]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
cifs: Fix reading into an ITER_FOLIOQ from the smbdirect code
authorDavid Howells <dhowells@redhat.com>
Wed, 2 Apr 2025 19:27:26 +0000 (20:27 +0100)
committerSteve French <stfrench@microsoft.com>
Thu, 26 Jun 2025 16:13:16 +0000 (11:13 -0500)
When performing a file read from RDMA, smbd_recv() prints an "Invalid msg
type 4" error and fails the I/O.  This is due to the switch-statement there
not handling the ITER_FOLIOQ handed down from netfslib.

Fix this by collapsing smbd_recv_buf() and smbd_recv_page() into
smbd_recv() and just using copy_to_iter() instead of memcpy().  This
future-proofs the function too, in case more ITER_* types are added.

Fixes: ee4cdf7ba857 ("netfs: Speed up buffered reading")
Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Tom Talpey <tom@talpey.com>
cc: Paulo Alcantara (Red Hat) <pc@manguebit.com>
cc: Matthew Wilcox <willy@infradead.org>
cc: linux-cifs@vger.kernel.org
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Steve French <stfrench@microsoft.com>
fs/smb/client/smbdirect.c

index 0a9fd6c399f666e9e9c7e40df5b1926cd9a2ed5c..754e94a0e07f502d70bb24fa06b4b8762be44a8b 100644 (file)
@@ -1778,35 +1778,39 @@ try_again:
 }
 
 /*
- * Receive data from receive reassembly queue
+ * Receive data from the transport's receive reassembly queue
  * All the incoming data packets are placed in reassembly queue
- * buf: the buffer to read data into
+ * iter: the buffer to read data into
  * size: the length of data to read
  * return value: actual data read
- * Note: this implementation copies the data from reassebmly queue to receive
+ *
+ * Note: this implementation copies the data from reassembly queue to receive
  * buffers used by upper layer. This is not the optimal code path. A better way
  * to do it is to not have upper layer allocate its receive buffers but rather
  * borrow the buffer from reassembly queue, and return it after data is
  * consumed. But this will require more changes to upper layer code, and also
  * need to consider packet boundaries while they still being reassembled.
  */
-static int smbd_recv_buf(struct smbd_connection *info, char *buf,
-               unsigned int size)
+int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
 {
        struct smbdirect_socket *sc = &info->socket;
        struct smbd_response *response;
        struct smbdirect_data_transfer *data_transfer;
+       size_t size = iov_iter_count(&msg->msg_iter);
        int to_copy, to_read, data_read, offset;
        u32 data_length, remaining_data_length, data_offset;
        int rc;
 
+       if (WARN_ON_ONCE(iov_iter_rw(&msg->msg_iter) == WRITE))
+               return -EINVAL; /* It's a bug in upper layer to get there */
+
 again:
        /*
         * No need to hold the reassembly queue lock all the time as we are
         * the only one reading from the front of the queue. The transport
         * may add more entries to the back of the queue at the same time
         */
-       log_read(INFO, "size=%d info->reassembly_data_length=%d\n", size,
+       log_read(INFO, "size=%zd info->reassembly_data_length=%d\n", size,
                info->reassembly_data_length);
        if (info->reassembly_data_length >= size) {
                int queue_length;
@@ -1844,7 +1848,10 @@ again:
                        if (response->first_segment && size == 4) {
                                unsigned int rfc1002_len =
                                        data_length + remaining_data_length;
-                               *((__be32 *)buf) = cpu_to_be32(rfc1002_len);
+                               __be32 rfc1002_hdr = cpu_to_be32(rfc1002_len);
+                               if (copy_to_iter(&rfc1002_hdr, sizeof(rfc1002_hdr),
+                                                &msg->msg_iter) != sizeof(rfc1002_hdr))
+                                       return -EFAULT;
                                data_read = 4;
                                response->first_segment = false;
                                log_read(INFO, "returning rfc1002 length %d\n",
@@ -1853,10 +1860,9 @@ again:
                        }
 
                        to_copy = min_t(int, data_length - offset, to_read);
-                       memcpy(
-                               buf + data_read,
-                               (char *)data_transfer + data_offset + offset,
-                               to_copy);
+                       if (copy_to_iter((char *)data_transfer + data_offset + offset,
+                                        to_copy, &msg->msg_iter) != to_copy)
+                               return -EFAULT;
 
                        /* move on to the next buffer? */
                        if (to_copy == data_length - offset) {
@@ -1921,90 +1927,6 @@ read_rfc1002_done:
        goto again;
 }
 
-/*
- * Receive a page from receive reassembly queue
- * page: the page to read data into
- * to_read: the length of data to read
- * return value: actual data read
- */
-static int smbd_recv_page(struct smbd_connection *info,
-               struct page *page, unsigned int page_offset,
-               unsigned int to_read)
-{
-       struct smbdirect_socket *sc = &info->socket;
-       int ret;
-       char *to_address;
-       void *page_address;
-
-       /* make sure we have the page ready for read */
-       ret = wait_event_interruptible(
-               info->wait_reassembly_queue,
-               info->reassembly_data_length >= to_read ||
-                       sc->status != SMBDIRECT_SOCKET_CONNECTED);
-       if (ret)
-               return ret;
-
-       /* now we can read from reassembly queue and not sleep */
-       page_address = kmap_atomic(page);
-       to_address = (char *) page_address + page_offset;
-
-       log_read(INFO, "reading from page=%p address=%p to_read=%d\n",
-               page, to_address, to_read);
-
-       ret = smbd_recv_buf(info, to_address, to_read);
-       kunmap_atomic(page_address);
-
-       return ret;
-}
-
-/*
- * Receive data from transport
- * msg: a msghdr point to the buffer, can be ITER_KVEC or ITER_BVEC
- * return: total bytes read, or 0. SMB Direct will not do partial read.
- */
-int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
-{
-       char *buf;
-       struct page *page;
-       unsigned int to_read, page_offset;
-       int rc;
-
-       if (iov_iter_rw(&msg->msg_iter) == WRITE) {
-               /* It's a bug in upper layer to get there */
-               cifs_dbg(VFS, "Invalid msg iter dir %u\n",
-                        iov_iter_rw(&msg->msg_iter));
-               rc = -EINVAL;
-               goto out;
-       }
-
-       switch (iov_iter_type(&msg->msg_iter)) {
-       case ITER_KVEC:
-               buf = msg->msg_iter.kvec->iov_base;
-               to_read = msg->msg_iter.kvec->iov_len;
-               rc = smbd_recv_buf(info, buf, to_read);
-               break;
-
-       case ITER_BVEC:
-               page = msg->msg_iter.bvec->bv_page;
-               page_offset = msg->msg_iter.bvec->bv_offset;
-               to_read = msg->msg_iter.bvec->bv_len;
-               rc = smbd_recv_page(info, page, page_offset, to_read);
-               break;
-
-       default:
-               /* It's a bug in upper layer to get there */
-               cifs_dbg(VFS, "Invalid msg type %d\n",
-                        iov_iter_type(&msg->msg_iter));
-               rc = -EINVAL;
-       }
-
-out:
-       /* SMBDirect will read it all or nothing */
-       if (rc > 0)
-               msg->msg_iter.count = 0;
-       return rc;
-}
-
 /*
  * Send data to transport
  * Each rqst is transported as a SMBDirect payload