ceph: only dirty ITER_IOVEC pages for direct read

If a page is already locked, attempting to dirty it leads to a deadlock
in lock_page().  This is what currently happens to ITER_BVEC pages when
a dio-enabled loop device is backed by ceph:

  $ losetup --direct-io /dev/loop0 /mnt/cephfs/img
  $ xfs_io -c 'pread 0 4k' /dev/loop0

Follow other file systems and only dirty ITER_IOVEC pages.

Cc: stable@kernel.org
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 6639926eed4e8461d0c50b6a686df236e1c96338..b67eec3532a125c0c8ca0c97b7b0f22c650f7966 100644 (file)
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -640,7 +640,8 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
 struct ceph_aio_request {
        struct kiocb *iocb;
        size_t total_len;
-       int write;
+       bool write;
+       bool should_dirty;
        int error;
        struct list_head osd_reqs;
        unsigned num_reqs;
@@ -750,7 +751,7 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
                }
        }
 
-       ceph_put_page_vector(osd_data->pages, num_pages, !aio_req->write);
+       ceph_put_page_vector(osd_data->pages, num_pages, aio_req->should_dirty);
        ceph_osdc_put_request(req);
 
        if (rc < 0)
@@ -847,6 +848,7 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
        size_t count = iov_iter_count(iter);
        loff_t pos = iocb->ki_pos;
        bool write = iov_iter_rw(iter) == WRITE;
+       bool should_dirty = !write && iter_is_iovec(iter);
 
        if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
                return -EROFS;
@@ -914,6 +916,7 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
                        if (aio_req) {
                                aio_req->iocb = iocb;
                                aio_req->write = write;
+                               aio_req->should_dirty = should_dirty;
                                INIT_LIST_HEAD(&aio_req->osd_reqs);
                                if (write) {
                                        aio_req->mtime = mtime;
@@ -971,7 +974,7 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
                                len = ret;
                }
 
-               ceph_put_page_vector(pages, num_pages, !write);
+               ceph_put_page_vector(pages, num_pages, should_dirty);
 
                ceph_osdc_put_request(req);
                if (ret < 0)