[PATCH] gfs2: Prevent recursive memory reclaim

[PATCH] gfs2: Prevent recursive memory reclaim

[Andreas Gruenbacher]

Function new_inode() returns a new inode with inode->i_mapping->gfp_mask
set to GFP_HIGHUSER_MOVABLE.  This value includes the __GFP_FS flag, so
allocations in that address space can recurse into filesystem memory
reclaim.  We don't want that to happen because it can consume a
significant amount of stack memory.

Worse than that is that it can also deadlock: for example, in several
places, gfs2_unstuff_dinode() is called inside filesystem transactions.
This calls calls filemap_grab_folio(), which can allocate a new folio,
which can trigger memory reclaim.  If memory reclaim recurses into the
filesystem and starts another transaction, a deadlock will ensue.

To fix these kinds of problems, prevent memory reclaim from recursing
into filesystem code by making sure that the gfp_mask of inode address
spaces doesn't include __GFP_FS.

The "meta" and resource group address spaces were already using GFP_NOFS
as their gfp_mask (which doesn't include __GFP_FS).  Inodes allocated by
new_inode() will have their inode->i_mapping->gfp_mask set to
GFP_HIGHUSER_MOVABLE, which is less restrictive than GFP_NOFS, though.
To avoid being overly limiting, use the default value and only knock off
the __GFP_FS flag.  I'm not sure if this will actually make a
difference, but it also shouldn't hurt.

This patch is loosely based on commit ad22c7a043c2 ("xfs: prevent stack
overflows from page cache allocation").

Fixes xfstest generic/273.

Reviewed-by: Andrew Price <anprice@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/glock.c      |  5 ++++-
 fs/gfs2/inode.c      | 15 +++++++++++++++
 fs/gfs2/inode.h      |  1 +
 fs/gfs2/ops_fstype.c |  2 +-
 4 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index b677c0e6b9ab..9f2eb7e38569 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1211,10 +1211,13 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 
 	mapping = gfs2_glock2aspace(gl);
 	if (mapping) {
+		gfp_t gfp_mask;
+
                 mapping->a_ops = &gfs2_meta_aops;
 		mapping->host = sdp->sd_inode;
 		mapping->flags = 0;
-		mapping_set_gfp_mask(mapping, GFP_NOFS);
+		gfp_mask = mapping_gfp_mask(sdp->sd_inode->i_mapping);
+		mapping_set_gfp_mask(mapping, gfp_mask);
 		mapping->i_private_data = NULL;
 		mapping->writeback_index = 0;
 	}
diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c
index 8a7ed80d9f2d..d7e35a05c161 100644
--- a/fs/gfs2/inode.c
+++ b/fs/gfs2/inode.c
@@ -89,6 +89,19 @@ static int iget_set(struct inode *inode, void *opaque)
 	return 0;
 }
 
+void gfs2_setup_inode(struct inode *inode)
+{
+	gfp_t gfp_mask;
+
+	/*
+	 * Ensure all page cache allocations are done from GFP_NOFS context to
+	 * prevent direct reclaim recursion back into the filesystem and blowing
+	 * stacks or deadlocking.
+	 */
+	gfp_mask = mapping_gfp_mask(inode->i_mapping);
+	mapping_set_gfp_mask(inode->i_mapping, gfp_mask & ~__GFP_FS);
+}
+
 /**
  * gfs2_inode_lookup - Lookup an inode
  * @sb: The super block
@@ -132,6 +145,7 @@ struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
 		struct gfs2_glock *io_gl;
 		int extra_flags = 0;
 
+		gfs2_setup_inode(inode);
 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE,
 				       &ip->i_gl);
 		if (unlikely(error))
@@ -752,6 +766,7 @@ static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
 	error = -ENOMEM;
 	if (!inode)
 		goto fail_gunlock;
+	gfs2_setup_inode(inode);
 	ip = GFS2_I(inode);
 
 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
diff --git a/fs/gfs2/inode.h b/fs/gfs2/inode.h
index e43f08eb26e7..2fcd96dd1361 100644
--- a/fs/gfs2/inode.h
+++ b/fs/gfs2/inode.h
@@ -86,6 +86,7 @@ static inline int gfs2_check_internal_file_size(struct inode *inode,
 	return -EIO;
 }
 
+void gfs2_setup_inode(struct inode *inode);
 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned type,
 			        u64 no_addr, u64 no_formal_ino,
 			        unsigned int blktype);
diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index aa15183f9a16..1a2db8053da0 100644
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -1183,7 +1183,7 @@ static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
 
 	mapping = gfs2_aspace(sdp);
 	mapping->a_ops = &gfs2_rgrp_aops;
-	mapping_set_gfp_mask(mapping, GFP_NOFS);
+	gfs2_setup_inode(sdp->sd_inode);
 
 	error = init_names(sdp, silent);
 	if (error)
-- 
2.51.0