| Message ID | 20181024195837.35532-2-olga.kornievskaia@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | client-side support for "inter" SSC copy | expand |
On Wed, Oct 24, 2018 at 10:58 PM Olga Kornievskaia <olga.kornievskaia@gmail.com> wrote: > > From: Olga Kornievskaia <kolga@netapp.com> > > This patch removes the check for source and destination files to > come from the same superblock. This feature was of interest to > NFS as well as CIFS communities. > > Specifically, this feature is needed to allow for NFSv4.2 copy offload > to be done between different NFSv4.2 servers. SMBv3 copy offload between > different servers would be able to use this as well. > > Removal of the check implies that passed in source and destination > files can come from different superblocks of the same file system > type or different. It is upto each individual copy_file_range() > file system implementation to decide what type of copy it is > capable of doing and return -EXDEV in cases support is lacking. > > There are 3 known implementator of copy_file_range() f_op: NFS, > CIFS, OverlayFS. NFS and CIFS are interested to support cross-device > copy offload but do not support cross file system types copy offload. > Following patches will add appropriate checks in each of the drivers. > That should be the other way around - first add limitation inside filesystems then relax vfs check. otherwise you leave a bug in the middle of bisect. > If the copy_file_range() errors with EXDEV, the code would fallback > on doing do_splice_direct() copying which in itself is beneficial. > > Adding wording to the vfs.txt and porting documentation about the > new support for cross-device copy offload. > > Signed-off-by: Olga Kornievskaia <kolga@netapp.com> > --- > Documentation/filesystems/porting | 7 +++++++ > Documentation/filesystems/vfs.txt | 6 +++++- > fs/read_write.c | 9 +++------ > 3 files changed, 15 insertions(+), 7 deletions(-) > > diff --git a/Documentation/filesystems/porting b/Documentation/filesystems/porting > index 7b7b845..ebb4954 100644 > --- a/Documentation/filesystems/porting > +++ b/Documentation/filesystems/porting > @@ -622,3 +622,10 @@ in your dentry operations instead. > alloc_file_clone(file, flags, ops) does not affect any caller's references. > On success you get a new struct file sharing the mount/dentry with the > original, on failure - ERR_PTR(). > +-- > +[mandatory] > + ->copy_file_range() may now be passed files which belong to two > + different superblocks of the same file system type or which belong > + to two different filesystems types all together. As before, the > + destination's copy_file_range() is the function which is called. > + If it cannot copy ranges from the source, it should return -EXDEV. > diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt > index a6c6a8a..34c0e8c 100644 > --- a/Documentation/filesystems/vfs.txt > +++ b/Documentation/filesystems/vfs.txt > @@ -1,5 +1,6 @@ > > Overview of the Linux Virtual File System > +- [fs] nfs: Don't let readdirplus revalidate an inode that was marked as stale (Benjamin Coddington) [1429514 1416532] > ??? Thanks, Amir.
diff --git a/Documentation/filesystems/porting b/Documentation/filesystems/porting index 7b7b845..ebb4954 100644 --- a/Documentation/filesystems/porting +++ b/Documentation/filesystems/porting @@ -622,3 +622,10 @@ in your dentry operations instead. alloc_file_clone(file, flags, ops) does not affect any caller's references. On success you get a new struct file sharing the mount/dentry with the original, on failure - ERR_PTR(). +-- +[mandatory] + ->copy_file_range() may now be passed files which belong to two + different superblocks of the same file system type or which belong + to two different filesystems types all together. As before, the + destination's copy_file_range() is the function which is called. + If it cannot copy ranges from the source, it should return -EXDEV. diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt index a6c6a8a..34c0e8c 100644 --- a/Documentation/filesystems/vfs.txt +++ b/Documentation/filesystems/vfs.txt @@ -1,5 +1,6 @@ Overview of the Linux Virtual File System +- [fs] nfs: Don't let readdirplus revalidate an inode that was marked as stale (Benjamin Coddington) [1429514 1416532] Original author: Richard Gooch <rgooch@atnf.csiro.au> @@ -958,7 +959,10 @@ otherwise noted. fallocate: called by the VFS to preallocate blocks or punch a hole. - copy_file_range: called by the copy_file_range(2) system call. + copy_file_range: called by copy_file_range(2) system call. This method + works on two file descriptors that might reside on + different superblocks which might belong to file systems + of different types. clone_file_range: called by the ioctl(2) system call for FICLONERANGE and FICLONE commands. diff --git a/fs/read_write.c b/fs/read_write.c index 39b4a21..fb4ffca 100644 --- a/fs/read_write.c +++ b/fs/read_write.c @@ -1575,10 +1575,6 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, (file_out->f_flags & O_APPEND)) return -EBADF; - /* this could be relaxed once a method supports cross-fs copies */ - if (inode_in->i_sb != inode_out->i_sb) - return -EXDEV; - if (len == 0) return 0; @@ -1588,7 +1584,8 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, * Try cloning first, this is supported by more file systems, and * more efficient if both clone and copy are supported (e.g. NFS). */ - if (file_in->f_op->clone_file_range) { + if (inode_in->i_sb == inode_out->i_sb && + file_in->f_op->clone_file_range) { ret = file_in->f_op->clone_file_range(file_in, pos_in, file_out, pos_out, len); if (ret == 0) { @@ -1600,7 +1597,7 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, if (file_out->f_op->copy_file_range) { ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out, pos_out, len, flags); - if (ret != -EOPNOTSUPP) + if (ret != -EOPNOTSUPP && ret != -EXDEV) goto done; }