On 02/21/2019 10:23 AM, Johannes Thumshirn wrote: > Some file-systems, like tmpfs, do not support direct IO, but file-backed > namespaces default to using direct IO. If direct IO is unavailable fall > back to using buffered IO for the file-backed namespace. > > This might not ultimately be a solution for production environments but > for test environments it sometimes is feasible to use tmpfs. > > Signed-off-by: Johannes Thumshirn <[email protected]> > --- > drivers/nvme/target/io-cmd-file.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/drivers/nvme/target/io-cmd-file.c > b/drivers/nvme/target/io-cmd-file.c > index 517522305e5c..8a861cc0160e 100644 > --- a/drivers/nvme/target/io-cmd-file.c > +++ b/drivers/nvme/target/io-cmd-file.c > @@ -38,11 +38,21 @@ int nvmet_file_ns_enable(struct nvmet_ns *ns) > > ns->file = filp_open(ns->device_path, flags, 0); > if (IS_ERR(ns->file)) { > + if (ns->file == ERR_PTR(-EINVAL) && (flags & O_DIRECT)) { > + flags &= ~O_DIRECT; > + ns->buffered_io = 0; This needs to be set to 1 when we enable buffered I/O. > + ns->file = filp_open(ns->device_path, flags, 0); > + if (!IS_ERR(ns->file)) { > + pr_info("direct I/O unavailable, falling back > to buffered I/O\n"); > + goto getattr; > + } > + } > pr_err("failed to open file %s: (%ld)\n", > ns->device_path, PTR_ERR(ns->file)); > return PTR_ERR(ns->file); > } > > +getattr: > ret = vfs_getattr(&ns->file->f_path, > &stat, STATX_SIZE, AT_STATX_FORCE_SYNC); > if (ret) > kk
As per specified in the patch, this is only useful for testing, then we should modify the test scripts so that on creation of the ctrl we switch to the buffered I/O before running fio. OR Similar result can be achieved by setting buffered I/O flag buffered_io=1 before enabling the name-space in the test script.
