[PATCH] f2fs: support atomic operations for database
From: Jaegeuk Kim
Date: Fri Sep 26 2014 - 00:54:45 EST
This patch introduces a very limited functionality for atomic write support.
In order to support atomic write, this patch adds two ioctls:
o F2FS_IOC_DB_OPEN
o F2FS_IOC_COMMIT
The database engine should be aware of the following sequence.
1. open
-> ioctl(F2FS_IOC_DB_OPEN);
2. writes
: all the written data will be treated as atomic pages.
3. commit
-> ioctl(F2FS_IOC_COMMIT);
: this flushes all the data blocks to the disk, which will be shown all or
nothing by f2fs recovery procedure.
The IO pattens should be:
CP | D D D D D D | FSYNC | D D D D | FSYNC ...
While supporting atomic writes for main database file, we can keep its journal
data temporarily in the page cache by the following sequence.
1. open
-> ioctl(F2FS_IOC_JOURNAL_OPEN);
2. writes
: keep all the data in the page cache.
3. flush to the database file with atomic writes
a. ioctl(F2FS_IOC_DB_OPEN);
b. writes
c. ioctl(F2FS_IOC_COMMIT);
4. close
-> drop the cached data
Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
---
fs/f2fs/data.c | 9 ++++++-
fs/f2fs/f2fs.h | 27 ++++++++++++++++---
fs/f2fs/file.c | 55 ++++++++++++++++++++++++++++++++++++++
fs/f2fs/inline.c | 3 +++
fs/f2fs/inode.c | 4 +++
fs/f2fs/segment.c | 79 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
fs/f2fs/segment.h | 10 +++++--
fs/f2fs/super.c | 2 ++
8 files changed, 183 insertions(+), 6 deletions(-)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 13ab7208..5869cfc 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -14,6 +14,7 @@
#include <linux/mpage.h>
#include <linux/aio.h>
#include <linux/writeback.h>
+#include <linux/mount.h>
#include <linux/backing-dev.h>
#include <linux/blkdev.h>
#include <linux/bio.h>
@@ -1052,7 +1053,10 @@ static int f2fs_write_end(struct file *file,
trace_f2fs_write_end(inode, pos, len, copied);
- set_page_dirty(page);
+ if (f2fs_is_db_file(inode))
+ register_db_page(inode, page);
+ else
+ set_page_dirty(page);
if (pos + copied > i_size_read(inode)) {
i_size_write(inode, pos + copied);
@@ -1116,6 +1120,9 @@ static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
if (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE)
return;
+ if (f2fs_is_db_file(inode))
+ invalidate_db_page(inode, page);
+
if (PageDirty(page))
inode_dec_dirty_pages(inode);
ClearPagePrivate(page);
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index a397f7a..f424ae7 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -192,8 +192,13 @@ static inline bool __has_cursum_space(struct f2fs_summary_block *sum, int size,
/*
* ioctl commands
*/
-#define F2FS_IOC_GETFLAGS FS_IOC_GETFLAGS
-#define F2FS_IOC_SETFLAGS FS_IOC_SETFLAGS
+#define F2FS_IOC_GETFLAGS FS_IOC_GETFLAGS
+#define F2FS_IOC_SETFLAGS FS_IOC_SETFLAGS
+
+#define F2FS_IOCTL_MAGIC 0xf5
+#define F2FS_IOC_DB_OPEN _IO(F2FS_IOCTL_MAGIC, 1)
+#define F2FS_IOC_JOURNAL_OPEN _IO(F2FS_IOCTL_MAGIC, 2)
+#define F2FS_IOC_COMMIT _IO(F2FS_IOCTL_MAGIC, 3)
#if defined(__KERNEL__) && defined(CONFIG_COMPAT)
/*
@@ -263,6 +268,9 @@ struct f2fs_inode_info {
unsigned long long xattr_ver; /* cp version of xattr modification */
struct extent_info ext; /* in-memory extent cache entry */
struct dir_inode_entry *dirty_dir; /* the pointer of dirty dir */
+
+ struct list_head db_pages; /* atomic page indexes */
+ struct mutex db_lock; /* lock for atomic pages */
};
static inline void get_extent_info(struct extent_info *ext,
@@ -1051,7 +1059,9 @@ enum {
FI_INLINE_DATA, /* used for inline data*/
FI_APPEND_WRITE, /* inode has appended data */
FI_UPDATE_WRITE, /* inode has in-place-update data */
- FI_NEED_IPU, /* used fo ipu for fdatasync */
+ FI_NEED_IPU, /* used for ipu for fdatasync */
+ FI_DB_FILE, /* indicate database file */
+ FI_JOURNAL_FILE, /* indicate journal file */
};
static inline void set_inode_flag(struct f2fs_inode_info *fi, int flag)
@@ -1111,6 +1121,14 @@ static inline int f2fs_has_inline_xattr(struct inode *inode)
return is_inode_flag_set(F2FS_I(inode), FI_INLINE_XATTR);
}
+static inline bool f2fs_is_db_file(struct inode *inode)
+{
+ if (is_inode_flag_set(F2FS_I(inode), FI_DB_FILE) ||
+ is_inode_flag_set(F2FS_I(inode), FI_JOURNAL_FILE))
+ return true;
+ return false;
+}
+
static inline unsigned int addrs_per_inode(struct f2fs_inode_info *fi)
{
if (f2fs_has_inline_xattr(&fi->vfs_inode))
@@ -1275,6 +1293,9 @@ void destroy_node_manager_caches(void);
/*
* segment.c
*/
+void register_db_page(struct inode *, struct page *);
+void invalidate_db_page(struct inode *, struct page *);
+void commit_db_pages(struct inode *, bool);
void f2fs_balance_fs(struct f2fs_sb_info *);
void f2fs_balance_fs_bg(struct f2fs_sb_info *);
int f2fs_issue_flush(struct f2fs_sb_info *);
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 735e9a2..de486ad 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -862,6 +862,55 @@ out:
return ret;
}
+static int f2fs_ioc_open_db_file(struct file *filp)
+{
+ struct inode *inode = file_inode(filp);
+ struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+
+ if (!inode_owner_or_capable(inode))
+ return -EACCES;
+
+ f2fs_balance_fs(sbi);
+
+ set_inode_flag(F2FS_I(inode), FI_DB_FILE);
+
+ return f2fs_convert_inline_data(inode, MAX_INLINE_DATA + 1, NULL);
+}
+
+static int f2fs_ioc_open_journal_file(struct file *filp)
+{
+ struct inode *inode = file_inode(filp);
+
+ if (!inode_owner_or_capable(inode))
+ return -EACCES;
+
+ set_inode_flag(F2FS_I(inode), FI_JOURNAL_FILE);
+ return 0;
+}
+
+static int f2fs_ioc_commit(struct file *filp)
+{
+ struct inode *inode = file_inode(filp);
+ int ret;
+
+ if (!inode_owner_or_capable(inode))
+ return -EACCES;
+
+ if (is_inode_flag_set(F2FS_I(inode), FI_JOURNAL_FILE))
+ return 0;
+
+ ret = mnt_want_write_file(filp);
+ if (ret)
+ return ret;
+
+ if (is_inode_flag_set(F2FS_I(inode), FI_DB_FILE))
+ commit_db_pages(inode, false);
+
+ ret = f2fs_sync_file(filp, 0, LONG_MAX, 0);
+ mnt_drop_write_file(filp);
+ return ret;
+}
+
static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
{
struct inode *inode = file_inode(filp);
@@ -899,6 +948,12 @@ long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
return f2fs_ioc_getflags(filp, arg);
case F2FS_IOC_SETFLAGS:
return f2fs_ioc_setflags(filp, arg);
+ case F2FS_IOC_DB_OPEN:
+ return f2fs_ioc_open_db_file(filp);
+ case F2FS_IOC_JOURNAL_OPEN:
+ return f2fs_ioc_open_journal_file(filp);
+ case F2FS_IOC_COMMIT:
+ return f2fs_ioc_commit(filp);
case FITRIM:
return f2fs_ioc_fitrim(filp, arg);
default:
diff --git a/fs/f2fs/inline.c b/fs/f2fs/inline.c
index 6aef11d..b47377b 100644
--- a/fs/f2fs/inline.c
+++ b/fs/f2fs/inline.c
@@ -21,6 +21,9 @@ bool f2fs_may_inline(struct inode *inode)
if (!test_opt(F2FS_I_SB(inode), INLINE_DATA))
return false;
+ if (is_inode_flag_set(F2FS_I(inode), FI_DB_FILE))
+ return false;
+
nr_blocks = F2FS_I(inode)->i_xattr_nid ? 3 : 2;
if (inode->i_blocks > nr_blocks)
return false;
diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index 63923ee..23f718e 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -269,6 +269,10 @@ void f2fs_evict_inode(struct inode *inode)
struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
nid_t xnid = F2FS_I(inode)->i_xattr_nid;
+ /* some remained atomic pages should discarded */
+ if (f2fs_is_db_file(inode))
+ commit_db_pages(inode, true);
+
trace_f2fs_evict_inode(inode);
truncate_inode_pages_final(&inode->i_data);
diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index 4d1c49a..b70e2ac 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -26,6 +26,7 @@
static struct kmem_cache *discard_entry_slab;
static struct kmem_cache *sit_entry_set_slab;
+static struct kmem_cache *aw_entry_slab;
/*
* __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
@@ -173,6 +174,77 @@ found_middle:
return result + __reverse_ffz(tmp);
}
+/* For atomic write support */
+void register_db_page(struct inode *inode, struct page *page)
+{
+ struct f2fs_inode_info *fi = F2FS_I(inode);
+ struct db_pages *new;
+
+ new = f2fs_kmem_cache_alloc(aw_entry_slab, GFP_NOFS);
+
+ /* add atomic page indices to the list */
+ new->page = page;
+ INIT_LIST_HEAD(&new->list);
+
+ /* increase reference count with clean state */
+ mutex_lock(&fi->db_lock);
+ get_page(page);
+ list_add_tail(&new->list, &fi->db_pages);
+ mutex_unlock(&fi->db_lock);
+}
+
+void invalidate_db_page(struct inode *inode, struct page *page)
+{
+ struct f2fs_inode_info *fi = F2FS_I(inode);
+ struct db_pages *cur, *tmp;
+
+ mutex_lock(&fi->db_lock);
+ list_for_each_entry_safe(cur, tmp, &fi->db_pages, list) {
+ if (cur->page == page) {
+ put_page(page);
+ list_del(&cur->list);
+ kmem_cache_free(aw_entry_slab, cur);
+ }
+ }
+ mutex_unlock(&fi->db_lock);
+}
+
+void commit_db_pages(struct inode *inode, bool abort)
+{
+ struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+ struct f2fs_inode_info *fi = F2FS_I(inode);
+ struct db_pages *cur, *tmp;
+ bool submit_bio = false;
+ struct f2fs_io_info fio = {
+ .type = DATA,
+ .rw = WRITE_SYNC,
+ };
+
+ f2fs_balance_fs(sbi);
+ f2fs_lock_op(sbi);
+
+ mutex_lock(&fi->db_lock);
+ list_for_each_entry_safe(cur, tmp, &fi->db_pages, list) {
+ lock_page(cur->page);
+ if (!abort && cur->page->mapping == inode->i_mapping) {
+ f2fs_wait_on_page_writeback(cur->page, DATA);
+ if (clear_page_dirty_for_io(cur->page))
+ inode_dec_dirty_pages(inode);
+ do_write_data_page(cur->page, &fio);
+ submit_bio = true;
+ }
+ f2fs_put_page(cur->page, 1);
+ list_del(&cur->list);
+ kmem_cache_free(aw_entry_slab, cur);
+ }
+ if (submit_bio)
+ f2fs_submit_merged_bio(sbi, DATA, WRITE);
+ mutex_unlock(&fi->db_lock);
+
+ filemap_fdatawait_range(inode->i_mapping, 0, LLONG_MAX);
+ f2fs_unlock_op(sbi);
+}
+
/*
* This function balances dirty node and dentry pages.
* In addition, it controls garbage collection.
@@ -2148,8 +2220,14 @@ int __init create_segment_manager_caches(void)
sizeof(struct nat_entry_set));
if (!sit_entry_set_slab)
goto destory_discard_entry;
+ aw_entry_slab = f2fs_kmem_cache_create("db_page_entry",
+ sizeof(struct db_pages));
+ if (!aw_entry_slab)
+ goto destroy_sit_entry_set;
return 0;
+destroy_sit_entry_set:
+ kmem_cache_destroy(sit_entry_set_slab);
destory_discard_entry:
kmem_cache_destroy(discard_entry_slab);
fail:
@@ -2160,4 +2238,5 @@ void destroy_segment_manager_caches(void)
{
kmem_cache_destroy(sit_entry_set_slab);
kmem_cache_destroy(discard_entry_slab);
+ kmem_cache_destroy(aw_entry_slab);
}
diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
index afb7362..45583a9 100644
--- a/fs/f2fs/segment.h
+++ b/fs/f2fs/segment.h
@@ -175,6 +175,11 @@ struct segment_allocation {
void (*allocate_segment)(struct f2fs_sb_info *, int, bool);
};
+struct db_pages {
+ struct list_head list;
+ struct page *page;
+};
+
struct sit_info {
const struct segment_allocation *s_ops;
@@ -502,9 +507,10 @@ static inline bool need_inplace_update(struct inode *inode)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
unsigned int policy = SM_I(sbi)->ipu_policy;
+ struct f2fs_inode_info *fi = F2FS_I(inode);
/* IPU can be done only for the user data */
- if (S_ISDIR(inode->i_mode))
+ if (S_ISDIR(inode->i_mode) || is_inode_flag_set(fi, FI_DB_FILE))
return false;
if (policy & (0x1 << F2FS_IPU_FORCE))
@@ -520,7 +526,7 @@ static inline bool need_inplace_update(struct inode *inode)
/* this is only set during fdatasync */
if (policy & (0x1 << F2FS_IPU_FSYNC) &&
- is_inode_flag_set(F2FS_I(inode), FI_NEED_IPU))
+ is_inode_flag_set(fi, FI_NEED_IPU))
return true;
return false;
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index bb6b568..68a5047 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -373,6 +373,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
fi->i_advise = 0;
rwlock_init(&fi->ext.ext_lock);
init_rwsem(&fi->i_sem);
+ INIT_LIST_HEAD(&fi->db_pages);
+ mutex_init(&fi->db_lock);
set_inode_flag(fi, FI_NEW_INODE);
--
2.1.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/