ufs: remove the BKL
This introduces a new per-superblock mutex in UFS to replace
the big kernel lock. I have been careful to avoid nested
calls to lock_ufs and to get the lock order right with
respect to other mutexes, in particular lock_super.
I did not make any attempt to prove that the big kernel
lock is not needed in a particular place in the code,
which is very possible.
The mutex has a significant performance impact, so it is only
used on SMP or PREEMPT configurations.
As Nick Piggin noticed, any allocation inside of the lock
may end up deadlocking when we get to ufs_getfrag_block
in the reclaim task, so we now use GFP_NOFS.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Nick Bowler <nbowler@elliptictech.com>
Cc: Evgeniy Dushistov <dushistov@mail.ru>
Cc: Nick Piggin <npiggin@gmail.com>
diff --git a/fs/ufs/super.c b/fs/ufs/super.c
index 2c61ac5..7693d62 100644
--- a/fs/ufs/super.c
+++ b/fs/ufs/super.c
@@ -84,7 +84,6 @@
#include <linux/blkdev.h>
#include <linux/init.h>
#include <linux/parser.h>
-#include <linux/smp_lock.h>
#include <linux/buffer_head.h>
#include <linux/vfs.h>
#include <linux/log2.h>
@@ -96,6 +95,26 @@
#include "swab.h"
#include "util.h"
+void lock_ufs(struct super_block *sb)
+{
+#if defined(CONFIG_SMP) || defined (CONFIG_PREEMPT)
+ struct ufs_sb_info *sbi = UFS_SB(sb);
+
+ mutex_lock(&sbi->mutex);
+ sbi->mutex_owner = current;
+#endif
+}
+
+void unlock_ufs(struct super_block *sb)
+{
+#if defined(CONFIG_SMP) || defined (CONFIG_PREEMPT)
+ struct ufs_sb_info *sbi = UFS_SB(sb);
+
+ sbi->mutex_owner = NULL;
+ mutex_unlock(&sbi->mutex);
+#endif
+}
+
static struct inode *ufs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
{
struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
@@ -313,7 +332,6 @@
struct ufs_super_block_first * usb1;
va_list args;
- lock_kernel();
uspi = UFS_SB(sb)->s_uspi;
usb1 = ubh_get_usb_first(uspi);
@@ -521,7 +539,7 @@
*/
size = uspi->s_cssize;
blks = (size + uspi->s_fsize - 1) >> uspi->s_fshift;
- base = space = kmalloc(size, GFP_KERNEL);
+ base = space = kmalloc(size, GFP_NOFS);
if (!base)
goto failed;
sbi->s_csp = (struct ufs_csum *)space;
@@ -546,7 +564,7 @@
* Read cylinder group (we read only first fragment from block
* at this time) and prepare internal data structures for cg caching.
*/
- if (!(sbi->s_ucg = kmalloc (sizeof(struct buffer_head *) * uspi->s_ncg, GFP_KERNEL)))
+ if (!(sbi->s_ucg = kmalloc (sizeof(struct buffer_head *) * uspi->s_ncg, GFP_NOFS)))
goto failed;
for (i = 0; i < uspi->s_ncg; i++)
sbi->s_ucg[i] = NULL;
@@ -564,7 +582,7 @@
ufs_print_cylinder_stuff(sb, (struct ufs_cylinder_group *) sbi->s_ucg[i]->b_data);
}
for (i = 0; i < UFS_MAX_GROUP_LOADED; i++) {
- if (!(sbi->s_ucpi[i] = kmalloc (sizeof(struct ufs_cg_private_info), GFP_KERNEL)))
+ if (!(sbi->s_ucpi[i] = kmalloc (sizeof(struct ufs_cg_private_info), GFP_NOFS)))
goto failed;
sbi->s_cgno[i] = UFS_CGNO_EMPTY;
}
@@ -646,8 +664,6 @@
UFSD("ENTER\n");
- lock_kernel();
-
ufs_put_cstotal(sb);
size = uspi->s_cssize;
blks = (size + uspi->s_fsize - 1) >> uspi->s_fshift;
@@ -676,8 +692,6 @@
kfree (sbi->s_ucg);
kfree (base);
- unlock_kernel();
-
UFSD("EXIT\n");
}
@@ -696,8 +710,6 @@
unsigned maxsymlen;
int ret = -EINVAL;
- lock_kernel();
-
uspi = NULL;
ubh = NULL;
flags = 0;
@@ -718,6 +730,7 @@
goto failed;
}
#endif
+ mutex_init(&sbi->mutex);
/*
* Set default mount options
* Parse mount options
@@ -1165,7 +1178,6 @@
goto failed;
UFSD("EXIT\n");
- unlock_kernel();
return 0;
dalloc_failed:
@@ -1177,12 +1189,10 @@
kfree(sbi);
sb->s_fs_info = NULL;
UFSD("EXIT (FAILED)\n");
- unlock_kernel();
return ret;
failed_nomem:
UFSD("EXIT (NOMEM)\n");
- unlock_kernel();
return -ENOMEM;
}
@@ -1193,8 +1203,8 @@
struct ufs_super_block_third * usb3;
unsigned flags;
+ lock_ufs(sb);
lock_super(sb);
- lock_kernel();
UFSD("ENTER\n");
@@ -1213,8 +1223,8 @@
sb->s_dirt = 0;
UFSD("EXIT\n");
- unlock_kernel();
unlock_super(sb);
+ unlock_ufs(sb);
return 0;
}
@@ -1256,7 +1266,7 @@
unsigned new_mount_opt, ufstype;
unsigned flags;
- lock_kernel();
+ lock_ufs(sb);
lock_super(sb);
uspi = UFS_SB(sb)->s_uspi;
flags = UFS_SB(sb)->s_flags;
@@ -1272,7 +1282,7 @@
ufs_set_opt (new_mount_opt, ONERROR_LOCK);
if (!ufs_parse_options (data, &new_mount_opt)) {
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return -EINVAL;
}
if (!(new_mount_opt & UFS_MOUNT_UFSTYPE)) {
@@ -1280,14 +1290,14 @@
} else if ((new_mount_opt & UFS_MOUNT_UFSTYPE) != ufstype) {
printk("ufstype can't be changed during remount\n");
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return -EINVAL;
}
if ((*mount_flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY)) {
UFS_SB(sb)->s_mount_opt = new_mount_opt;
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return 0;
}
@@ -1313,7 +1323,7 @@
printk("ufs was compiled with read-only support, "
"can't be mounted as read-write\n");
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return -EINVAL;
#else
if (ufstype != UFS_MOUNT_UFSTYPE_SUN &&
@@ -1323,13 +1333,13 @@
ufstype != UFS_MOUNT_UFSTYPE_UFS2) {
printk("this ufstype is read-only supported\n");
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return -EINVAL;
}
if (!ufs_read_cylinder_structures(sb)) {
printk("failed during remounting\n");
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return -EPERM;
}
sb->s_flags &= ~MS_RDONLY;
@@ -1337,7 +1347,7 @@
}
UFS_SB(sb)->s_mount_opt = new_mount_opt;
unlock_super(sb);
- unlock_kernel();
+ unlock_ufs(sb);
return 0;
}
@@ -1371,7 +1381,7 @@
struct ufs_super_block_third *usb3;
u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
- lock_kernel();
+ lock_ufs(sb);
usb1 = ubh_get_usb_first(uspi);
usb2 = ubh_get_usb_second(uspi);
@@ -1395,7 +1405,7 @@
buf->f_fsid.val[0] = (u32)id;
buf->f_fsid.val[1] = (u32)(id >> 32);
- unlock_kernel();
+ unlock_ufs(sb);
return 0;
}
@@ -1405,7 +1415,7 @@
static struct inode *ufs_alloc_inode(struct super_block *sb)
{
struct ufs_inode_info *ei;
- ei = (struct ufs_inode_info *)kmem_cache_alloc(ufs_inode_cachep, GFP_KERNEL);
+ ei = (struct ufs_inode_info *)kmem_cache_alloc(ufs_inode_cachep, GFP_NOFS);
if (!ei)
return NULL;
ei->vfs_inode.i_version = 1;