Re: [PATCH] mm: Use GFP_KERNEL allocation for the page cache in page_cache_read

[Michal Hocko <mhocko@suse.cz>]

On Fri 20-03-15 14:48:20, Dave Chinner wrote:
> On Thu, Mar 19, 2015 at 01:44:41PM +0100, Michal Hocko wrote:
> > On Thu 19-03-15 18:14:39, Dave Chinner wrote:
> > > On Wed, Mar 18, 2015 at 03:55:28PM +0100, Michal Hocko wrote:
> > > > On Wed 18-03-15 10:44:11, Rik van Riel wrote:
> > > > > On 03/18/2015 10:09 AM, Michal Hocko wrote:
> > > > > > page_cache_read has been historically using page_cache_alloc_cold to
> > > > > > allocate a new page. This means that mapping_gfp_mask is used as the
> > > > > > base for the gfp_mask. Many filesystems are setting this mask to
> > > > > > GFP_NOFS to prevent from fs recursion issues. page_cache_read is,
> > > > > > however, not called from the fs layer 
> > > > > 
> > > > > Is that true for filesystems that have directories in
> > > > > the page cache?
> > > > 
> > > > I haven't found any explicit callers of filemap_fault except for ocfs2
> > > > and ceph and those seem OK to me. Which filesystems you have in mind?
> > > 
> > > Just about every major filesystem calls filemap_fault through the
> > > .fault callout.
> > 
> > That is right but the callback is called from the VM layer where we
> > obviously do not take any fs locks (we are holding only mmap_sem
> > for reading).
> > Those who call filemap_fault directly (ocfs2 and ceph) and those
> > who call the callback directly: qxl_ttm_fault, radeon_ttm_fault,
> > kernfs_vma_fault, shm_fault seem to be safe from the reclaim recursion
> > POV. radeon_ttm_fault takes a lock for reading but that one doesn't seem
> > to be used from the reclaim context.
> > 
> > Or did I miss your point? Are you concerned about some fs overloading
> > filemap_fault and do some locking before delegating to filemap_fault?
> 
> The latter:
> 
> https://git.kernel.org/cgit/linux/kernel/git/dgc/linux-xfs.git/commit/?h=xfs-mmap-lock&id=de0e8c20ba3a65b0f15040aabbefdc1999876e6b

I will have a look at the code to see what we can do about it.
 
> > > GFP_KERNEL allocation for mappings is simply wrong. All mapping
> > > allocations where the caller cannot pass a gfp_mask need to obey
> > > the mapping_gfp_mask that is set by the mapping owner....
> > 
> > Hmm, I thought this is true only when the function might be called from
> > the fs path.
> 
> How do you know in, say, mpage_readpages, you aren't being called
> from a fs path that holds locks? e.g. we can get there from ext4
> doing readdir, so it is holding an i_mutex lock at that point.
> 
> Many other paths into mpages_readpages don't hold locks, but there
> are some that do, and those that do need functionals like this to
> obey the mapping_gfp_mask because it is set appropriately for the
> allocation context of the inode that owns the mapping....

What about the following?
---
>From 5d905cb291138d61bbab056845d6e53bc4451ec8 Mon Sep 17 00:00:00 2001
From: Michal Hocko <mhocko@suse.cz>
Date: Thu, 19 Mar 2015 14:56:56 +0100
Subject: [PATCH 1/2] mm: do not ignore mapping_gfp_mask in page cache
 allocation paths

page_cache_read, do_generic_file_read, __generic_file_splice_read and
__ntfs_grab_cache_pages currently ignore mapping_gfp_mask when calling
add_to_page_cache_lru which might cause recursion into fs down in the
direct reclaim path if the mapping really relies on GFP_NOFS semantic.

This doesn't seem to be the case now because page_cache_read (page fault
path) doesn't seem to suffer from the reclaim recursion issues and
do_generic_file_read and __generic_file_splice_read also shouldn't be
called under fs locks which would deadlock in the reclaim path. Anyway
it is better to obey mapping gfp mask and prevent from later breakage.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
---
 fs/ntfs/file.c | 2 +-
 fs/splice.c    | 2 +-
 mm/filemap.c   | 6 ++++--
 3 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/fs/ntfs/file.c b/fs/ntfs/file.c
index 1da9b2d184dc..568c9dbc7e61 100644
--- a/fs/ntfs/file.c
+++ b/fs/ntfs/file.c
@@ -422,7 +422,7 @@ static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
 				}
 			}
 			err = add_to_page_cache_lru(*cached_page, mapping, index,
-					GFP_KERNEL);
+					GFP_KERNEL & mapping_gfp_mask(mapping));
 			if (unlikely(err)) {
 				if (err == -EEXIST)
 					continue;
diff --git a/fs/splice.c b/fs/splice.c
index 75c6058eabf2..71f6c51f019a 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -360,7 +360,7 @@ __generic_file_splice_read(struct file *in, loff_t *ppos,
 				break;
 
 			error = add_to_page_cache_lru(page, mapping, index,
-						GFP_KERNEL);
+						GFP_KERNEL & mapping_gfp_mask(mapping));
 			if (unlikely(error)) {
 				page_cache_release(page);
 				if (error == -EEXIST)
diff --git a/mm/filemap.c b/mm/filemap.c
index 968cd8e03d2e..4756cba51655 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -1656,7 +1656,8 @@ no_cached_page:
 			goto out;
 		}
 		error = add_to_page_cache_lru(page, mapping,
-						index, GFP_KERNEL);
+						index,
+						GFP_KERNEL & mapping_gfp_mask(mapping));
 		if (error) {
 			page_cache_release(page);
 			if (error == -EEXIST) {
@@ -1756,7 +1757,8 @@ static int page_cache_read(struct file *file, pgoff_t offset)
 		if (!page)
 			return -ENOMEM;
 
-		ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
+		ret = add_to_page_cache_lru(page, mapping, offset,
+				GFP_KERNEL & mapping_gfp_mask(mapping));
 		if (ret == 0)
 			ret = mapping->a_ops->readpage(file, page);
 		else if (ret == -EEXIST)
-- 
2.1.4


-- 
Michal Hocko
SUSE Labs