summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJens Axboe <axboe@suse.de>2006-03-30 15:16:46 +0200
committerLinus Torvalds <torvalds@g5.osdl.org>2006-03-30 12:28:18 -0800
commit5abc97aa25b2c41413b3a520faee83f2282d9f18 (patch)
tree4ba13ae0e91f15d02986df7cdca5e9455212d7d4
parent5274f052e7b3dbd81935772eb551dfd0325dfa9d (diff)
downloadlinux-5abc97aa25b2c41413b3a520faee83f2282d9f18.tar.gz
linux-5abc97aa25b2c41413b3a520faee83f2282d9f18.tar.bz2
linux-5abc97aa25b2c41413b3a520faee83f2282d9f18.zip
[PATCH] splice: add support for SPLICE_F_MOVE flag
This enables the caller to migrate pages from one address space page cache to another. In buzz word marketing, you can do zero-copy file copies! Signed-off-by: Jens Axboe <axboe@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
-rw-r--r--fs/pipe.c8
-rw-r--r--fs/splice.c121
-rw-r--r--include/linux/pipe_fs_i.h8
3 files changed, 100 insertions, 37 deletions
diff --git a/fs/pipe.c b/fs/pipe.c
index 2414bf270db6..109a102c150d 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -121,11 +121,19 @@ static void anon_pipe_buf_unmap(struct pipe_inode_info *info, struct pipe_buffer
kunmap(buf->page);
}
+static int anon_pipe_buf_steal(struct pipe_inode_info *info,
+ struct pipe_buffer *buf)
+{
+ buf->stolen = 1;
+ return 0;
+}
+
static struct pipe_buf_operations anon_pipe_buf_ops = {
.can_merge = 1,
.map = anon_pipe_buf_map,
.unmap = anon_pipe_buf_unmap,
.release = anon_pipe_buf_release,
+ .steal = anon_pipe_buf_steal,
};
static ssize_t
diff --git a/fs/splice.c b/fs/splice.c
index efa47c1c4e13..4a026f95884f 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -21,6 +21,7 @@
#include <linux/pagemap.h>
#include <linux/pipe_fs_i.h>
#include <linux/mm_inline.h>
+#include <linux/swap.h>
/*
* Passed to the actors
@@ -32,11 +33,37 @@ struct splice_desc {
loff_t pos; /* file position */
};
+static int page_cache_pipe_buf_steal(struct pipe_inode_info *info,
+ struct pipe_buffer *buf)
+{
+ struct page *page = buf->page;
+
+ WARN_ON(!PageLocked(page));
+ WARN_ON(!PageUptodate(page));
+
+ if (!remove_mapping(page_mapping(page), page))
+ return 1;
+
+ if (PageLRU(page)) {
+ struct zone *zone = page_zone(page);
+
+ spin_lock_irq(&zone->lru_lock);
+ BUG_ON(!PageLRU(page));
+ __ClearPageLRU(page);
+ del_page_from_lru(zone, page);
+ spin_unlock_irq(&zone->lru_lock);
+ }
+
+ buf->stolen = 1;
+ return 0;
+}
+
static void page_cache_pipe_buf_release(struct pipe_inode_info *info,
struct pipe_buffer *buf)
{
page_cache_release(buf->page);
buf->page = NULL;
+ buf->stolen = 0;
}
static void *page_cache_pipe_buf_map(struct file *file,
@@ -63,7 +90,8 @@ static void *page_cache_pipe_buf_map(struct file *file,
static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info,
struct pipe_buffer *buf)
{
- unlock_page(buf->page);
+ if (!buf->stolen)
+ unlock_page(buf->page);
kunmap(buf->page);
}
@@ -72,6 +100,7 @@ static struct pipe_buf_operations page_cache_pipe_buf_ops = {
.map = page_cache_pipe_buf_map,
.unmap = page_cache_pipe_buf_unmap,
.release = page_cache_pipe_buf_release,
+ .steal = page_cache_pipe_buf_steal,
};
static ssize_t move_to_pipe(struct inode *inode, struct page **pages,
@@ -336,8 +365,8 @@ static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
struct address_space *mapping = file->f_mapping;
unsigned int offset;
struct page *page;
- char *src, *dst;
pgoff_t index;
+ char *src;
int ret;
/*
@@ -350,40 +379,54 @@ static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
index = sd->pos >> PAGE_CACHE_SHIFT;
offset = sd->pos & ~PAGE_CACHE_MASK;
-find_page:
- ret = -ENOMEM;
- page = find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
- if (!page)
- goto out;
-
/*
- * If the page is uptodate, it is also locked. If it isn't
- * uptodate, we can mark it uptodate if we are filling the
- * full page. Otherwise we need to read it in first...
+ * reuse buf page, if SPLICE_F_MOVE is set
*/
- if (!PageUptodate(page)) {
- if (sd->len < PAGE_CACHE_SIZE) {
- ret = mapping->a_ops->readpage(file, page);
- if (unlikely(ret))
- goto out;
-
- lock_page(page);
-
- if (!PageUptodate(page)) {
- /*
- * page got invalidated, repeat
- */
- if (!page->mapping) {
- unlock_page(page);
- page_cache_release(page);
- goto find_page;
+ if (sd->flags & SPLICE_F_MOVE) {
+ if (buf->ops->steal(info, buf))
+ goto find_page;
+
+ page = buf->page;
+ if (add_to_page_cache_lru(page, mapping, index,
+ mapping_gfp_mask(mapping)))
+ goto find_page;
+ } else {
+find_page:
+ ret = -ENOMEM;
+ page = find_or_create_page(mapping, index,
+ mapping_gfp_mask(mapping));
+ if (!page)
+ goto out;
+
+ /*
+ * If the page is uptodate, it is also locked. If it isn't
+ * uptodate, we can mark it uptodate if we are filling the
+ * full page. Otherwise we need to read it in first...
+ */
+ if (!PageUptodate(page)) {
+ if (sd->len < PAGE_CACHE_SIZE) {
+ ret = mapping->a_ops->readpage(file, page);
+ if (unlikely(ret))
+ goto out;
+
+ lock_page(page);
+
+ if (!PageUptodate(page)) {
+ /*
+ * page got invalidated, repeat
+ */
+ if (!page->mapping) {
+ unlock_page(page);
+ page_cache_release(page);
+ goto find_page;
+ }
+ ret = -EIO;
+ goto out;
}
- ret = -EIO;
- goto out;
+ } else {
+ WARN_ON(!PageLocked(page));
+ SetPageUptodate(page);
}
- } else {
- WARN_ON(!PageLocked(page));
- SetPageUptodate(page);
}
}
@@ -391,10 +434,13 @@ find_page:
if (ret)
goto out;
- dst = kmap_atomic(page, KM_USER0);
- memcpy(dst + offset, src + buf->offset, sd->len);
- flush_dcache_page(page);
- kunmap_atomic(dst, KM_USER0);
+ if (!buf->stolen) {
+ char *dst = kmap_atomic(page, KM_USER0);
+
+ memcpy(dst + offset, src + buf->offset, sd->len);
+ flush_dcache_page(page);
+ kunmap_atomic(dst, KM_USER0);
+ }
ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
if (ret < 0)
@@ -405,7 +451,8 @@ find_page:
out:
if (ret < 0)
unlock_page(page);
- page_cache_release(page);
+ if (!buf->stolen)
+ page_cache_release(page);
buf->ops->unmap(info, buf);
return ret;
}
diff --git a/include/linux/pipe_fs_i.h b/include/linux/pipe_fs_i.h
index b12e59c75752..75c7f55023ab 100644
--- a/include/linux/pipe_fs_i.h
+++ b/include/linux/pipe_fs_i.h
@@ -9,6 +9,7 @@ struct pipe_buffer {
struct page *page;
unsigned int offset, len;
struct pipe_buf_operations *ops;
+ unsigned int stolen;
};
struct pipe_buf_operations {
@@ -16,6 +17,7 @@ struct pipe_buf_operations {
void * (*map)(struct file *, struct pipe_inode_info *, struct pipe_buffer *);
void (*unmap)(struct pipe_inode_info *, struct pipe_buffer *);
void (*release)(struct pipe_inode_info *, struct pipe_buffer *);
+ int (*steal)(struct pipe_inode_info *, struct pipe_buffer *);
};
struct pipe_inode_info {
@@ -53,4 +55,10 @@ void pipe_wait(struct inode * inode);
struct inode* pipe_new(struct inode* inode);
void free_pipe_info(struct inode* inode);
+/*
+ * splice is tied to pipes as a transport (at least for now), so we'll just
+ * add the splice flags here.
+ */
+#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
+
#endif