| Message ID | 20221110155522.556225-1-amir73il@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | vfs: fix copy_file_range() averts filesystem freeze protection | expand |
2022-11-11 0:55 GMT+09:00, Amir Goldstein <amir73il@gmail.com>: > Commit 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs > copies") removed fallback to generic_copy_file_range() for cross-fs > cases inside vfs_copy_file_range(). > > To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to > generic_copy_file_range() was added in nfsd and ksmbd code, but that > call is missing sb_start_write(), fsnotify hooks and more. > > Ideally, nfsd and ksmbd would pass a flag to vfs_copy_file_range() that > will take care of the fallback, but that code would be subtle and we got > vfs_copy_file_range() logic wrong too many times already. > > Instead, add a flag to explicitly request vfs_copy_file_range() to > perform only generic_copy_file_range() and let nfsd and ksmbd use this > flag only in the fallback path. > > This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code > paths to reduce the risk of further regressions. > > Fixes: 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs > copies") > Signed-off-by: Amir Goldstein <amir73il@gmail.com> > --- > > Hi Al, > > Another fix for the long tradition of copy_file_range() regressions. > This one only affected cross-fs server-side-copy from nfsd/ksmbd. > > I ran the copy_range fstests group on ext4/xfs/overlay to verify no > regressions in local fs and nfsv3/nfsv4 to test server-side-copy. > > I also patched copy_file_range() to test the nfsd fallback code on > local fs. > > Namje, could you please test ksmbd. Works fine. You can add tested-by tag for ksmbd. Tested-by: Namjae Jeon <linkinjeon@kernel.org> > > Thanks, > Amir.
Amir Goldstein <amir73il@gmail.com> writes: > Commit 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs > copies") removed fallback to generic_copy_file_range() for cross-fs > cases inside vfs_copy_file_range(). > > To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to > generic_copy_file_range() was added in nfsd and ksmbd code, but that > call is missing sb_start_write(), fsnotify hooks and more. > > Ideally, nfsd and ksmbd would pass a flag to vfs_copy_file_range() that > will take care of the fallback, but that code would be subtle and we got > vfs_copy_file_range() logic wrong too many times already. > > Instead, add a flag to explicitly request vfs_copy_file_range() to > perform only generic_copy_file_range() and let nfsd and ksmbd use this > flag only in the fallback path. > > This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code > paths to reduce the risk of further regressions. > > Fixes: 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies") > Signed-off-by: Amir Goldstein <amir73il@gmail.com> > --- > > Hi Al, > > Another fix for the long tradition of copy_file_range() regressions. > This one only affected cross-fs server-side-copy from nfsd/ksmbd. > > I ran the copy_range fstests group on ext4/xfs/overlay to verify no > regressions in local fs and nfsv3/nfsv4 to test server-side-copy. > > I also patched copy_file_range() to test the nfsd fallback code on > local fs. > > Namje, could you please test ksmbd. For what is worth, I've also done some testing with ceph and I didn't saw any regression either. So, feel free to add my Tested-by: Luís Henriques <lhenriques@suse.de> Cheers,
diff --git a/fs/ksmbd/vfs.c b/fs/ksmbd/vfs.c index 8de970d6146f..94b8ed4ef870 100644 --- a/fs/ksmbd/vfs.c +++ b/fs/ksmbd/vfs.c @@ -1794,9 +1794,9 @@ int ksmbd_vfs_copy_file_ranges(struct ksmbd_work *work, ret = vfs_copy_file_range(src_fp->filp, src_off, dst_fp->filp, dst_off, len, 0); if (ret == -EOPNOTSUPP || ret == -EXDEV) - ret = generic_copy_file_range(src_fp->filp, src_off, - dst_fp->filp, dst_off, - len, 0); + ret = vfs_copy_file_range(src_fp->filp, src_off, + dst_fp->filp, dst_off, len, + COPY_FILE_SPLICE); if (ret < 0) return ret; diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c index f650afedd67f..5cf11cde51f8 100644 --- a/fs/nfsd/vfs.c +++ b/fs/nfsd/vfs.c @@ -596,8 +596,8 @@ ssize_t nfsd_copy_file_range(struct file *src, u64 src_pos, struct file *dst, ret = vfs_copy_file_range(src, src_pos, dst, dst_pos, count, 0); if (ret == -EOPNOTSUPP || ret == -EXDEV) - ret = generic_copy_file_range(src, src_pos, dst, dst_pos, - count, 0); + ret = vfs_copy_file_range(src, src_pos, dst, dst_pos, count, + COPY_FILE_SPLICE); return ret; } diff --git a/fs/read_write.c b/fs/read_write.c index 328ce8cf9a85..24b9668d6377 100644 --- a/fs/read_write.c +++ b/fs/read_write.c @@ -1388,6 +1388,8 @@ ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in, struct file *file_out, loff_t pos_out, size_t len, unsigned int flags) { + lockdep_assert(sb_write_started(file_inode(file_out)->i_sb)); + return do_splice_direct(file_in, &pos_in, file_out, &pos_out, len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0); } @@ -1424,7 +1426,9 @@ static int generic_copy_file_checks(struct file *file_in, loff_t pos_in, * and several different sets of file_operations, but they all end up * using the same ->copy_file_range() function pointer. */ - if (file_out->f_op->copy_file_range) { + if (flags & COPY_FILE_SPLICE) { + /* cross sb splice is allowed */ + } else if (file_out->f_op->copy_file_range) { if (file_in->f_op->copy_file_range != file_out->f_op->copy_file_range) return -EXDEV; @@ -1474,8 +1478,9 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, size_t len, unsigned int flags) { ssize_t ret; + bool splice = flags & COPY_FILE_SPLICE; - if (flags != 0) + if (flags & ~COPY_FILE_SPLICE) return -EINVAL; ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len, @@ -1501,14 +1506,14 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, * same sb using clone, but for filesystems where both clone and copy * are supported (e.g. nfs,cifs), we only call the copy method. */ - if (file_out->f_op->copy_file_range) { + if (!splice && 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); goto done; } - if (file_in->f_op->remap_file_range && + if (!splice && file_in->f_op->remap_file_range && file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) { ret = file_in->f_op->remap_file_range(file_in, pos_in, file_out, pos_out, @@ -1528,6 +1533,8 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, * consistent story about which filesystems support copy_file_range() * and which filesystems do not, that will allow userspace tools to * make consistent desicions w.r.t using copy_file_range(). + * + * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE. */ ret = generic_copy_file_range(file_in, pos_in, file_out, pos_out, len, flags); @@ -1582,6 +1589,10 @@ SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, pos_out = f_out.file->f_pos; } + ret = -EINVAL; + if (flags != 0) + goto out; + ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len, flags); if (ret > 0) { diff --git a/include/linux/fs.h b/include/linux/fs.h index e654435f1651..59ae95ddb679 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -2089,6 +2089,14 @@ struct dir_context { */ #define REMAP_FILE_ADVISORY (REMAP_FILE_CAN_SHORTEN) +/* + * These flags control the behavior of vfs_copy_file_range(). + * They are not available to the user via syscall. + * + * COPY_FILE_SPLICE: call splice direct instead of fs clone/copy ops + */ +#define COPY_FILE_SPLICE (1 << 0) + struct iov_iter; struct io_uring_cmd;
Commit 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies") removed fallback to generic_copy_file_range() for cross-fs cases inside vfs_copy_file_range(). To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to generic_copy_file_range() was added in nfsd and ksmbd code, but that call is missing sb_start_write(), fsnotify hooks and more. Ideally, nfsd and ksmbd would pass a flag to vfs_copy_file_range() that will take care of the fallback, but that code would be subtle and we got vfs_copy_file_range() logic wrong too many times already. Instead, add a flag to explicitly request vfs_copy_file_range() to perform only generic_copy_file_range() and let nfsd and ksmbd use this flag only in the fallback path. This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code paths to reduce the risk of further regressions. Fixes: 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies") Signed-off-by: Amir Goldstein <amir73il@gmail.com> --- Hi Al, Another fix for the long tradition of copy_file_range() regressions. This one only affected cross-fs server-side-copy from nfsd/ksmbd. I ran the copy_range fstests group on ext4/xfs/overlay to verify no regressions in local fs and nfsv3/nfsv4 to test server-side-copy. I also patched copy_file_range() to test the nfsd fallback code on local fs. Namje, could you please test ksmbd. Thanks, Amir. fs/ksmbd/vfs.c | 6 +++--- fs/nfsd/vfs.c | 4 ++-- fs/read_write.c | 19 +++++++++++++++---- include/linux/fs.h | 8 ++++++++ 4 files changed, 28 insertions(+), 9 deletions(-)