f2fs: introduce a new global lock scheme
In the previous version, f2fs uses global locks according to the usage types,
such as directory operations, block allocation, block write, and so on.
Reference the following lock types in f2fs.h.
enum lock_type {
RENAME, /* for renaming operations */
DENTRY_OPS, /* for directory operations */
DATA_WRITE, /* for data write */
DATA_NEW, /* for data allocation */
DATA_TRUNC, /* for data truncate */
NODE_NEW, /* for node allocation */
NODE_TRUNC, /* for node truncate */
NODE_WRITE, /* for node write */
NR_LOCK_TYPE,
};
In that case, we lose the performance under the multi-threading environment,
since every types of operations must be conducted one at a time.
In order to address the problem, let's share the locks globally with a mutex
array regardless of any types.
So, let users grab a mutex and perform their jobs in parallel as much as
possbile.
For this, I propose a new global lock scheme as follows.
0. Data structure
- f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS]
- f2fs_sb_info -> node_write
1. mutex_lock_op(sbi)
- try to get an avaiable lock from the array.
- returns the index of the gottern lock variable.
2. mutex_unlock_op(sbi, index of the lock)
- unlock the given index of the lock.
3. mutex_lock_all(sbi)
- grab all the locks in the array before the checkpoint.
4. mutex_unlock_all(sbi)
- release all the locks in the array after checkpoint.
5. block_operations()
- call mutex_lock_all()
- sync_dirty_dir_inodes()
- grab node_write
- sync_node_pages()
Note that,
the pairs of mutex_lock_op()/mutex_unlock_op() and
mutex_lock_all()/mutex_unlock_all() should be used together.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index 93fd57d..be6aa2ee 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -543,54 +543,39 @@
*/
static void block_operations(struct f2fs_sb_info *sbi)
{
- int t;
struct writeback_control wbc = {
.sync_mode = WB_SYNC_ALL,
.nr_to_write = LONG_MAX,
.for_reclaim = 0,
};
+retry_flush_dents:
+ mutex_lock_all(sbi);
- /* Stop renaming operation */
- mutex_lock_op(sbi, RENAME);
- mutex_lock_op(sbi, DENTRY_OPS);
-
-retry_dents:
/* write all the dirty dentry pages */
- sync_dirty_dir_inodes(sbi);
-
- mutex_lock_op(sbi, DATA_WRITE);
if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
- mutex_unlock_op(sbi, DATA_WRITE);
- goto retry_dents;
+ mutex_unlock_all(sbi);
+ sync_dirty_dir_inodes(sbi);
+ goto retry_flush_dents;
}
- /* block all the operations */
- for (t = DATA_NEW; t <= NODE_TRUNC; t++)
- mutex_lock_op(sbi, t);
-
- mutex_lock(&sbi->write_inode);
-
/*
* POR: we should ensure that there is no dirty node pages
* until finishing nat/sit flush.
*/
-retry:
- sync_node_pages(sbi, 0, &wbc);
-
- mutex_lock_op(sbi, NODE_WRITE);
+retry_flush_nodes:
+ mutex_lock(&sbi->node_write);
if (get_pages(sbi, F2FS_DIRTY_NODES)) {
- mutex_unlock_op(sbi, NODE_WRITE);
- goto retry;
+ mutex_unlock(&sbi->node_write);
+ sync_node_pages(sbi, 0, &wbc);
+ goto retry_flush_nodes;
}
- mutex_unlock(&sbi->write_inode);
}
static void unblock_operations(struct f2fs_sb_info *sbi)
{
- int t;
- for (t = NODE_WRITE; t >= RENAME; t--)
- mutex_unlock_op(sbi, t);
+ mutex_unlock(&sbi->node_write);
+ mutex_unlock_all(sbi);
}
static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index cf9ff5f..72a1b30 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -260,6 +260,9 @@
/*
* Caller ensures that this data page is never allocated.
* A new zero-filled data page is allocated in the page cache.
+ *
+ * Also, caller should grab and release a mutex by calling mutex_lock_op() and
+ * mutex_unlock_op().
*/
struct page *get_new_data_page(struct inode *inode, pgoff_t index,
bool new_i_size)
@@ -479,10 +482,11 @@
const pgoff_t end_index = ((unsigned long long) i_size)
>> PAGE_CACHE_SHIFT;
unsigned offset;
+ bool need_balance_fs = false;
int err = 0;
if (page->index < end_index)
- goto out;
+ goto write;
/*
* If the offset is out-of-range of file size,
@@ -494,50 +498,46 @@
dec_page_count(sbi, F2FS_DIRTY_DENTS);
inode_dec_dirty_dents(inode);
}
- goto unlock_out;
+ goto out;
}
zero_user_segment(page, offset, PAGE_CACHE_SIZE);
-out:
- if (sbi->por_doing)
+write:
+ if (sbi->por_doing) {
+ err = AOP_WRITEPAGE_ACTIVATE;
goto redirty_out;
+ }
- if (wbc->for_reclaim && !S_ISDIR(inode->i_mode) && !is_cold_data(page))
- goto redirty_out;
-
- mutex_lock_op(sbi, DATA_WRITE);
+ /* Dentry blocks are controlled by checkpoint */
if (S_ISDIR(inode->i_mode)) {
dec_page_count(sbi, F2FS_DIRTY_DENTS);
inode_dec_dirty_dents(inode);
+ err = do_write_data_page(page);
+ } else {
+ int ilock = mutex_lock_op(sbi);
+ err = do_write_data_page(page);
+ mutex_unlock_op(sbi, ilock);
+ need_balance_fs = true;
}
- err = do_write_data_page(page);
- if (err && err != -ENOENT) {
- wbc->pages_skipped++;
- set_page_dirty(page);
- }
- mutex_unlock_op(sbi, DATA_WRITE);
+ if (err == -ENOENT)
+ goto out;
+ else if (err)
+ goto redirty_out;
if (wbc->for_reclaim)
f2fs_submit_bio(sbi, DATA, true);
- if (err == -ENOENT)
- goto unlock_out;
-
clear_cold_data(page);
+out:
unlock_page(page);
-
- if (!wbc->for_reclaim && !S_ISDIR(inode->i_mode))
+ if (need_balance_fs)
f2fs_balance_fs(sbi);
return 0;
-unlock_out:
- unlock_page(page);
- return (err == -ENOENT) ? 0 : err;
-
redirty_out:
wbc->pages_skipped++;
set_page_dirty(page);
- return AOP_WRITEPAGE_ACTIVATE;
+ return err;
}
#define MAX_DESIRED_PAGES_WP 4096
@@ -592,6 +592,7 @@
pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
struct dnode_of_data dn;
int err = 0;
+ int ilock;
/* for nobh_write_end */
*fsdata = NULL;
@@ -603,28 +604,21 @@
return -ENOMEM;
*pagep = page;
- mutex_lock_op(sbi, DATA_NEW);
+ ilock = mutex_lock_op(sbi);
set_new_dnode(&dn, inode, NULL, NULL, 0);
err = get_dnode_of_data(&dn, index, ALLOC_NODE);
- if (err) {
- mutex_unlock_op(sbi, DATA_NEW);
- f2fs_put_page(page, 1);
- return err;
- }
+ if (err)
+ goto err;
- if (dn.data_blkaddr == NULL_ADDR) {
+ if (dn.data_blkaddr == NULL_ADDR)
err = reserve_new_block(&dn);
- if (err) {
- f2fs_put_dnode(&dn);
- mutex_unlock_op(sbi, DATA_NEW);
- f2fs_put_page(page, 1);
- return err;
- }
- }
- f2fs_put_dnode(&dn);
- mutex_unlock_op(sbi, DATA_NEW);
+ f2fs_put_dnode(&dn);
+ if (err)
+ goto err;
+
+ mutex_unlock_op(sbi, ilock);
if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
return 0;
@@ -654,6 +648,11 @@
SetPageUptodate(page);
clear_cold_data(page);
return 0;
+
+err:
+ mutex_unlock_op(sbi, ilock);
+ f2fs_put_page(page, 1);
+ return err;
}
static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
index 2851ae6..cd3342d 100644
--- a/fs/f2fs/dir.c
+++ b/fs/f2fs/dir.c
@@ -249,9 +249,6 @@
void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
struct page *page, struct inode *inode)
{
- struct f2fs_sb_info *sbi = F2FS_SB(dir->i_sb);
-
- mutex_lock_op(sbi, DENTRY_OPS);
lock_page(page);
wait_on_page_writeback(page);
de->ino = cpu_to_le32(inode->i_ino);
@@ -265,7 +262,6 @@
F2FS_I(inode)->i_pino = dir->i_ino;
f2fs_put_page(page, 1);
- mutex_unlock_op(sbi, DENTRY_OPS);
}
void init_dent_inode(const struct qstr *name, struct page *ipage)
@@ -284,6 +280,43 @@
set_page_dirty(ipage);
}
+static int make_empty_dir(struct inode *inode, struct inode *parent)
+{
+ struct page *dentry_page;
+ struct f2fs_dentry_block *dentry_blk;
+ struct f2fs_dir_entry *de;
+ void *kaddr;
+
+ dentry_page = get_new_data_page(inode, 0, true);
+ if (IS_ERR(dentry_page))
+ return PTR_ERR(dentry_page);
+
+ kaddr = kmap_atomic(dentry_page);
+ dentry_blk = (struct f2fs_dentry_block *)kaddr;
+
+ de = &dentry_blk->dentry[0];
+ de->name_len = cpu_to_le16(1);
+ de->hash_code = 0;
+ de->ino = cpu_to_le32(inode->i_ino);
+ memcpy(dentry_blk->filename[0], ".", 1);
+ set_de_type(de, inode);
+
+ de = &dentry_blk->dentry[1];
+ de->hash_code = 0;
+ de->name_len = cpu_to_le16(2);
+ de->ino = cpu_to_le32(parent->i_ino);
+ memcpy(dentry_blk->filename[1], "..", 2);
+ set_de_type(de, inode);
+
+ test_and_set_bit_le(0, &dentry_blk->dentry_bitmap);
+ test_and_set_bit_le(1, &dentry_blk->dentry_bitmap);
+ kunmap_atomic(kaddr);
+
+ set_page_dirty(dentry_page);
+ f2fs_put_page(dentry_page, 1);
+ return 0;
+}
+
static int init_inode_metadata(struct inode *inode,
struct inode *dir, const struct qstr *name)
{
@@ -294,7 +327,7 @@
return err;
if (S_ISDIR(inode->i_mode)) {
- err = f2fs_make_empty(inode, dir);
+ err = make_empty_dir(inode, dir);
if (err) {
remove_inode_page(inode);
return err;
@@ -317,7 +350,7 @@
}
if (is_inode_flag_set(F2FS_I(inode), FI_INC_LINK)) {
inc_nlink(inode);
- f2fs_write_inode(inode, NULL);
+ update_inode_page(inode);
}
return 0;
}
@@ -341,7 +374,7 @@
}
if (need_dir_update)
- f2fs_write_inode(dir, NULL);
+ update_inode_page(dir);
else
mark_inode_dirty(dir);
@@ -373,6 +406,10 @@
goto next;
}
+/*
+ * Caller should grab and release a mutex by calling mutex_lock_op() and
+ * mutex_unlock_op().
+ */
int __f2fs_add_link(struct inode *dir, const struct qstr *name, struct inode *inode)
{
unsigned int bit_pos;
@@ -382,7 +419,6 @@
f2fs_hash_t dentry_hash;
struct f2fs_dir_entry *de;
unsigned int nbucket, nblock;
- struct f2fs_sb_info *sbi = F2FS_SB(dir->i_sb);
size_t namelen = name->len;
struct page *dentry_page = NULL;
struct f2fs_dentry_block *dentry_blk = NULL;
@@ -412,12 +448,9 @@
bidx = dir_block_index(level, (le32_to_cpu(dentry_hash) % nbucket));
for (block = bidx; block <= (bidx + nblock - 1); block++) {
- mutex_lock_op(sbi, DENTRY_OPS);
dentry_page = get_new_data_page(dir, block, true);
- if (IS_ERR(dentry_page)) {
- mutex_unlock_op(sbi, DENTRY_OPS);
+ if (IS_ERR(dentry_page))
return PTR_ERR(dentry_page);
- }
dentry_blk = kmap(dentry_page);
bit_pos = room_for_filename(dentry_blk, slots);
@@ -426,7 +459,6 @@
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
- mutex_unlock_op(sbi, DENTRY_OPS);
}
/* Move to next level to find the empty slot for new dentry */
@@ -456,7 +488,6 @@
fail:
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
- mutex_unlock_op(sbi, DENTRY_OPS);
return err;
}
@@ -476,8 +507,6 @@
void *kaddr = page_address(page);
int i;
- mutex_lock_op(sbi, DENTRY_OPS);
-
lock_page(page);
wait_on_page_writeback(page);
@@ -497,7 +526,7 @@
if (inode && S_ISDIR(inode->i_mode)) {
drop_nlink(dir);
- f2fs_write_inode(dir, NULL);
+ update_inode_page(dir);
} else {
mark_inode_dirty(dir);
}
@@ -509,7 +538,8 @@
drop_nlink(inode);
i_size_write(inode, 0);
}
- f2fs_write_inode(inode, NULL);
+ update_inode_page(inode);
+
if (inode->i_nlink == 0)
add_orphan_inode(sbi, inode->i_ino);
}
@@ -522,45 +552,6 @@
inode_dec_dirty_dents(dir);
}
f2fs_put_page(page, 1);
-
- mutex_unlock_op(sbi, DENTRY_OPS);
-}
-
-int f2fs_make_empty(struct inode *inode, struct inode *parent)
-{
- struct page *dentry_page;
- struct f2fs_dentry_block *dentry_blk;
- struct f2fs_dir_entry *de;
- void *kaddr;
-
- dentry_page = get_new_data_page(inode, 0, true);
- if (IS_ERR(dentry_page))
- return PTR_ERR(dentry_page);
-
- kaddr = kmap_atomic(dentry_page);
- dentry_blk = (struct f2fs_dentry_block *)kaddr;
-
- de = &dentry_blk->dentry[0];
- de->name_len = cpu_to_le16(1);
- de->hash_code = f2fs_dentry_hash(".", 1);
- de->ino = cpu_to_le32(inode->i_ino);
- memcpy(dentry_blk->filename[0], ".", 1);
- set_de_type(de, inode);
-
- de = &dentry_blk->dentry[1];
- de->hash_code = f2fs_dentry_hash("..", 2);
- de->name_len = cpu_to_le16(2);
- de->ino = cpu_to_le32(parent->i_ino);
- memcpy(dentry_blk->filename[1], "..", 2);
- set_de_type(de, inode);
-
- test_and_set_bit_le(0, &dentry_blk->dentry_bitmap);
- test_and_set_bit_le(1, &dentry_blk->dentry_bitmap);
- kunmap_atomic(kaddr);
-
- set_page_dirty(dentry_page);
- f2fs_put_page(dentry_page, 1);
- return 0;
}
bool f2fs_empty_dir(struct inode *dir)
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 71eacd3..06cc75c 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -309,23 +309,12 @@
};
/*
- * FS_LOCK nesting subclasses for the lock validator:
- *
- * The locking order between these classes is
- * RENAME -> DENTRY_OPS -> DATA_WRITE -> DATA_NEW
- * -> DATA_TRUNC -> NODE_WRITE -> NODE_NEW -> NODE_TRUNC
+ * Uses as sbi->fs_lock[NR_GLOBAL_LOCKS].
+ * The checkpoint procedure blocks all the locks in this fs_lock array.
+ * Some FS operations grab free locks, and if there is no free lock,
+ * then wait to grab a lock in a round-robin manner.
*/
-enum lock_type {
- RENAME, /* for renaming operations */
- DENTRY_OPS, /* for directory operations */
- DATA_WRITE, /* for data write */
- DATA_NEW, /* for data allocation */
- DATA_TRUNC, /* for data truncate */
- NODE_NEW, /* for node allocation */
- NODE_TRUNC, /* for node truncate */
- NODE_WRITE, /* for node write */
- NR_LOCK_TYPE,
-};
+#define NR_GLOBAL_LOCKS 8
/*
* The below are the page types of bios used in submti_bio().
@@ -365,10 +354,11 @@
/* for checkpoint */
struct f2fs_checkpoint *ckpt; /* raw checkpoint pointer */
struct inode *meta_inode; /* cache meta blocks */
- struct mutex cp_mutex; /* for checkpoint procedure */
- struct mutex fs_lock[NR_LOCK_TYPE]; /* for blocking FS operations */
- struct mutex write_inode; /* mutex for write inode */
+ struct mutex cp_mutex; /* checkpoint procedure lock */
+ struct mutex fs_lock[NR_GLOBAL_LOCKS]; /* blocking FS operations */
+ struct mutex node_write; /* locking node writes */
struct mutex writepages; /* mutex for writepages() */
+ unsigned char next_lock_num; /* round-robin global locks */
int por_doing; /* recovery is doing or not */
/* for orphan inode management */
@@ -503,14 +493,40 @@
cp->ckpt_flags = cpu_to_le32(ckpt_flags);
}
-static inline void mutex_lock_op(struct f2fs_sb_info *sbi, enum lock_type t)
+static inline void mutex_lock_all(struct f2fs_sb_info *sbi)
{
- mutex_lock_nested(&sbi->fs_lock[t], t);
+ int i = 0;
+ for (; i < NR_GLOBAL_LOCKS; i++)
+ mutex_lock(&sbi->fs_lock[i]);
}
-static inline void mutex_unlock_op(struct f2fs_sb_info *sbi, enum lock_type t)
+static inline void mutex_unlock_all(struct f2fs_sb_info *sbi)
{
- mutex_unlock(&sbi->fs_lock[t]);
+ int i = 0;
+ for (; i < NR_GLOBAL_LOCKS; i++)
+ mutex_unlock(&sbi->fs_lock[i]);
+}
+
+static inline int mutex_lock_op(struct f2fs_sb_info *sbi)
+{
+ unsigned char next_lock = sbi->next_lock_num % NR_GLOBAL_LOCKS;
+ int i = 0;
+
+ for (; i < NR_GLOBAL_LOCKS; i++)
+ if (mutex_trylock(&sbi->fs_lock[i]))
+ return i;
+
+ mutex_lock(&sbi->fs_lock[next_lock]);
+ sbi->next_lock_num++;
+ return next_lock;
+}
+
+static inline void mutex_unlock_op(struct f2fs_sb_info *sbi, int ilock)
+{
+ if (ilock < 0)
+ return;
+ BUG_ON(ilock >= NR_GLOBAL_LOCKS);
+ mutex_unlock(&sbi->fs_lock[ilock]);
}
/*
@@ -879,6 +895,7 @@
void f2fs_set_inode_flags(struct inode *);
struct inode *f2fs_iget(struct super_block *, unsigned long);
void update_inode(struct inode *, struct page *);
+int update_inode_page(struct inode *);
int f2fs_write_inode(struct inode *, struct writeback_control *);
void f2fs_evict_inode(struct inode *);
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 155b362..07be88d 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -34,19 +34,18 @@
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
block_t old_blk_addr;
struct dnode_of_data dn;
- int err;
+ int err, ilock;
f2fs_balance_fs(sbi);
sb_start_pagefault(inode->i_sb);
- mutex_lock_op(sbi, DATA_NEW);
-
/* block allocation */
+ ilock = mutex_lock_op(sbi);
set_new_dnode(&dn, inode, NULL, NULL, 0);
err = get_dnode_of_data(&dn, page->index, ALLOC_NODE);
if (err) {
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
goto out;
}
@@ -56,13 +55,12 @@
err = reserve_new_block(&dn);
if (err) {
f2fs_put_dnode(&dn);
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
goto out;
}
}
f2fs_put_dnode(&dn);
-
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
lock_page(page);
if (page->mapping != inode->i_mapping ||
@@ -223,20 +221,19 @@
unsigned int blocksize = inode->i_sb->s_blocksize;
struct dnode_of_data dn;
pgoff_t free_from;
- int count = 0;
+ int count = 0, ilock = -1;
int err;
free_from = (pgoff_t)
((from + blocksize - 1) >> (sbi->log_blocksize));
- mutex_lock_op(sbi, DATA_TRUNC);
-
+ ilock = mutex_lock_op(sbi);
set_new_dnode(&dn, inode, NULL, NULL, 0);
err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE);
if (err) {
if (err == -ENOENT)
goto free_next;
- mutex_unlock_op(sbi, DATA_TRUNC);
+ mutex_unlock_op(sbi, ilock);
return err;
}
@@ -247,6 +244,7 @@
count -= dn.ofs_in_node;
BUG_ON(count < 0);
+
if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
truncate_data_blocks_range(&dn, count);
free_from += count;
@@ -255,7 +253,7 @@
f2fs_put_dnode(&dn);
free_next:
err = truncate_inode_blocks(inode, free_from);
- mutex_unlock_op(sbi, DATA_TRUNC);
+ mutex_unlock_op(sbi, ilock);
/* lastly zero out the first data page */
truncate_partial_data_page(inode, from);
@@ -363,15 +361,16 @@
{
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
struct page *page;
+ int ilock;
if (!len)
return;
f2fs_balance_fs(sbi);
- mutex_lock_op(sbi, DATA_NEW);
+ ilock = mutex_lock_op(sbi);
page = get_new_data_page(inode, index, false);
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
if (!IS_ERR(page)) {
wait_on_page_writeback(page);
@@ -388,13 +387,10 @@
for (index = pg_start; index < pg_end; index++) {
struct dnode_of_data dn;
- struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
- mutex_lock_op(sbi, DATA_TRUNC);
set_new_dnode(&dn, inode, NULL, NULL, 0);
err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
if (err) {
- mutex_unlock_op(sbi, DATA_TRUNC);
if (err == -ENOENT)
continue;
return err;
@@ -403,7 +399,6 @@
if (dn.data_blkaddr != NULL_ADDR)
truncate_data_blocks_range(&dn, 1);
f2fs_put_dnode(&dn);
- mutex_unlock_op(sbi, DATA_TRUNC);
}
return 0;
}
@@ -434,6 +429,7 @@
struct address_space *mapping = inode->i_mapping;
loff_t blk_start, blk_end;
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+ int ilock;
f2fs_balance_fs(sbi);
@@ -441,7 +437,10 @@
blk_end = pg_end << PAGE_CACHE_SHIFT;
truncate_inode_pages_range(mapping, blk_start,
blk_end - 1);
+
+ ilock = mutex_lock_op(sbi);
ret = truncate_hole(inode, pg_start, pg_end);
+ mutex_unlock_op(sbi, ilock);
}
}
@@ -475,13 +474,13 @@
for (index = pg_start; index <= pg_end; index++) {
struct dnode_of_data dn;
+ int ilock;
- mutex_lock_op(sbi, DATA_NEW);
-
+ ilock = mutex_lock_op(sbi);
set_new_dnode(&dn, inode, NULL, NULL, 0);
ret = get_dnode_of_data(&dn, index, ALLOC_NODE);
if (ret) {
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
break;
}
@@ -489,13 +488,12 @@
ret = reserve_new_block(&dn);
if (ret) {
f2fs_put_dnode(&dn);
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
break;
}
}
f2fs_put_dnode(&dn);
-
- mutex_unlock_op(sbi, DATA_NEW);
+ mutex_unlock_op(sbi, ilock);
if (pg_start == pg_end)
new_size = offset + len;
diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index e97f301..83cec8f 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -510,7 +510,6 @@
wait_on_page_writeback(page);
}
- mutex_lock_op(sbi, DATA_WRITE);
if (clear_page_dirty_for_io(page) &&
S_ISDIR(inode->i_mode)) {
dec_page_count(sbi, F2FS_DIRTY_DENTS);
@@ -518,7 +517,6 @@
}
set_cold_data(page);
do_write_data_page(page);
- mutex_unlock_op(sbi, DATA_WRITE);
clear_cold_data(page);
}
out:
diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index f798ddf..60105b7 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -195,11 +195,24 @@
set_page_dirty(node_page);
}
-int f2fs_write_inode(struct inode *inode, struct writeback_control *wbc)
+int update_inode_page(struct inode *inode)
{
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
struct page *node_page;
- bool need_lock = false;
+
+ node_page = get_node_page(sbi, inode->i_ino);
+ if (IS_ERR(node_page))
+ return PTR_ERR(node_page);
+
+ update_inode(inode, node_page);
+ f2fs_put_page(node_page, 1);
+ return 0;
+}
+
+int f2fs_write_inode(struct inode *inode, struct writeback_control *wbc)
+{
+ struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+ int ret, ilock;
if (inode->i_ino == F2FS_NODE_INO(sbi) ||
inode->i_ino == F2FS_META_INO(sbi))
@@ -208,25 +221,14 @@
if (wbc)
f2fs_balance_fs(sbi);
- node_page = get_node_page(sbi, inode->i_ino);
- if (IS_ERR(node_page))
- return PTR_ERR(node_page);
-
- if (!PageDirty(node_page)) {
- need_lock = true;
- f2fs_put_page(node_page, 1);
- mutex_lock(&sbi->write_inode);
- node_page = get_node_page(sbi, inode->i_ino);
- if (IS_ERR(node_page)) {
- mutex_unlock(&sbi->write_inode);
- return PTR_ERR(node_page);
- }
- }
- update_inode(inode, node_page);
- f2fs_put_page(node_page, 1);
- if (need_lock)
- mutex_unlock(&sbi->write_inode);
- return 0;
+ /*
+ * We need to lock here to prevent from producing dirty node pages
+ * during the urgent cleaning time when runing out of free sections.
+ */
+ ilock = mutex_lock_op(sbi);
+ ret = update_inode_page(inode);
+ mutex_unlock_op(sbi, ilock);
+ return ret;
}
/*
@@ -235,6 +237,7 @@
void f2fs_evict_inode(struct inode *inode)
{
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
+ int ilock;
truncate_inode_pages(&inode->i_data, 0);
@@ -255,7 +258,10 @@
if (F2FS_HAS_BLOCKS(inode))
f2fs_truncate(inode);
+ ilock = mutex_lock_op(sbi);
remove_inode_page(inode);
+ mutex_unlock_op(sbi, ilock);
+
sb_end_intwrite(inode->i_sb);
no_delete:
clear_inode(inode);
diff --git a/fs/f2fs/namei.c b/fs/f2fs/namei.c
index 7c6e219..841f6b4 100644
--- a/fs/f2fs/namei.c
+++ b/fs/f2fs/namei.c
@@ -26,19 +26,19 @@
nid_t ino;
struct inode *inode;
bool nid_free = false;
- int err;
+ int err, ilock;
inode = new_inode(sb);
if (!inode)
return ERR_PTR(-ENOMEM);
- mutex_lock_op(sbi, NODE_NEW);
+ ilock = mutex_lock_op(sbi);
if (!alloc_nid(sbi, &ino)) {
- mutex_unlock_op(sbi, NODE_NEW);
+ mutex_unlock_op(sbi, ilock);
err = -ENOSPC;
goto fail;
}
- mutex_unlock_op(sbi, NODE_NEW);
+ mutex_unlock_op(sbi, ilock);
inode->i_uid = current_fsuid();
@@ -122,7 +122,7 @@
struct f2fs_sb_info *sbi = F2FS_SB(sb);
struct inode *inode;
nid_t ino = 0;
- int err;
+ int err, ilock;
f2fs_balance_fs(sbi);
@@ -138,7 +138,9 @@
inode->i_mapping->a_ops = &f2fs_dblock_aops;
ino = inode->i_ino;
+ ilock = mutex_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
+ mutex_unlock_op(sbi, ilock);
if (err)
goto out;
@@ -162,7 +164,7 @@
struct inode *inode = old_dentry->d_inode;
struct super_block *sb = dir->i_sb;
struct f2fs_sb_info *sbi = F2FS_SB(sb);
- int err;
+ int err, ilock;
f2fs_balance_fs(sbi);
@@ -170,7 +172,9 @@
atomic_inc(&inode->i_count);
set_inode_flag(F2FS_I(inode), FI_INC_LINK);
+ ilock = mutex_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
+ mutex_unlock_op(sbi, ilock);
if (err)
goto out;
@@ -229,6 +233,7 @@
struct f2fs_dir_entry *de;
struct page *page;
int err = -ENOENT;
+ int ilock;
f2fs_balance_fs(sbi);
@@ -243,7 +248,9 @@
goto fail;
}
+ ilock = mutex_lock_op(sbi);
f2fs_delete_entry(de, page, inode);
+ mutex_unlock_op(sbi, ilock);
/* In order to evict this inode, we set it dirty */
mark_inode_dirty(inode);
@@ -258,7 +265,7 @@
struct f2fs_sb_info *sbi = F2FS_SB(sb);
struct inode *inode;
size_t symlen = strlen(symname) + 1;
- int err;
+ int err, ilock;
f2fs_balance_fs(sbi);
@@ -269,7 +276,9 @@
inode->i_op = &f2fs_symlink_inode_operations;
inode->i_mapping->a_ops = &f2fs_dblock_aops;
+ ilock = mutex_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
+ mutex_unlock_op(sbi, ilock);
if (err)
goto out;
@@ -291,7 +300,7 @@
{
struct f2fs_sb_info *sbi = F2FS_SB(dir->i_sb);
struct inode *inode;
- int err;
+ int err, ilock;
f2fs_balance_fs(sbi);
@@ -305,7 +314,9 @@
mapping_set_gfp_mask(inode->i_mapping, GFP_F2FS_ZERO);
set_inode_flag(F2FS_I(inode), FI_INC_LINK);
+ ilock = mutex_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
+ mutex_unlock_op(sbi, ilock);
if (err)
goto out_fail;
@@ -340,6 +351,7 @@
struct f2fs_sb_info *sbi = F2FS_SB(sb);
struct inode *inode;
int err = 0;
+ int ilock;
if (!new_valid_dev(rdev))
return -EINVAL;
@@ -353,7 +365,9 @@
init_special_inode(inode, inode->i_mode, rdev);
inode->i_op = &f2fs_special_inode_operations;
+ ilock = mutex_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
+ mutex_unlock_op(sbi, ilock);
if (err)
goto out;
@@ -381,7 +395,7 @@
struct f2fs_dir_entry *old_dir_entry = NULL;
struct f2fs_dir_entry *old_entry;
struct f2fs_dir_entry *new_entry;
- int err = -ENOENT;
+ int err = -ENOENT, ilock = -1;
f2fs_balance_fs(sbi);
@@ -396,7 +410,7 @@
goto out_old;
}
- mutex_lock_op(sbi, RENAME);
+ ilock = mutex_lock_op(sbi);
if (new_inode) {
struct page *new_page;
@@ -419,7 +433,7 @@
drop_nlink(new_inode);
if (!new_inode->i_nlink)
add_orphan_inode(sbi, new_inode->i_ino);
- f2fs_write_inode(new_inode, NULL);
+ update_inode_page(new_inode);
} else {
err = f2fs_add_link(new_dentry, old_inode);
if (err)
@@ -427,7 +441,7 @@
if (old_dir_entry) {
inc_nlink(new_dir);
- f2fs_write_inode(new_dir, NULL);
+ update_inode_page(new_dir);
}
}
@@ -445,10 +459,10 @@
f2fs_put_page(old_dir_page, 0);
}
drop_nlink(old_dir);
- f2fs_write_inode(old_dir, NULL);
+ update_inode_page(old_dir);
}
- mutex_unlock_op(sbi, RENAME);
+ mutex_unlock_op(sbi, ilock);
return 0;
out_dir:
@@ -456,7 +470,7 @@
kunmap(old_dir_page);
f2fs_put_page(old_dir_page, 0);
}
- mutex_unlock_op(sbi, RENAME);
+ mutex_unlock_op(sbi, ilock);
out_old:
kunmap(old_page);
f2fs_put_page(old_page, 0);
diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index ad3adbe..5a7edf9 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -385,6 +385,9 @@
/*
* Caller should call f2fs_put_dnode(dn).
+ * Also, it should grab and release a mutex by calling mutex_lock_op() and
+ * mutex_unlock_op() only if ro is not set RDONLY_NODE.
+ * In the case of RDONLY_NODE, we don't need to care about mutex.
*/
int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
{
@@ -415,11 +418,8 @@
bool done = false;
if (!nids[i] && mode == ALLOC_NODE) {
- mutex_lock_op(sbi, NODE_NEW);
-
/* alloc new node */
if (!alloc_nid(sbi, &(nids[i]))) {
- mutex_unlock_op(sbi, NODE_NEW);
err = -ENOSPC;
goto release_pages;
}
@@ -428,14 +428,12 @@
npage[i] = new_node_page(dn, noffset[i]);
if (IS_ERR(npage[i])) {
alloc_nid_failed(sbi, nids[i]);
- mutex_unlock_op(sbi, NODE_NEW);
err = PTR_ERR(npage[i]);
goto release_pages;
}
set_nid(parent, offset[i - 1], nids[i], i == 1);
alloc_nid_done(sbi, nids[i]);
- mutex_unlock_op(sbi, NODE_NEW);
done = true;
} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
npage[i] = get_node_page_ra(parent, offset[i - 1]);
@@ -745,6 +743,10 @@
return err > 0 ? 0 : err;
}
+/*
+ * Caller should grab and release a mutex by calling mutex_lock_op() and
+ * mutex_unlock_op().
+ */
int remove_inode_page(struct inode *inode)
{
struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
@@ -752,21 +754,16 @@
nid_t ino = inode->i_ino;
struct dnode_of_data dn;
- mutex_lock_op(sbi, NODE_TRUNC);
page = get_node_page(sbi, ino);
- if (IS_ERR(page)) {
- mutex_unlock_op(sbi, NODE_TRUNC);
+ if (IS_ERR(page))
return PTR_ERR(page);
- }
if (F2FS_I(inode)->i_xattr_nid) {
nid_t nid = F2FS_I(inode)->i_xattr_nid;
struct page *npage = get_node_page(sbi, nid);
- if (IS_ERR(npage)) {
- mutex_unlock_op(sbi, NODE_TRUNC);
+ if (IS_ERR(npage))
return PTR_ERR(npage);
- }
F2FS_I(inode)->i_xattr_nid = 0;
set_new_dnode(&dn, inode, page, npage, nid);
@@ -778,23 +775,18 @@
BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
set_new_dnode(&dn, inode, page, page, ino);
truncate_node(&dn);
-
- mutex_unlock_op(sbi, NODE_TRUNC);
return 0;
}
int new_inode_page(struct inode *inode, const struct qstr *name)
{
- struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
struct page *page;
struct dnode_of_data dn;
/* allocate inode page for new inode */
set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
- mutex_lock_op(sbi, NODE_NEW);
page = new_node_page(&dn, 0);
init_dent_inode(name, page);
- mutex_unlock_op(sbi, NODE_NEW);
if (IS_ERR(page))
return PTR_ERR(page);
f2fs_put_page(page, 1);
@@ -985,7 +977,7 @@
if (!dn->inode_page_locked)
unlock_page(dn->inode_page);
} else {
- f2fs_write_inode(dn->inode, NULL);
+ update_inode_page(dn->inode);
}
}
@@ -1102,8 +1094,6 @@
wait_on_page_writeback(page);
- mutex_lock_op(sbi, NODE_WRITE);
-
/* get old block addr of this node page */
nid = nid_of_node(page);
BUG_ON(page->index != nid);
@@ -1111,25 +1101,25 @@
get_node_info(sbi, nid, &ni);
/* This page is already truncated */
- if (ni.blk_addr == NULL_ADDR)
- goto out;
+ if (ni.blk_addr == NULL_ADDR) {
+ dec_page_count(sbi, F2FS_DIRTY_NODES);
+ unlock_page(page);
+ return 0;
+ }
if (wbc->for_reclaim) {
dec_page_count(sbi, F2FS_DIRTY_NODES);
wbc->pages_skipped++;
set_page_dirty(page);
- mutex_unlock_op(sbi, NODE_WRITE);
return AOP_WRITEPAGE_ACTIVATE;
}
+ mutex_lock(&sbi->node_write);
set_page_writeback(page);
-
- /* insert node offset */
write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
set_node_addr(sbi, &ni, new_addr);
-out:
dec_page_count(sbi, F2FS_DIRTY_NODES);
- mutex_unlock_op(sbi, NODE_WRITE);
+ mutex_unlock(&sbi->node_write);
unlock_page(page);
return 0;
}
diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
index 61bdaa7..f16d12d 100644
--- a/fs/f2fs/recovery.c
+++ b/fs/f2fs/recovery.c
@@ -242,6 +242,7 @@
struct f2fs_summary sum;
struct node_info ni;
int err = 0;
+ int ilock;
start = start_bidx_of_node(ofs_of_node(page));
if (IS_INODE(page))
@@ -249,10 +250,14 @@
else
end = start + ADDRS_PER_BLOCK;
+ ilock = mutex_lock_op(sbi);
set_new_dnode(&dn, inode, NULL, NULL, 0);
+
err = get_dnode_of_data(&dn, start, ALLOC_NODE);
- if (err)
+ if (err) {
+ mutex_unlock_op(sbi, ilock);
return err;
+ }
wait_on_page_writeback(dn.node_page);
@@ -297,6 +302,7 @@
recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
f2fs_put_dnode(&dn);
+ mutex_unlock_op(sbi, ilock);
return 0;
}
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index ca54133..2643c49 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -556,11 +556,11 @@
sbi->raw_super = raw_super;
sbi->raw_super_buf = raw_super_buf;
mutex_init(&sbi->gc_mutex);
- mutex_init(&sbi->write_inode);
mutex_init(&sbi->writepages);
mutex_init(&sbi->cp_mutex);
- for (i = 0; i < NR_LOCK_TYPE; i++)
+ for (i = 0; i < NR_GLOBAL_LOCKS; i++)
mutex_init(&sbi->fs_lock[i]);
+ mutex_init(&sbi->node_write);
sbi->por_doing = 0;
spin_lock_init(&sbi->stat_lock);
init_rwsem(&sbi->bio_sem);
diff --git a/fs/f2fs/xattr.c b/fs/f2fs/xattr.c
index 3bfea80..0b02dce 100644
--- a/fs/f2fs/xattr.c
+++ b/fs/f2fs/xattr.c
@@ -307,6 +307,7 @@
int error, found, free, newsize;
size_t name_len;
char *pval;
+ int ilock;
if (name == NULL)
return -EINVAL;
@@ -321,7 +322,8 @@
f2fs_balance_fs(sbi);
- mutex_lock_op(sbi, NODE_NEW);
+ ilock = mutex_lock_op(sbi);
+
if (!fi->i_xattr_nid) {
/* Allocate new attribute block */
struct dnode_of_data dn;
@@ -433,13 +435,13 @@
inode->i_ctime = CURRENT_TIME;
clear_inode_flag(fi, FI_ACL_MODE);
}
- f2fs_write_inode(inode, NULL);
- mutex_unlock_op(sbi, NODE_NEW);
+ update_inode_page(inode);
+ mutex_unlock_op(sbi, ilock);
return 0;
cleanup:
f2fs_put_page(page, 1);
exit:
- mutex_unlock_op(sbi, NODE_NEW);
+ mutex_unlock_op(sbi, ilock);
return error;
}