| Message ID | 20210102101805.355106-1-cgxu519@mykernel.net |
|---|---|
| State | Not Applicable |
| Headers | show |
| Series | ext2: discard block reservation on last writable file release | expand |
On Sat 02-01-21 18:18:05, Chengguang Xu wrote: > Currently reserved blocks are discarded on every writable > file release, it's not efficient for multiple writer case. > > Signed-off-by: Chengguang Xu <cgxu519@mykernel.net> Thanks for the patch. I agree that in principle something like this is desirable but there's a small catch. i_writecount is also elevated from vfs_truncate() which does not have inode open. So it can happen that ->release() gets called, sees inode->i_writecount > 1, but never gets called again (and thus reservation is not properly released). So I prefer to leave ext2 as is until this gets resolved - especially since ext2 fs driver isn't really used on any performance sensitive multi-writer workloads AFAIK (ext4 driver is usually used in such cases). Honza > --- > fs/ext2/file.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/fs/ext2/file.c b/fs/ext2/file.c > index 96044f5dbc0e..9a19d8fe7ffd 100644 > --- a/fs/ext2/file.c > +++ b/fs/ext2/file.c > @@ -141,7 +141,7 @@ static int ext2_file_mmap(struct file *file, struct vm_area_struct *vma) > */ > static int ext2_release_file (struct inode * inode, struct file * filp) > { > - if (filp->f_mode & FMODE_WRITE) { > + if (filp->f_mode & FMODE_WRITE && (atomic_read(&inode->i_writecount) == 1)) { > mutex_lock(&EXT2_I(inode)->truncate_mutex); > ext2_discard_reservation(inode); > mutex_unlock(&EXT2_I(inode)->truncate_mutex); > -- > 2.18.4 >
diff --git a/fs/ext2/file.c b/fs/ext2/file.c index 96044f5dbc0e..9a19d8fe7ffd 100644 --- a/fs/ext2/file.c +++ b/fs/ext2/file.c @@ -141,7 +141,7 @@ static int ext2_file_mmap(struct file *file, struct vm_area_struct *vma) */ static int ext2_release_file (struct inode * inode, struct file * filp) { - if (filp->f_mode & FMODE_WRITE) { + if (filp->f_mode & FMODE_WRITE && (atomic_read(&inode->i_writecount) == 1)) { mutex_lock(&EXT2_I(inode)->truncate_mutex); ext2_discard_reservation(inode); mutex_unlock(&EXT2_I(inode)->truncate_mutex);
Currently reserved blocks are discarded on every writable file release, it's not efficient for multiple writer case. Signed-off-by: Chengguang Xu <cgxu519@mykernel.net> --- fs/ext2/file.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)