linux_old1/fs/f2fs/namei.c

1100 lines
25 KiB
C
Raw Normal View History

/*
* fs/f2fs/namei.c
*
* Copyright (c) 2012 Samsung Electronics Co., Ltd.
* http://www.samsung.com/
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/fs.h>
#include <linux/f2fs_fs.h>
#include <linux/pagemap.h>
#include <linux/sched.h>
#include <linux/ctype.h>
#include <linux/dcache.h>
#include <linux/namei.h>
#include "f2fs.h"
#include "node.h"
#include "xattr.h"
#include "acl.h"
#include <trace/events/f2fs.h>
static struct inode *f2fs_new_inode(struct inode *dir, umode_t mode)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
nid_t ino;
struct inode *inode;
bool nid_free = false;
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err;
inode = new_inode(dir->i_sb);
if (!inode)
return ERR_PTR(-ENOMEM);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
if (!alloc_nid(sbi, &ino)) {
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_unlock_op(sbi);
err = -ENOSPC;
goto fail;
}
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_unlock_op(sbi);
inode_init_owner(inode, dir, mode);
inode->i_ino = ino;
inode->i_blocks = 0;
inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
inode->i_generation = sbi->s_next_generation++;
err = insert_inode_locked(inode);
if (err) {
err = -EINVAL;
nid_free = true;
goto fail;
}
/* If the directory encrypted, then we should encrypt the inode. */
if (f2fs_encrypted_inode(dir) && f2fs_may_encrypt(inode))
f2fs_set_encrypted_inode(inode);
if (test_opt(sbi, INLINE_DATA) && f2fs_may_inline_data(inode))
set_inode_flag(F2FS_I(inode), FI_INLINE_DATA);
if (f2fs_may_inline_dentry(inode))
set_inode_flag(F2FS_I(inode), FI_INLINE_DENTRY);
f2fs_init_extent_tree(inode, NULL);
stat_inc_inline_xattr(inode);
stat_inc_inline_inode(inode);
stat_inc_inline_dir(inode);
trace_f2fs_new_inode(inode, 0);
mark_inode_dirty(inode);
return inode;
fail:
trace_f2fs_new_inode(inode, err);
make_bad_inode(inode);
if (nid_free)
set_inode_flag(F2FS_I(inode), FI_FREE_NID);
iput(inode);
return ERR_PTR(err);
}
static int is_multimedia_file(const unsigned char *s, const char *sub)
{
size_t slen = strlen(s);
size_t sublen = strlen(sub);
/*
* filename format of multimedia file should be defined as:
* "filename + '.' + extension".
*/
if (slen < sublen + 2)
return 0;
if (s[slen - sublen - 1] != '.')
return 0;
return !strncasecmp(s + slen - sublen, sub, sublen);
}
/*
* Set multimedia files as cold files for hot/cold data separation
*/
static inline void set_cold_files(struct f2fs_sb_info *sbi, struct inode *inode,
const unsigned char *name)
{
int i;
__u8 (*extlist)[8] = sbi->raw_super->extension_list;
int count = le32_to_cpu(sbi->raw_super->extension_count);
for (i = 0; i < count; i++) {
if (is_multimedia_file(name, extlist[i])) {
file_set_cold(inode);
break;
}
}
}
static int f2fs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
bool excl)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode;
nid_t ino = 0;
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err;
inode = f2fs_new_inode(dir, mode);
if (IS_ERR(inode))
return PTR_ERR(inode);
if (!test_opt(sbi, DISABLE_EXT_IDENTIFY))
set_cold_files(sbi, inode, dentry->d_name.name);
inode->i_op = &f2fs_file_inode_operations;
inode->i_fop = &f2fs_file_operations;
inode->i_mapping->a_ops = &f2fs_dblock_aops;
ino = inode->i_ino;
f2fs_balance_fs(sbi, true);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
if (err)
goto out;
f2fs_unlock_op(sbi);
alloc_nid_done(sbi, ino);
d_instantiate(dentry, inode);
unlock_new_inode(inode);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
out:
handle_failed_inode(inode);
return err;
}
static int f2fs_link(struct dentry *old_dentry, struct inode *dir,
struct dentry *dentry)
{
struct inode *inode = d_inode(old_dentry);
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err;
if (f2fs_encrypted_inode(dir) &&
!f2fs_is_child_context_consistent_with_parent(dir, inode))
return -EPERM;
f2fs_balance_fs(sbi, true);
inode->i_ctime = CURRENT_TIME;
ihold(inode);
set_inode_flag(F2FS_I(inode), FI_INC_LINK);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
if (err)
goto out;
f2fs_unlock_op(sbi);
d_instantiate(dentry, inode);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
out:
clear_inode_flag(F2FS_I(inode), FI_INC_LINK);
iput(inode);
f2fs_unlock_op(sbi);
return err;
}
struct dentry *f2fs_get_parent(struct dentry *child)
{
struct qstr dotdot = QSTR_INIT("..", 2);
unsigned long ino = f2fs_inode_by_name(d_inode(child), &dotdot);
if (!ino)
return ERR_PTR(-ENOENT);
return d_obtain_alias(f2fs_iget(d_inode(child)->i_sb, ino));
}
static int __recover_dot_dentries(struct inode *dir, nid_t pino)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct qstr dot = QSTR_INIT(".", 1);
struct qstr dotdot = QSTR_INIT("..", 2);
struct f2fs_dir_entry *de;
struct page *page;
int err = 0;
if (f2fs_readonly(sbi->sb)) {
f2fs_msg(sbi->sb, KERN_INFO,
"skip recovering inline_dots inode (ino:%lu, pino:%u) "
"in readonly mountpoint", dir->i_ino, pino);
return 0;
}
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
de = f2fs_find_entry(dir, &dot, &page);
if (de) {
f2fs_dentry_kunmap(dir, page);
f2fs_put_page(page, 0);
} else {
err = __f2fs_add_link(dir, &dot, NULL, dir->i_ino, S_IFDIR);
if (err)
goto out;
}
de = f2fs_find_entry(dir, &dotdot, &page);
if (de) {
f2fs_dentry_kunmap(dir, page);
f2fs_put_page(page, 0);
} else {
err = __f2fs_add_link(dir, &dotdot, NULL, pino, S_IFDIR);
}
out:
if (!err) {
clear_inode_flag(F2FS_I(dir), FI_INLINE_DOTS);
mark_inode_dirty(dir);
}
f2fs_unlock_op(sbi);
return err;
}
static struct dentry *f2fs_lookup(struct inode *dir, struct dentry *dentry,
unsigned int flags)
{
struct inode *inode = NULL;
struct f2fs_dir_entry *de;
struct page *page;
nid_t ino;
int err = 0;
if (dentry->d_name.len > F2FS_NAME_LEN)
return ERR_PTR(-ENAMETOOLONG);
de = f2fs_find_entry(dir, &dentry->d_name, &page);
if (!de)
return d_splice_alias(inode, dentry);
ino = le32_to_cpu(de->ino);
f2fs_dentry_kunmap(dir, page);
f2fs_put_page(page, 0);
inode = f2fs_iget(dir->i_sb, ino);
if (IS_ERR(inode))
return ERR_CAST(inode);
if (f2fs_has_inline_dots(inode)) {
err = __recover_dot_dentries(inode, dir->i_ino);
if (err)
goto err_out;
}
return d_splice_alias(inode, dentry);
err_out:
iget_failed(inode);
return ERR_PTR(err);
}
static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode = d_inode(dentry);
struct f2fs_dir_entry *de;
struct page *page;
int err = -ENOENT;
trace_f2fs_unlink_enter(dir, dentry);
de = f2fs_find_entry(dir, &dentry->d_name, &page);
if (!de)
goto fail;
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
err = acquire_orphan_inode(sbi);
if (err) {
f2fs_unlock_op(sbi);
f2fs_dentry_kunmap(dir, page);
f2fs_put_page(page, 0);
goto fail;
}
f2fs_delete_entry(de, page, dir, inode);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_unlock_op(sbi);
/* In order to evict this inode, we set it dirty */
mark_inode_dirty(inode);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
fail:
trace_f2fs_unlink_exit(inode, err);
return err;
}
static const char *f2fs_get_link(struct dentry *dentry,
struct inode *inode,
struct delayed_call *done)
{
const char *link = page_get_link(dentry, inode, done);
if (!IS_ERR(link) && !*link) {
/* this is broken symlink case */
do_delayed_call(done);
clear_delayed_call(done);
link = ERR_PTR(-ENOENT);
}
return link;
}
static int f2fs_symlink(struct inode *dir, struct dentry *dentry,
const char *symname)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode;
size_t len = strlen(symname);
struct f2fs_str disk_link = FSTR_INIT((char *)symname, len + 1);
struct f2fs_encrypted_symlink_data *sd = NULL;
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err;
if (f2fs_encrypted_inode(dir)) {
err = f2fs_get_encryption_info(dir);
if (err)
return err;
if (!f2fs_encrypted_inode(dir))
return -EPERM;
disk_link.len = (f2fs_fname_encrypted_size(dir, len) +
sizeof(struct f2fs_encrypted_symlink_data));
}
if (disk_link.len > dir->i_sb->s_blocksize)
return -ENAMETOOLONG;
inode = f2fs_new_inode(dir, S_IFLNK | S_IRWXUGO);
if (IS_ERR(inode))
return PTR_ERR(inode);
if (f2fs_encrypted_inode(inode))
inode->i_op = &f2fs_encrypted_symlink_inode_operations;
else
inode->i_op = &f2fs_symlink_inode_operations;
inode_nohighmem(inode);
inode->i_mapping->a_ops = &f2fs_dblock_aops;
f2fs_balance_fs(sbi, true);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
if (err)
goto out;
f2fs_unlock_op(sbi);
alloc_nid_done(sbi, inode->i_ino);
if (f2fs_encrypted_inode(inode)) {
struct qstr istr = QSTR_INIT(symname, len);
struct f2fs_str ostr;
sd = kzalloc(disk_link.len, GFP_NOFS);
if (!sd) {
err = -ENOMEM;
goto err_out;
}
err = f2fs_get_encryption_info(inode);
if (err)
goto err_out;
if (!f2fs_encrypted_inode(inode)) {
err = -EPERM;
goto err_out;
}
ostr.name = sd->encrypted_path;
ostr.len = disk_link.len;
err = f2fs_fname_usr_to_disk(inode, &istr, &ostr);
if (err < 0)
goto err_out;
sd->len = cpu_to_le16(ostr.len);
disk_link.name = (char *)sd;
}
err = page_symlink(inode, disk_link.name, disk_link.len);
err_out:
d_instantiate(dentry, inode);
unlock_new_inode(inode);
/*
* Let's flush symlink data in order to avoid broken symlink as much as
* possible. Nevertheless, fsyncing is the best way, but there is no
* way to get a file descriptor in order to flush that.
*
* Note that, it needs to do dir->fsync to make this recoverable.
* If the symlink path is stored into inline_data, there is no
* performance regression.
*/
if (!err) {
filemap_write_and_wait_range(inode->i_mapping, 0,
disk_link.len - 1);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
} else {
f2fs_unlink(dir, dentry);
}
kfree(sd);
return err;
out:
handle_failed_inode(inode);
return err;
}
static int f2fs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode;
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err;
inode = f2fs_new_inode(dir, S_IFDIR | mode);
if (IS_ERR(inode))
return PTR_ERR(inode);
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_F2FS_HIGH_ZERO);
f2fs_balance_fs(sbi, true);
set_inode_flag(F2FS_I(inode), FI_INC_LINK);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
if (err)
goto out_fail;
f2fs_unlock_op(sbi);
alloc_nid_done(sbi, inode->i_ino);
d_instantiate(dentry, inode);
unlock_new_inode(inode);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
out_fail:
clear_inode_flag(F2FS_I(inode), FI_INC_LINK);
handle_failed_inode(inode);
return err;
}
static int f2fs_rmdir(struct inode *dir, struct dentry *dentry)
{
struct inode *inode = d_inode(dentry);
if (f2fs_empty_dir(inode))
return f2fs_unlink(dir, dentry);
return -ENOTEMPTY;
}
static int f2fs_mknod(struct inode *dir, struct dentry *dentry,
umode_t mode, dev_t rdev)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode;
int err = 0;
inode = f2fs_new_inode(dir, mode);
if (IS_ERR(inode))
return PTR_ERR(inode);
init_special_inode(inode, inode->i_mode, rdev);
inode->i_op = &f2fs_special_inode_operations;
f2fs_balance_fs(sbi, true);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_lock_op(sbi);
err = f2fs_add_link(dentry, inode);
if (err)
goto out;
f2fs_unlock_op(sbi);
alloc_nid_done(sbi, inode->i_ino);
d_instantiate(dentry, inode);
unlock_new_inode(inode);
if (IS_DIRSYNC(dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
out:
handle_failed_inode(inode);
return err;
}
static int __f2fs_tmpfile(struct inode *dir, struct dentry *dentry,
umode_t mode, struct inode **whiteout)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
struct inode *inode;
int err;
inode = f2fs_new_inode(dir, mode);
if (IS_ERR(inode))
return PTR_ERR(inode);
if (whiteout) {
init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
inode->i_op = &f2fs_special_inode_operations;
} else {
inode->i_op = &f2fs_file_inode_operations;
inode->i_fop = &f2fs_file_operations;
inode->i_mapping->a_ops = &f2fs_dblock_aops;
}
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
err = acquire_orphan_inode(sbi);
if (err)
goto out;
err = f2fs_do_tmpfile(inode, dir);
if (err)
goto release_out;
/*
* add this non-linked tmpfile to orphan list, in this way we could
* remove all unused data of tmpfile after abnormal power-off.
*/
add_orphan_inode(sbi, inode->i_ino);
f2fs_unlock_op(sbi);
alloc_nid_done(sbi, inode->i_ino);
if (whiteout) {
inode_dec_link_count(inode);
*whiteout = inode;
} else {
d_tmpfile(dentry, inode);
}
unlock_new_inode(inode);
return 0;
release_out:
release_orphan_inode(sbi);
out:
handle_failed_inode(inode);
return err;
}
static int f2fs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
{
if (f2fs_encrypted_inode(dir)) {
int err = f2fs_get_encryption_info(dir);
if (err)
return err;
}
return __f2fs_tmpfile(dir, dentry, mode, NULL);
}
static int f2fs_create_whiteout(struct inode *dir, struct inode **whiteout)
{
return __f2fs_tmpfile(dir, NULL, S_IFCHR | WHITEOUT_MODE, whiteout);
}
static int f2fs_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry,
unsigned int flags)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(old_dir);
struct inode *old_inode = d_inode(old_dentry);
struct inode *new_inode = d_inode(new_dentry);
struct inode *whiteout = NULL;
struct page *old_dir_page;
struct page *old_page, *new_page = NULL;
struct f2fs_dir_entry *old_dir_entry = NULL;
struct f2fs_dir_entry *old_entry;
struct f2fs_dir_entry *new_entry;
f2fs: fix to delete old dirent in converted inline directory in ->rename When doing test with fstests/generic/068 in inline_dentry enabled f2fs, following oops dmesg will be reported: ------------[ cut here ]------------ WARNING: CPU: 5 PID: 11841 at fs/inode.c:273 drop_nlink+0x49/0x50() Modules linked in: f2fs(O) ip6table_filter ip6_tables ebtable_nat ebtables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state CPU: 5 PID: 11841 Comm: fsstress Tainted: G O 4.5.0-rc1 #45 Hardware name: Hewlett-Packard HP Z220 CMT Workstation/1790, BIOS K51 v01.61 05/16/2013 0000000000000111 ffff88009cdf7ae8 ffffffff813e5944 0000000000002e41 0000000000000000 0000000000000111 0000000000000000 ffff88009cdf7b28 ffffffff8106a587 ffff88009cdf7b58 ffff8804078fe180 ffff880374a64e00 Call Trace: [<ffffffff813e5944>] dump_stack+0x48/0x64 [<ffffffff8106a587>] warn_slowpath_common+0x97/0xe0 [<ffffffff8106a5ea>] warn_slowpath_null+0x1a/0x20 [<ffffffff81231039>] drop_nlink+0x49/0x50 [<ffffffffa07b95b4>] f2fs_rename2+0xe04/0x10c0 [f2fs] [<ffffffff81231ff1>] ? lock_two_nondirectories+0x81/0x90 [<ffffffff813f454d>] ? lockref_get+0x1d/0x30 [<ffffffff81220f70>] vfs_rename+0x2e0/0x640 [<ffffffff8121f9db>] ? lookup_dcache+0x3b/0xd0 [<ffffffff810b8e41>] ? update_fast_ctr+0x21/0x40 [<ffffffff8134ff12>] ? security_path_rename+0xa2/0xd0 [<ffffffff81224af6>] SYSC_renameat2+0x4b6/0x540 [<ffffffff810ba8ed>] ? trace_hardirqs_off+0xd/0x10 [<ffffffff810022ba>] ? exit_to_usermode_loop+0x7a/0xd0 [<ffffffff817e0ade>] ? int_ret_from_sys_call+0x52/0x9f [<ffffffff810bdc90>] ? trace_hardirqs_on_caller+0x100/0x1c0 [<ffffffff81224b8e>] SyS_renameat2+0xe/0x10 [<ffffffff8121f08e>] SyS_rename+0x1e/0x20 [<ffffffff817e0957>] entry_SYSCALL_64_fastpath+0x12/0x6f ---[ end trace 2b31e17995404e42 ]--- This is because: in the same inline directory, when we renaming one file from source name to target name which is not existed, once space of inline dentry is not enough, inline conversion will be triggered, after that all data in inline dentry will be moved to normal dentry page. After attaching the new entry in coverted dentry page, still we try to remove old entry in original inline dentry, since old entry has been moved, so it obviously doesn't make any effect, result in remaining old entry in converted dentry page. Now, we have two valid dentries pointed to the same inode which has nlink value of 1, deleting them both, above warning appears. This issue can be reproduced easily as below steps: 1. mount f2fs with inline_dentry option 2. mkdir dir 3. touch 180 files named [001-180] in dir 4. rename dir/180 dir/181 5. rm dir/180 dir/181 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2016-02-17 16:47:05 +08:00
bool is_old_inline = f2fs_has_inline_dentry(old_dir);
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
int err = -ENOENT;
if ((old_dir != new_dir) && f2fs_encrypted_inode(new_dir) &&
!f2fs_is_child_context_consistent_with_parent(new_dir,
old_inode)) {
err = -EPERM;
goto out;
}
old_entry = f2fs_find_entry(old_dir, &old_dentry->d_name, &old_page);
if (!old_entry)
goto out;
if (S_ISDIR(old_inode->i_mode)) {
err = -EIO;
old_dir_entry = f2fs_parent_dir(old_inode, &old_dir_page);
if (!old_dir_entry)
goto out_old;
}
if (flags & RENAME_WHITEOUT) {
err = f2fs_create_whiteout(old_dir, &whiteout);
if (err)
goto out_dir;
}
if (new_inode) {
err = -ENOTEMPTY;
if (old_dir_entry && !f2fs_empty_dir(new_inode))
goto out_whiteout;
err = -ENOENT;
new_entry = f2fs_find_entry(new_dir, &new_dentry->d_name,
&new_page);
if (!new_entry)
goto out_whiteout;
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
err = acquire_orphan_inode(sbi);
if (err)
goto put_out_dir;
err = update_dent_inode(old_inode, new_inode,
&new_dentry->d_name);
if (err) {
release_orphan_inode(sbi);
goto put_out_dir;
}
f2fs_set_link(new_dir, new_entry, new_page, old_inode);
new_inode->i_ctime = CURRENT_TIME;
down_write(&F2FS_I(new_inode)->i_sem);
if (old_dir_entry)
drop_nlink(new_inode);
drop_nlink(new_inode);
up_write(&F2FS_I(new_inode)->i_sem);
mark_inode_dirty(new_inode);
if (!new_inode->i_nlink)
add_orphan_inode(sbi, new_inode->i_ino);
else
release_orphan_inode(sbi);
update_inode_page(old_inode);
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>
2012-11-22 15:21:29 +08:00
update_inode_page(new_inode);
} else {
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
err = f2fs_add_link(new_dentry, old_inode);
if (err) {
f2fs_unlock_op(sbi);
goto out_whiteout;
}
if (old_dir_entry) {
inc_nlink(new_dir);
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>
2012-11-22 15:21:29 +08:00
update_inode_page(new_dir);
}
f2fs: fix to delete old dirent in converted inline directory in ->rename When doing test with fstests/generic/068 in inline_dentry enabled f2fs, following oops dmesg will be reported: ------------[ cut here ]------------ WARNING: CPU: 5 PID: 11841 at fs/inode.c:273 drop_nlink+0x49/0x50() Modules linked in: f2fs(O) ip6table_filter ip6_tables ebtable_nat ebtables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state CPU: 5 PID: 11841 Comm: fsstress Tainted: G O 4.5.0-rc1 #45 Hardware name: Hewlett-Packard HP Z220 CMT Workstation/1790, BIOS K51 v01.61 05/16/2013 0000000000000111 ffff88009cdf7ae8 ffffffff813e5944 0000000000002e41 0000000000000000 0000000000000111 0000000000000000 ffff88009cdf7b28 ffffffff8106a587 ffff88009cdf7b58 ffff8804078fe180 ffff880374a64e00 Call Trace: [<ffffffff813e5944>] dump_stack+0x48/0x64 [<ffffffff8106a587>] warn_slowpath_common+0x97/0xe0 [<ffffffff8106a5ea>] warn_slowpath_null+0x1a/0x20 [<ffffffff81231039>] drop_nlink+0x49/0x50 [<ffffffffa07b95b4>] f2fs_rename2+0xe04/0x10c0 [f2fs] [<ffffffff81231ff1>] ? lock_two_nondirectories+0x81/0x90 [<ffffffff813f454d>] ? lockref_get+0x1d/0x30 [<ffffffff81220f70>] vfs_rename+0x2e0/0x640 [<ffffffff8121f9db>] ? lookup_dcache+0x3b/0xd0 [<ffffffff810b8e41>] ? update_fast_ctr+0x21/0x40 [<ffffffff8134ff12>] ? security_path_rename+0xa2/0xd0 [<ffffffff81224af6>] SYSC_renameat2+0x4b6/0x540 [<ffffffff810ba8ed>] ? trace_hardirqs_off+0xd/0x10 [<ffffffff810022ba>] ? exit_to_usermode_loop+0x7a/0xd0 [<ffffffff817e0ade>] ? int_ret_from_sys_call+0x52/0x9f [<ffffffff810bdc90>] ? trace_hardirqs_on_caller+0x100/0x1c0 [<ffffffff81224b8e>] SyS_renameat2+0xe/0x10 [<ffffffff8121f08e>] SyS_rename+0x1e/0x20 [<ffffffff817e0957>] entry_SYSCALL_64_fastpath+0x12/0x6f ---[ end trace 2b31e17995404e42 ]--- This is because: in the same inline directory, when we renaming one file from source name to target name which is not existed, once space of inline dentry is not enough, inline conversion will be triggered, after that all data in inline dentry will be moved to normal dentry page. After attaching the new entry in coverted dentry page, still we try to remove old entry in original inline dentry, since old entry has been moved, so it obviously doesn't make any effect, result in remaining old entry in converted dentry page. Now, we have two valid dentries pointed to the same inode which has nlink value of 1, deleting them both, above warning appears. This issue can be reproduced easily as below steps: 1. mount f2fs with inline_dentry option 2. mkdir dir 3. touch 180 files named [001-180] in dir 4. rename dir/180 dir/181 5. rm dir/180 dir/181 Signed-off-by: Chao Yu <chao2.yu@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2016-02-17 16:47:05 +08:00
/*
* old entry and new entry can locate in the same inline
* dentry in inode, when attaching new entry in inline dentry,
* it could force inline dentry conversion, after that,
* old_entry and old_page will point to wrong address, in
* order to avoid this, let's do the check and update here.
*/
if (is_old_inline && !f2fs_has_inline_dentry(old_dir)) {
f2fs_put_page(old_page, 0);
old_page = NULL;
old_entry = f2fs_find_entry(old_dir,
&old_dentry->d_name, &old_page);
if (!old_entry) {
err = -EIO;
f2fs_unlock_op(sbi);
goto out_whiteout;
}
}
}
down_write(&F2FS_I(old_inode)->i_sem);
file_lost_pino(old_inode);
if (new_inode && file_enc_name(new_inode))
file_set_enc_name(old_inode);
up_write(&F2FS_I(old_inode)->i_sem);
old_inode->i_ctime = CURRENT_TIME;
mark_inode_dirty(old_inode);
f2fs_delete_entry(old_entry, old_page, old_dir, NULL);
if (whiteout) {
whiteout->i_state |= I_LINKABLE;
set_inode_flag(F2FS_I(whiteout), FI_INC_LINK);
err = f2fs_add_link(old_dentry, whiteout);
if (err)
goto put_out_dir;
whiteout->i_state &= ~I_LINKABLE;
iput(whiteout);
}
if (old_dir_entry) {
if (old_dir != new_dir && !whiteout) {
f2fs_set_link(old_inode, old_dir_entry,
old_dir_page, new_dir);
update_inode_page(old_inode);
} else {
f2fs_dentry_kunmap(old_inode, old_dir_page);
f2fs_put_page(old_dir_page, 0);
}
drop_nlink(old_dir);
mark_inode_dirty(old_dir);
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>
2012-11-22 15:21:29 +08:00
update_inode_page(old_dir);
}
f2fs: use rw_sem instead of fs_lock(locks mutex) The fs_locks is used to block other ops(ex, recovery) when doing checkpoint. And each other operate routine(besides checkpoint) needs to acquire a fs_lock, there is a terrible problem here, if these are too many concurrency threads acquiring fs_lock, so that they will block each other and may lead to some performance problem, but this is not the phenomenon we want to see. Though there are some optimization patches introduced to enhance the usage of fs_lock, but the thorough solution is using a *rw_sem* to replace the fs_lock. Checkpoint routine takes write_sem, and other ops take read_sem, so that we can block other ops(ex, recovery) when doing checkpoint, and other ops will not disturb each other, this can avoid the problem described above completely. Because of the weakness of rw_sem, the above change may introduce a potential problem that the checkpoint thread might get starved if other threads are intensively locking the read semaphore for I/O.(Pointed out by Xu Jin) In order to avoid this, a wait_list is introduced, the appending read semaphore ops will be dropped into the wait_list if checkpoint thread is waiting for write semaphore, and will be waked up when checkpoint thread gives up write semaphore. Thanks to Kim's previous review and test, and will be very glad to see other guys' performance tests about this patch. V2: -fix the potential starvation problem. -use more suitable func name suggested by Xu Jin. Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com> [Jaegeuk Kim: adjust minor coding standard] Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-09-27 18:08:30 +08:00
f2fs_unlock_op(sbi);
if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
put_out_dir:
f2fs_unlock_op(sbi);
if (new_page) {
f2fs_dentry_kunmap(new_dir, new_page);
f2fs_put_page(new_page, 0);
}
out_whiteout:
if (whiteout)
iput(whiteout);
out_dir:
if (old_dir_entry) {
f2fs_dentry_kunmap(old_inode, old_dir_page);
f2fs_put_page(old_dir_page, 0);
}
out_old:
f2fs_dentry_kunmap(old_dir, old_page);
f2fs_put_page(old_page, 0);
out:
return err;
}
static int f2fs_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(old_dir);
struct inode *old_inode = d_inode(old_dentry);
struct inode *new_inode = d_inode(new_dentry);
struct page *old_dir_page, *new_dir_page;
struct page *old_page, *new_page;
struct f2fs_dir_entry *old_dir_entry = NULL, *new_dir_entry = NULL;
struct f2fs_dir_entry *old_entry, *new_entry;
int old_nlink = 0, new_nlink = 0;
int err = -ENOENT;
if ((f2fs_encrypted_inode(old_dir) || f2fs_encrypted_inode(new_dir)) &&
(old_dir != new_dir) &&
(!f2fs_is_child_context_consistent_with_parent(new_dir,
old_inode) ||
!f2fs_is_child_context_consistent_with_parent(old_dir,
new_inode)))
return -EPERM;
old_entry = f2fs_find_entry(old_dir, &old_dentry->d_name, &old_page);
if (!old_entry)
goto out;
new_entry = f2fs_find_entry(new_dir, &new_dentry->d_name, &new_page);
if (!new_entry)
goto out_old;
/* prepare for updating ".." directory entry info later */
if (old_dir != new_dir) {
if (S_ISDIR(old_inode->i_mode)) {
err = -EIO;
old_dir_entry = f2fs_parent_dir(old_inode,
&old_dir_page);
if (!old_dir_entry)
goto out_new;
}
if (S_ISDIR(new_inode->i_mode)) {
err = -EIO;
new_dir_entry = f2fs_parent_dir(new_inode,
&new_dir_page);
if (!new_dir_entry)
goto out_old_dir;
}
}
/*
* If cross rename between file and directory those are not
* in the same directory, we will inc nlink of file's parent
* later, so we should check upper boundary of its nlink.
*/
if ((!old_dir_entry || !new_dir_entry) &&
old_dir_entry != new_dir_entry) {
old_nlink = old_dir_entry ? -1 : 1;
new_nlink = -old_nlink;
err = -EMLINK;
if ((old_nlink > 0 && old_inode->i_nlink >= F2FS_LINK_MAX) ||
(new_nlink > 0 && new_inode->i_nlink >= F2FS_LINK_MAX))
goto out_new_dir;
}
f2fs_balance_fs(sbi, true);
f2fs_lock_op(sbi);
err = update_dent_inode(old_inode, new_inode, &new_dentry->d_name);
if (err)
goto out_unlock;
if (file_enc_name(new_inode))
file_set_enc_name(old_inode);
err = update_dent_inode(new_inode, old_inode, &old_dentry->d_name);
if (err)
goto out_undo;
if (file_enc_name(old_inode))
file_set_enc_name(new_inode);
/* update ".." directory entry info of old dentry */
if (old_dir_entry)
f2fs_set_link(old_inode, old_dir_entry, old_dir_page, new_dir);
/* update ".." directory entry info of new dentry */
if (new_dir_entry)
f2fs_set_link(new_inode, new_dir_entry, new_dir_page, old_dir);
/* update directory entry info of old dir inode */
f2fs_set_link(old_dir, old_entry, old_page, new_inode);
down_write(&F2FS_I(old_inode)->i_sem);
file_lost_pino(old_inode);
up_write(&F2FS_I(old_inode)->i_sem);
update_inode_page(old_inode);
old_dir->i_ctime = CURRENT_TIME;
if (old_nlink) {
down_write(&F2FS_I(old_dir)->i_sem);
if (old_nlink < 0)
drop_nlink(old_dir);
else
inc_nlink(old_dir);
up_write(&F2FS_I(old_dir)->i_sem);
}
mark_inode_dirty(old_dir);
update_inode_page(old_dir);
/* update directory entry info of new dir inode */
f2fs_set_link(new_dir, new_entry, new_page, old_inode);
down_write(&F2FS_I(new_inode)->i_sem);
file_lost_pino(new_inode);
up_write(&F2FS_I(new_inode)->i_sem);
update_inode_page(new_inode);
new_dir->i_ctime = CURRENT_TIME;
if (new_nlink) {
down_write(&F2FS_I(new_dir)->i_sem);
if (new_nlink < 0)
drop_nlink(new_dir);
else
inc_nlink(new_dir);
up_write(&F2FS_I(new_dir)->i_sem);
}
mark_inode_dirty(new_dir);
update_inode_page(new_dir);
f2fs_unlock_op(sbi);
if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
f2fs_sync_fs(sbi->sb, 1);
return 0;
out_undo:
/*
* Still we may fail to recover name info of f2fs_inode here
* Drop it, once its name is set as encrypted
*/
update_dent_inode(old_inode, old_inode, &old_dentry->d_name);
out_unlock:
f2fs_unlock_op(sbi);
out_new_dir:
if (new_dir_entry) {
f2fs_dentry_kunmap(new_inode, new_dir_page);
f2fs_put_page(new_dir_page, 0);
}
out_old_dir:
if (old_dir_entry) {
f2fs_dentry_kunmap(old_inode, old_dir_page);
f2fs_put_page(old_dir_page, 0);
}
out_new:
f2fs_dentry_kunmap(new_dir, new_page);
f2fs_put_page(new_page, 0);
out_old:
f2fs_dentry_kunmap(old_dir, old_page);
f2fs_put_page(old_page, 0);
out:
return err;
}
static int f2fs_rename2(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry,
unsigned int flags)
{
if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
return -EINVAL;
if (flags & RENAME_EXCHANGE) {
return f2fs_cross_rename(old_dir, old_dentry,
new_dir, new_dentry);
}
/*
* VFS has already handled the new dentry existence case,
* here, we just deal with "RENAME_NOREPLACE" as regular rename.
*/
return f2fs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
}
#ifdef CONFIG_F2FS_FS_ENCRYPTION
static const char *f2fs_encrypted_get_link(struct dentry *dentry,
struct inode *inode,
struct delayed_call *done)
{
struct page *cpage = NULL;
char *caddr, *paddr = NULL;
struct f2fs_str cstr = FSTR_INIT(NULL, 0);
struct f2fs_str pstr = FSTR_INIT(NULL, 0);
struct f2fs_encrypted_symlink_data *sd;
loff_t size = min_t(loff_t, i_size_read(inode), PAGE_SIZE - 1);
u32 max_size = inode->i_sb->s_blocksize;
int res;
if (!dentry)
return ERR_PTR(-ECHILD);
res = f2fs_get_encryption_info(inode);
if (res)
return ERR_PTR(res);
cpage = read_mapping_page(inode->i_mapping, 0, NULL);
if (IS_ERR(cpage))
Merge tag 'for-f2fs-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs Pull f2fs updates from Jaegeuk Kim: "New features: - per-file encryption (e.g., ext4) - FALLOC_FL_ZERO_RANGE - FALLOC_FL_COLLAPSE_RANGE - RENAME_WHITEOUT Major enhancement/fixes: - recovery broken superblocks - enhance f2fs_trim_fs with a discard_map - fix a race condition on dentry block allocation - fix a deadlock during summary operation - fix a missing fiemap result .. and many minor bug fixes and clean-ups were done" * tag 'for-f2fs-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs: (83 commits) f2fs: do not trim preallocated blocks when truncating after i_size f2fs crypto: add alloc_bounce_page f2fs crypto: fix to handle errors likewise ext4 f2fs: drop the volatile_write flag only f2fs: skip committing valid superblock f2fs: setting discard option in parse_options() f2fs: fix to return exact trimmed size f2fs: support FALLOC_FL_INSERT_RANGE f2fs: hide common code in f2fs_replace_block f2fs: disable the discard option when device doesn't support f2fs crypto: remove alloc_page for bounce_page f2fs: fix a deadlock for summary page lock vs. sentry_lock f2fs crypto: clean up error handling in f2fs_fname_setup_filename f2fs crypto: avoid f2fs_inherit_context for symlink f2fs crypto: do not set encryption policy for non-directory by ioctl f2fs crypto: allow setting encryption policy once f2fs crypto: check context consistent for rename2 f2fs: avoid duplicated code by reusing f2fs_read_end_io f2fs crypto: use per-inode tfm structure f2fs: recovering broken superblock during mount ...
2015-06-25 11:38:29 +08:00
return ERR_CAST(cpage);
caddr = page_address(cpage);
caddr[size] = 0;
/* Symlink is encrypted */
sd = (struct f2fs_encrypted_symlink_data *)caddr;
cstr.name = sd->encrypted_path;
cstr.len = le16_to_cpu(sd->len);
/* this is broken symlink case */
if (unlikely(cstr.len == 0)) {
res = -ENOENT;
goto errout;
}
/* this is broken symlink case */
if (unlikely(cstr.name[0] == 0)) {
res = -ENOENT;
goto errout;
}
if ((cstr.len + sizeof(struct f2fs_encrypted_symlink_data) - 1) >
max_size) {
/* Symlink data on the disk is corrupted */
res = -EIO;
goto errout;
}
res = f2fs_fname_crypto_alloc_buffer(inode, cstr.len, &pstr);
if (res)
goto errout;
res = f2fs_fname_disk_to_usr(inode, NULL, &cstr, &pstr);
if (res < 0)
goto errout;
paddr = pstr.name;
/* Null-terminate the name */
paddr[res] = '\0';
page_cache_release(cpage);
set_delayed_call(done, kfree_link, paddr);
return paddr;
errout:
f2fs_fname_crypto_free_buffer(&pstr);
page_cache_release(cpage);
return ERR_PTR(res);
}
const struct inode_operations f2fs_encrypted_symlink_inode_operations = {
.readlink = generic_readlink,
.get_link = f2fs_encrypted_get_link,
.getattr = f2fs_getattr,
.setattr = f2fs_setattr,
#ifdef CONFIG_F2FS_FS_XATTR
.setxattr = generic_setxattr,
.getxattr = generic_getxattr,
.listxattr = f2fs_listxattr,
.removexattr = generic_removexattr,
#endif
};
#endif
const struct inode_operations f2fs_dir_inode_operations = {
.create = f2fs_create,
.lookup = f2fs_lookup,
.link = f2fs_link,
.unlink = f2fs_unlink,
.symlink = f2fs_symlink,
.mkdir = f2fs_mkdir,
.rmdir = f2fs_rmdir,
.mknod = f2fs_mknod,
.rename2 = f2fs_rename2,
.tmpfile = f2fs_tmpfile,
.getattr = f2fs_getattr,
.setattr = f2fs_setattr,
.get_acl = f2fs_get_acl,
.set_acl = f2fs_set_acl,
#ifdef CONFIG_F2FS_FS_XATTR
.setxattr = generic_setxattr,
.getxattr = generic_getxattr,
.listxattr = f2fs_listxattr,
.removexattr = generic_removexattr,
#endif
};
const struct inode_operations f2fs_symlink_inode_operations = {
.readlink = generic_readlink,
.get_link = f2fs_get_link,
.getattr = f2fs_getattr,
.setattr = f2fs_setattr,
#ifdef CONFIG_F2FS_FS_XATTR
.setxattr = generic_setxattr,
.getxattr = generic_getxattr,
.listxattr = f2fs_listxattr,
.removexattr = generic_removexattr,
#endif
};
const struct inode_operations f2fs_special_inode_operations = {
.getattr = f2fs_getattr,
.setattr = f2fs_setattr,
.get_acl = f2fs_get_acl,
.set_acl = f2fs_set_acl,
#ifdef CONFIG_F2FS_FS_XATTR
.setxattr = generic_setxattr,
.getxattr = generic_getxattr,
.listxattr = f2fs_listxattr,
.removexattr = generic_removexattr,
#endif
};