2016-11-01 21:06:18

by Jan Kara

Subject: [PATCH 08/11] ext4: Convert DAX faults to iomap infrastructure

Convert DAX faults to use iomap infrastructure. We would not have to start
transaction in ext4_dax_fault() anymore since ext4_iomap_begin takes
care of that but so far we do that to avoid lock inversion of
transaction start with DAX entry lock which gets acquired in
dax_iomap_fault() before calling ->iomap_begin handler.

Signed-off-by: Jan Kara <[email protected]>
---
fs/ext4/ext4.h | 1 +
fs/ext4/file.c | 9 +++++----
fs/ext4/inode.c | 18 ++++++++++++++++++
3 files changed, 24 insertions(+), 4 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 098b39910001..2714eb6174ab 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -3272,6 +3272,7 @@ static inline bool ext4_aligned_io(struct inode *inode, loff_t off, loff_t len)
}

extern struct iomap_ops ext4_iomap_ops;
+extern struct iomap_ops ext4_iomap_fault_ops;

#endif /* __KERNEL__ */

diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index d7ab0e90d1b8..da44e49c8276 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -273,7 +273,7 @@ static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
if (IS_ERR(handle))
result = VM_FAULT_SIGBUS;
else
- result = dax_fault(vma, vmf, ext4_dax_get_block);
+ result = dax_iomap_fault(vma, vmf, &ext4_iomap_fault_ops);

if (write) {
if (!IS_ERR(handle))
@@ -307,9 +307,10 @@ static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,

if (IS_ERR(handle))
result = VM_FAULT_SIGBUS;
- else
- result = dax_pmd_fault(vma, addr, pmd, flags,
- ext4_dax_get_block);
+ else {
+ result = dax_iomap_pmd_fault(vma, addr, pmd, flags,
+ &ext4_iomap_fault_ops);
+ }

if (write) {
if (!IS_ERR(handle))
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 635518dde20e..001fef06ea97 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3419,11 +3419,29 @@ static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
return 0;
}

+/*
+ * For faults we don't allocate any blocks outside of isize and we don't want
+ * to change it so we use a dedicated function for it...
+ */
+static int ext4_iomap_fault_end(struct inode *inode, loff_t offset,
+ loff_t length, ssize_t written, unsigned flags,
+ struct iomap *iomap)
+{
+ if (flags & IOMAP_WRITE)
+ ext4_journal_stop(ext4_journal_current_handle());
+ return 0;
+}
+
struct iomap_ops ext4_iomap_ops = {
.iomap_begin = ext4_iomap_begin,
.iomap_end = ext4_iomap_end,
};

+struct iomap_ops ext4_iomap_fault_ops = {
+ .iomap_begin = ext4_iomap_begin,
+ .iomap_end = ext4_iomap_fault_end,
+};
+
#else
/* Just define empty function, it will never get called. */
int ext4_dax_get_block(struct inode *inode, sector_t iblock,
--
2.6.6

2016-11-02 14:30:06

by Christoph Hellwig

Subject: Re: [PATCH 08/11] ext4: Convert DAX faults to iomap infrastructure

On Tue, Nov 01, 2016 at 10:06:18PM +0100, Jan Kara wrote:
> Convert DAX faults to use iomap infrastructure. We would not have to start
> transaction in ext4_dax_fault() anymore since ext4_iomap_begin takes
> care of that but so far we do that to avoid lock inversion of
> transaction start with DAX entry lock which gets acquired in
> dax_iomap_fault() before calling ->iomap_begin handler.

So far I've tried to avoid the need to avoid having multiple iomap_ops
for the same fs. Would you be fine with a new IOMAP_FAULT flag for write
faults?

2016-11-04 00:02:30

by Jan Kara

Subject: Re: [PATCH 08/11] ext4: Convert DAX faults to iomap infrastructure

On Wed 02-11-16 07:30:06, Christoph Hellwig wrote:
> On Tue, Nov 01, 2016 at 10:06:18PM +0100, Jan Kara wrote:
> > Convert DAX faults to use iomap infrastructure. We would not have to start
> > transaction in ext4_dax_fault() anymore since ext4_iomap_begin takes
> > care of that but so far we do that to avoid lock inversion of
> > transaction start with DAX entry lock which gets acquired in
> > dax_iomap_fault() before calling ->iomap_begin handler.
>
> So far I've tried to avoid the need to avoid having multiple iomap_ops
> for the same fs. Would you be fine with a new IOMAP_FAULT flag for write
> faults?

Yes, that's another option. If that's better, I'll redo the patch like
that.
Honza
--
Jan Kara <jack-IBi9RG/[email protected]>
SUSE Labs, CR