kmap() is being deprecated in favor of kmap_local_page().
There are two main problems with kmap(): (1) It comes with an overhead as
mapping space is restricted and protected by a global lock for
synchronization and (2) it also requires global TLB invalidation when the
kmap’s pool wraps and it might block when the mapping space is fully
utilized until a slot becomes available.
With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
the tasks can be preempted and, when they are scheduled to run again, the
kernel virtual addresses are restored and still valid.
Since its use in fs/hfsplus is safe everywhere, it should be preferred.
Therefore, replace kmap() with kmap_local_page() in fs/hfsplus. Where
possible, use the suited standard helpers (memzero_page(), memcpy_page())
instead of open coding kmap_local_page() plus memset() or memcpy().
Fix a bug due to a page being not unmapped if the code jumps to the
"fail_page" label (1/4).
Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.
Fabio M. De Francesco (4):
hfsplus: Unmap the page in the "fail_page" label
hfsplus: Convert kmap() to kmap_local_page() in bnode.c
hfsplus: Convert kmap() to kmap_local_page() in bitmap.c
hfsplus: Convert kmap() to kmap_local_page() in btree.c
fs/hfsplus/bitmap.c | 20 ++++-----
fs/hfsplus/bnode.c | 105 ++++++++++++++++++++------------------------
fs/hfsplus/btree.c | 27 ++++++------
3 files changed, 72 insertions(+), 80 deletions(-)
--
2.37.1
Several paths within hfs_btree_open() jump to the "fail_page" label
where put_page() is called while the page is still mapped.
Call kunmap() to unmap the page soon before put_page().
Cc: Viacheslav Dubeyko <[email protected]>
Reviewed-by: Ira Weiny <[email protected]>
Signed-off-by: Fabio M. De Francesco <[email protected]>
---
fs/hfsplus/btree.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/fs/hfsplus/btree.c b/fs/hfsplus/btree.c
index 66774f4cb4fd..3a917a9a4edd 100644
--- a/fs/hfsplus/btree.c
+++ b/fs/hfsplus/btree.c
@@ -245,6 +245,7 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
return tree;
fail_page:
+ kunmap(page);
put_page(page);
free_inode:
tree->inode->i_mapping->a_ops = &hfsplus_aops;
--
2.37.1
kmap() is being deprecated in favor of kmap_local_page().
There are two main problems with kmap(): (1) It comes with an overhead as
mapping space is restricted and protected by a global lock for
synchronization and (2) it also requires global TLB invalidation when the
kmap’s pool wraps and it might block when the mapping space is fully
utilized until a slot becomes available.
With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
the tasks can be preempted and, when they are scheduled to run again, the
kernel virtual addresses are restored and are still valid.
Since its use in btree.c is safe everywhere, it should be preferred.
Therefore, replace kmap() with kmap_local_page() in btree.c.
Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.
Cc: Viacheslav Dubeyko <[email protected]>
Suggested-by: Ira Weiny <[email protected]>
Reviewed-by: Ira Weiny <[email protected]>
Signed-off-by: Fabio M. De Francesco <[email protected]>
---
fs/hfsplus/btree.c | 28 ++++++++++++++--------------
1 file changed, 14 insertions(+), 14 deletions(-)
diff --git a/fs/hfsplus/btree.c b/fs/hfsplus/btree.c
index 3a917a9a4edd..9e1732a2b92a 100644
--- a/fs/hfsplus/btree.c
+++ b/fs/hfsplus/btree.c
@@ -163,7 +163,7 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
goto free_inode;
/* Load the header */
- head = (struct hfs_btree_header_rec *)(kmap(page) +
+ head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
sizeof(struct hfs_bnode_desc));
tree->root = be32_to_cpu(head->root);
tree->leaf_count = be32_to_cpu(head->leaf_count);
@@ -240,12 +240,12 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
(tree->node_size + PAGE_SIZE - 1) >>
PAGE_SHIFT;
- kunmap(page);
+ kunmap_local(head);
put_page(page);
return tree;
fail_page:
- kunmap(page);
+ kunmap_local(head);
put_page(page);
free_inode:
tree->inode->i_mapping->a_ops = &hfsplus_aops;
@@ -292,7 +292,7 @@ int hfs_btree_write(struct hfs_btree *tree)
return -EIO;
/* Load the header */
page = node->page[0];
- head = (struct hfs_btree_header_rec *)(kmap(page) +
+ head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
sizeof(struct hfs_bnode_desc));
head->root = cpu_to_be32(tree->root);
@@ -304,7 +304,7 @@ int hfs_btree_write(struct hfs_btree *tree)
head->attributes = cpu_to_be32(tree->attributes);
head->depth = cpu_to_be16(tree->depth);
- kunmap(page);
+ kunmap_local(head);
set_page_dirty(page);
hfs_bnode_put(node);
return 0;
@@ -395,7 +395,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
off += node->page_offset;
pagep = node->page + (off >> PAGE_SHIFT);
- data = kmap(*pagep);
+ data = kmap_local_page(*pagep);
off &= ~PAGE_MASK;
idx = 0;
@@ -408,7 +408,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
idx += i;
data[off] |= m;
set_page_dirty(*pagep);
- kunmap(*pagep);
+ kunmap_local(data);
tree->free_nodes--;
mark_inode_dirty(tree->inode);
hfs_bnode_put(node);
@@ -418,14 +418,14 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
}
}
if (++off >= PAGE_SIZE) {
- kunmap(*pagep);
- data = kmap(*++pagep);
+ kunmap_local(data);
+ data = kmap_local_page(*++pagep);
off = 0;
}
idx += 8;
len--;
}
- kunmap(*pagep);
+ kunmap_local(data);
nidx = node->next;
if (!nidx) {
hfs_dbg(BNODE_MOD, "create new bmap node\n");
@@ -441,7 +441,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
off = off16;
off += node->page_offset;
pagep = node->page + (off >> PAGE_SHIFT);
- data = kmap(*pagep);
+ data = kmap_local_page(*pagep);
off &= ~PAGE_MASK;
}
}
@@ -491,7 +491,7 @@ void hfs_bmap_free(struct hfs_bnode *node)
}
off += node->page_offset + nidx / 8;
page = node->page[off >> PAGE_SHIFT];
- data = kmap(page);
+ data = kmap_local_page(page);
off &= ~PAGE_MASK;
m = 1 << (~nidx & 7);
byte = data[off];
@@ -499,13 +499,13 @@ void hfs_bmap_free(struct hfs_bnode *node)
pr_crit("trying to free free bnode "
"%u(%d)\n",
node->this, node->type);
- kunmap(page);
+ kunmap_local(data);
hfs_bnode_put(node);
return;
}
data[off] = byte & ~m;
set_page_dirty(page);
- kunmap(page);
+ kunmap_local(data);
hfs_bnode_put(node);
tree->free_nodes++;
mark_inode_dirty(tree->inode);
--
2.37.1
> On Aug 9, 2022, at 1:31 PM, Fabio M. De Francesco <[email protected]> wrote:
>
> kmap() is being deprecated in favor of kmap_local_page().
>
> There are two main problems with kmap(): (1) It comes with an overhead as
> mapping space is restricted and protected by a global lock for
> synchronization and (2) it also requires global TLB invalidation when the
> kmap’s pool wraps and it might block when the mapping space is fully
> utilized until a slot becomes available.
>
> With kmap_local_page() the mappings are per thread, CPU local, can take
> page faults, and can be called from any context (including interrupts).
> It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
> the tasks can be preempted and, when they are scheduled to run again, the
> kernel virtual addresses are restored and are still valid.
>
> Since its use in btree.c is safe everywhere, it should be preferred.
>
> Therefore, replace kmap() with kmap_local_page() in btree.c.
>
> Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
> HIGHMEM64GB enabled.
>
> Cc: Viacheslav Dubeyko <[email protected]>
> Suggested-by: Ira Weiny <[email protected]>
> Reviewed-by: Ira Weiny <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>
> ---
Looks good.
Reviewed-by: Viacheslav Dubeyko <[email protected]>
Thanks,
Slava.
> fs/hfsplus/btree.c | 28 ++++++++++++++--------------
> 1 file changed, 14 insertions(+), 14 deletions(-)
>
> diff --git a/fs/hfsplus/btree.c b/fs/hfsplus/btree.c
> index 3a917a9a4edd..9e1732a2b92a 100644
> --- a/fs/hfsplus/btree.c
> +++ b/fs/hfsplus/btree.c
> @@ -163,7 +163,7 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
> goto free_inode;
>
> /* Load the header */
> - head = (struct hfs_btree_header_rec *)(kmap(page) +
> + head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
> sizeof(struct hfs_bnode_desc));
> tree->root = be32_to_cpu(head->root);
> tree->leaf_count = be32_to_cpu(head->leaf_count);
> @@ -240,12 +240,12 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
> (tree->node_size + PAGE_SIZE - 1) >>
> PAGE_SHIFT;
>
> - kunmap(page);
> + kunmap_local(head);
> put_page(page);
> return tree;
>
> fail_page:
> - kunmap(page);
> + kunmap_local(head);
> put_page(page);
> free_inode:
> tree->inode->i_mapping->a_ops = &hfsplus_aops;
> @@ -292,7 +292,7 @@ int hfs_btree_write(struct hfs_btree *tree)
> return -EIO;
> /* Load the header */
> page = node->page[0];
> - head = (struct hfs_btree_header_rec *)(kmap(page) +
> + head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
> sizeof(struct hfs_bnode_desc));
>
> head->root = cpu_to_be32(tree->root);
> @@ -304,7 +304,7 @@ int hfs_btree_write(struct hfs_btree *tree)
> head->attributes = cpu_to_be32(tree->attributes);
> head->depth = cpu_to_be16(tree->depth);
>
> - kunmap(page);
> + kunmap_local(head);
> set_page_dirty(page);
> hfs_bnode_put(node);
> return 0;
> @@ -395,7 +395,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
>
> off += node->page_offset;
> pagep = node->page + (off >> PAGE_SHIFT);
> - data = kmap(*pagep);
> + data = kmap_local_page(*pagep);
> off &= ~PAGE_MASK;
> idx = 0;
>
> @@ -408,7 +408,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
> idx += i;
> data[off] |= m;
> set_page_dirty(*pagep);
> - kunmap(*pagep);
> + kunmap_local(data);
> tree->free_nodes--;
> mark_inode_dirty(tree->inode);
> hfs_bnode_put(node);
> @@ -418,14 +418,14 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
> }
> }
> if (++off >= PAGE_SIZE) {
> - kunmap(*pagep);
> - data = kmap(*++pagep);
> + kunmap_local(data);
> + data = kmap_local_page(*++pagep);
> off = 0;
> }
> idx += 8;
> len--;
> }
> - kunmap(*pagep);
> + kunmap_local(data);
> nidx = node->next;
> if (!nidx) {
> hfs_dbg(BNODE_MOD, "create new bmap node\n");
> @@ -441,7 +441,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
> off = off16;
> off += node->page_offset;
> pagep = node->page + (off >> PAGE_SHIFT);
> - data = kmap(*pagep);
> + data = kmap_local_page(*pagep);
> off &= ~PAGE_MASK;
> }
> }
> @@ -491,7 +491,7 @@ void hfs_bmap_free(struct hfs_bnode *node)
> }
> off += node->page_offset + nidx / 8;
> page = node->page[off >> PAGE_SHIFT];
> - data = kmap(page);
> + data = kmap_local_page(page);
> off &= ~PAGE_MASK;
> m = 1 << (~nidx & 7);
> byte = data[off];
> @@ -499,13 +499,13 @@ void hfs_bmap_free(struct hfs_bnode *node)
> pr_crit("trying to free free bnode "
> "%u(%d)\n",
> node->this, node->type);
> - kunmap(page);
> + kunmap_local(data);
> hfs_bnode_put(node);
> return;
> }
> data[off] = byte & ~m;
> set_page_dirty(page);
> - kunmap(page);
> + kunmap_local(data);
> hfs_bnode_put(node);
> tree->free_nodes++;
> mark_inode_dirty(tree->inode);
> --
> 2.37.1
>
> On Aug 9, 2022, at 1:31 PM, Fabio M. De Francesco <[email protected]> wrote:
>
> Several paths within hfs_btree_open() jump to the "fail_page" label
> where put_page() is called while the page is still mapped.
>
> Call kunmap() to unmap the page soon before put_page().
>
> Cc: Viacheslav Dubeyko <[email protected]>
> Reviewed-by: Ira Weiny <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>
> ---
> fs/hfsplus/btree.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/hfsplus/btree.c b/fs/hfsplus/btree.c
> index 66774f4cb4fd..3a917a9a4edd 100644
> --- a/fs/hfsplus/btree.c
> +++ b/fs/hfsplus/btree.c
> @@ -245,6 +245,7 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
> return tree;
>
> fail_page:
> + kunmap(page);
> put_page(page);
> free_inode:
> tree->inode->i_mapping->a_ops = &hfsplus_aops;
> --
> 2.37.1
>
Looks good.
Reviewed-by: Viacheslav Dubeyko <[email protected]>
Thanks,
Slava.