This patch tries to avoid the following deadlock condition of which the reclaim
path can trigger f2fs_balance_fs again.
=================================
[ INFO: inconsistent lock state ]
---------------------------------
inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-W} usage.
kswapd0/41 [HC0[0]:SC0[0]:HE1:SE1] takes:
(&sbi->gc_mutex){+.+.?.}, at: f2fs_balance_fs+0xe6/0x100 [f2fs]
{RECLAIM_FS-ON-W} state was registered at:
[<ffffffff810aa5a9>] mark_held_locks+0xb9/0x140
[<ffffffff810aae85>] lockdep_trace_alloc+0x85/0xf0
[<ffffffff8113ab2c>] __alloc_pages_nodemask+0x7c/0x9b0
[<ffffffff81175aa8>] alloc_pages_current+0xb8/0x180
[<ffffffff811319cf>] __page_cache_alloc+0xaf/0xd0
[<ffffffff8113225c>] find_or_create_page+0x4c/0xb0
[<ffffffffa021359e>] find_data_page+0x14e/0x210 [f2fs]
[<ffffffffa021161b>] f2fs_gc+0x9eb/0xd90 [f2fs]
[<ffffffffa0218fae>] f2fs_balance_fs+0xee/0x100 [f2fs]
[<ffffffffa020848c>] f2fs_setattr+0x6c/0x200 [f2fs]
[<ffffffff811ae51b>] notify_change+0x1db/0x3a0
[<ffffffff8118fbd0>] do_truncate+0x60/0xa0
[<ffffffff8118fd95>] vfs_truncate+0x185/0x1b0
[<ffffffff8118fe1c>] do_sys_truncate+0x5c/0xa0
[<ffffffff8118ffee>] SyS_truncate+0xe/0x10
[<ffffffff816e2b42>] system_call_fastpath+0x16/0x1b
Signed-off-by: Jaegeuk Kim <[email protected]>
---
fs/f2fs/data.c | 4 ++--
fs/f2fs/inode.c | 3 +--
2 files changed, 3 insertions(+), 4 deletions(-)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index c320f7f..1644fff 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -199,7 +199,7 @@ struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
if (dn.data_blkaddr == NEW_ADDR)
return ERR_PTR(-EINVAL);
- page = grab_cache_page(mapping, index);
+ page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
if (!page)
return ERR_PTR(-ENOMEM);
@@ -234,7 +234,7 @@ struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
int err;
repeat:
- page = grab_cache_page(mapping, index);
+ page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
if (!page)
return ERR_PTR(-ENOMEM);
diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index 91ac7f9..a18946e 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -130,8 +130,7 @@ make_now:
inode->i_op = &f2fs_dir_inode_operations;
inode->i_fop = &f2fs_dir_operations;
inode->i_mapping->a_ops = &f2fs_dblock_aops;
- mapping_set_gfp_mask(inode->i_mapping, GFP_HIGHUSER_MOVABLE |
- __GFP_ZERO);
+ mapping_set_gfp_mask(inode->i_mapping, GFP_F2FS_ZERO);
} else if (S_ISLNK(inode->i_mode)) {
inode->i_op = &f2fs_symlink_inode_operations;
inode->i_mapping->a_ops = &f2fs_dblock_aops;
--
1.8.1.3.566.gaa39828
While an orphan inode has zero link_count, f2fs_gc is able to select the inode
for foreground gc.
- f2fs_gc
- do_garbage_collect
- gc_data_segment
: f2fs_iget is failed
: get_valid_blocks() != 0, so that retry
--> here we got the infinite loop.
This patch resolved this issue.
Signed-off-by: Jaegeuk Kim <[email protected]>
---
fs/f2fs/inode.c | 6 ------
1 file changed, 6 deletions(-)
diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index a18946e..b44a4c1 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -109,12 +109,6 @@ struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
ret = do_read_inode(inode);
if (ret)
goto bad_inode;
-
- if (!sbi->por_doing && inode->i_nlink == 0) {
- ret = -ENOENT;
- goto bad_inode;
- }
-
make_now:
if (ino == F2FS_NODE_INO(sbi)) {
inode->i_mapping->a_ops = &f2fs_node_aops;
--
1.8.1.3.566.gaa39828
Introduce a simple macro function for readability.
Signed-off-by: Jaegeuk Kim <[email protected]>
---
fs/f2fs/f2fs.h | 5 +++++
fs/f2fs/file.c | 2 +-
fs/f2fs/super.c | 2 +-
3 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 9182b27..6594ce1 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -887,6 +887,11 @@ static inline int cond_clear_inode_flag(struct f2fs_inode_info *fi, int flag)
return 0;
}
+static inline int f2fs_readonly(struct super_block *sb)
+{
+ return sb->s_flags & MS_RDONLY;
+}
+
/*
* file.c
*/
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 9937ba1..316bcfe 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -114,7 +114,7 @@ int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
.for_reclaim = 0,
};
- if (inode->i_sb->s_flags & MS_RDONLY)
+ if (f2fs_readonly(inode->i_sb))
return 0;
trace_f2fs_sync_file_enter(inode);
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 8555f7d..3ac305d 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -170,7 +170,7 @@ static int f2fs_freeze(struct super_block *sb)
{
int err;
- if (sb->s_flags & MS_RDONLY)
+ if (f2fs_readonly(sb))
return 0;
err = f2fs_sync_fs(sb, 1);
--
1.8.1.3.566.gaa39828