| Message ID | 20221214102409.1857526-1-agruenba@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] iomap: Move page_done callback under the folio lock | expand |
On Wed, Dec 14, 2022 at 11:24 AM Andreas Gruenbacher <agruenba@redhat.com> wrote: > > Move the ->page_done() call in iomap_write_end() under the folio lock. > This closes a race between journaled data writes and the shrinker in > gfs2. What's happening is that gfs2_iomap_page_done() is called after > the page has been unlocked, so try_to_free_buffers() can come in and > free the buffers while gfs2_iomap_page_done() is trying to add them to > the current transaction. The folio lock prevents that from happening. > > The only current user of ->page_done() is gfs2, so other filesystems are > not affected. Still, to catch out any new users, switch from page to > folio in ->page_done(). > > Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com> > --- > fs/gfs2/bmap.c | 7 ++++--- > fs/iomap/buffered-io.c | 4 ++-- > include/linux/iomap.h | 10 +++++----- > 3 files changed, 11 insertions(+), 10 deletions(-) > > diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c > index e7537fd305dd..c4ee47f8e499 100644 > --- a/fs/gfs2/bmap.c > +++ b/fs/gfs2/bmap.c > @@ -968,14 +968,15 @@ static int gfs2_iomap_page_prepare(struct inode *inode, loff_t pos, > } > > static void gfs2_iomap_page_done(struct inode *inode, loff_t pos, > - unsigned copied, struct page *page) > + unsigned copied, struct folio *folio) > { > struct gfs2_trans *tr = current->journal_info; > struct gfs2_inode *ip = GFS2_I(inode); > struct gfs2_sbd *sdp = GFS2_SB(inode); > > - if (page && !gfs2_is_stuffed(ip)) > - gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); > + if (folio && !gfs2_is_stuffed(ip)) > + gfs2_page_add_databufs(ip, &folio->page, offset_in_page(pos), > + copied); > > if (tr->tr_num_buf_new) > __mark_inode_dirty(inode, I_DIRTY_DATASYNC); This is still screwed up. We really need to unlock the page before calling into __mark_inode_dirty() and ending the transaction. The current page_done() hook would force us to then re-lock the page just so that the caller can unlock it again. This just doesn't make sense, particularly since the page_prepare and page_done hooks only exist to allow gfs2 to do data journaling via iomap. I'll follow up with a more useful approach ... > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c > index 91ee0b308e13..d988c1bedf70 100644 > --- a/fs/iomap/buffered-io.c > +++ b/fs/iomap/buffered-io.c > @@ -714,12 +714,12 @@ static size_t iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len, > i_size_write(iter->inode, pos + ret); > iter->iomap.flags |= IOMAP_F_SIZE_CHANGED; > } > + if (page_ops && page_ops->page_done) > + page_ops->page_done(iter->inode, pos, ret, folio); > folio_unlock(folio); > > if (old_size < pos) > pagecache_isize_extended(iter->inode, old_size, pos); > - if (page_ops && page_ops->page_done) > - page_ops->page_done(iter->inode, pos, ret, &folio->page); > folio_put(folio); > > if (ret < len) > diff --git a/include/linux/iomap.h b/include/linux/iomap.h > index 238a03087e17..bd6d80453726 100644 > --- a/include/linux/iomap.h > +++ b/include/linux/iomap.h > @@ -116,18 +116,18 @@ static inline bool iomap_inline_data_valid(const struct iomap *iomap) > > /* > * When a filesystem sets page_ops in an iomap mapping it returns, page_prepare > - * and page_done will be called for each page written to. This only applies to > - * buffered writes as unbuffered writes will not typically have pages > + * and page_done will be called for each folio written to. This only applies > + * to buffered writes as unbuffered writes will not typically have folios > * associated with them. > * > * When page_prepare succeeds, page_done will always be called to do any > - * cleanup work necessary. In that page_done call, @page will be NULL if the > - * associated page could not be obtained. > + * cleanup work necessary. In that page_done call, @folio will be NULL if the > + * associated folio could not be obtained. > */ > struct iomap_page_ops { > int (*page_prepare)(struct inode *inode, loff_t pos, unsigned len); > void (*page_done)(struct inode *inode, loff_t pos, unsigned copied, > - struct page *page); > + struct folio *folio); > }; > > /* > -- > 2.38.1 >
On Thu, Dec 15, 2022 at 09:13:50PM +0100, Andreas Gruenbacher wrote: > This is still screwed up. We really need to unlock the page before > calling into __mark_inode_dirty() and ending the transaction. The > current page_done() hook would force us to then re-lock the page just > so that the caller can unlock it again. This just doesn't make sense, > particularly since the page_prepare and page_done hooks only exist to > allow gfs2 to do data journaling via iomap. I'll follow up with a more > useful approach ... Yes. And it would make sense to include the gfs2 patches.
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c index e7537fd305dd..c4ee47f8e499 100644 --- a/fs/gfs2/bmap.c +++ b/fs/gfs2/bmap.c @@ -968,14 +968,15 @@ static int gfs2_iomap_page_prepare(struct inode *inode, loff_t pos, } static void gfs2_iomap_page_done(struct inode *inode, loff_t pos, - unsigned copied, struct page *page) + unsigned copied, struct folio *folio) { struct gfs2_trans *tr = current->journal_info; struct gfs2_inode *ip = GFS2_I(inode); struct gfs2_sbd *sdp = GFS2_SB(inode); - if (page && !gfs2_is_stuffed(ip)) - gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); + if (folio && !gfs2_is_stuffed(ip)) + gfs2_page_add_databufs(ip, &folio->page, offset_in_page(pos), + copied); if (tr->tr_num_buf_new) __mark_inode_dirty(inode, I_DIRTY_DATASYNC); diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c index 91ee0b308e13..d988c1bedf70 100644 --- a/fs/iomap/buffered-io.c +++ b/fs/iomap/buffered-io.c @@ -714,12 +714,12 @@ static size_t iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len, i_size_write(iter->inode, pos + ret); iter->iomap.flags |= IOMAP_F_SIZE_CHANGED; } + if (page_ops && page_ops->page_done) + page_ops->page_done(iter->inode, pos, ret, folio); folio_unlock(folio); if (old_size < pos) pagecache_isize_extended(iter->inode, old_size, pos); - if (page_ops && page_ops->page_done) - page_ops->page_done(iter->inode, pos, ret, &folio->page); folio_put(folio); if (ret < len) diff --git a/include/linux/iomap.h b/include/linux/iomap.h index 238a03087e17..bd6d80453726 100644 --- a/include/linux/iomap.h +++ b/include/linux/iomap.h @@ -116,18 +116,18 @@ static inline bool iomap_inline_data_valid(const struct iomap *iomap) /* * When a filesystem sets page_ops in an iomap mapping it returns, page_prepare - * and page_done will be called for each page written to. This only applies to - * buffered writes as unbuffered writes will not typically have pages + * and page_done will be called for each folio written to. This only applies + * to buffered writes as unbuffered writes will not typically have folios * associated with them. * * When page_prepare succeeds, page_done will always be called to do any - * cleanup work necessary. In that page_done call, @page will be NULL if the - * associated page could not be obtained. + * cleanup work necessary. In that page_done call, @folio will be NULL if the + * associated folio could not be obtained. */ struct iomap_page_ops { int (*page_prepare)(struct inode *inode, loff_t pos, unsigned len); void (*page_done)(struct inode *inode, loff_t pos, unsigned copied, - struct page *page); + struct folio *folio); }; /*
Move the ->page_done() call in iomap_write_end() under the folio lock. This closes a race between journaled data writes and the shrinker in gfs2. What's happening is that gfs2_iomap_page_done() is called after the page has been unlocked, so try_to_free_buffers() can come in and free the buffers while gfs2_iomap_page_done() is trying to add them to the current transaction. The folio lock prevents that from happening. The only current user of ->page_done() is gfs2, so other filesystems are not affected. Still, to catch out any new users, switch from page to folio in ->page_done(). Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com> --- fs/gfs2/bmap.c | 7 ++++--- fs/iomap/buffered-io.c | 4 ++-- include/linux/iomap.h | 10 +++++----- 3 files changed, 11 insertions(+), 10 deletions(-)