit_write() fatal.
> 
> Free the page array (and reset locked_pages) in ceph_submit_write()'s
> error-handling 'if' block so that the caller's invariant (that the array
> does not remain in ceph_wbc) is maintained unconditionally, making
> failures in ceph_submit_write() recoverable as originally intended.
> 
> Fixes: 1551ec61dc55 ("ceph: introduce ceph_submit_write() method")
> Cc: stable@vger.kernel.org
> Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> ---
>  fs/ceph/addr.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> index 63b75d214210..c3e0b5b429ea 100644
> --- a/fs/ceph/addr.c
> +++ b/fs/ceph/addr.c
> @@ -1470,6 +1470,14 @@ int ceph_submit_write(struct address_space *mapping,
>  			unlock_page(page);
>  		}
>  
> +		if (ceph_wbc->from_pool) {
> +			mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
> +			ceph_wbc->from_pool = false;
> +		} else
> +			kfree(ceph_wbc->pages);
> +		ceph_wbc->pages = NULL;
> +		ceph_wbc->locked_pages = 0;
> +


I see the completely identical code pattern in two patches:

+	if (ceph_wbc->from_pool) {
+		mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
+		ceph_wbc->from_pool = false;
+	} else
+		kfree(ceph_wbc->pages);
+	ceph_wbc->pages = NULL;
+	ceph_wbc->locked_pages = 0;

I believe we need to introduce the inline function that can be reused in two
places.

Thanks,
Slava.

>  		ceph_osdc_put_request(req);
>  		return -EIO;
>  	}

Re: [PATCH 1/2] ceph: free page array when ceph_submit_write() fails

[Sam Edwards]

On Mon, Jan 26, 2026 at 2:51 PM Viacheslav Dubeyko
<Slava.Dubeyko@ibm.com> wrote:
>
> On Sun, 2026-01-25 at 18:27 -0800, Sam Edwards wrote:
> > If `locked_pages` is zero, the page array must not be allocated:
> > ceph_process_folio_batch() uses `locked_pages` to decide when to
> > allocate `pages`, and redundant allocations trigger
> > ceph_allocate_page_array()'s BUG_ON(), resulting in a worker oops (and
> > writeback stall) or even a kernel panic. Consequently, the main loop in
> > ceph_writepages_start() assumes that the lifetime of `pages` is confined
> > to a single iteration.
> >
> > The ceph_submit_write() function claims ownership of the page array on
> > success (it is later freed when the write concludes). But failures only
> > redirty/unlock the pages and fail to free the array, making the failure
> > case in ceph_submit_write() fatal.
> >
> > Free the page array (and reset locked_pages) in ceph_submit_write()'s
> > error-handling 'if' block so that the caller's invariant (that the array
> > does not remain in ceph_wbc) is maintained unconditionally, making
> > failures in ceph_submit_write() recoverable as originally intended.
> >
> > Fixes: 1551ec61dc55 ("ceph: introduce ceph_submit_write() method")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> > ---
> >  fs/ceph/addr.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> > index 63b75d214210..c3e0b5b429ea 100644
> > --- a/fs/ceph/addr.c
> > +++ b/fs/ceph/addr.c
> > @@ -1470,6 +1470,14 @@ int ceph_submit_write(struct address_space *mapping,
> >                       unlock_page(page);
> >               }
> >
> > +             if (ceph_wbc->from_pool) {
> > +                     mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
> > +                     ceph_wbc->from_pool = false;
> > +             } else
> > +                     kfree(ceph_wbc->pages);
> > +             ceph_wbc->pages = NULL;
> > +             ceph_wbc->locked_pages = 0;
> > +
>
>
> I see the completely identical code pattern in two patches:

The second patch only contains that pattern because it is moving it to
a separate function, patch 2 isn't introducing any *new* code.

>
> +       if (ceph_wbc->from_pool) {
> +               mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
> +               ceph_wbc->from_pool = false;
> +       } else
> +               kfree(ceph_wbc->pages);
> +       ceph_wbc->pages = NULL;
> +       ceph_wbc->locked_pages = 0;
>
> I believe we need to introduce the inline function that can be reused in two
> places.

Patch 2 is introducing that inline function as requested -- but that
function is not actually used in two places: for now (in this series),
it is only split out for better readability.

These patches are organized like this because of kernel development
norms: bugfixes intended for stable (such as this patch) should
consist of minimal, backport-friendly and correctness-focused changes.
Moving existing code to a new function is a separate change and does
not constitute a bugfix, so it needs to go in its own patch that isn't
Cc: stable.

Cheers,
Sam

>
> Thanks,
> Slava.
>
> >               ceph_osdc_put_request(req);
> >               return -EIO;
> >       }

Re: [PATCH 1/2] ceph: free page array when ceph_submit_write() fails

[Ilya Dryomov]

On Mon, Jan 26, 2026 at 3:27 AM Sam Edwards <cfsworks@gmail.com> wrote:
>
> If `locked_pages` is zero, the page array must not be allocated:
> ceph_process_folio_batch() uses `locked_pages` to decide when to
> allocate `pages`, and redundant allocations trigger
> ceph_allocate_page_array()'s BUG_ON(), resulting in a worker oops (and
> writeback stall) or even a kernel panic. Consequently, the main loop in
> ceph_writepages_start() assumes that the lifetime of `pages` is confined
> to a single iteration.
>
> The ceph_submit_write() function claims ownership of the page array on
> success (it is later freed when the write concludes). But failures only
> redirty/unlock the pages and fail to free the array, making the failure
> case in ceph_submit_write() fatal.
>
> Free the page array (and reset locked_pages) in ceph_submit_write()'s
> error-handling 'if' block so that the caller's invariant (that the array
> does not remain in ceph_wbc) is maintained unconditionally, making
> failures in ceph_submit_write() recoverable as originally intended.
>
> Fixes: 1551ec61dc55 ("ceph: introduce ceph_submit_write() method")
> Cc: stable@vger.kernel.org
> Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> ---
>  fs/ceph/addr.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> index 63b75d214210..c3e0b5b429ea 100644
> --- a/fs/ceph/addr.c
> +++ b/fs/ceph/addr.c
> @@ -1470,6 +1470,14 @@ int ceph_submit_write(struct address_space *mapping,
>                         unlock_page(page);
>                 }
>
> +               if (ceph_wbc->from_pool) {
> +                       mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
> +                       ceph_wbc->from_pool = false;
> +               } else
> +                       kfree(ceph_wbc->pages);
> +               ceph_wbc->pages = NULL;
> +               ceph_wbc->locked_pages = 0;

Hi Sam,

While I don't see anything wrong with the patch per se, I can't help
but question the existence of this entire branch along with the meaning
of the error.

ceph_writepages_start() is the only caller of ceph_submit_write() and
it already calls ceph_inc_osd_stopping_blocker() at the top where the
error can be handled naturally -- nothing needs to be unlocked or freed
at that point.  Since mdsc->stopping_blockers is just a counter, all
calls made by ceph_submit_write() invocations in a loop would be
"contained" within that ceph_writepages_start() call.  The only benefit
achieved is potentially faster response to the MDS client moving to
CEPH_MDSC_STOPPING_FLUSHING state, but it's rather dubious because
sneaking in/having to wait for some more OSD requests isn't really the
end of the world.

Rather than patching the error path, I wonder if instead of calling
ceph_inc_osd_stopping_blocker() the counter could just be incremented
unconditionally, with the check for CEPH_MDSC_STOPPING_FLUSHING bypassed
there?  This could be wrapped into a new helper that could also assert
that the counter is already elevated before the increment.

Thanks,

                Ilya

Re: [PATCH 1/2] ceph: free page array when ceph_submit_write() fails

[Ilya Dryomov]

On Wed, Feb 11, 2026 at 3:10 PM Ilya Dryomov <idryomov@gmail.com> wrote:
>
> On Mon, Jan 26, 2026 at 3:27 AM Sam Edwards <cfsworks@gmail.com> wrote:
> >
> > If `locked_pages` is zero, the page array must not be allocated:
> > ceph_process_folio_batch() uses `locked_pages` to decide when to
> > allocate `pages`, and redundant allocations trigger
> > ceph_allocate_page_array()'s BUG_ON(), resulting in a worker oops (and
> > writeback stall) or even a kernel panic. Consequently, the main loop in
> > ceph_writepages_start() assumes that the lifetime of `pages` is confined
> > to a single iteration.
> >
> > The ceph_submit_write() function claims ownership of the page array on
> > success (it is later freed when the write concludes). But failures only
> > redirty/unlock the pages and fail to free the array, making the failure
> > case in ceph_submit_write() fatal.
> >
> > Free the page array (and reset locked_pages) in ceph_submit_write()'s
> > error-handling 'if' block so that the caller's invariant (that the array
> > does not remain in ceph_wbc) is maintained unconditionally, making
> > failures in ceph_submit_write() recoverable as originally intended.
> >
> > Fixes: 1551ec61dc55 ("ceph: introduce ceph_submit_write() method")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> > ---
> >  fs/ceph/addr.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> > index 63b75d214210..c3e0b5b429ea 100644
> > --- a/fs/ceph/addr.c
> > +++ b/fs/ceph/addr.c
> > @@ -1470,6 +1470,14 @@ int ceph_submit_write(struct address_space *mapping,
> >                         unlock_page(page);
> >                 }
> >
> > +               if (ceph_wbc->from_pool) {
> > +                       mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
> > +                       ceph_wbc->from_pool = false;
> > +               } else
> > +                       kfree(ceph_wbc->pages);
> > +               ceph_wbc->pages = NULL;
> > +               ceph_wbc->locked_pages = 0;
>
> Hi Sam,
>
> While I don't see anything wrong with the patch per se, I can't help
> but question the existence of this entire branch along with the meaning
> of the error.
>
> ceph_writepages_start() is the only caller of ceph_submit_write() and
> it already calls ceph_inc_osd_stopping_blocker() at the top where the
> error can be handled naturally -- nothing needs to be unlocked or freed
> at that point.  Since mdsc->stopping_blockers is just a counter, all
> calls made by ceph_submit_write() invocations in a loop would be
> "contained" within that ceph_writepages_start() call.  The only benefit
> achieved is potentially faster response to the MDS client moving to
> CEPH_MDSC_STOPPING_FLUSHING state, but it's rather dubious because
> sneaking in/having to wait for some more OSD requests isn't really the
> end of the world.
>
> Rather than patching the error path, I wonder if instead of calling
> ceph_inc_osd_stopping_blocker() the counter could just be incremented
> unconditionally, with the check for CEPH_MDSC_STOPPING_FLUSHING bypassed
> there?  This could be wrapped into a new helper that could also assert
> that the counter is already elevated before the increment.

Something along the lines of

void __ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
{
        BUG_ON(!atomic_inc_not_zero(&mdsc->stopping_blockers));
}

and switching to __ceph_inc_osd_stopping_blocker() in place of
ceph_inc_osd_stopping_blocker() in ceph_submit_write().

Thanks,

                Ilya

[PATCH 2/2] ceph: split out page-array discarding to a function

[Sam Edwards]

Discarding a page array (i.e. after failure to submit it) is a little
complex:
- Every folio in the batch needs to be redirtied and unlocked.
- Some folios are bounce pages created for fscrypt; the underlying
  plaintext folios also need to be redirtied and unlocked.
- The array itself can come either from the mempool or general kalloc,
  so different free functions need to be used depending on which.

Although currently only ceph_submit_write() does this, this logic is
complex enough to warrant its own function. Move it to a new
ceph_discard_page_array() function that is called by ceph_submit_write()
instead.

Suggested-by: Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>
Reviewed-by: Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>
Signed-off-by: Sam Edwards <CFSworks@gmail.com>
---
 fs/ceph/addr.c | 67 ++++++++++++++++++++++++++++----------------------
 1 file changed, 38 insertions(+), 29 deletions(-)

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index c3e0b5b429ea..28463b9a3bae 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -1222,6 +1222,43 @@ void ceph_allocate_page_array(struct address_space *mapping,
 	ceph_wbc->len = 0;
 }
 
+static inline
+void ceph_discard_page_array(struct writeback_control *wbc,
+			     struct ceph_writeback_ctl *ceph_wbc)
+{
+	int i;
+	struct page *page;
+
+	for (i = 0; i < folio_batch_count(&ceph_wbc->fbatch); i++) {
+		struct folio *folio = ceph_wbc->fbatch.folios[i];
+
+		if (!folio)
+			continue;
+
+		page = &folio->page;
+		redirty_page_for_writepage(wbc, page);
+		unlock_page(page);
+	}
+
+	for (i = 0; i < ceph_wbc->locked_pages; i++) {
+		page = ceph_fscrypt_pagecache_page(ceph_wbc->pages[i]);
+
+		if (!page)
+			continue;
+
+		redirty_page_for_writepage(wbc, page);
+		unlock_page(page);
+	}
+
+	if (ceph_wbc->from_pool) {
+		mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
+		ceph_wbc->from_pool = false;
+	} else
+		kfree(ceph_wbc->pages);
+	ceph_wbc->pages = NULL;
+	ceph_wbc->locked_pages = 0;
+}
+
 static inline
 bool is_folio_index_contiguous(const struct ceph_writeback_ctl *ceph_wbc,
 			      const struct folio *folio)
@@ -1449,35 +1486,7 @@ int ceph_submit_write(struct address_space *mapping,
 	BUG_ON(len < ceph_fscrypt_page_offset(page) + thp_size(page) - offset);
 
 	if (!ceph_inc_osd_stopping_blocker(fsc->mdsc)) {
-		for (i = 0; i < folio_batch_count(&ceph_wbc->fbatch); i++) {
-			struct folio *folio = ceph_wbc->fbatch.folios[i];
-
-			if (!folio)
-				continue;
-
-			page = &folio->page;
-			redirty_page_for_writepage(wbc, page);
-			unlock_page(page);
-		}
-
-		for (i = 0; i < ceph_wbc->locked_pages; i++) {
-			page = ceph_fscrypt_pagecache_page(ceph_wbc->pages[i]);
-
-			if (!page)
-				continue;
-
-			redirty_page_for_writepage(wbc, page);
-			unlock_page(page);
-		}
-
-		if (ceph_wbc->from_pool) {
-			mempool_free(ceph_wbc->pages, ceph_wb_pagevec_pool);
-			ceph_wbc->from_pool = false;
-		} else
-			kfree(ceph_wbc->pages);
-		ceph_wbc->pages = NULL;
-		ceph_wbc->locked_pages = 0;
-
+		ceph_discard_page_array(wbc, ceph_wbc);
 		ceph_osdc_put_request(req);
 		return -EIO;
 	}
-- 
2.52.0