Ling, Xiaofeng
2004-Jun-08 04:20 UTC
[Ocfs2-devel] [Patch] for bug 88, touch&chmod file cause hang.
When you touch a file, in the call sequence,
sys_utime->notify_change->ocfs_setattr, before notify_change,
there is a "down(&inode->i_sem)", so in ocfs_setattr, when down
again,
it causes dead lock.
I guess it should use the priv_sem instead. see the patch, it can work.
Index: file.c
==================================================================--- file.c
(revision 1022)
+++ file.c (working copy)
@@ -1347,7 +1347,7 @@
/* NOTE: Other filesystems get away without locking this, but
* we're clustered and this has to hit disk now... */
if (!(attr->ia_valid & ATTR_SIZE))
- down(&inode->i_sem);
+ down (&(OCFS_I(inode)->priv_sem));
if (!dentry->d_parent || !dentry->d_parent->d_inode) {
LOG_ERROR_STR ("bad inode or root inode");
@@ -1449,7 +1449,7 @@
bail:
if (!(attr->ia_valid & ATTR_SIZE))
- up(&inode->i_sem);
+ up (&(OCFS_I(inode)->priv_sem));
#ifndef BH_SEM_LEAK_CHECKING
if (error < 0)
bug 88 description
kernel 2.6.6
mount a ocfs2 file system and then do:
cd /ocfs
touch /ocfs/a
the command hangs. system not hang.
The call trace:
touch D C036B74C 0 1593 1227 (NOTLB)
ccf4fe6c 00000082 00000246 c036b74c ccc9c484 00000000 d08cc25c d08cc25c
cd1fa200 cfebf680 ccc9c484 00000001 cd086400 0001f061 805bbf22
000000b1
cd8fefd8 cd1da868 00000286 cd8fee30 cd1da870 c0354af0 00000001
cd8fee30
Call Trace:
[<d08cc25c>] ocfs_bh_sem_lookup+0x1d7/0x3ad [ocfs2]
[<d08cc25c>] ocfs_bh_sem_lookup+0x1d7/0x3ad [ocfs2]
[<c0354af0>] __down+0x7c/0xc7
[<c011269c>] default_wake_function+0x0/0x12
[<c0354c44>] __down_failed+0x8/0xc
[<d08cbe3f>] .text.lock.file+0x181/0x186 [ocfs2]
[<c015c95c>] notify_change+0x153/0x187
[<c0143ecb>] sys_utime+0x105/0x181
[<d08c6fd8>] ocfs_file_open+0x0/0x7aa [ocfs2]
[<d08c7b4f>] ocfs_file_release+0x3cd/0x415 [ocfs2]
[<c0146304>] __fput+0xa5/0xfb
[<c0103e5f>] syscall_call+0x7/0xb
Mark Fasheh
2004-Jun-08 14:15 UTC
[Ocfs2-devel] [Patch] for bug 88, touch&chmod file cause hang.
Yeah, I'll have a fix for this in svn soon. Right now it's actually just better that you remove the down() / up() combination altogether in your private tree. It shouldn't hurt anything as the buffer semaphore hash is still enabled in svn. --Mark On Tue, Jun 08, 2004 at 05:20:48PM +0800, Ling, Xiaofeng wrote:> > When you touch a file, in the call sequence, > sys_utime->notify_change->ocfs_setattr, before notify_change, > there is a "down(&inode->i_sem)", so in ocfs_setattr, when down again, > it causes dead lock. > I guess it should use the priv_sem instead. see the patch, it can work. > > Index: file.c > ==================================================================> --- file.c (revision 1022) > +++ file.c (working copy) > @@ -1347,7 +1347,7 @@ > /* NOTE: Other filesystems get away without locking this, but > * we're clustered and this has to hit disk now... */ > if (!(attr->ia_valid & ATTR_SIZE)) > - down(&inode->i_sem); > + down (&(OCFS_I(inode)->priv_sem)); > > if (!dentry->d_parent || !dentry->d_parent->d_inode) { > LOG_ERROR_STR ("bad inode or root inode"); > @@ -1449,7 +1449,7 @@ > > bail: > if (!(attr->ia_valid & ATTR_SIZE)) > - up(&inode->i_sem); > + up (&(OCFS_I(inode)->priv_sem)); > > #ifndef BH_SEM_LEAK_CHECKING > if (error < 0) > > bug 88 description > > kernel 2.6.6 > mount a ocfs2 file system and then do: > cd /ocfs > touch /ocfs/a > the command hangs. system not hang. > > The call trace: > > touch D C036B74C 0 1593 1227 (NOTLB) > ccf4fe6c 00000082 00000246 c036b74c ccc9c484 00000000 d08cc25c d08cc25c > cd1fa200 cfebf680 ccc9c484 00000001 cd086400 0001f061 805bbf22 > 000000b1 > cd8fefd8 cd1da868 00000286 cd8fee30 cd1da870 c0354af0 00000001 > cd8fee30 > Call Trace: > [<d08cc25c>] ocfs_bh_sem_lookup+0x1d7/0x3ad [ocfs2] > [<d08cc25c>] ocfs_bh_sem_lookup+0x1d7/0x3ad [ocfs2] > [<c0354af0>] __down+0x7c/0xc7 > [<c011269c>] default_wake_function+0x0/0x12 > [<c0354c44>] __down_failed+0x8/0xc > [<d08cbe3f>] .text.lock.file+0x181/0x186 [ocfs2] > [<c015c95c>] notify_change+0x153/0x187 > [<c0143ecb>] sys_utime+0x105/0x181 > [<d08c6fd8>] ocfs_file_open+0x0/0x7aa [ocfs2] > [<d08c7b4f>] ocfs_file_release+0x3cd/0x415 [ocfs2] > [<c0146304>] __fput+0xa5/0xfb > [<c0103e5f>] syscall_call+0x7/0xb > _______________________________________________ > Ocfs2-devel mailing list > Ocfs2-devel@oss.oracle.com > http://oss.oracle.com/mailman/listinfo/ocfs2-devel-- Mark Fasheh Software Developer, Oracle Corp mark.fasheh@oracle.com