Merge tag '5.19-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd

Pull ksmbd server updates from Steve French:

 - rdma (smbdirect) fixes, cleanup and optimizations

 - crediting (flow control) fix for mounts from Windows client

 - ACL fix

 - Windows client query dir fix

 - write validation fix

 - cleanups

* tag '5.19-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd:
  ksmbd: smbd: relax the count of sges required
  ksmbd: fix outstanding credits related bugs
  ksmbd: smbd: fix connection dropped issue
  ksmbd: Fix some kernel-doc comments
  ksmbd: fix wrong smbd max read/write size check
  ksmbd: add smbd max io size parameter
  ksmbd: handle smb2 query dir request for OutputBufferLength that is too small
  ksmbd: smbd: handle multiple Buffer descriptors
  ksmbd: smbd: change the return value of get_sg_list
  ksmbd: smbd: simplify tracking pending packets
  ksmbd: smbd: introduce read/write credits for RDMA read/write
  ksmbd: smbd: change prototypes of RDMA read/write related functions
  ksmbd: validate length in smb2_write()
  ksmbd: fix reference count leak in smb_check_perm_dacl()

fs/ksmbd/connection.c

@@ -62,7 +62,7 @@ struct ksmbd_conn *ksmbd_conn_alloc(void)
 	atomic_set(&conn->req_running, 0);
 	atomic_set(&conn->r_count, 0);
 	conn->total_credits = 1;
-	conn->outstanding_credits = 1;
+	conn->outstanding_credits = 0;
 
 	init_waitqueue_head(&conn->req_running_q);
 	INIT_LIST_HEAD(&conn->conns_list);
@@ -205,31 +205,31 @@ int ksmbd_conn_write(struct ksmbd_work *work)
 	return 0;
 }
 
-int ksmbd_conn_rdma_read(struct ksmbd_conn *conn, void *buf,
-			 unsigned int buflen, u32 remote_key, u64 remote_offset,
-			 u32 remote_len)
+int ksmbd_conn_rdma_read(struct ksmbd_conn *conn,
+			 void *buf, unsigned int buflen,
+			 struct smb2_buffer_desc_v1 *desc,
+			 unsigned int desc_len)
 {
 	int ret = -EINVAL;
 
 	if (conn->transport->ops->rdma_read)
 		ret = conn->transport->ops->rdma_read(conn->transport,
 						      buf, buflen,
-						      remote_key, remote_offset,
-						      remote_len);
+						      desc, desc_len);
 	return ret;
 }
 
-int ksmbd_conn_rdma_write(struct ksmbd_conn *conn, void *buf,
-			  unsigned int buflen, u32 remote_key,
-			  u64 remote_offset, u32 remote_len)
+int ksmbd_conn_rdma_write(struct ksmbd_conn *conn,
+			  void *buf, unsigned int buflen,
+			  struct smb2_buffer_desc_v1 *desc,
+			  unsigned int desc_len)
 {
 	int ret = -EINVAL;
 
 	if (conn->transport->ops->rdma_write)
 		ret = conn->transport->ops->rdma_write(conn->transport,
 						       buf, buflen,
-						       remote_key, remote_offset,
-						       remote_len);
+						       desc, desc_len);
 	return ret;
 }
 

fs/ksmbd/connection.h

@@ -122,11 +122,14 @@ struct ksmbd_transport_ops {
 	int (*writev)(struct ksmbd_transport *t, struct kvec *iovs, int niov,
 		      int size, bool need_invalidate_rkey,
 		      unsigned int remote_key);
-	int (*rdma_read)(struct ksmbd_transport *t, void *buf, unsigned int len,
-			 u32 remote_key, u64 remote_offset, u32 remote_len);
-	int (*rdma_write)(struct ksmbd_transport *t, void *buf,
-			  unsigned int len, u32 remote_key, u64 remote_offset,
-			  u32 remote_len);
+	int (*rdma_read)(struct ksmbd_transport *t,
+			 void *buf, unsigned int len,
+			 struct smb2_buffer_desc_v1 *desc,
+			 unsigned int desc_len);
+	int (*rdma_write)(struct ksmbd_transport *t,
+			  void *buf, unsigned int len,
+			  struct smb2_buffer_desc_v1 *desc,
+			  unsigned int desc_len);
 };
 
 struct ksmbd_transport {
@@ -148,12 +151,14 @@ struct ksmbd_conn *ksmbd_conn_alloc(void);
 void ksmbd_conn_free(struct ksmbd_conn *conn);
 bool ksmbd_conn_lookup_dialect(struct ksmbd_conn *c);
 int ksmbd_conn_write(struct ksmbd_work *work);
-int ksmbd_conn_rdma_read(struct ksmbd_conn *conn, void *buf,
-			 unsigned int buflen, u32 remote_key, u64 remote_offset,
-			 u32 remote_len);
-int ksmbd_conn_rdma_write(struct ksmbd_conn *conn, void *buf,
-			  unsigned int buflen, u32 remote_key, u64 remote_offset,
-			  u32 remote_len);
+int ksmbd_conn_rdma_read(struct ksmbd_conn *conn,
+			 void *buf, unsigned int buflen,
+			 struct smb2_buffer_desc_v1 *desc,
+			 unsigned int desc_len);
+int ksmbd_conn_rdma_write(struct ksmbd_conn *conn,
+			  void *buf, unsigned int buflen,
+			  struct smb2_buffer_desc_v1 *desc,
+			  unsigned int desc_len);
 void ksmbd_conn_enqueue_request(struct ksmbd_work *work);
 int ksmbd_conn_try_dequeue_request(struct ksmbd_work *work);
 void ksmbd_conn_init_server_callbacks(struct ksmbd_conn_ops *ops);

fs/ksmbd/ksmbd_netlink.h

@@ -104,7 +104,8 @@ struct ksmbd_startup_request {
 					 */
 	__u32	sub_auth[3];		/* Subauth value for Security ID */
 	__u32	smb2_max_credits;	/* MAX credits */
-	__u32	reserved[128];		/* Reserved room */
+	__u32	smbd_max_io_size;	/* smbd read write size */
+	__u32	reserved[127];		/* Reserved room */
 	__u32	ifc_list_sz;		/* interfaces list size */
 	__s8	____payload[];
 };

fs/ksmbd/misc.c

@@ -20,7 +20,7 @@
  * wildcard '*' and '?'
  * TODO : implement consideration about DOS_DOT, DOS_QM and DOS_STAR
  *
- * @string:	string to compare with a pattern
+ * @str:	string to compare with a pattern
  * @len:	string length
  * @pattern:	pattern string which might include wildcard '*' and '?'
  *
@@ -152,8 +152,8 @@ int parse_stream_name(char *filename, char **stream_name, int *s_type)
 /**
  * convert_to_nt_pathname() - extract and return windows path string
  *      whose share directory prefix was removed from file path
- * @filename : unix filename
- * @sharepath: share path string
+ * @share: ksmbd_share_config pointer
+ * @path: path to report
  *
  * Return : windows path string or error
  */
@@ -250,8 +250,8 @@ char *ksmbd_extract_sharename(char *treename)
 
 /**
  * convert_to_unix_name() - convert windows name to unix format
- * @path:	name to be converted
- * @tid:	tree id of mathing share
+ * @share:	ksmbd_share_config pointer
+ * @name:	file name that is relative to share
  *
  * Return:	converted name on success, otherwise NULL
  */

fs/ksmbd/smb2misc.c

@@ -338,7 +338,7 @@ static int smb2_validate_credit_charge(struct ksmbd_conn *conn,
 		ret = 1;
 	}
 
-	if ((u64)conn->outstanding_credits + credit_charge > conn->vals->max_credits) {
+	if ((u64)conn->outstanding_credits + credit_charge > conn->total_credits) {
 		ksmbd_debug(SMB, "Limits exceeding the maximum allowable outstanding requests, given : %u, pending : %u\n",
 			    credit_charge, conn->outstanding_credits);
 		ret = 1;

fs/ksmbd/smb2pdu.c

@@ -3938,6 +3938,12 @@ int smb2_query_dir(struct ksmbd_work *work)
 	set_ctx_actor(&dir_fp->readdir_data.ctx, __query_dir);
 
 	rc = iterate_dir(dir_fp->filp, &dir_fp->readdir_data.ctx);
+	/*
+	 * req->OutputBufferLength is too small to contain even one entry.
+	 * In this case, it immediately returns OutputBufferLength 0 to client.
+	 */
+	if (!d_info.out_buf_len && !d_info.num_entry)
+		goto no_buf_len;
 	if (rc == 0)
 		restart_ctx(&dir_fp->readdir_data.ctx);
 	if (rc == -ENOSPC)
@@ -3964,9 +3970,11 @@ int smb2_query_dir(struct ksmbd_work *work)
 		rsp->Buffer[0] = 0;
 		inc_rfc1001_len(work->response_buf, 9);
 	} else {
+no_buf_len:
 		((struct file_directory_info *)
 		((char *)rsp->Buffer + d_info.last_entry_offset))
 		->NextEntryOffset = 0;
+		if (d_info.data_count >= d_info.last_entry_off_align)
 			d_info.data_count -= d_info.last_entry_off_align;
 
 		rsp->StructureSize = cpu_to_le16(9);
@@ -6116,7 +6124,6 @@ static noinline int smb2_read_pipe(struct ksmbd_work *work)
 static int smb2_set_remote_key_for_rdma(struct ksmbd_work *work,
 					struct smb2_buffer_desc_v1 *desc,
 					__le32 Channel,
-					__le16 ChannelInfoOffset,
 					__le16 ChannelInfoLength)
 {
 	unsigned int i, ch_count;
@@ -6134,14 +6141,12 @@ static int smb2_set_remote_key_for_rdma(struct ksmbd_work *work,
 				le32_to_cpu(desc[i].length));
 		}
 	}
-	if (ch_count != 1) {
-		ksmbd_debug(RDMA, "RDMA multiple buffer descriptors %d are not supported yet\n",
-			    ch_count);
+	if (!ch_count)
 		return -EINVAL;
-	}
 
 	work->need_invalidate_rkey =
 		(Channel == SMB2_CHANNEL_RDMA_V1_INVALIDATE);
+	if (Channel == SMB2_CHANNEL_RDMA_V1_INVALIDATE)
 		work->remote_key = le32_to_cpu(desc->token);
 	return 0;
 }
@@ -6150,14 +6155,12 @@ static ssize_t smb2_read_rdma_channel(struct ksmbd_work *work,
 				      struct smb2_read_req *req, void *data_buf,
 				      size_t length)
 {
-	struct smb2_buffer_desc_v1 *desc =
-		(struct smb2_buffer_desc_v1 *)&req->Buffer[0];
 	int err;
 
 	err = ksmbd_conn_rdma_write(work->conn, data_buf, length,
-				    le32_to_cpu(desc->token),
-				    le64_to_cpu(desc->offset),
-				    le32_to_cpu(desc->length));
+				    (struct smb2_buffer_desc_v1 *)
+				    ((char *)req + le16_to_cpu(req->ReadChannelInfoOffset)),
+				    le16_to_cpu(req->ReadChannelInfoLength));
 	if (err)
 		return err;
 
@@ -6180,6 +6183,8 @@ int smb2_read(struct ksmbd_work *work)
 	size_t length, mincount;
 	ssize_t nbytes = 0, remain_bytes = 0;
 	int err = 0;
+	bool is_rdma_channel = false;
+	unsigned int max_read_size = conn->vals->max_read_size;
 
 	WORK_BUFFERS(work, req, rsp);
 
@@ -6191,6 +6196,11 @@ int smb2_read(struct ksmbd_work *work)
 
 	if (req->Channel == SMB2_CHANNEL_RDMA_V1_INVALIDATE ||
 	    req->Channel == SMB2_CHANNEL_RDMA_V1) {
+		is_rdma_channel = true;
+		max_read_size = get_smbd_max_read_write_size();
+	}
+
+	if (is_rdma_channel == true) {
 		unsigned int ch_offset = le16_to_cpu(req->ReadChannelInfoOffset);
 
 		if (ch_offset < offsetof(struct smb2_read_req, Buffer)) {
@@ -6201,7 +6211,6 @@ int smb2_read(struct ksmbd_work *work)
 						   (struct smb2_buffer_desc_v1 *)
 						   ((char *)req + ch_offset),
 						   req->Channel,
-						   req->ReadChannelInfoOffset,
 						   req->ReadChannelInfoLength);
 		if (err)
 			goto out;
@@ -6223,9 +6232,9 @@ int smb2_read(struct ksmbd_work *work)
 	length = le32_to_cpu(req->Length);
 	mincount = le32_to_cpu(req->MinimumCount);
 
-	if (length > conn->vals->max_read_size) {
+	if (length > max_read_size) {
 		ksmbd_debug(SMB, "limiting read size to max size(%u)\n",
-			    conn->vals->max_read_size);
+			    max_read_size);
 		err = -EINVAL;
 		goto out;
 	}
@@ -6257,8 +6266,7 @@ int smb2_read(struct ksmbd_work *work)
 	ksmbd_debug(SMB, "nbytes %zu, offset %lld mincount %zu\n",
 		    nbytes, offset, mincount);
 
-	if (req->Channel == SMB2_CHANNEL_RDMA_V1_INVALIDATE ||
-	    req->Channel == SMB2_CHANNEL_RDMA_V1) {
+	if (is_rdma_channel == true) {
 		/* write data to the client using rdma channel */
 		remain_bytes = smb2_read_rdma_channel(work, req,
 						      work->aux_payload_buf,
@@ -6328,10 +6336,6 @@ static noinline int smb2_write_pipe(struct ksmbd_work *work)
 	length = le32_to_cpu(req->Length);
 	id = req->VolatileFileId;
 
-	if (le16_to_cpu(req->DataOffset) ==
-	    offsetof(struct smb2_write_req, Buffer)) {
-		data_buf = (char *)&req->Buffer[0];
-	} else {
 	if ((u64)le16_to_cpu(req->DataOffset) + length >
 	    get_rfc1002_len(work->request_buf)) {
 		pr_err("invalid write data offset %u, smb_len %u\n",
@@ -6343,7 +6347,6 @@ static noinline int smb2_write_pipe(struct ksmbd_work *work)
 
 	data_buf = (char *)(((char *)&req->hdr.ProtocolId) +
 			   le16_to_cpu(req->DataOffset));
-	}
 
 	rpc_resp = ksmbd_rpc_write(work->sess, id, data_buf, length);
 	if (rpc_resp) {
@@ -6384,21 +6387,18 @@ static ssize_t smb2_write_rdma_channel(struct ksmbd_work *work,
 				       struct ksmbd_file *fp,
 				       loff_t offset, size_t length, bool sync)
 {
-	struct smb2_buffer_desc_v1 *desc;
 	char *data_buf;
 	int ret;
 	ssize_t nbytes;
 
-	desc = (struct smb2_buffer_desc_v1 *)&req->Buffer[0];
-
 	data_buf = kvmalloc(length, GFP_KERNEL | __GFP_ZERO);
 	if (!data_buf)
 		return -ENOMEM;
 
 	ret = ksmbd_conn_rdma_read(work->conn, data_buf, length,
-				   le32_to_cpu(desc->token),
-				   le64_to_cpu(desc->offset),
-				   le32_to_cpu(desc->length));
+				   (struct smb2_buffer_desc_v1 *)
+				   ((char *)req + le16_to_cpu(req->WriteChannelInfoOffset)),
+				   le16_to_cpu(req->WriteChannelInfoLength));
 	if (ret < 0) {
 		kvfree(data_buf);
 		return ret;
@@ -6427,8 +6427,9 @@ int smb2_write(struct ksmbd_work *work)
 	size_t length;
 	ssize_t nbytes;
 	char *data_buf;
-	bool writethrough = false;
+	bool writethrough = false, is_rdma_channel = false;
 	int err = 0;
+	unsigned int max_write_size = work->conn->vals->max_write_size;
 
 	WORK_BUFFERS(work, req, rsp);
 
@@ -6437,8 +6438,17 @@ int smb2_write(struct ksmbd_work *work)
 		return smb2_write_pipe(work);
 	}
 
+	offset = le64_to_cpu(req->Offset);
+	length = le32_to_cpu(req->Length);
+
 	if (req->Channel == SMB2_CHANNEL_RDMA_V1 ||
 	    req->Channel == SMB2_CHANNEL_RDMA_V1_INVALIDATE) {
+		is_rdma_channel = true;
+		max_write_size = get_smbd_max_read_write_size();
+		length = le32_to_cpu(req->RemainingBytes);
+	}
+
+	if (is_rdma_channel == true) {
 		unsigned int ch_offset = le16_to_cpu(req->WriteChannelInfoOffset);
 
 		if (req->Length != 0 || req->DataOffset != 0 ||
@@ -6450,7 +6460,6 @@ int smb2_write(struct ksmbd_work *work)
 						   (struct smb2_buffer_desc_v1 *)
 						   ((char *)req + ch_offset),
 						   req->Channel,
-						   req->WriteChannelInfoOffset,
 						   req->WriteChannelInfoLength);
 		if (err)
 			goto out;
@@ -6474,12 +6483,9 @@ int smb2_write(struct ksmbd_work *work)
 		goto out;
 	}
 
-	offset = le64_to_cpu(req->Offset);
-	length = le32_to_cpu(req->Length);
-
-	if (length > work->conn->vals->max_write_size) {
+	if (length > max_write_size) {
 		ksmbd_debug(SMB, "limiting write size to max size(%u)\n",
-			    work->conn->vals->max_write_size);
+			    max_write_size);
 		err = -EINVAL;
 		goto out;
 	}
@@ -6487,12 +6493,7 @@ int smb2_write(struct ksmbd_work *work)
 	if (le32_to_cpu(req->Flags) & SMB2_WRITEFLAG_WRITE_THROUGH)
 		writethrough = true;
 
-	if (req->Channel != SMB2_CHANNEL_RDMA_V1 &&
-	    req->Channel != SMB2_CHANNEL_RDMA_V1_INVALIDATE) {
-		if (le16_to_cpu(req->DataOffset) ==
-		    offsetof(struct smb2_write_req, Buffer)) {
-			data_buf = (char *)&req->Buffer[0];
-		} else {
+	if (is_rdma_channel == false) {
 		if ((u64)le16_to_cpu(req->DataOffset) + length >
 		    get_rfc1002_len(work->request_buf)) {
 			pr_err("invalid write data offset %u, smb_len %u\n",
@@ -6501,10 +6502,8 @@ int smb2_write(struct ksmbd_work *work)
 			err = -EINVAL;
 			goto out;
 		}
-
 		data_buf = (char *)(((char *)&req->hdr.ProtocolId) +
 				    le16_to_cpu(req->DataOffset));
-		}
 
 		ksmbd_debug(SMB, "flags %u\n", le32_to_cpu(req->Flags));
 		if (le32_to_cpu(req->Flags) & SMB2_WRITEFLAG_WRITE_THROUGH)
@@ -6520,8 +6519,7 @@ int smb2_write(struct ksmbd_work *work)
 		/* read data from the client using rdma channel, and
 		 * write the data.
 		 */
-		nbytes = smb2_write_rdma_channel(work, req, fp, offset,
-						 le32_to_cpu(req->RemainingBytes),
+		nbytes = smb2_write_rdma_channel(work, req, fp, offset, length,
 						 writethrough);
 		if (nbytes < 0) {
 			err = (int)nbytes;

fs/ksmbd/smb_common.c

@@ -140,8 +140,10 @@ int ksmbd_verify_smb_message(struct ksmbd_work *work)
 
 	hdr = work->request_buf;
 	if (*(__le32 *)hdr->Protocol == SMB1_PROTO_NUMBER &&
-	    hdr->Command == SMB_COM_NEGOTIATE)
+	    hdr->Command == SMB_COM_NEGOTIATE) {
+		work->conn->outstanding_credits++;
 		return 0;
+	}
 
 	return -EINVAL;
 }

fs/ksmbd/smbacl.c

@@ -1261,6 +1261,7 @@ int smb_check_perm_dacl(struct ksmbd_conn *conn, struct path *path,
 					if (!access_bits)
 						access_bits =
 							SET_MINIMUM_RIGHTS;
+					posix_acl_release(posix_acls);
 					goto check_access_bits;
 				}
 			}

fs/ksmbd/transport_ipc.c

@@ -26,6 +26,7 @@
 #include "mgmt/ksmbd_ida.h"
 #include "connection.h"
 #include "transport_tcp.h"
+#include "transport_rdma.h"
 
 #define IPC_WAIT_TIMEOUT	(2 * HZ)
 
@@ -303,6 +304,8 @@ static int ipc_server_config_on_startup(struct ksmbd_startup_request *req)
 		init_smb2_max_trans_size(req->smb2_max_trans);
 	if (req->smb2_max_credits)
 		init_smb2_max_credits(req->smb2_max_credits);
+	if (req->smbd_max_io_size)
+		init_smbd_max_io_size(req->smbd_max_io_size);
 
 	ret = ksmbd_set_netbios_name(req->netbios_name);
 	ret |= ksmbd_set_server_string(req->server_string);

fs/ksmbd/transport_rdma.c

@@ -80,9 +80,7 @@ static int smb_direct_max_fragmented_recv_size = 1024 * 1024;
 /*  The maximum single-message size which can be received */
 static int smb_direct_max_receive_size = 8192;
 
-static int smb_direct_max_read_write_size = 524224;
-
-static int smb_direct_max_outstanding_rw_ops = 8;
+static int smb_direct_max_read_write_size = SMBD_DEFAULT_IOSIZE;
 
 static LIST_HEAD(smb_direct_device_list);
 static DEFINE_RWLOCK(smb_direct_device_lock);
@@ -147,18 +145,18 @@ struct smb_direct_transport {
 	atomic_t		send_credits;
 	spinlock_t		lock_new_recv_credits;
 	int			new_recv_credits;
-	atomic_t		rw_avail_ops;
+	int			max_rw_credits;
+	int			pages_per_rw_credit;
+	atomic_t		rw_credits;
 
 	wait_queue_head_t	wait_send_credits;
-	wait_queue_head_t	wait_rw_avail_ops;
+	wait_queue_head_t	wait_rw_credits;
 
 	mempool_t		*sendmsg_mempool;
 	struct kmem_cache	*sendmsg_cache;
 	mempool_t		*recvmsg_mempool;
 	struct kmem_cache	*recvmsg_cache;
 
-	wait_queue_head_t	wait_send_payload_pending;
-	atomic_t		send_payload_pending;
 	wait_queue_head_t	wait_send_pending;
 	atomic_t		send_pending;
 
@@ -208,12 +206,25 @@ struct smb_direct_recvmsg {
 struct smb_direct_rdma_rw_msg {
 	struct smb_direct_transport	*t;
 	struct ib_cqe		cqe;
+	int			status;
 	struct completion	*completion;
+	struct list_head	list;
 	struct rdma_rw_ctx	rw_ctx;
 	struct sg_table		sgt;
 	struct scatterlist	sg_list[];
 };
 
+void init_smbd_max_io_size(unsigned int sz)
+{
+	sz = clamp_val(sz, SMBD_MIN_IOSIZE, SMBD_MAX_IOSIZE);
+	smb_direct_max_read_write_size = sz;
+}
+
+unsigned int get_smbd_max_read_write_size(void)
+{
+	return smb_direct_max_read_write_size;
+}
+
 static inline int get_buf_page_count(void *buf, int size)
 {
 	return DIV_ROUND_UP((uintptr_t)buf + size, PAGE_SIZE) -
@@ -377,7 +388,7 @@ static struct smb_direct_transport *alloc_transport(struct rdma_cm_id *cm_id)
 	t->reassembly_queue_length = 0;
 	init_waitqueue_head(&t->wait_reassembly_queue);
 	init_waitqueue_head(&t->wait_send_credits);
-	init_waitqueue_head(&t->wait_rw_avail_ops);
+	init_waitqueue_head(&t->wait_rw_credits);
 
 	spin_lock_init(&t->receive_credit_lock);
 	spin_lock_init(&t->recvmsg_queue_lock);
@@ -386,8 +397,6 @@ static struct smb_direct_transport *alloc_transport(struct rdma_cm_id *cm_id)
 	spin_lock_init(&t->empty_recvmsg_queue_lock);
 	INIT_LIST_HEAD(&t->empty_recvmsg_queue);
 
-	init_waitqueue_head(&t->wait_send_payload_pending);
-	atomic_set(&t->send_payload_pending, 0);
 	init_waitqueue_head(&t->wait_send_pending);
 	atomic_set(&t->send_pending, 0);
 
@@ -417,8 +426,6 @@ static void free_transport(struct smb_direct_transport *t)
 	wake_up_interruptible(&t->wait_send_credits);
 
 	ksmbd_debug(RDMA, "wait for all send posted to IB to finish\n");
-	wait_event(t->wait_send_payload_pending,
-		   atomic_read(&t->send_payload_pending) == 0);
 	wait_event(t->wait_send_pending,
 		   atomic_read(&t->send_pending) == 0);
 
@@ -569,6 +576,7 @@ static void recv_done(struct ib_cq *cq, struct ib_wc *wc)
 		}
 		t->negotiation_requested = true;
 		t->full_packet_received = true;
+		t->status = SMB_DIRECT_CS_CONNECTED;
 		enqueue_reassembly(t, recvmsg, 0);
 		wake_up_interruptible(&t->wait_status);
 		break;
@@ -873,13 +881,8 @@ static void send_done(struct ib_cq *cq, struct ib_wc *wc)
 		smb_direct_disconnect_rdma_connection(t);
 	}
 
-	if (sendmsg->num_sge > 1) {
-		if (atomic_dec_and_test(&t->send_payload_pending))
-			wake_up(&t->wait_send_payload_pending);
-	} else {
 	if (atomic_dec_and_test(&t->send_pending))
 		wake_up(&t->wait_send_pending);
-	}
 
 	/* iterate and free the list of messages in reverse. the list's head
 	 * is invalid.
@@ -911,21 +914,12 @@ static int smb_direct_post_send(struct smb_direct_transport *t,
 {
 	int ret;
 
-	if (wr->num_sge > 1)
-		atomic_inc(&t->send_payload_pending);
-	else
 	atomic_inc(&t->send_pending);
-
 	ret = ib_post_send(t->qp, wr, NULL);
 	if (ret) {
 		pr_err("failed to post send: %d\n", ret);
-		if (wr->num_sge > 1) {
-			if (atomic_dec_and_test(&t->send_payload_pending))
-				wake_up(&t->wait_send_payload_pending);
-		} else {
 		if (atomic_dec_and_test(&t->send_pending))
 			wake_up(&t->wait_send_pending);
-		}
 		smb_direct_disconnect_rdma_connection(t);
 	}
 	return ret;
@@ -983,17 +977,18 @@ static int smb_direct_flush_send_list(struct smb_direct_transport *t,
 }
 
 static int wait_for_credits(struct smb_direct_transport *t,
-			    wait_queue_head_t *waitq, atomic_t *credits)
+			    wait_queue_head_t *waitq, atomic_t *total_credits,
+			    int needed)
 {
 	int ret;
 
 	do {
-		if (atomic_dec_return(credits) >= 0)
+		if (atomic_sub_return(needed, total_credits) >= 0)
 			return 0;
 
-		atomic_inc(credits);
+		atomic_add(needed, total_credits);
 		ret = wait_event_interruptible(*waitq,
-					       atomic_read(credits) > 0 ||
+					       atomic_read(total_credits) >= needed ||
 					       t->status != SMB_DIRECT_CS_CONNECTED);
 
 		if (t->status != SMB_DIRECT_CS_CONNECTED)
@@ -1015,7 +1010,19 @@ static int wait_for_send_credits(struct smb_direct_transport *t,
 			return ret;
 	}
 
-	return wait_for_credits(t, &t->wait_send_credits, &t->send_credits);
+	return wait_for_credits(t, &t->wait_send_credits, &t->send_credits, 1);
+}
+
+static int wait_for_rw_credits(struct smb_direct_transport *t, int credits)
+{
+	return wait_for_credits(t, &t->wait_rw_credits, &t->rw_credits, credits);
+}
+
+static int calc_rw_credits(struct smb_direct_transport *t,
+			   char *buf, unsigned int len)
+{
+	return DIV_ROUND_UP(get_buf_page_count(buf, len),
+			    t->pages_per_rw_credit);
 }
 
 static int smb_direct_create_header(struct smb_direct_transport *t,
@@ -1086,7 +1093,7 @@ static int get_sg_list(void *buf, int size, struct scatterlist *sg_list, int nen
 	int offset, len;
 	int i = 0;
 
-	if (nentries < get_buf_page_count(buf, size))
+	if (size <= 0 || nentries < get_buf_page_count(buf, size))
 		return -EINVAL;
 
 	offset = offset_in_page(buf);
@@ -1118,7 +1125,7 @@ static int get_mapped_sg_list(struct ib_device *device, void *buf, int size,
 	int npages;
 
 	npages = get_sg_list(buf, size, sg_list, nentries);
-	if (npages <= 0)
+	if (npages < 0)
 		return -EINVAL;
 	return ib_dma_map_sg(device, sg_list, npages, dir);
 }
@@ -1313,11 +1320,21 @@ static int smb_direct_writev(struct ksmbd_transport *t,
 	 * that means all the I/Os have been out and we are good to return
 	 */
 
-	wait_event(st->wait_send_payload_pending,
-		   atomic_read(&st->send_payload_pending) == 0);
+	wait_event(st->wait_send_pending,
+		   atomic_read(&st->send_pending) == 0);
 	return ret;
 }
 
+static void smb_direct_free_rdma_rw_msg(struct smb_direct_transport *t,
+					struct smb_direct_rdma_rw_msg *msg,
+					enum dma_data_direction dir)
+{
+	rdma_rw_ctx_destroy(&msg->rw_ctx, t->qp, t->qp->port,
+			    msg->sgt.sgl, msg->sgt.nents, dir);
+	sg_free_table_chained(&msg->sgt, SG_CHUNK_SIZE);
+	kfree(msg);
+}
+
 static void read_write_done(struct ib_cq *cq, struct ib_wc *wc,
 			    enum dma_data_direction dir)
 {
@@ -1326,19 +1343,14 @@ static void read_write_done(struct ib_cq *cq, struct ib_wc *wc,
 	struct smb_direct_transport *t = msg->t;
 
 	if (wc->status != IB_WC_SUCCESS) {
+		msg->status = -EIO;
 		pr_err("read/write error. opcode = %d, status = %s(%d)\n",
 		       wc->opcode, ib_wc_status_msg(wc->status), wc->status);
+		if (wc->status != IB_WC_WR_FLUSH_ERR)
 			smb_direct_disconnect_rdma_connection(t);
 	}
 
-	if (atomic_inc_return(&t->rw_avail_ops) > 0)
-		wake_up(&t->wait_rw_avail_ops);
-
-	rdma_rw_ctx_destroy(&msg->rw_ctx, t->qp, t->qp->port,
-			    msg->sg_list, msg->sgt.nents, dir);
-	sg_free_table_chained(&msg->sgt, SG_CHUNK_SIZE);
 	complete(msg->completion);
-	kfree(msg);
 }
 
 static void read_done(struct ib_cq *cq, struct ib_wc *wc)
@@ -1351,94 +1363,141 @@ static void write_done(struct ib_cq *cq, struct ib_wc *wc)
 	read_write_done(cq, wc, DMA_TO_DEVICE);
 }
 
-static int smb_direct_rdma_xmit(struct smb_direct_transport *t, void *buf,
-				int buf_len, u32 remote_key, u64 remote_offset,
-				u32 remote_len, bool is_read)
+static int smb_direct_rdma_xmit(struct smb_direct_transport *t,
+				void *buf, int buf_len,
+				struct smb2_buffer_desc_v1 *desc,
+				unsigned int desc_len,
+				bool is_read)
 {
-	struct smb_direct_rdma_rw_msg *msg;
-	int ret;
+	struct smb_direct_rdma_rw_msg *msg, *next_msg;
+	int i, ret;
 	DECLARE_COMPLETION_ONSTACK(completion);
-	struct ib_send_wr *first_wr = NULL;
+	struct ib_send_wr *first_wr;
+	LIST_HEAD(msg_list);
+	char *desc_buf;
+	int credits_needed;
+	unsigned int desc_buf_len;
+	size_t total_length = 0;
+
+	if (t->status != SMB_DIRECT_CS_CONNECTED)
+		return -ENOTCONN;
+
+	/* calculate needed credits */
+	credits_needed = 0;
+	desc_buf = buf;
+	for (i = 0; i < desc_len / sizeof(*desc); i++) {
+		desc_buf_len = le32_to_cpu(desc[i].length);
+
+		credits_needed += calc_rw_credits(t, desc_buf, desc_buf_len);
+		desc_buf += desc_buf_len;
+		total_length += desc_buf_len;
+		if (desc_buf_len == 0 || total_length > buf_len ||
+		    total_length > t->max_rdma_rw_size)
+			return -EINVAL;
+	}
+
+	ksmbd_debug(RDMA, "RDMA %s, len %#x, needed credits %#x\n",
+		    is_read ? "read" : "write", buf_len, credits_needed);
 
-	ret = wait_for_credits(t, &t->wait_rw_avail_ops, &t->rw_avail_ops);
+	ret = wait_for_rw_credits(t, credits_needed);
 	if (ret < 0)
 		return ret;
 
-	/* TODO: mempool */
-	msg = kmalloc(offsetof(struct smb_direct_rdma_rw_msg, sg_list) +
+	/* build rdma_rw_ctx for each descriptor */
+	desc_buf = buf;
+	for (i = 0; i < desc_len / sizeof(*desc); i++) {
+		msg = kzalloc(offsetof(struct smb_direct_rdma_rw_msg, sg_list) +
 			      sizeof(struct scatterlist) * SG_CHUNK_SIZE, GFP_KERNEL);
 		if (!msg) {
-		atomic_inc(&t->rw_avail_ops);
-		return -ENOMEM;
+			ret = -ENOMEM;
+			goto out;
 		}
 
+		desc_buf_len = le32_to_cpu(desc[i].length);
+
+		msg->t = t;
+		msg->cqe.done = is_read ? read_done : write_done;
+		msg->completion = &completion;
+
 		msg->sgt.sgl = &msg->sg_list[0];
 		ret = sg_alloc_table_chained(&msg->sgt,
-				     get_buf_page_count(buf, buf_len),
+					     get_buf_page_count(desc_buf, desc_buf_len),
 					     msg->sg_list, SG_CHUNK_SIZE);
 		if (ret) {
-		atomic_inc(&t->rw_avail_ops);
 			kfree(msg);
-		return -ENOMEM;
+			ret = -ENOMEM;
+			goto out;
 		}
 
-	ret = get_sg_list(buf, buf_len, msg->sgt.sgl, msg->sgt.orig_nents);
-	if (ret <= 0) {
-		pr_err("failed to get pages\n");
-		goto err;
+		ret = get_sg_list(desc_buf, desc_buf_len,
+				  msg->sgt.sgl, msg->sgt.orig_nents);
+		if (ret < 0) {
+			sg_free_table_chained(&msg->sgt, SG_CHUNK_SIZE);
+			kfree(msg);
+			goto out;
 		}
 
 		ret = rdma_rw_ctx_init(&msg->rw_ctx, t->qp, t->qp->port,
-			       msg->sg_list, get_buf_page_count(buf, buf_len),
-			       0, remote_offset, remote_key,
+				       msg->sgt.sgl,
+				       get_buf_page_count(desc_buf, desc_buf_len),
+				       0,
+				       le64_to_cpu(desc[i].offset),
+				       le32_to_cpu(desc[i].token),
 				       is_read ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
 		if (ret < 0) {
 			pr_err("failed to init rdma_rw_ctx: %d\n", ret);
-		goto err;
+			sg_free_table_chained(&msg->sgt, SG_CHUNK_SIZE);
+			kfree(msg);
+			goto out;
 		}
 
-	msg->t = t;
-	msg->cqe.done = is_read ? read_done : write_done;
-	msg->completion = &completion;
+		list_add_tail(&msg->list, &msg_list);
+		desc_buf += desc_buf_len;
+	}
+
+	/* concatenate work requests of rdma_rw_ctxs */
+	first_wr = NULL;
+	list_for_each_entry_reverse(msg, &msg_list, list) {
 		first_wr = rdma_rw_ctx_wrs(&msg->rw_ctx, t->qp, t->qp->port,
-				   &msg->cqe, NULL);
+					   &msg->cqe, first_wr);
+	}
 
 	ret = ib_post_send(t->qp, first_wr, NULL);
 	if (ret) {
-		pr_err("failed to post send wr: %d\n", ret);
-		goto err;
+		pr_err("failed to post send wr for RDMA R/W: %d\n", ret);
+		goto out;
 	}
 
+	msg = list_last_entry(&msg_list, struct smb_direct_rdma_rw_msg, list);
 	wait_for_completion(&completion);
-	return 0;
-
-err:
-	atomic_inc(&t->rw_avail_ops);
-	if (first_wr)
-		rdma_rw_ctx_destroy(&msg->rw_ctx, t->qp, t->qp->port,
-				    msg->sg_list, msg->sgt.nents,
+	ret = msg->status;
+out:
+	list_for_each_entry_safe(msg, next_msg, &msg_list, list) {
+		list_del(&msg->list);
+		smb_direct_free_rdma_rw_msg(t, msg,
 					    is_read ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-	sg_free_table_chained(&msg->sgt, SG_CHUNK_SIZE);
-	kfree(msg);
+	}
+	atomic_add(credits_needed, &t->rw_credits);
+	wake_up(&t->wait_rw_credits);
 	return ret;
 }
 
-static int smb_direct_rdma_write(struct ksmbd_transport *t, void *buf,
-				 unsigned int buflen, u32 remote_key,
-				 u64 remote_offset, u32 remote_len)
+static int smb_direct_rdma_write(struct ksmbd_transport *t,
+				 void *buf, unsigned int buflen,
+				 struct smb2_buffer_desc_v1 *desc,
+				 unsigned int desc_len)
 {
 	return smb_direct_rdma_xmit(smb_trans_direct_transfort(t), buf, buflen,
-				    remote_key, remote_offset,
-				    remote_len, false);
+				    desc, desc_len, false);
 }
 
-static int smb_direct_rdma_read(struct ksmbd_transport *t, void *buf,
-				unsigned int buflen, u32 remote_key,
-				u64 remote_offset, u32 remote_len)
+static int smb_direct_rdma_read(struct ksmbd_transport *t,
+				void *buf, unsigned int buflen,
+				struct smb2_buffer_desc_v1 *desc,
+				unsigned int desc_len)
 {
 	return smb_direct_rdma_xmit(smb_trans_direct_transfort(t), buf, buflen,
-				    remote_key, remote_offset,
-				    remote_len, true);
+				    desc, desc_len, true);
 }
 
 static void smb_direct_disconnect(struct ksmbd_transport *t)
@@ -1638,41 +1697,57 @@ static int smb_direct_prepare_negotiation(struct smb_direct_transport *t)
 	return ret;
 }
 
+static unsigned int smb_direct_get_max_fr_pages(struct smb_direct_transport *t)
+{
+	return min_t(unsigned int,
+		     t->cm_id->device->attrs.max_fast_reg_page_list_len,
+		     256);
+}
+
 static int smb_direct_init_params(struct smb_direct_transport *t,
 				  struct ib_qp_cap *cap)
 {
 	struct ib_device *device = t->cm_id->device;
-	int max_send_sges, max_pages, max_rw_wrs, max_send_wrs;
+	int max_send_sges, max_rw_wrs, max_send_wrs;
+	unsigned int max_sge_per_wr, wrs_per_credit;
 
-	/* need 2 more sge. because a SMB_DIRECT header will be mapped,
-	 * and maybe a send buffer could be not page aligned.
+	/* need 3 more sge. because a SMB_DIRECT header, SMB2 header,
+	 * SMB2 response could be mapped.
 	 */
 	t->max_send_size = smb_direct_max_send_size;
-	max_send_sges = DIV_ROUND_UP(t->max_send_size, PAGE_SIZE) + 2;
+	max_send_sges = DIV_ROUND_UP(t->max_send_size, PAGE_SIZE) + 3;
 	if (max_send_sges > SMB_DIRECT_MAX_SEND_SGES) {
 		pr_err("max_send_size %d is too large\n", t->max_send_size);
 		return -EINVAL;
 	}
 
-	/*
-	 * allow smb_direct_max_outstanding_rw_ops of in-flight RDMA
-	 * read/writes. HCA guarantees at least max_send_sge of sges for
-	 * a RDMA read/write work request, and if memory registration is used,
-	 * we need reg_mr, local_inv wrs for each read/write.
+	/* Calculate the number of work requests for RDMA R/W.
+	 * The maximum number of pages which can be registered
+	 * with one Memory region can be transferred with one
+	 * R/W credit. And at least 4 work requests for each credit
+	 * are needed for MR registration, RDMA R/W, local & remote
+	 * MR invalidation.
 	 */
 	t->max_rdma_rw_size = smb_direct_max_read_write_size;
-	max_pages = DIV_ROUND_UP(t->max_rdma_rw_size, PAGE_SIZE) + 1;
-	max_rw_wrs = DIV_ROUND_UP(max_pages, SMB_DIRECT_MAX_SEND_SGES);
-	max_rw_wrs += rdma_rw_mr_factor(device, t->cm_id->port_num,
-			max_pages) * 2;
-	max_rw_wrs *= smb_direct_max_outstanding_rw_ops;
+	t->pages_per_rw_credit = smb_direct_get_max_fr_pages(t);
+	t->max_rw_credits = DIV_ROUND_UP(t->max_rdma_rw_size,
+					 (t->pages_per_rw_credit - 1) *
+					 PAGE_SIZE);
+
+	max_sge_per_wr = min_t(unsigned int, device->attrs.max_send_sge,
+			       device->attrs.max_sge_rd);
+	max_sge_per_wr = max_t(unsigned int, max_sge_per_wr,
+			       max_send_sges);
+	wrs_per_credit = max_t(unsigned int, 4,
+			       DIV_ROUND_UP(t->pages_per_rw_credit,
+					    max_sge_per_wr) + 1);
+	max_rw_wrs = t->max_rw_credits * wrs_per_credit;
 
 	max_send_wrs = smb_direct_send_credit_target + max_rw_wrs;
 	if (max_send_wrs > device->attrs.max_cqe ||
 	    max_send_wrs > device->attrs.max_qp_wr) {
-		pr_err("consider lowering send_credit_target = %d, or max_outstanding_rw_ops = %d\n",
-		       smb_direct_send_credit_target,
-		       smb_direct_max_outstanding_rw_ops);
+		pr_err("consider lowering send_credit_target = %d\n",
+		       smb_direct_send_credit_target);
 		pr_err("Possible CQE overrun, device reporting max_cqe %d max_qp_wr %d\n",
 		       device->attrs.max_cqe, device->attrs.max_qp_wr);
 		return -EINVAL;
@@ -1687,11 +1762,6 @@ static int smb_direct_init_params(struct smb_direct_transport *t,
 		return -EINVAL;
 	}
 
-	if (device->attrs.max_send_sge < SMB_DIRECT_MAX_SEND_SGES) {
-		pr_err("warning: device max_send_sge = %d too small\n",
-		       device->attrs.max_send_sge);
-		return -EINVAL;
-	}
 	if (device->attrs.max_recv_sge < SMB_DIRECT_MAX_RECV_SGES) {
 		pr_err("warning: device max_recv_sge = %d too small\n",
 		       device->attrs.max_recv_sge);
@@ -1707,7 +1777,7 @@ static int smb_direct_init_params(struct smb_direct_transport *t,
 
 	t->send_credit_target = smb_direct_send_credit_target;
 	atomic_set(&t->send_credits, 0);
-	atomic_set(&t->rw_avail_ops, smb_direct_max_outstanding_rw_ops);
+	atomic_set(&t->rw_credits, t->max_rw_credits);
 
 	t->max_send_size = smb_direct_max_send_size;
 	t->max_recv_size = smb_direct_max_receive_size;
@@ -1715,12 +1785,10 @@ static int smb_direct_init_params(struct smb_direct_transport *t,
 
 	cap->max_send_wr = max_send_wrs;
 	cap->max_recv_wr = t->recv_credit_max;
-	cap->max_send_sge = SMB_DIRECT_MAX_SEND_SGES;
+	cap->max_send_sge = max_sge_per_wr;
 	cap->max_recv_sge = SMB_DIRECT_MAX_RECV_SGES;
 	cap->max_inline_data = 0;
-	cap->max_rdma_ctxs =
-		rdma_rw_mr_factor(device, t->cm_id->port_num, max_pages) *
-		smb_direct_max_outstanding_rw_ops;
+	cap->max_rdma_ctxs = t->max_rw_credits;
 	return 0;
 }
 
@@ -1813,7 +1881,8 @@ static int smb_direct_create_qpair(struct smb_direct_transport *t,
 	}
 
 	t->send_cq = ib_alloc_cq(t->cm_id->device, t,
-				 t->send_credit_target, 0, IB_POLL_WORKQUEUE);
+				 smb_direct_send_credit_target + cap->max_rdma_ctxs,
+				 0, IB_POLL_WORKQUEUE);
 	if (IS_ERR(t->send_cq)) {
 		pr_err("Can't create RDMA send CQ\n");
 		ret = PTR_ERR(t->send_cq);
@@ -1822,8 +1891,7 @@ static int smb_direct_create_qpair(struct smb_direct_transport *t,
 	}
 
 	t->recv_cq = ib_alloc_cq(t->cm_id->device, t,
-				 cap->max_send_wr + cap->max_rdma_ctxs,
-				 0, IB_POLL_WORKQUEUE);
+				 t->recv_credit_max, 0, IB_POLL_WORKQUEUE);
 	if (IS_ERR(t->recv_cq)) {
 		pr_err("Can't create RDMA recv CQ\n");
 		ret = PTR_ERR(t->recv_cq);
@@ -1852,17 +1920,12 @@ static int smb_direct_create_qpair(struct smb_direct_transport *t,
 
 	pages_per_rw = DIV_ROUND_UP(t->max_rdma_rw_size, PAGE_SIZE) + 1;
 	if (pages_per_rw > t->cm_id->device->attrs.max_sgl_rd) {
-		int pages_per_mr, mr_count;
-
-		pages_per_mr = min_t(int, pages_per_rw,
-				     t->cm_id->device->attrs.max_fast_reg_page_list_len);
-		mr_count = DIV_ROUND_UP(pages_per_rw, pages_per_mr) *
-			atomic_read(&t->rw_avail_ops);
-		ret = ib_mr_pool_init(t->qp, &t->qp->rdma_mrs, mr_count,
-				      IB_MR_TYPE_MEM_REG, pages_per_mr, 0);
+		ret = ib_mr_pool_init(t->qp, &t->qp->rdma_mrs,
+				      t->max_rw_credits, IB_MR_TYPE_MEM_REG,
+				      t->pages_per_rw_credit, 0);
 		if (ret) {
 			pr_err("failed to init mr pool count %d pages %d\n",
-			       mr_count, pages_per_mr);
+			       t->max_rw_credits, t->pages_per_rw_credit);
 			goto err;
 		}
 	}

fs/ksmbd/transport_rdma.h

@@ -7,6 +7,10 @@
 #ifndef __KSMBD_TRANSPORT_RDMA_H__
 #define __KSMBD_TRANSPORT_RDMA_H__
 
+#define SMBD_DEFAULT_IOSIZE (8 * 1024 * 1024)
+#define SMBD_MIN_IOSIZE (512 * 1024)
+#define SMBD_MAX_IOSIZE (16 * 1024 * 1024)
+
 /* SMB DIRECT negotiation request packet [MS-SMBD] 2.2.1 */
 struct smb_direct_negotiate_req {
 	__le16 min_version;
@@ -52,10 +56,14 @@ struct smb_direct_data_transfer {
 int ksmbd_rdma_init(void);
 void ksmbd_rdma_destroy(void);
 bool ksmbd_rdma_capable_netdev(struct net_device *netdev);
+void init_smbd_max_io_size(unsigned int sz);
+unsigned int get_smbd_max_read_write_size(void);
 #else
 static inline int ksmbd_rdma_init(void) { return 0; }
 static inline int ksmbd_rdma_destroy(void) { return 0; }
 static inline bool ksmbd_rdma_capable_netdev(struct net_device *netdev) { return false; }
+static inline void init_smbd_max_io_size(unsigned int sz) { }
+static inline unsigned int get_smbd_max_read_write_size(void) { return 0; }
 #endif
 
 #endif /* __KSMBD_TRANSPORT_RDMA_H__ */