[PATCH 0/2 v3] ext4: Fix ENOSPC handling for DAX faults

[PATCH 0/2 v3] ext4: Fix ENOSPC handling for DAX faults

[Jan Kara]

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.

Ted, can you please merge these patches through your tree? Ext4 is the part
that is touched by these patches the most.

Changes since v2:
* Added Reviewed-by tags

Changes since v1:
* Made error passed back from dax_iomap_fault() explicitely be only the error
  from ->iomap_begin()

Comments are welcome.

								Honza

[PATCH 1/2] dax: Pass detailed error code from dax_iomap_fault()

[Jan Kara]

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.

Reviewed-by: Ross Zwisler <ross.zwisler-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
Signed-off-by: Jan Kara <jack-AlSwsSmVLrQ@public.gmane.org>
---
 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

Re: [PATCH 1/2] dax: Pass detailed error code from dax_iomap_fault()

[Theodore Ts'o]

On Wed, Jan 03, 2018 at 11:08:43AM +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.
> 
> Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> Signed-off-by: Jan Kara <jack@suse.cz>

Applied, thanks.

						- Ted

[PATCH 2/2] ext4: Fix ENOSPC handling in DAX page fault handler

[Jan Kara]

When allocation of underlying block for a page fault fails, we fail the
fault with SIGBUS. However we may well hit ENOSPC just due to lots of
free blocks being held by the running / committing transaction. So
propagate the error from ext4_iomap_begin() and implement do standard
allocation retry loop in ext4_dax_huge_fault().

Reviewed-by: Ross Zwisler <ross.zwisler-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
Signed-off-by: Jan Kara <jack-AlSwsSmVLrQ@public.gmane.org>
---
 fs/ext4/file.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 1c7cd882d998..fb6f023622fe 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -280,7 +280,8 @@ ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 static int ext4_dax_huge_fault(struct vm_fault *vmf,
 		enum page_entry_size pe_size)
 {
-	int result;
+	int result, error = 0;
+	int retries = 0;
 	handle_t *handle = NULL;
 	struct inode *inode = file_inode(vmf->vma->vm_file);
 	struct super_block *sb = inode->i_sb;
@@ -304,6 +305,7 @@ static int ext4_dax_huge_fault(struct vm_fault *vmf,
 		sb_start_pagefault(sb);
 		file_update_time(vmf->vma->vm_file);
 		down_read(&EXT4_I(inode)->i_mmap_sem);
+retry:
 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 					       EXT4_DATA_TRANS_BLOCKS(sb));
 		if (IS_ERR(handle)) {
@@ -314,9 +316,13 @@ 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, NULL, &ext4_iomap_ops);
+	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
 	if (write) {
 		ext4_journal_stop(handle);
+
+		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
+		    ext4_should_retry_alloc(sb, &retries))
+			goto retry;
 		/* Handling synchronous page fault? */
 		if (result & VM_FAULT_NEEDDSYNC)
 			result = dax_finish_sync_fault(vmf, pe_size, pfn);
-- 
2.12.3

Re: [PATCH 2/2] ext4: Fix ENOSPC handling in DAX page fault handler

[Theodore Ts'o]

On Wed, Jan 03, 2018 at 11:08:44AM +0100, Jan Kara wrote:
> When allocation of underlying block for a page fault fails, we fail the
> fault with SIGBUS. However we may well hit ENOSPC just due to lots of
> free blocks being held by the running / committing transaction. So
> propagate the error from ext4_iomap_begin() and implement do standard
> allocation retry loop in ext4_dax_huge_fault().
> 
> Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> Signed-off-by: Jan Kara <jack@suse.cz>

Applied, thanks.

					- Ted