jim owens
2010-Jan-08 17:10 UTC
[RFC PATCH 01/01] Btrfs: add decompression code for direct I/O.
Direct I/O needs to inflate compressed temp pages with the
uncompressed result stored in not-necessarily-aligned user
memory. Add btrfs_zlib_inflate() to do this and expose
struct workspace, find_zlib_workspace(), and free_workspace()
so that multiple extents can be processed using one workspace.
Signed-off-by: jim owens <jowens@hp.com>
---
fs/btrfs/compression.h | 20 +++++++
fs/btrfs/zlib.c | 143 +++++++++++++++++++++++++++++++++++++++++++++---
2 files changed, 155 insertions(+), 8 deletions(-)
diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
index 421f5b4..afd262d 100644
--- a/fs/btrfs/compression.h
+++ b/fs/btrfs/compression.h
@@ -19,6 +19,26 @@
#ifndef __BTRFS_COMPRESSION_
#define __BTRFS_COMPRESSION_
+#include <linux/zlib.h>
+struct workspace {
+ z_stream z_strm;
+ char *buf;
+ struct list_head list;
+};
+
+struct btrfs_inflate {
+ struct workspace *workspace;
+ int (*get_next_in)(struct bio_vec *vec, struct btrfs_inflate *icb);
+ int (*get_next_out)(struct bio_vec *vec, struct btrfs_inflate *icb);
+ void (*done_with_out)(struct bio_vec *vec, struct btrfs_inflate *icb);
+ u32 out_start;
+ u32 out_len;
+};
+
+struct workspace *find_zlib_workspace(void);
+int free_workspace(struct workspace *workspace);
+int btrfs_zlib_inflate(struct btrfs_inflate *icb);
+
int btrfs_zlib_decompress(unsigned char *data_in,
struct page *dest_page,
unsigned long start_byte,
diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c
index d7bdce5..779a259 100644
--- a/fs/btrfs/zlib.c
+++ b/fs/btrfs/zlib.c
@@ -41,12 +41,6 @@
*/
#define STREAM_END_SPACE 12
-struct workspace {
- z_stream z_strm;
- char *buf;
- struct list_head list;
-};
-
static LIST_HEAD(idle_workspace);
static DEFINE_SPINLOCK(workspace_lock);
static unsigned long num_workspace;
@@ -57,7 +51,7 @@ static DECLARE_WAIT_QUEUE_HEAD(workspace_wait);
* this finds an available zlib workspace or allocates a new one
* NULL or an ERR_PTR is returned if things go bad.
*/
-static struct workspace *find_zlib_workspace(void)
+struct workspace *find_zlib_workspace(void)
{
struct workspace *workspace;
int ret;
@@ -90,6 +84,7 @@ again:
goto fail;
}
+printk("JIM workspace allocated %d deflate %d inflate
%d\n",max(zlib_deflate_workspacesize(),
zlib_inflate_workspacesize()),zlib_deflate_workspacesize(),zlib_inflate_workspacesize());
workspace->z_strm.workspace = vmalloc(max(zlib_deflate_workspacesize(),
zlib_inflate_workspacesize()));
if (!workspace->z_strm.workspace) {
@@ -117,7 +112,7 @@ fail:
* put a workspace struct back on the list or free it if we have enough
* idle ones sitting around
*/
-static int free_workspace(struct workspace *workspace)
+int free_workspace(struct workspace *workspace)
{
spin_lock(&workspace_lock);
if (num_workspace < num_online_cpus()) {
@@ -622,3 +617,135 @@ void btrfs_zlib_exit(void)
{
free_workspaces();
}
+
+/* inflate compressed data for one contiguous file range from directio */
+int btrfs_zlib_inflate(struct btrfs_inflate *icb)
+{
+ struct workspace *workspace = icb->workspace;
+ unsigned long out_start = icb->out_start;
+ unsigned long total_len = icb->out_len + out_start;
+ struct bio_vec ivec;
+ struct bio_vec ovec;
+ char *in;
+ char *out;
+ int err;
+ int wbits;
+
+ icb->out_len = 0;
+ ivec.bv_len = 0;
+ ovec.bv_len = 0;
+ if (!workspace) {
+ workspace = find_zlib_workspace();
+ if (IS_ERR(workspace))
+ return -ENOMEM;
+ }
+
+ err = icb->get_next_in(&ivec, icb);
+ if (err)
+ goto fail;
+ in = kmap_atomic(ivec.bv_page, KM_USER0);
+ workspace->z_strm.next_in = in + ivec.bv_offset;
+ workspace->z_strm.avail_in = ivec.bv_len;
+ workspace->z_strm.total_in = 0;
+ workspace->z_strm.total_out = 0;
+ workspace->z_strm.next_out = workspace->buf;
+ workspace->z_strm.avail_out = PAGE_CACHE_SIZE;
+
+ /* with no preset dictionary, tell zlib to skip the adler32 check */
+ if (!(in[ivec.bv_offset+1] & PRESET_DICT) &&
+ ((in[ivec.bv_offset] & 0x0f) == Z_DEFLATED) &&
+ !(((in[ivec.bv_offset]<<8) + in[ivec.bv_offset+1]) % 31)) {
+
+ wbits = -((in[ivec.bv_offset] >> 4) + 8);
+ workspace->z_strm.next_in += 2;
+ workspace->z_strm.avail_in -= 2;
+ } else {
+ wbits = MAX_WBITS;
+ }
+
+ err = zlib_inflateInit2(&workspace->z_strm, wbits);
+ if (err) {
+ kunmap_atomic(in, KM_USER0);
+ goto fail;
+ }
+
+ ivec.bv_len = workspace->z_strm.avail_in;
+ ivec.bv_offset = (char *)workspace->z_strm.next_in - in;
+ kunmap_atomic(in, KM_USER0);
+
+ /* use temp buf to toss everything before the real data we want */
+ while (workspace->z_strm.total_out < out_start) {
+ workspace->z_strm.next_out = workspace->buf;
+ workspace->z_strm.avail_out = min(PAGE_CACHE_SIZE,
+ out_start - workspace->z_strm.total_out);
+
+ if (!ivec.bv_len) {
+ err = icb->get_next_in(&ivec, icb);
+ if (err)
+ goto fail;
+ }
+ in = kmap_atomic(ivec.bv_page, KM_USER0);
+ workspace->z_strm.next_in = in + ivec.bv_offset;
+ workspace->z_strm.avail_in = ivec.bv_len;
+
+ err = zlib_inflate(&workspace->z_strm, Z_NO_FLUSH);
+
+ ivec.bv_len = workspace->z_strm.avail_in;
+ ivec.bv_offset = (char *)workspace->z_strm.next_in - in;
+ kunmap_atomic(in, KM_USER0);
+
+ if (err != Z_OK) /* Z_STREAM_END is no-user-data failure here */
+ goto fail;
+ cond_resched();
+ }
+
+ while (workspace->z_strm.total_out < total_len) {
+ if (!ivec.bv_len) {
+ err = icb->get_next_in(&ivec, icb);
+ if (err)
+ goto fail;
+ }
+ if (!ovec.bv_len) {
+ err = icb->get_next_out(&ovec, icb);
+ if (err)
+ goto fail;
+ }
+
+ in = kmap_atomic(ivec.bv_page, KM_USER0);
+ workspace->z_strm.next_in = in + ivec.bv_offset;
+ workspace->z_strm.avail_in = ivec.bv_len;
+
+ out = kmap_atomic(ovec.bv_page, KM_USER1);
+ workspace->z_strm.next_out = out + ovec.bv_offset;
+ workspace->z_strm.avail_out = ovec.bv_len;
+
+ err = zlib_inflate(&workspace->z_strm, Z_NO_FLUSH);
+
+ icb->out_len += (ovec.bv_len - workspace->z_strm.avail_out);
+ ovec.bv_len = workspace->z_strm.avail_out;
+ ovec.bv_offset = (char *)workspace->z_strm.next_out - out;
+ kunmap_atomic(out, KM_USER1);
+
+ ivec.bv_len = workspace->z_strm.avail_in;
+ ivec.bv_offset = (char *)workspace->z_strm.next_in - in;
+ kunmap_atomic(in, KM_USER0);
+
+ if (!ovec.bv_len)
+ icb->done_with_out(&ovec, icb);
+ else
+ flush_dcache_page(ovec.bv_page);
+
+ if (err != Z_OK)
+ goto fail;
+ cond_resched();
+ }
+
+fail:
+ if (ovec.bv_len)
+ icb->done_with_out(&ovec, icb);
+ if (!icb->workspace)
+ free_workspace(workspace);
+ if (err == Z_OK || err == Z_STREAM_END)
+ return 0;
+ return err;
+}
--
1.5.6.3
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html