I've started looking into resurrecting splice F_MOVE support, and it
seems ->steal for anon pages is completly bogus at the moment:
- the page count check is incorrect
- it doesn't isolate the mapping from the lru
- it sets the PIPE_BUF_FLAG_LRU flag, which doesn't get the file
added to the file lru
Currently on fuse calls ->steal, but I'm not sure it could work on
a vmspliced buffered at all.
Below is a patch that attempts to fix / paper over ->steal, and the
second is the unfinished F_MOVE resurrection patch which shows
what additional workarouns we need for ->steal from anon pages.
>From 94958673ed6d0add7f5d95cc17fb0c9fa8f58c03 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@xxxxxx>
Date: Wed, 1 Apr 2015 14:28:50 +0200
Subject: fix ->steal for anon pages
---
fs/splice.c | 20 ++++++++++++++++++--
include/linux/swap.h | 1 +
mm/internal.h | 1 -
3 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/fs/splice.c b/fs/splice.c
index 41cbb16..36aa4a9 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -145,11 +145,27 @@ const struct pipe_buf_operations page_cache_pipe_buf_ops = {
static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
struct pipe_buffer *buf)
{
+ struct page *page = buf->page;
+
if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
return 1;
- buf->flags |= PIPE_BUF_FLAG_LRU;
- return generic_pipe_buf_steal(pipe, buf);
+ /*
+ * We should have three references to the page: gup, lru, and
+ * one for being mapped into page tables.
+ */
+ if (page_count(page) != 3)
+ return 1;
+
+ lock_page(page);
+
+ if (!isolate_lru_page(page)) {
+ ClearPageActive(page);
+ ClearPageUnevictable(page);
+ page_cache_release(page);
+ }
+
+ return 0;
}
static const struct pipe_buf_operations user_page_pipe_buf_ops = {
diff --git a/include/linux/swap.h b/include/linux/swap.h
index 7067eca..a3742f3 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -318,6 +318,7 @@ extern void lru_cache_add_active_or_unevictable(struct page *page,
/* linux/mm/vmscan.c */
extern unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
gfp_t gfp_mask, nodemask_t *mask);
+extern int isolate_lru_page(struct page *page);
extern int __isolate_lru_page(struct page *page, isolate_mode_t mode);
extern unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
unsigned long nr_pages,
diff --git a/mm/internal.h b/mm/internal.h
index a96da5b..48c5731 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -96,7 +96,6 @@ extern unsigned long highest_memmap_pfn;
/*
* in mm/vmscan.c:
*/
-extern int isolate_lru_page(struct page *page);
extern void putback_lru_page(struct page *page);
extern bool zone_reclaimable(struct zone *zone);
--
1.9.1
>From b2b9ed7a41d6d629abefedbfafcdbbd1b1094491 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@xxxxxx>
Date: Wed, 1 Apr 2015 14:23:09 +0200
Subject: splice F_MOVE WIP
---
fs/splice.c | 38 ++++++++++++++++++++++++++++++++++++++
1 file changed, 38 insertions(+)
diff --git a/fs/splice.c b/fs/splice.c
index 36aa4a9..33f611e 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -939,6 +939,36 @@ ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
return ret;
}
+static void steal_page(struct pipe_inode_info *pipe,
+ struct splice_desc *sd, struct pipe_buffer *buf)
+{
+ struct page *page = buf->page;
+ bool was_anon = PageAnon(page);
+
+ if (buf->ops->steal(pipe, buf)) {
+ printk("failed to steal page..\n");
+ return;
+ }
+
+ if (add_to_page_cache_lru(page, sd->u.file->f_mapping,
+ sd->pos >> PAGE_CACHE_SHIFT, GFP_KERNEL)) {
+ /*
+ * This is harmless as we fall back to simply
+ * copying the page. Of course that will most
+ * likely fail to add to the page cache as well,
+ * but at that point it's not our problem..
+ */
+ goto out_unlock;
+ }
+
+ if (was_anon) {
+ inc_mm_counter(current->mm, MM_FILEPAGES);
+ dec_mm_counter(current->mm, MM_ANONPAGES);
+ }
+out_unlock:
+ unlock_page(page);
+}
+
/**
* iter_file_splice_write - splice data from a pipe to a file
* @pipe: pipe info
@@ -1013,6 +1043,14 @@ iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
goto done;
}
+ /*
+ * XXX: hch: we should allow page steals
+ * around EOF a well.
+ */
+ if ((sd.flags & SPLICE_F_MOVE) &&
+ this_len == PAGE_CACHE_SIZE)
+ steal_page(pipe, &sd, buf);
+
array[n].bv_page = buf->page;
array[n].bv_len = this_len;
array[n].bv_offset = buf->offset;
--
1.9.1
