Hello,
these two patches fix handling of ENOSPC during DAX faults. The problem is
that currently running transaction may be holding lots of already freed
blocks which can be reallocated only once the transaction commits. Standard
retry logic in ext4_iomap_end() does not work for DAX page fault handler
since we hold current transaction open in ext4_dax_huge_fault() and thus
retry logic cannot force the running transaction and as a result application
gets SIGBUS due to premature ENOSPC error.
Changes since v1:
* Made error passed back from dax_iomap_fault() explicitely be only the error
from ->iomap_begin()
Comments are welcome.
Honza
Ext4 needs to pass through error from its iomap handler to the page
fault handler so that it can properly detect ENOSPC and force
transaction commit and retry the fault (and block allocation). Add
argument to dax_iomap_fault() for passing such error.
Signed-off-by: Jan Kara <[email protected]>
---
fs/dax.c | 9 ++++++---
fs/ext2/file.c | 2 +-
fs/ext4/file.c | 2 +-
fs/xfs/xfs_file.c | 2 +-
include/linux/dax.h | 2 +-
5 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/fs/dax.c b/fs/dax.c
index 95981591977a..f3afa1d6156c 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -1096,7 +1096,7 @@ static bool dax_fault_is_synchronous(unsigned long flags,
}
static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
- const struct iomap_ops *ops)
+ int *iomap_errp, const struct iomap_ops *ops)
{
struct vm_area_struct *vma = vmf->vma;
struct address_space *mapping = vma->vm_file->f_mapping;
@@ -1149,6 +1149,8 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
* that we never have to deal with more than a single extent here.
*/
error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap);
+ if (iomap_errp)
+ *iomap_errp = error;
if (error) {
vmf_ret = dax_fault_return(error);
goto unlock_entry;
@@ -1488,6 +1490,7 @@ static int dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
* @vmf: The description of the fault
* @pe_size: Size of the page to fault in
* @pfnp: PFN to insert for synchronous faults if fsync is required
+ * @iomap_errp: Storage for detailed error code in case of error
* @ops: Iomap ops passed from the file system
*
* When a page fault occurs, filesystems may call this helper in
@@ -1496,11 +1499,11 @@ static int dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
* successfully.
*/
int dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size,
- pfn_t *pfnp, const struct iomap_ops *ops)
+ pfn_t *pfnp, int *iomap_errp, const struct iomap_ops *ops)
{
switch (pe_size) {
case PE_SIZE_PTE:
- return dax_iomap_pte_fault(vmf, pfnp, ops);
+ return dax_iomap_pte_fault(vmf, pfnp, iomap_errp, ops);
case PE_SIZE_PMD:
return dax_iomap_pmd_fault(vmf, pfnp, ops);
default:
diff --git a/fs/ext2/file.c b/fs/ext2/file.c
index 2da67699dc33..09640220fda8 100644
--- a/fs/ext2/file.c
+++ b/fs/ext2/file.c
@@ -100,7 +100,7 @@ static int ext2_dax_fault(struct vm_fault *vmf)
}
down_read(&ei->dax_sem);
- ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, &ext2_iomap_ops);
+ ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, NULL, &ext2_iomap_ops);
up_read(&ei->dax_sem);
if (vmf->flags & FAULT_FLAG_WRITE)
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index a0ae27b1bc66..1c7cd882d998 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -314,7 +314,7 @@ static int ext4_dax_huge_fault(struct vm_fault *vmf,
} else {
down_read(&EXT4_I(inode)->i_mmap_sem);
}
- result = dax_iomap_fault(vmf, pe_size, &pfn, &ext4_iomap_ops);
+ result = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &ext4_iomap_ops);
if (write) {
ext4_journal_stop(handle);
/* Handling synchronous page fault? */
diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 8601275cc5e6..9ea08326f876 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -1048,7 +1048,7 @@ __xfs_filemap_fault(
if (IS_DAX(inode)) {
pfn_t pfn;
- ret = dax_iomap_fault(vmf, pe_size, &pfn, &xfs_iomap_ops);
+ ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
if (ret & VM_FAULT_NEEDDSYNC)
ret = dax_finish_sync_fault(vmf, pe_size, pfn);
} else {
diff --git a/include/linux/dax.h b/include/linux/dax.h
index 5258346c558c..0185ecdae135 100644
--- a/include/linux/dax.h
+++ b/include/linux/dax.h
@@ -96,7 +96,7 @@ bool dax_write_cache_enabled(struct dax_device *dax_dev);
ssize_t dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
const struct iomap_ops *ops);
int dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size,
- pfn_t *pfnp, const struct iomap_ops *ops);
+ pfn_t *pfnp, int *errp, const struct iomap_ops *ops);
int dax_finish_sync_fault(struct vm_fault *vmf, enum page_entry_size pe_size,
pfn_t pfn);
int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index);
--
2.12.3
On Thu, Dec 21, 2017 at 8:30 AM, Jan Kara <[email protected]> wrote:
> Ext4 needs to pass through error from its iomap handler to the page
> fault handler so that it can properly detect ENOSPC and force
> transaction commit and retry the fault (and block allocation). Add
> argument to dax_iomap_fault() for passing such error.
>
> Signed-off-by: Jan Kara <[email protected]>
> ---
> fs/dax.c | 9 ++++++---
> fs/ext2/file.c | 2 +-
> fs/ext4/file.c | 2 +-
> fs/xfs/xfs_file.c | 2 +-
> include/linux/dax.h | 2 +-
> 5 files changed, 10 insertions(+), 7 deletions(-)
>
> diff --git a/fs/dax.c b/fs/dax.c
> index 95981591977a..f3afa1d6156c 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -1096,7 +1096,7 @@ static bool dax_fault_is_synchronous(unsigned long flags,
> }
>
> static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> - const struct iomap_ops *ops)
> + int *iomap_errp, const struct iomap_ops *ops)
> {
> struct vm_area_struct *vma = vmf->vma;
> struct address_space *mapping = vma->vm_file->f_mapping;
> @@ -1149,6 +1149,8 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> * that we never have to deal with more than a single extent here.
> */
> error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap);
> + if (iomap_errp)
> + *iomap_errp = error;
Since we already have 'struct iomap' tracking the state of the iomap
should we track the error status there as well? I.e. move the on
stack allocation of struct iomap to the per-fs dax fault handlers.
On Thu, Dec 21, 2017 at 05:30:54PM +0100, Jan Kara wrote:
> Ext4 needs to pass through error from its iomap handler to the page
> fault handler so that it can properly detect ENOSPC and force
> transaction commit and retry the fault (and block allocation). Add
> argument to dax_iomap_fault() for passing such error.
>
> Signed-off-by: Jan Kara <[email protected]>
I like how much simpler this version is. Looks good.
Reviewed-by: Ross Zwisler <[email protected]>
On Thu 21-12-17 09:12:52, Dan Williams wrote:
> On Thu, Dec 21, 2017 at 8:30 AM, Jan Kara <[email protected]> wrote:
> > Ext4 needs to pass through error from its iomap handler to the page
> > fault handler so that it can properly detect ENOSPC and force
> > transaction commit and retry the fault (and block allocation). Add
> > argument to dax_iomap_fault() for passing such error.
> >
> > Signed-off-by: Jan Kara <[email protected]>
> > ---
> > fs/dax.c | 9 ++++++---
> > fs/ext2/file.c | 2 +-
> > fs/ext4/file.c | 2 +-
> > fs/xfs/xfs_file.c | 2 +-
> > include/linux/dax.h | 2 +-
> > 5 files changed, 10 insertions(+), 7 deletions(-)
> >
> > diff --git a/fs/dax.c b/fs/dax.c
> > index 95981591977a..f3afa1d6156c 100644
> > --- a/fs/dax.c
> > +++ b/fs/dax.c
> > @@ -1096,7 +1096,7 @@ static bool dax_fault_is_synchronous(unsigned long flags,
> > }
> >
> > static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > - const struct iomap_ops *ops)
> > + int *iomap_errp, const struct iomap_ops *ops)
> > {
> > struct vm_area_struct *vma = vmf->vma;
> > struct address_space *mapping = vma->vm_file->f_mapping;
> > @@ -1149,6 +1149,8 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > * that we never have to deal with more than a single extent here.
> > */
> > error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap);
> > + if (iomap_errp)
> > + *iomap_errp = error;
>
> Since we already have 'struct iomap' tracking the state of the iomap
> should we track the error status there as well? I.e. move the on
> stack allocation of struct iomap to the per-fs dax fault handlers.
I don't think that's really adequate. Firstly because at least part of
struct iomap needs to be initialized inside dax_iomap_fault() and secondly
because by the time dax_iomap_fault() returns, mapping information in struct
iomap may be invalid (as we unlocked radix tree entry). So I think it is
really better to pass back only the error code and keep struct iomap
internal to dax_iomap_fault().
Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR
On Tue, Jan 2, 2018 at 10:54 AM, Jan Kara <[email protected]> wrote:
> On Thu 21-12-17 09:12:52, Dan Williams wrote:
>> On Thu, Dec 21, 2017 at 8:30 AM, Jan Kara <[email protected]> wrote:
>> > Ext4 needs to pass through error from its iomap handler to the page
>> > fault handler so that it can properly detect ENOSPC and force
>> > transaction commit and retry the fault (and block allocation). Add
>> > argument to dax_iomap_fault() for passing such error.
>> >
>> > Signed-off-by: Jan Kara <[email protected]>
>> > ---
>> > fs/dax.c | 9 ++++++---
>> > fs/ext2/file.c | 2 +-
>> > fs/ext4/file.c | 2 +-
>> > fs/xfs/xfs_file.c | 2 +-
>> > include/linux/dax.h | 2 +-
>> > 5 files changed, 10 insertions(+), 7 deletions(-)
>> >
>> > diff --git a/fs/dax.c b/fs/dax.c
>> > index 95981591977a..f3afa1d6156c 100644
>> > --- a/fs/dax.c
>> > +++ b/fs/dax.c
>> > @@ -1096,7 +1096,7 @@ static bool dax_fault_is_synchronous(unsigned long flags,
>> > }
>> >
>> > static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
>> > - const struct iomap_ops *ops)
>> > + int *iomap_errp, const struct iomap_ops *ops)
>> > {
>> > struct vm_area_struct *vma = vmf->vma;
>> > struct address_space *mapping = vma->vm_file->f_mapping;
>> > @@ -1149,6 +1149,8 @@ static int dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
>> > * that we never have to deal with more than a single extent here.
>> > */
>> > error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap);
>> > + if (iomap_errp)
>> > + *iomap_errp = error;
>>
>> Since we already have 'struct iomap' tracking the state of the iomap
>> should we track the error status there as well? I.e. move the on
>> stack allocation of struct iomap to the per-fs dax fault handlers.
>
> I don't think that's really adequate. Firstly because at least part of
> struct iomap needs to be initialized inside dax_iomap_fault() and secondly
> because by the time dax_iomap_fault() returns, mapping information in struct
> iomap may be invalid (as we unlocked radix tree entry). So I think it is
> really better to pass back only the error code and keep struct iomap
> internal to dax_iomap_fault().
Ok, that makes sense to me.