* [Patch v4 01/22] CIFS: SMBD: Add SMBDirect protocol initial values and constants
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 02/22] CIFS: SMBD: Establish SMBDirect connection Long Li
` (14 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
To prepare for protocol implementation, add constants and user-configurable
values in the SMBDirect protocol.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smbdirect.c | 77 +++++++++++++++++++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 21 +++++++++++++++
2 files changed, 98 insertions(+)
create mode 100644 fs/cifs/smbdirect.c
create mode 100644 fs/cifs/smbdirect.h
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
new file mode 100644
index 0000000..d3c16f8
--- /dev/null
+++ b/fs/cifs/smbdirect.c
@@ -0,0 +1,77 @@
+/*
+ * Copyright (C) 2017, Microsoft Corporation.
+ *
+ * Author(s): Long Li <longli@microsoft.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
+ * the GNU General Public License for more details.
+ */
+#include "smbdirect.h"
+
+/* SMBD version number */
+#define SMBD_V1 0x0100
+
+/* Port numbers for SMBD transport */
+#define SMB_PORT 445
+#define SMBD_PORT 5445
+
+/* Address lookup and resolve timeout in ms */
+#define RDMA_RESOLVE_TIMEOUT 5000
+
+/* SMBD negotiation timeout in seconds */
+#define SMBD_NEGOTIATE_TIMEOUT 120
+
+/* SMBD minimum receive size and fragmented sized defined in [MS-SMBD] */
+#define SMBD_MIN_RECEIVE_SIZE 128
+#define SMBD_MIN_FRAGMENTED_SIZE 131072
+
+/*
+ * Default maximum number of RDMA read/write outstanding on this connection
+ * This value is possibly decreased during QP creation on hardware limit
+ */
+#define SMBD_CM_RESPONDER_RESOURCES 32
+
+/* Maximum number of retries on data transfer operations */
+#define SMBD_CM_RETRY 6
+/* No need to retry on Receiver Not Ready since SMBD manages credits */
+#define SMBD_CM_RNR_RETRY 0
+
+/*
+ * User configurable initial values per SMBD transport connection
+ * as defined in [MS-SMBD] 3.1.1.1
+ * Those may change after a SMBD negotiation
+ */
+/* The local peer's maximum number of credits to grant to the peer */
+int smbd_receive_credit_max = 255;
+
+/* The remote peer's credit request of local peer */
+int smbd_send_credit_target = 255;
+
+/* The maximum single message size can be sent to remote peer */
+int smbd_max_send_size = 1364;
+
+/* The maximum fragmented upper-layer payload receive size supported */
+int smbd_max_fragmented_recv_size = 1024 * 1024;
+
+/* The maximum single-message size which can be received */
+int smbd_max_receive_size = 8192;
+
+/* The timeout to initiate send of a keepalive message on idle */
+int smbd_keep_alive_interval = 120;
+
+/*
+ * User configurable initial values for RDMA transport
+ * The actual values used may be lower and are limited to hardware capabilities
+ */
+/* Default maximum number of SGEs in a RDMA write/read */
+int smbd_max_frmr_depth = 2048;
+
+/* If payload is less than this byte, use RDMA send/recv not read/write */
+int rdma_readwrite_threshold = 4096;
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
new file mode 100644
index 0000000..c55f28b
--- /dev/null
+++ b/fs/cifs/smbdirect.h
@@ -0,0 +1,21 @@
+/*
+ * Copyright (C) 2017, Microsoft Corporation.
+ *
+ * Author(s): Long Li <longli@microsoft.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
+ * the GNU General Public License for more details.
+ */
+#ifndef _SMBDIRECT_H
+#define _SMBDIRECT_H
+
+/* Default maximum number of SGEs in a RDMA send/recv */
+#define SMBDIRECT_MAX_SGE 16
+#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 02/22] CIFS: SMBD: Establish SMBDirect connection
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
2017-10-02 2:30 ` [Patch v4 01/22] CIFS: SMBD: Add SMBDirect protocol initial values and constants Long Li
@ 2017-10-02 2:30 ` Long Li
[not found] ` <20171002023030.3582-3-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 03/22] CIFS: SMBD: export protocol initial values Long Li
` (13 subsequent siblings)
15 siblings, 1 reply; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Add code to implement the core functions to establish a SMBDirect connection.
1. Establish an RDMA connection to SMB server.
2. Negotiate and setup SMBDirect protocol.
3. Implement idle connection timer and credit management.
Add to Makefile.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/Makefile | 2 +-
fs/cifs/smbdirect.c | 1600 +++++++++++++++++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 229 ++++++++
3 files changed, 1830 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/Makefile b/fs/cifs/Makefile
index 5e853a3..bb54662b 100644
--- a/fs/cifs/Makefile
+++ b/fs/cifs/Makefile
@@ -8,7 +8,7 @@ cifs-y := cifsfs.o cifssmb.o cifs_debug.o connect.o dir.o file.o inode.o \
cifs_unicode.o nterr.o cifsencrypt.o \
readdir.o ioctl.o sess.o export.o smb1ops.o winucase.o \
smb2ops.o smb2maperror.o smb2transport.o \
- smb2misc.o smb2pdu.o smb2inode.o smb2file.o
+ smb2misc.o smb2pdu.o smb2inode.o smb2file.o smbdirect.o
cifs-$(CONFIG_CIFS_XATTR) += xattr.o
cifs-$(CONFIG_CIFS_ACL) += cifsacl.o
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index d3c16f8..e8f976f 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -13,7 +13,35 @@
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
* the GNU General Public License for more details.
*/
+#include <linux/module.h>
#include "smbdirect.h"
+#include "cifs_debug.h"
+#include <asm/msr.h>
+
+static struct smbd_response *get_empty_queue_buffer(
+ struct smbd_connection *info);
+static struct smbd_response *get_receive_buffer(
+ struct smbd_connection *info);
+static void put_receive_buffer(
+ struct smbd_connection *info,
+ struct smbd_response *response,
+ bool lock);
+static int allocate_receive_buffers(struct smbd_connection *info, int num_buf);
+static void destroy_receive_buffers(struct smbd_connection *info);
+
+static void put_empty_packet(
+ struct smbd_connection *info, struct smbd_response *response);
+static void enqueue_reassembly(
+ struct smbd_connection *info,
+ struct smbd_response *response, int data_length);
+static struct smbd_response *_get_first_reassembly(
+ struct smbd_connection *info);
+
+static int smbd_post_recv(
+ struct smbd_connection *info,
+ struct smbd_response *response);
+
+static int smbd_post_send_empty(struct smbd_connection *info);
/* SMBD version number */
#define SMBD_V1 0x0100
@@ -75,3 +103,1575 @@ int smbd_max_frmr_depth = 2048;
/* If payload is less than this byte, use RDMA send/recv not read/write */
int rdma_readwrite_threshold = 4096;
+
+/* Transport logging functions
+ * Logging are defined as classes. They can be OR'ed to define the actual
+ * logging level via module parameter smbd_logging_class
+ * e.g. cifs.smbd_logging_class=0x500 will log all log_rdma_recv() and
+ * log_rdma_event()
+ */
+#define LOG_OUTGOING 0x1
+#define LOG_INCOMING 0x2
+#define LOG_READ 0x4
+#define LOG_WRITE 0x8
+#define LOG_RDMA_SEND 0x10
+#define LOG_RDMA_RECV 0x20
+#define LOG_KEEP_ALIVE 0x40
+#define LOG_RDMA_EVENT 0x80
+#define LOG_RDMA_MR 0x100
+static unsigned int smbd_logging_class = 0;
+module_param(smbd_logging_class, uint, 0644);
+MODULE_PARM_DESC(smbd_logging_class,
+ "Logging class for SMBD transport 0x0 to 0x100");
+
+#define ERR 0x0
+#define INFO 0x1
+static unsigned int smbd_logging_level = ERR;
+module_param(smbd_logging_level, uint, 0644);
+MODULE_PARM_DESC(smbd_logging_level,
+ "Logging level for SMBD transport, 0 (default): error, 1: info");
+
+#define log_rdma(level, class, fmt, args...) \
+do { \
+ if (level <= smbd_logging_level || class & smbd_logging_class) \
+ cifs_dbg(VFS, "%s:%d " fmt, __func__, __LINE__, ##args);\
+} while (0)
+
+#define log_outgoing(level, fmt, args...) \
+ log_rdma(level, LOG_OUTGOING, fmt, ##args)
+#define log_incoming(level, fmt, args...) \
+ log_rdma(level, LOG_INCOMING, fmt, ##args)
+#define log_read(level, fmt, args...) log_rdma(level, LOG_READ, fmt, ##args)
+#define log_write(level, fmt, args...) log_rdma(level, LOG_WRITE, fmt, ##args)
+#define log_rdma_send(level, fmt, args...) \
+ log_rdma(level, LOG_RDMA_SEND, fmt, ##args)
+#define log_rdma_recv(level, fmt, args...) \
+ log_rdma(level, LOG_RDMA_RECV, fmt, ##args)
+#define log_keep_alive(level, fmt, args...) \
+ log_rdma(level, LOG_KEEP_ALIVE, fmt, ##args)
+#define log_rdma_event(level, fmt, args...) \
+ log_rdma(level, LOG_RDMA_EVENT, fmt, ##args)
+#define log_rdma_mr(level, fmt, args...) \
+ log_rdma(level, LOG_RDMA_MR, fmt, ##args)
+
+void profiling_display_histogram(
+ struct seq_file *m, unsigned long long array[])
+{
+ int i;
+ unsigned long long total = 0;
+
+ for (i = 0; i < SIZE_LONG_LONG; i++)
+ total += array[i];
+
+ for (i = 0; i < SIZE_LONG_LONG; i++)
+ if (array[i])
+ seq_printf(m, "%d:%llu ", i, array[i] * 10000 / total);
+}
+
+static inline void profiling_add_histogram(
+ unsigned long long value, unsigned long long array[])
+{
+ array[__builtin_clzll(value)]++;
+}
+
+/*
+ * Destroy the transport and related RDMA and memory resources
+ * Need to go through all the pending counters and make sure on one is using
+ * the transport while it is destroyed
+ */
+static void smbd_destroy_rdma_work(struct work_struct *work)
+{
+ struct smbd_response *response;
+ struct smbd_connection *info =
+ container_of(work, struct smbd_connection, destroy_work);
+ unsigned long flags;
+
+ log_rdma_event(INFO, "destroying qp\n");
+ ib_drain_qp(info->id->qp);
+ rdma_destroy_qp(info->id);
+
+ /* Unblock all I/O waiting on the send queue */
+ wake_up_interruptible_all(&info->wait_send_queue);
+
+ log_rdma_event(INFO, "cancelling idle timer\n");
+ cancel_delayed_work_sync(&info->idle_timer_work);
+ log_rdma_event(INFO, "cancelling send immediate work\n");
+ cancel_delayed_work_sync(&info->send_immediate_work);
+
+ log_rdma_event(INFO, "wait for all recv to finish\n");
+ wake_up_interruptible(&info->wait_reassembly_queue);
+
+ log_rdma_event(INFO, "wait for all send posted to IB to finish\n");
+ wait_event(info->wait_send_pending,
+ atomic_read(&info->send_pending) == 0);
+ wait_event(info->wait_send_payload_pending,
+ atomic_read(&info->send_payload_pending) == 0);
+
+ /* It's not posssible for upper layer to get to reassembly */
+ log_rdma_event(INFO, "drain the reassembly queue\n");
+ do {
+ spin_lock_irqsave(&info->reassembly_queue_lock, flags);
+ response = _get_first_reassembly(info);
+ if (response) {
+ list_del(&response->list);
+ spin_unlock_irqrestore(
+ &info->reassembly_queue_lock, flags);
+ put_receive_buffer(info, response, true);
+ }
+ } while (response);
+ spin_unlock_irqrestore(&info->reassembly_queue_lock, flags);
+ info->reassembly_data_length = 0;
+
+ log_rdma_event(INFO, "free receive buffers\n");
+ wait_event(info->wait_receive_queues,
+ info->count_receive_queue + info->count_empty_packet_queue
+ == info->receive_credit_max);
+ destroy_receive_buffers(info);
+
+ ib_free_cq(info->send_cq);
+ ib_free_cq(info->recv_cq);
+ ib_dealloc_pd(info->pd);
+ rdma_destroy_id(info->id);
+
+ /* free mempools */
+ mempool_destroy(info->request_mempool);
+ kmem_cache_destroy(info->request_cache);
+
+ mempool_destroy(info->response_mempool);
+ kmem_cache_destroy(info->response_cache);
+
+ info->transport_status = SMBD_DESTROYED;
+ wake_up_all(&info->wait_destroy);
+}
+
+static int smbd_process_disconnected(struct smbd_connection *info)
+{
+// queue_work(info->workqueue, &info->destroy_work);
+ schedule_work(&info->destroy_work);
+ return 0;
+}
+
+static void smbd_disconnect_rdma_work(struct work_struct *work)
+{
+ struct smbd_connection *info =
+ container_of(work, struct smbd_connection, disconnect_work);
+
+ if (info->transport_status == SMBD_CONNECTED) {
+ info->transport_status = SMBD_DISCONNECTING;
+ rdma_disconnect(info->id);
+ }
+}
+
+static void smbd_disconnect_rdma_connection(struct smbd_connection *info)
+{
+ queue_work(info->workqueue, &info->disconnect_work);
+}
+
+/* Upcall from RDMA CM */
+static int smbd_conn_upcall(
+ struct rdma_cm_id *id, struct rdma_cm_event *event)
+{
+ struct smbd_connection *info = id->context;
+
+ log_rdma_event(INFO, "event=%d status=%d\n",
+ event->event, event->status);
+
+ switch (event->event) {
+ case RDMA_CM_EVENT_ADDR_RESOLVED:
+ case RDMA_CM_EVENT_ROUTE_RESOLVED:
+ info->ri_rc = 0;
+ complete(&info->ri_done);
+ break;
+
+ case RDMA_CM_EVENT_ADDR_ERROR:
+ info->ri_rc = -EHOSTUNREACH;
+ complete(&info->ri_done);
+ break;
+
+ case RDMA_CM_EVENT_ROUTE_ERROR:
+ info->ri_rc = -ENETUNREACH;
+ complete(&info->ri_done);
+ break;
+
+ case RDMA_CM_EVENT_ESTABLISHED:
+ log_rdma_event(INFO, "connected event=%d\n", event->event);
+ info->transport_status = SMBD_CONNECTED;
+ wake_up_interruptible(&info->conn_wait);
+ break;
+
+ case RDMA_CM_EVENT_CONNECT_ERROR:
+ case RDMA_CM_EVENT_UNREACHABLE:
+ case RDMA_CM_EVENT_REJECTED:
+ log_rdma_event(INFO, "connecting failed event=%d\n", event->event);
+ info->transport_status = SMBD_DISCONNECTED;
+ wake_up_interruptible(&info->conn_wait);
+ break;
+
+ case RDMA_CM_EVENT_DEVICE_REMOVAL:
+ case RDMA_CM_EVENT_DISCONNECTED:
+ /* This happenes when we fail the negotiation */
+ if (info->transport_status == SMBD_NEGOTIATE_FAILED) {
+ info->transport_status = SMBD_DISCONNECTED;
+ wake_up(&info->conn_wait);
+ break;
+ }
+
+ info->transport_status = SMBD_DISCONNECTED;
+ smbd_process_disconnected(info);
+ break;
+
+ default:
+ break;
+ }
+
+ return 0;
+}
+
+/* Upcall from RDMA QP */
+static void
+smbd_qp_async_error_upcall(struct ib_event *event, void *context)
+{
+ struct smbd_connection *info = context;
+
+ log_rdma_event(ERR, "%s on device %s info %p\n",
+ ib_event_msg(event->event), event->device->name, info);
+
+ switch (event->event) {
+ case IB_EVENT_CQ_ERR:
+ case IB_EVENT_QP_FATAL:
+ smbd_disconnect_rdma_connection(info);
+
+ default:
+ break;
+ }
+}
+
+static inline void *smbd_request_payload(struct smbd_request *request)
+{
+ return (void *)request->packet;
+}
+
+static inline void *smbd_response_payload(struct smbd_response *response)
+{
+ return (void *)response->packet;
+}
+
+/* Called when a RDMA send is done */
+static void send_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ int i;
+ struct smbd_request *request =
+ container_of(wc->wr_cqe, struct smbd_request, cqe);
+
+ log_rdma_send(INFO, "smbd_request %p completed wc->status=%d\n",
+ request, wc->status);
+
+ if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_SEND) {
+ log_rdma_send(ERR, "wc->status=%d wc->opcode=%d\n",
+ wc->status, wc->opcode);
+ smbd_disconnect_rdma_connection(request->info);
+ }
+
+ for (i = 0; i < request->num_sge; i++)
+ ib_dma_unmap_single(request->info->id->device,
+ request->sge[i].addr,
+ request->sge[i].length,
+ DMA_TO_DEVICE);
+
+ if (request->has_payload) {
+ if (atomic_dec_and_test(&request->info->send_payload_pending))
+ wake_up(&request->info->wait_send_payload_pending);
+ } else {
+ if (atomic_dec_and_test(&request->info->send_pending))
+ wake_up(&request->info->wait_send_pending);
+ }
+
+ mempool_free(request, request->info->request_mempool);
+}
+
+static void dump_smbd_negotiate_resp(struct smbd_negotiate_resp *resp)
+{
+ log_rdma_event(INFO, "resp message min_version %u max_version %u "
+ "negotiated_version %u credits_requested %u "
+ "credits_granted %u status %u max_readwrite_size %u "
+ "preferred_send_size %u max_receive_size %u "
+ "max_fragmented_size %u\n",
+ resp->min_version, resp->max_version, resp->negotiated_version,
+ resp->credits_requested, resp->credits_granted, resp->status,
+ resp->max_readwrite_size, resp->preferred_send_size,
+ resp->max_receive_size, resp->max_fragmented_size);
+}
+
+/*
+ * Process a negotiation response message, according to [MS-SMBD]3.1.5.7
+ * response, packet_length: the negotiation response message
+ * return value: true if negotiation is a success, false if failed
+ */
+static bool process_negotiation_response(
+ struct smbd_response *response, int packet_length)
+{
+ struct smbd_connection *info = response->info;
+ struct smbd_negotiate_resp *packet = smbd_response_payload(response);
+
+ if (packet_length < sizeof(struct smbd_negotiate_resp)) {
+ log_rdma_event(ERR,
+ "error: packet_length=%d\n", packet_length);
+ return false;
+ }
+
+ if (le16_to_cpu(packet->negotiated_version) != SMBD_V1) {
+ log_rdma_event(ERR, "error: negotiated_version=%x\n",
+ le16_to_cpu(packet->negotiated_version));
+ return false;
+ }
+ info->protocol = le16_to_cpu(packet->negotiated_version);
+
+ if (packet->credits_requested == 0) {
+ log_rdma_event(ERR, "error: credits_requested==0\n");
+ return false;
+ }
+ info->receive_credit_target = le16_to_cpu(packet->credits_requested);
+
+ if (packet->credits_granted == 0) {
+ log_rdma_event(ERR, "error: credits_granted==0\n");
+ return false;
+ }
+ atomic_set(&info->send_credits, le16_to_cpu(packet->credits_granted));
+
+ atomic_set(&info->receive_credits, 0);
+
+ if (le32_to_cpu(packet->preferred_send_size) > info->max_receive_size) {
+ log_rdma_event(ERR, "error: preferred_send_size=%d\n",
+ le32_to_cpu(packet->preferred_send_size));
+ return false;
+ }
+ info->max_receive_size = le32_to_cpu(packet->preferred_send_size);
+
+ if (le32_to_cpu(packet->max_receive_size) < SMBD_MIN_RECEIVE_SIZE) {
+ log_rdma_event(ERR, "error: max_receive_size=%d\n",
+ le32_to_cpu(packet->max_receive_size));
+ return false;
+ }
+ info->max_send_size = min_t(int, info->max_send_size,
+ le32_to_cpu(packet->max_receive_size));
+
+ if (le32_to_cpu(packet->max_fragmented_size) <
+ SMBD_MIN_FRAGMENTED_SIZE) {
+ log_rdma_event(ERR, "error: max_fragmented_size=%d\n",
+ le32_to_cpu(packet->max_fragmented_size));
+ return false;
+ }
+ info->max_fragmented_send_size =
+ le32_to_cpu(packet->max_fragmented_size);
+
+ return true;
+}
+
+/*
+ * Check and schedule to send an immediate packet
+ * This is used to extend credtis to remote peer to keep the transport busy
+ */
+static void check_and_send_immediate(struct smbd_connection *info)
+{
+ if (info->transport_status != SMBD_CONNECTED)
+ return;
+
+ info->send_immediate = true;
+
+ /*
+ * Promptly send a packet if our peer is running low on receive
+ * credits
+ */
+ if (atomic_read(&info->receive_credits) <
+ info->receive_credit_target - 1)
+ queue_delayed_work(
+ info->workqueue, &info->send_immediate_work, 0);
+}
+
+static void smbd_post_send_credits(struct work_struct *work)
+{
+ int ret = 0;
+ int use_receive_queue = 1;
+ int rc;
+ struct smbd_response *response;
+ struct smbd_connection *info =
+ container_of(work, struct smbd_connection,
+ post_send_credits_work);
+
+ if (info->transport_status != SMBD_CONNECTED) {
+ wake_up(&info->wait_receive_queues);
+ return;
+ }
+
+ if (info->receive_credit_target >
+ atomic_read(&info->receive_credits)) {
+ while (true) {
+ if (use_receive_queue)
+ response = get_receive_buffer(info);
+ else
+ response = get_empty_queue_buffer(info);
+ if (!response) {
+ /* now switch to emtpy packet queue */
+ if (use_receive_queue) {
+ use_receive_queue = 0;
+ continue;
+ } else
+ break;
+ }
+
+ response->type = SMBD_TRANSFER_DATA;
+ response->first_segment = false;
+ rc = smbd_post_recv(info, response);
+ if (rc) {
+ log_rdma_recv(ERR,
+ "post_recv failed rc=%d\n", rc);
+ put_receive_buffer(info, response, true);
+ break;
+ }
+
+ ret++;
+ }
+ }
+
+ spin_lock(&info->lock_new_credits_offered);
+ info->new_credits_offered += ret;
+ spin_unlock(&info->lock_new_credits_offered);
+
+ atomic_add(ret, &info->receive_credits);
+
+ /* Check if we can post new receive and grant credits to peer */
+ check_and_send_immediate(info);
+}
+
+static void smbd_recv_done_work(struct work_struct *work)
+{
+ struct smbd_connection *info =
+ container_of(work, struct smbd_connection, recv_done_work);
+
+ /*
+ * We may have new send credits granted from remote peer
+ * If any sender is blcoked on lack of credets, unblock it
+ */
+ if (atomic_read(&info->send_credits))
+ wake_up_interruptible(&info->wait_send_queue);
+
+ /*
+ * Check if we need to send something to remote peer to
+ * grant more credits or respond to KEEP_ALIVE packet
+ */
+ check_and_send_immediate(info);
+}
+
+/* Called from softirq, when recv is done */
+static void recv_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct smbd_data_transfer *data_transfer;
+ struct smbd_response *response =
+ container_of(wc->wr_cqe, struct smbd_response, cqe);
+ struct smbd_connection *info = response->info;
+ unsigned long long t1 = rdtsc();
+ int data_length = 0;
+
+ log_rdma_recv(INFO, "response=%p type=%d wc status=%d wc opcode %d "
+ "byte_len=%d pkey_index=%x\n",
+ response, response->type, wc->status, wc->opcode,
+ wc->byte_len, wc->pkey_index);
+
+ if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_RECV) {
+ log_rdma_recv(INFO, "wc->status=%d opcode=%d\n",
+ wc->status, wc->opcode);
+ smbd_disconnect_rdma_connection(info);
+ goto error;
+ }
+
+ ib_dma_sync_single_for_cpu(
+ wc->qp->device,
+ response->sge.addr,
+ response->sge.length,
+ DMA_FROM_DEVICE);
+
+ switch (response->type) {
+ /* SMBD negotiation response */
+ case SMBD_NEGOTIATE_RESP:
+ dump_smbd_negotiate_resp(smbd_response_payload(response));
+ info->full_packet_received = true;
+ info->negotiate_done =
+ process_negotiation_response(response, wc->byte_len);
+ complete(&info->negotiate_completion);
+ break;
+
+ /* SMBD data transfer packet */
+ case SMBD_TRANSFER_DATA:
+ data_transfer = smbd_response_payload(response);
+ data_length = le32_to_cpu(data_transfer->data_length);
+
+ /*
+ * If this is a packet with data playload place the data in
+ * reassembly queue and wake up the reading thread
+ */
+ if (data_length) {
+ if (info->full_packet_received)
+ response->first_segment = true;
+
+ if (le32_to_cpu(data_transfer->remaining_data_length))
+ info->full_packet_received = false;
+ else
+ info->full_packet_received = true;
+
+ enqueue_reassembly(
+ info,
+ response,
+ data_length);
+ } else
+ put_empty_packet(info, response);
+
+ if (data_length) {
+ wake_up_interruptible(&info->wait_reassembly_queue);
+ profiling_add_histogram(rdtsc()-t1, info->recv_done_cycles);
+ }
+
+ atomic_dec(&info->receive_credits);
+ info->receive_credit_target =
+ le16_to_cpu(data_transfer->credits_requested);
+ atomic_add(le16_to_cpu(data_transfer->credits_granted),
+ &info->send_credits);
+
+ log_incoming(INFO, "data flags %d data_offset %d "
+ "data_length %d remaining_data_length %d\n",
+ le16_to_cpu(data_transfer->flags),
+ le32_to_cpu(data_transfer->data_offset),
+ le32_to_cpu(data_transfer->data_length),
+ le32_to_cpu(data_transfer->remaining_data_length));
+
+ /* Send a KEEP_ALIVE response right away if requested */
+ info->keep_alive_requested = KEEP_ALIVE_NONE;
+ if (le16_to_cpu(data_transfer->flags) &
+ le16_to_cpu(SMB_DIRECT_RESPONSE_REQUESTED)) {
+ info->keep_alive_requested = KEEP_ALIVE_PENDING;
+ }
+
+ queue_work(info->workqueue, &info->recv_done_work);
+ return;
+
+ default:
+ log_rdma_recv(ERR,
+ "unexpected response type=%d\n", response->type);
+ }
+
+error:
+ put_receive_buffer(info, response, true);
+}
+
+static struct rdma_cm_id *smbd_create_id(
+ struct smbd_connection *info,
+ struct sockaddr *dstaddr, int port)
+{
+ struct rdma_cm_id *id;
+ int rc;
+ __be16 *sport;
+
+ id = rdma_create_id(&init_net, smbd_conn_upcall, info,
+ RDMA_PS_TCP, IB_QPT_RC);
+ if (IS_ERR(id)) {
+ rc = PTR_ERR(id);
+ log_rdma_event(ERR, "rdma_create_id() failed %i\n", rc);
+ return id;
+ }
+
+ if (dstaddr->sa_family == AF_INET6)
+ sport = &((struct sockaddr_in6 *)dstaddr)->sin6_port;
+ else
+ sport = &((struct sockaddr_in *)dstaddr)->sin_port;
+
+ *sport = htons(port);
+
+ init_completion(&info->ri_done);
+ info->ri_rc = -ETIMEDOUT;
+
+ rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)dstaddr,
+ RDMA_RESOLVE_TIMEOUT);
+ if (rc) {
+ log_rdma_event(ERR, "rdma_resolve_addr() failed %i\n", rc);
+ goto out;
+ }
+ wait_for_completion_interruptible_timeout(
+ &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
+ rc = info->ri_rc;
+ if (rc) {
+ log_rdma_event(ERR, "rdma_resolve_addr() completed %i\n", rc);
+ goto out;
+ }
+
+ info->ri_rc = -ETIMEDOUT;
+ rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
+ if (rc) {
+ log_rdma_event(ERR, "rdma_resolve_route() failed %i\n", rc);
+ goto out;
+ }
+ wait_for_completion_interruptible_timeout(
+ &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
+ rc = info->ri_rc;
+ if (rc) {
+ log_rdma_event(ERR, "rdma_resolve_route() completed %i\n", rc);
+ goto out;
+ }
+
+ return id;
+
+out:
+ rdma_destroy_id(id);
+ return ERR_PTR(rc);
+}
+
+/*
+ * Test if FRWR (Fast Registration Work Requests) is supported on the device
+ * This implementation requries FRWR on RDMA read/write
+ * return value: true if it is supported
+ */
+static bool frwr_is_supported(struct ib_device_attr *attrs)
+{
+ if (!(attrs->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
+ return false;
+ if (attrs->max_fast_reg_page_list_len == 0)
+ return false;
+ return true;
+}
+
+static int smbd_ia_open(
+ struct smbd_connection *info,
+ struct sockaddr *dstaddr, int port)
+{
+ int rc;
+
+ info->id = smbd_create_id(info, dstaddr, port);
+ if (IS_ERR(info->id)) {
+ rc = PTR_ERR(info->id);
+ goto out1;
+ }
+
+ if (!frwr_is_supported(&info->id->device->attrs)) {
+ log_rdma_event(ERR,
+ "Fast Registration Work Requests "
+ "(FRWR) is not supported\n");
+ log_rdma_event(ERR,
+ "Device capability flags = %llx "
+ "max_fast_reg_page_list_len = %u\n",
+ info->id->device->attrs.device_cap_flags,
+ info->id->device->attrs.max_fast_reg_page_list_len);
+ rc = -EPROTONOSUPPORT;
+ goto out2;
+ }
+
+ info->pd = ib_alloc_pd(info->id->device, 0);
+ if (IS_ERR(info->pd)) {
+ rc = PTR_ERR(info->pd);
+ log_rdma_event(ERR, "ib_alloc_pd() returned %d\n", rc);
+ goto out2;
+ }
+
+ return 0;
+
+out2:
+ rdma_destroy_id(info->id);
+ info->id = NULL;
+
+out1:
+ return rc;
+}
+
+/*
+ * Send a negotiation request message to the peer
+ * The negotiation procedure is in [MS-SMBD] 3.1.5.2 and 3.1.5.3
+ * After negotiation, the transport is connected and ready for
+ * carrying upper layer SMB payload
+ */
+static int smbd_post_send_negotiate_req(struct smbd_connection *info)
+{
+ struct ib_send_wr send_wr, *send_wr_fail;
+ int rc = -ENOMEM;
+ struct smbd_request *request;
+ struct smbd_negotiate_req *packet;
+
+ request = mempool_alloc(info->request_mempool, GFP_KERNEL);
+ if (!request)
+ return rc;
+
+ request->info = info;
+
+ packet = smbd_request_payload(request);
+ packet->min_version = cpu_to_le16(SMBD_V1);
+ packet->max_version = cpu_to_le16(SMBD_V1);
+ packet->reserved = 0;
+ packet->credits_requested = cpu_to_le16(info->send_credit_target);
+ packet->preferred_send_size = cpu_to_le32(info->max_send_size);
+ packet->max_receive_size = cpu_to_le32(info->max_receive_size);
+ packet->max_fragmented_size =
+ cpu_to_le32(info->max_fragmented_recv_size);
+
+ request->num_sge = 1;
+ request->sge[0].addr = ib_dma_map_single(
+ info->id->device, (void *)packet,
+ sizeof(*packet), DMA_TO_DEVICE);
+ if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
+ rc = -EIO;
+ goto dma_mapping_failed;
+ }
+
+ request->sge[0].length = sizeof(*packet);
+ request->sge[0].lkey = info->pd->local_dma_lkey;
+
+ ib_dma_sync_single_for_device(
+ info->id->device, request->sge[0].addr,
+ request->sge[0].length, DMA_TO_DEVICE);
+
+ request->cqe.done = send_done;
+
+ send_wr.next = NULL;
+ send_wr.wr_cqe = &request->cqe;
+ send_wr.sg_list = request->sge;
+ send_wr.num_sge = request->num_sge;
+ send_wr.opcode = IB_WR_SEND;
+ send_wr.send_flags = IB_SEND_SIGNALED;
+
+ log_rdma_send(INFO, "sge addr=%llx length=%x lkey=%x\n",
+ request->sge[0].addr,
+ request->sge[0].length, request->sge[0].lkey);
+
+ request->has_payload = false;
+ atomic_inc(&info->send_pending);
+ rc = ib_post_send(info->id->qp, &send_wr, &send_wr_fail);
+ if (!rc)
+ return 0;
+
+ /* if we reach here, post send failed */
+ log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
+ atomic_dec(&info->send_pending);
+ ib_dma_unmap_single(info->id->device, request->sge[0].addr,
+ request->sge[0].length, DMA_TO_DEVICE);
+
+dma_mapping_failed:
+ mempool_free(request, info->request_mempool);
+ return rc;
+}
+
+/*
+ * Extend the credits to remote peer
+ * This implements [MS-SMBD] 3.1.5.9
+ * The idea is that we should extend credits to remote peer as quickly as
+ * it's allowed, to maintain data flow. We allocate as much receive
+ * buffer as possible, and extend the receive credits to remote peer
+ * return value: the new credtis being granted.
+ */
+static int manage_credits_prior_sending(struct smbd_connection *info)
+{
+ int new_credits;
+
+ spin_lock(&info->lock_new_credits_offered);
+ new_credits = info->new_credits_offered;
+ info->new_credits_offered = 0;
+ spin_unlock(&info->lock_new_credits_offered);
+
+ return new_credits;
+}
+
+/*
+ * Check if we need to send a KEEP_ALIVE message
+ * The idle connection timer triggers a KEEP_ALIVE message when expires
+ * SMB_DIRECT_RESPONSE_REQUESTED is set in the message flag to have peer send
+ * back a response.
+ * return value:
+ * 1 if SMB_DIRECT_RESPONSE_REQUESTED needs to be set
+ * 0: otherwise
+ */
+static int manage_keep_alive_before_sending(struct smbd_connection *info)
+{
+ if (info->keep_alive_requested == KEEP_ALIVE_PENDING) {
+ info->keep_alive_requested = KEEP_ALIVE_SENT;
+ return 1;
+ }
+ return 0;
+}
+
+/*
+ * Build and prepare the SMBD packet header
+ * This function waits for avaialbe send credits and build a SMBD packet
+ * header. The caller then optional append payload to the packet after
+ * the header
+ * intput values
+ * size: the size of the payload
+ * remaining_data_length: remaining data to send if this is part of a
+ * fragmented packet
+ * output values
+ * request_out: the request allocated from this function
+ * return values: 0 on success, otherwise actual error code returned
+ */
+static int smbd_create_header(struct smbd_connection *info,
+ int size, int remaining_data_length,
+ struct smbd_request **request_out)
+{
+ struct smbd_request *request;
+ struct smbd_data_transfer *packet;
+ int header_length;
+ int rc;
+
+ /* Wait for send credits. A SMBD packet needs one credit */
+ rc = wait_event_interruptible(info->wait_send_queue,
+ atomic_read(&info->send_credits) > 0 ||
+ info->transport_status != SMBD_CONNECTED);
+ if (rc)
+ return rc;
+
+ if (info->transport_status != SMBD_CONNECTED) {
+ log_outgoing(ERR, "disconnected not sending\n");
+ return -ENOENT;
+ }
+ atomic_dec(&info->send_credits);
+
+ request = mempool_alloc(info->request_mempool, GFP_KERNEL);
+ if (!request) {
+ rc = -ENOMEM;
+ goto err;
+ }
+
+ request->info = info;
+
+ /* Fill in the packet header */
+ packet = smbd_request_payload(request);
+ packet->credits_requested = cpu_to_le16(info->send_credit_target);
+ packet->credits_granted =
+ cpu_to_le16(manage_credits_prior_sending(info));
+ info->send_immediate = false;
+
+ packet->flags = 0;
+ if (manage_keep_alive_before_sending(info))
+ packet->flags |= cpu_to_le16(SMB_DIRECT_RESPONSE_REQUESTED);
+
+ packet->reserved = 0;
+ if (!size)
+ packet->data_offset = 0;
+ else
+ packet->data_offset = cpu_to_le32(24);
+ packet->data_length = cpu_to_le32(size);
+ packet->remaining_data_length = cpu_to_le32(remaining_data_length);
+ packet->padding = 0;
+
+ log_outgoing(INFO, "credits_requested=%d credits_granted=%d "
+ "data_offset=%d data_length=%d remaining_data_length=%d\n",
+ le16_to_cpu(packet->credits_requested),
+ le16_to_cpu(packet->credits_granted),
+ le32_to_cpu(packet->data_offset),
+ le32_to_cpu(packet->data_length),
+ le32_to_cpu(packet->remaining_data_length));
+
+ /* Map the packet to DMA */
+ header_length = sizeof(struct smbd_data_transfer);
+ /* If this is a packet without payload, don't send padding */
+ if (!size)
+ header_length = offsetof(struct smbd_data_transfer, padding);
+
+ request->num_sge = 1;
+ request->sge[0].addr = ib_dma_map_single(info->id->device,
+ (void *)packet,
+ header_length,
+ DMA_BIDIRECTIONAL);
+ if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
+ mempool_free(request, info->request_mempool);
+ rc = -EIO;
+ goto err;
+ }
+
+ request->sge[0].length = header_length;
+ request->sge[0].lkey = info->pd->local_dma_lkey;
+
+ *request_out = request;
+ return 0;
+
+err:
+ atomic_inc(&info->send_credits);
+ return rc;
+}
+
+static void smbd_destroy_header(struct smbd_connection *info,
+ struct smbd_request *request)
+{
+
+ ib_dma_unmap_single(info->id->device,
+ request->sge[0].addr,
+ request->sge[0].length,
+ DMA_TO_DEVICE);
+ mempool_free(request, info->request_mempool);
+ atomic_inc(&info->send_credits);
+}
+
+/* Post the send request */
+static int smbd_post_send(struct smbd_connection *info,
+ struct smbd_request *request, bool has_payload)
+{
+ struct ib_send_wr send_wr, *send_wr_fail;
+ int rc, i;
+
+ for (i = 0; i < request->num_sge; i++) {
+ log_rdma_send(INFO,
+ "rdma_request sge[%d] addr=%llu legnth=%u\n",
+ i, request->sge[0].addr, request->sge[0].length);
+ ib_dma_sync_single_for_device(
+ info->id->device,
+ request->sge[i].addr,
+ request->sge[i].length,
+ DMA_TO_DEVICE);
+ }
+
+ request->cqe.done = send_done;
+
+ send_wr.next = NULL;
+ send_wr.wr_cqe = &request->cqe;
+ send_wr.sg_list = request->sge;
+ send_wr.num_sge = request->num_sge;
+ send_wr.opcode = IB_WR_SEND;
+ send_wr.send_flags = IB_SEND_SIGNALED;
+
+ if (has_payload) {
+ request->has_payload = true;
+ atomic_inc(&info->send_payload_pending);
+ } else {
+ request->has_payload = false;
+ atomic_inc(&info->send_pending);
+ }
+
+ rc = ib_post_send(info->id->qp, &send_wr, &send_wr_fail);
+ if (rc) {
+ log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
+ if (has_payload) {
+ if (atomic_dec_and_test(&info->send_payload_pending))
+ wake_up(&info->wait_send_payload_pending);
+ } else {
+ if (atomic_dec_and_test(&info->send_pending))
+ wake_up(&info->wait_send_pending);
+ }
+ } else
+ /* Reset timer for idle connection after packet is sent */
+ mod_delayed_work(info->workqueue, &info->idle_timer_work,
+ info->keep_alive_interval*HZ);
+
+ return rc;
+}
+
+static int smbd_post_send_sgl(struct smbd_connection *info,
+ struct scatterlist *sgl, int data_length, int remaining_data_length)
+{
+ int num_sgs;
+ int i, rc;
+ struct smbd_request *request;
+ struct scatterlist *sg;
+
+ rc = smbd_create_header(
+ info, data_length, remaining_data_length, &request);
+ if (rc)
+ return rc;
+
+ num_sgs = sgl ? sg_nents(sgl) : 0;
+ for_each_sg(sgl, sg, num_sgs, i) {
+ request->sge[i+1].addr =
+ ib_dma_map_page(info->id->device, sg_page(sg),
+ sg->offset, sg->length, DMA_BIDIRECTIONAL);
+ if (ib_dma_mapping_error(
+ info->id->device, request->sge[i+1].addr)) {
+ rc = -EIO;
+ request->sge[i+1].addr = 0;
+ goto dma_mapping_failure;
+ }
+ request->sge[i+1].length = sg->length;
+ request->sge[i+1].lkey = info->pd->local_dma_lkey;
+ request->num_sge++;
+ }
+
+ rc = smbd_post_send(info, request, data_length);
+ if (!rc)
+ return 0;
+
+dma_mapping_failure:
+ for (i = 1; i < request->num_sge; i++)
+ if (request->sge[i].addr)
+ ib_dma_unmap_single(info->id->device,
+ request->sge[i].addr,
+ request->sge[i].length,
+ DMA_TO_DEVICE);
+ smbd_destroy_header(info, request);
+ return rc;
+}
+
+/*
+ * Send an empty message
+ * Empty message is used to extend credits to peer to for keep live
+ * while there is no upper layer payload to send at the time
+ */
+static int smbd_post_send_empty(struct smbd_connection *info)
+{
+ info->count_send_empty++;
+ return smbd_post_send_sgl(info, NULL, 0, 0);
+}
+
+/*
+ * Post a receive request to the transport
+ * The remote peer can only send data when a receive request is posted
+ * The interaction is controlled by send/receive credit system
+ */
+static int smbd_post_recv(
+ struct smbd_connection *info, struct smbd_response *response)
+{
+ struct ib_recv_wr recv_wr, *recv_wr_fail = NULL;
+ int rc = -EIO;
+
+ response->sge.addr = ib_dma_map_single(
+ info->id->device, response->packet,
+ info->max_receive_size, DMA_FROM_DEVICE);
+ if (ib_dma_mapping_error(info->id->device, response->sge.addr))
+ return rc;
+
+ response->sge.length = info->max_receive_size;
+ response->sge.lkey = info->pd->local_dma_lkey;
+
+ response->cqe.done = recv_done;
+
+ recv_wr.wr_cqe = &response->cqe;
+ recv_wr.next = NULL;
+ recv_wr.sg_list = &response->sge;
+ recv_wr.num_sge = 1;
+
+ rc = ib_post_recv(info->id->qp, &recv_wr, &recv_wr_fail);
+ if (rc) {
+ ib_dma_unmap_single(info->id->device, response->sge.addr,
+ response->sge.length, DMA_FROM_DEVICE);
+
+ log_rdma_recv(ERR, "ib_post_recv failed rc=%d\n", rc);
+ }
+
+ return rc;
+}
+
+/* Perform SMBD negotiate according to [MS-SMBD] 3.1.5.2 */
+static int smbd_negotiate(struct smbd_connection *info)
+{
+ int rc;
+ struct smbd_response *response = get_receive_buffer(info);
+
+ response->type = SMBD_NEGOTIATE_RESP;
+ rc = smbd_post_recv(info, response);
+ log_rdma_event(INFO,
+ "smbd_post_recv rc=%d iov.addr=%llx iov.length=%x "
+ "iov.lkey=%x\n",
+ rc, response->sge.addr,
+ response->sge.length, response->sge.lkey);
+ if (rc)
+ return rc;
+
+ init_completion(&info->negotiate_completion);
+ info->negotiate_done = false;
+ rc = smbd_post_send_negotiate_req(info);
+ if (rc)
+ return rc;
+
+ rc = wait_for_completion_interruptible_timeout(
+ &info->negotiate_completion, SMBD_NEGOTIATE_TIMEOUT * HZ);
+ log_rdma_event(INFO, "wait_for_completion_timeout rc=%d\n", rc);
+
+ if (info->negotiate_done)
+ return 0;
+
+ if (rc == 0)
+ rc = -ETIMEDOUT;
+ else if (rc == -ERESTARTSYS)
+ rc = -EINTR;
+ else
+ rc = -ENOTCONN;
+
+ return rc;
+}
+
+static void put_empty_packet(
+ struct smbd_connection *info, struct smbd_response *response)
+{
+ spin_lock(&info->empty_packet_queue_lock);
+ list_add_tail(&response->list, &info->empty_packet_queue);
+ info->count_empty_packet_queue++;
+ spin_unlock(&info->empty_packet_queue_lock);
+
+ queue_work(info->workqueue, &info->post_send_credits_work);
+}
+
+/*
+ * Implement Connection.FragmentReassemblyBuffer defined in [MS-SMBD] 3.1.1.1
+ * This is a queue for reassembling upper layer payload and present to upper
+ * layer. All the inncoming payload go to the reassembly queue, regardless of
+ * if reassembly is required. The uuper layer code reads from the queue for all
+ * incoming payloads.
+ * Put a received packet to the reassembly queue
+ * response: the packet received
+ * data_length: the size of payload in this packet
+ */
+static void enqueue_reassembly(
+ struct smbd_connection *info,
+ struct smbd_response *response,
+ int data_length)
+{
+ spin_lock(&info->reassembly_queue_lock);
+ list_add_tail(&response->list, &info->reassembly_queue);
+ info->reassembly_queue_length++;
+ /*
+ * Make sure reassembly_data_length is updated after list and
+ * reassembly_queue_length are updated. On the dequeue side
+ * reassembly_data_length is checked without a lock to determine
+ * if reassembly_queue_length and list is up to date
+ */
+ virt_wmb();
+ info->reassembly_data_length += data_length;
+ spin_unlock(&info->reassembly_queue_lock);
+ info->count_reassembly_queue++;
+ info->count_enqueue_reassembly_queue++;
+}
+
+/*
+ * Get the first entry at the front of reassembly queue
+ * Caller is responsible for locking
+ * return value: the first entry if any, NULL if queue is empty
+ */
+static struct smbd_response *_get_first_reassembly(struct smbd_connection *info)
+{
+ struct smbd_response *ret = NULL;
+
+ if (!list_empty(&info->reassembly_queue)) {
+ ret = list_first_entry(
+ &info->reassembly_queue,
+ struct smbd_response, list);
+ }
+ return ret;
+}
+
+static struct smbd_response *get_empty_queue_buffer(
+ struct smbd_connection *info)
+{
+ struct smbd_response *ret = NULL;
+ unsigned long flags;
+
+ spin_lock_irqsave(&info->empty_packet_queue_lock, flags);
+ if (!list_empty(&info->empty_packet_queue)) {
+ ret = list_first_entry(
+ &info->empty_packet_queue,
+ struct smbd_response, list);
+ list_del(&ret->list);
+ info->count_empty_packet_queue--;
+ }
+ spin_unlock_irqrestore(&info->empty_packet_queue_lock, flags);
+
+ return ret;
+}
+
+/*
+ * Get a receive buffer
+ * For each remote send, we need to post a receive. The receive buffers are
+ * pre-allocated in advance.
+ * return value: the receive buffer, NULL if none is available
+ */
+static struct smbd_response *get_receive_buffer(struct smbd_connection *info)
+{
+ struct smbd_response *ret = NULL;
+ unsigned long flags;
+
+ spin_lock_irqsave(&info->receive_queue_lock, flags);
+ if (!list_empty(&info->receive_queue)) {
+ ret = list_first_entry(
+ &info->receive_queue,
+ struct smbd_response, list);
+ list_del(&ret->list);
+ info->count_receive_queue--;
+ info->count_get_receive_buffer++;
+ }
+ spin_unlock_irqrestore(&info->receive_queue_lock, flags);
+
+ return ret;
+}
+
+/*
+ * Return a receive buffer
+ * Upon returning of a receive buffer, we can post new receive and extend
+ * more receive credits to remote peer. This is done immediately after a
+ * receive buffer is returned.
+ */
+static void put_receive_buffer(
+ struct smbd_connection *info, struct smbd_response *response,
+ bool lock)
+{
+ unsigned long flags;
+
+ ib_dma_unmap_single(info->id->device, response->sge.addr,
+ response->sge.length, DMA_FROM_DEVICE);
+
+ if (lock)
+ spin_lock_irqsave(&info->receive_queue_lock, flags);
+ list_add_tail(&response->list, &info->receive_queue);
+ info->count_receive_queue++;
+ info->count_put_receive_buffer++;
+ if (lock)
+ spin_unlock_irqrestore(&info->receive_queue_lock, flags);
+
+ queue_work(info->workqueue, &info->post_send_credits_work);
+}
+
+/* Preallocate all receive buffer on transport establishment */
+static int allocate_receive_buffers(struct smbd_connection *info, int num_buf)
+{
+ int i;
+ struct smbd_response *response;
+
+ INIT_LIST_HEAD(&info->reassembly_queue);
+ spin_lock_init(&info->reassembly_queue_lock);
+ info->reassembly_data_length = 0;
+ info->reassembly_queue_length = 0;
+
+ INIT_LIST_HEAD(&info->receive_queue);
+ spin_lock_init(&info->receive_queue_lock);
+ info->count_receive_queue = 0;
+
+ INIT_LIST_HEAD(&info->empty_packet_queue);
+ spin_lock_init(&info->empty_packet_queue_lock);
+ info->count_empty_packet_queue = 0;
+
+ init_waitqueue_head(&info->wait_receive_queues);
+
+ for (i = 0; i < num_buf; i++) {
+ response = mempool_alloc(info->response_mempool, GFP_KERNEL);
+ if (!response)
+ goto allocate_failed;
+
+ response->info = info;
+ list_add_tail(&response->list, &info->receive_queue);
+ info->count_receive_queue++;
+ }
+
+ return 0;
+
+allocate_failed:
+ while (!list_empty(&info->receive_queue)) {
+ response = list_first_entry(
+ &info->receive_queue,
+ struct smbd_response, list);
+ list_del(&response->list);
+ info->count_receive_queue--;
+
+ mempool_free(response, info->response_mempool);
+ }
+ return -ENOMEM;
+}
+
+static void destroy_receive_buffers(struct smbd_connection *info)
+{
+ struct smbd_response *response;
+
+ while ((response = get_receive_buffer(info)))
+ mempool_free(response, info->response_mempool);
+
+ while ((response = get_empty_queue_buffer(info)))
+ mempool_free(response, info->response_mempool);
+}
+
+/*
+ * Check and send an immediate or keep alive packet
+ * The condition to send those packets are defined in [MS-SMBD] 3.1.1.1
+ * Connection.KeepaliveRequested and Connection.SendImmediate
+ * The idea is to extend credits to server as soon as it becomes available
+ */
+static void send_immediate_work(struct work_struct *work)
+{
+ struct smbd_connection *info = container_of(
+ work, struct smbd_connection,
+ send_immediate_work.work);
+
+ if (info->keep_alive_requested == KEEP_ALIVE_PENDING ||
+ info->send_immediate) {
+ log_keep_alive(INFO, "send an empty message\n");
+ smbd_post_send_empty(info);
+ }
+}
+
+/* Implement idle connection timer [MS-SMBD] 3.1.6.2 */
+static void idle_connection_timer(struct work_struct *work)
+{
+ struct smbd_connection *info = container_of(
+ work, struct smbd_connection,
+ idle_timer_work.work);
+
+ if (info->keep_alive_requested != KEEP_ALIVE_NONE) {
+ log_keep_alive(ERR,
+ "error status info->keep_alive_requested=%d\n",
+ info->keep_alive_requested);
+ smbd_disconnect_rdma_connection(info);
+ return;
+ }
+
+ log_keep_alive(INFO, "about to send an empty idle message\n");
+ smbd_post_send_empty(info);
+
+ /* Setup the next idle timeout work */
+ queue_delayed_work(info->workqueue, &info->idle_timer_work,
+ info->keep_alive_interval*HZ);
+}
+
+static void destroy_caches_and_workqueue(struct smbd_connection *info)
+{
+ destroy_receive_buffers(info);
+ destroy_workqueue(info->workqueue);
+ mempool_destroy(info->response_mempool);
+ kmem_cache_destroy(info->response_cache);
+ mempool_destroy(info->request_mempool);
+ kmem_cache_destroy(info->request_cache);
+}
+
+#define MAX_NAME_LEN 80
+static int allocate_caches_and_workqueue(struct smbd_connection *info)
+{
+ char name[MAX_NAME_LEN];
+ int rc;
+
+ snprintf(name, MAX_NAME_LEN, "smbd_request_%p", info);
+ info->request_cache =
+ kmem_cache_create(
+ name,
+ sizeof(struct smbd_request) +
+ sizeof(struct smbd_data_transfer),
+ 0, SLAB_HWCACHE_ALIGN, NULL);
+ if (!info->request_cache)
+ return -ENOMEM;
+
+ info->request_mempool =
+ mempool_create(info->send_credit_target, mempool_alloc_slab,
+ mempool_free_slab, info->request_cache);
+ if (!info->request_mempool)
+ goto out1;
+
+ snprintf(name, MAX_NAME_LEN, "smbd_response_%p", info);
+ info->response_cache =
+ kmem_cache_create(
+ name,
+ sizeof(struct smbd_response) +
+ info->max_receive_size,
+ 0, SLAB_HWCACHE_ALIGN, NULL);
+ if (!info->response_cache)
+ goto out2;
+
+ info->response_mempool =
+ mempool_create(info->receive_credit_max, mempool_alloc_slab,
+ mempool_free_slab, info->response_cache);
+ if (!info->response_mempool)
+ goto out3;
+
+ snprintf(name, MAX_NAME_LEN, "smbd_%p", info);
+ info->workqueue = create_workqueue(name);
+ if (!info->workqueue)
+ goto out4;
+
+ rc = allocate_receive_buffers(info, info->receive_credit_max);
+ if (rc) {
+ log_rdma_event(ERR, "failed to allocate receive buffers\n");
+ goto out5;
+ }
+
+ return 0;
+
+out5:
+ destroy_workqueue(info->workqueue);
+out4:
+ mempool_destroy(info->response_mempool);
+out3:
+ kmem_cache_destroy(info->response_cache);
+out2:
+ mempool_destroy(info->request_mempool);
+out1:
+ kmem_cache_destroy(info->request_cache);
+ return -ENOMEM;
+}
+
+/* Create a SMBD connection, called by upper layer */
+struct smbd_connection *_smbd_get_connection(
+ struct TCP_Server_Info *server, struct sockaddr *dstaddr, int port)
+{
+ int rc;
+ struct smbd_connection *info;
+ struct rdma_conn_param conn_param;
+ struct ib_qp_init_attr qp_attr;
+ struct sockaddr_in *addr_in = (struct sockaddr_in *) dstaddr;
+
+ info = kzalloc(sizeof(struct smbd_connection), GFP_KERNEL);
+ if (!info)
+ return NULL;
+
+ info->transport_status = SMBD_CONNECTING;
+ rc = smbd_ia_open(info, dstaddr, port);
+ if (rc) {
+ log_rdma_event(INFO, "smbd_ia_open rc=%d\n", rc);
+ goto create_id_failed;
+ }
+
+ if (smbd_send_credit_target > info->id->device->attrs.max_cqe ||
+ smbd_send_credit_target > info->id->device->attrs.max_qp_wr) {
+ log_rdma_event(ERR,
+ "consider lowering send_credit_target = %d. "
+ "Possible CQE overrun, device "
+ "reporting max_cpe %d max_qp_wr %d\n",
+ smbd_send_credit_target,
+ info->id->device->attrs.max_cqe,
+ info->id->device->attrs.max_qp_wr);
+ goto config_failed;
+ }
+
+ if (smbd_receive_credit_max > info->id->device->attrs.max_cqe ||
+ smbd_receive_credit_max > info->id->device->attrs.max_qp_wr) {
+ log_rdma_event(ERR,
+ "consider lowering receive_credit_max = %d. "
+ "Possible CQE overrun, device "
+ "reporting max_cpe %d max_qp_wr %d\n",
+ smbd_receive_credit_max,
+ info->id->device->attrs.max_cqe,
+ info->id->device->attrs.max_qp_wr);
+ goto config_failed;
+ }
+
+ info->receive_credit_max = smbd_receive_credit_max;
+ info->send_credit_target = smbd_send_credit_target;
+ info->max_send_size = smbd_max_send_size;
+ info->max_fragmented_recv_size = smbd_max_fragmented_recv_size;
+ info->max_receive_size = smbd_max_receive_size;
+ info->keep_alive_interval = smbd_keep_alive_interval;
+
+ if (SMBDIRECT_MAX_SGE > info->id->device->attrs.max_sge) {
+ log_rdma_event(ERR, "warning: device max_sge = %d too small\n",
+ info->id->device->attrs.max_sge);
+ log_rdma_event(ERR, "Queue Pair creation may fail\n");
+ }
+
+ info->send_cq = NULL;
+ info->recv_cq = NULL;
+ info->send_cq = ib_alloc_cq(info->id->device, info,
+ info->send_credit_target, 0, IB_POLL_SOFTIRQ);
+ if (IS_ERR(info->send_cq)) {
+ info->send_cq = NULL;
+ goto alloc_cq_failed;
+ }
+
+ info->recv_cq = ib_alloc_cq(info->id->device, info,
+ info->receive_credit_max, 0, IB_POLL_SOFTIRQ);
+ if (IS_ERR(info->recv_cq)) {
+ info->recv_cq = NULL;
+ goto alloc_cq_failed;
+ }
+
+ memset(&qp_attr, 0, sizeof(qp_attr));
+ qp_attr.event_handler = smbd_qp_async_error_upcall;
+ qp_attr.qp_context = info;
+ qp_attr.cap.max_send_wr = info->send_credit_target;
+ qp_attr.cap.max_recv_wr = info->receive_credit_max;
+ qp_attr.cap.max_send_sge = SMBDIRECT_MAX_SGE;
+ qp_attr.cap.max_recv_sge = SMBDIRECT_MAX_SGE;
+ qp_attr.cap.max_inline_data = 0;
+ qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
+ qp_attr.qp_type = IB_QPT_RC;
+ qp_attr.send_cq = info->send_cq;
+ qp_attr.recv_cq = info->recv_cq;
+ qp_attr.port_num = ~0;
+
+ rc = rdma_create_qp(info->id, info->pd, &qp_attr);
+ if (rc) {
+ log_rdma_event(ERR, "rdma_create_qp failed %i\n", rc);
+ goto create_qp_failed;
+ }
+
+ memset(&conn_param, 0, sizeof(conn_param));
+ conn_param.initiator_depth = 0;
+
+ conn_param.retry_count = SMBD_CM_RETRY;
+ conn_param.rnr_retry_count = SMBD_CM_RNR_RETRY;
+ conn_param.flow_control = 0;
+ init_waitqueue_head(&info->wait_destroy);
+
+ log_rdma_event(INFO, "connecting to IP %pI4 port %d\n",
+ &addr_in->sin_addr, port);
+
+ init_waitqueue_head(&info->conn_wait);
+ rc = rdma_connect(info->id, &conn_param);
+ if (rc) {
+ log_rdma_event(ERR, "rdma_connect() failed with %i\n", rc);
+ goto rdma_connect_failed;
+ }
+
+ wait_event_interruptible(
+ info->conn_wait, info->transport_status != SMBD_CONNECTING);
+
+ if (info->transport_status != SMBD_CONNECTED) {
+ log_rdma_event(ERR, "rdma_connect failed port=%d\n", port);
+ goto rdma_connect_failed;
+ }
+
+ log_rdma_event(INFO, "rdma_connect connected\n");
+
+ rc = allocate_caches_and_workqueue(info);
+ if (rc) {
+ log_rdma_event(ERR, "cache allocation failed\n");
+ goto allocate_cache_failed;
+ }
+
+ init_waitqueue_head(&info->wait_send_queue);
+ init_waitqueue_head(&info->wait_reassembly_queue);
+
+ INIT_DELAYED_WORK(&info->idle_timer_work, idle_connection_timer);
+ INIT_DELAYED_WORK(&info->send_immediate_work, send_immediate_work);
+ queue_delayed_work(info->workqueue, &info->idle_timer_work,
+ info->keep_alive_interval*HZ);
+
+ init_waitqueue_head(&info->wait_send_pending);
+ atomic_set(&info->send_pending, 0);
+
+ init_waitqueue_head(&info->wait_send_payload_pending);
+ atomic_set(&info->send_payload_pending, 0);
+
+ INIT_WORK(&info->disconnect_work, smbd_disconnect_rdma_work);
+ INIT_WORK(&info->destroy_work, smbd_destroy_rdma_work);
+ INIT_WORK(&info->recv_done_work, smbd_recv_done_work);
+ INIT_WORK(&info->post_send_credits_work, smbd_post_send_credits);
+ info->new_credits_offered = 0;
+ spin_lock_init(&info->lock_new_credits_offered);
+
+ rc = smbd_negotiate(info);
+ if (rc) {
+ log_rdma_event(ERR, "smbd_negotiate rc=%d\n", rc);
+ goto negotiation_failed;
+ }
+
+ return info;
+
+negotiation_failed:
+ cancel_delayed_work_sync(&info->idle_timer_work);
+ destroy_caches_and_workqueue(info);
+ info->transport_status = SMBD_NEGOTIATE_FAILED;
+ init_waitqueue_head(&info->conn_wait);
+ rdma_disconnect(info->id);
+ wait_event(info->conn_wait,
+ info->transport_status == SMBD_DISCONNECTED);
+
+allocate_cache_failed:
+rdma_connect_failed:
+ rdma_destroy_qp(info->id);
+
+create_qp_failed:
+alloc_cq_failed:
+ if (info->send_cq)
+ ib_free_cq(info->send_cq);
+ if (info->recv_cq)
+ ib_free_cq(info->recv_cq);
+
+config_failed:
+ ib_dealloc_pd(info->pd);
+ rdma_destroy_id(info->id);
+
+create_id_failed:
+ kfree(info);
+ return NULL;
+}
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index c55f28b..ca60700 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -16,6 +16,235 @@
#ifndef _SMBDIRECT_H
#define _SMBDIRECT_H
+#include "cifsglob.h"
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_cm.h>
+#include <linux/mempool.h>
+
+enum keep_alive_status {
+ KEEP_ALIVE_NONE,
+ KEEP_ALIVE_PENDING,
+ KEEP_ALIVE_SENT,
+};
+
+enum smbd_connection_status {
+ SMBD_CREATED,
+ SMBD_CONNECTING,
+ SMBD_CONNECTED,
+ SMBD_NEGOTIATE_FAILED,
+ SMBD_DISCONNECTING,
+ SMBD_DISCONNECTED,
+ SMBD_DESTROYED
+};
+
+/*
+ * The context for the SMBDirect transport
+ * Everything related to the transport is here. It has several logical parts
+ * 1. RDMA related structures
+ * 2. SMBDirect connection parameters
+ * 3. Memory registrations
+ * 4. Receive and reassembly queues for data receive path
+ * 5. mempools for allocating packets
+ */
+struct smbd_connection {
+ enum smbd_connection_status transport_status;
+
+ /* RDMA related */
+ struct rdma_cm_id *id;
+ struct ib_qp_init_attr qp_attr;
+ struct ib_pd *pd;
+ struct ib_cq *send_cq, *recv_cq;
+ struct ib_device_attr dev_attr;
+ int ri_rc;
+ struct completion ri_done;
+ wait_queue_head_t conn_wait;
+ wait_queue_head_t wait_destroy;
+
+ struct completion negotiate_completion;
+ bool negotiate_done;
+
+ struct work_struct destroy_work;
+ struct work_struct disconnect_work;
+ struct work_struct recv_done_work;
+ struct work_struct post_send_credits_work;
+
+ spinlock_t lock_new_credits_offered;
+ int new_credits_offered;
+
+ /* Connection parameters defined in [MS-SMBD] 3.1.1.1 */
+ int receive_credit_max;
+ int send_credit_target;
+ int max_send_size;
+ int max_fragmented_recv_size;
+ int max_fragmented_send_size;
+ int max_receive_size;
+ int keep_alive_interval;
+ int max_readwrite_size;
+ enum keep_alive_status keep_alive_requested;
+ int protocol;
+ atomic_t send_credits;
+ atomic_t receive_credits;
+ int receive_credit_target;
+ int fragment_reassembly_remaining;
+
+ /* Activity accoutning */
+
+ atomic_t send_pending;
+ wait_queue_head_t wait_send_pending;
+ atomic_t send_payload_pending;
+ wait_queue_head_t wait_send_payload_pending;
+
+ /* Receive queue */
+ struct list_head receive_queue;
+ int count_receive_queue;
+ spinlock_t receive_queue_lock;
+
+ struct list_head empty_packet_queue;
+ int count_empty_packet_queue;
+ spinlock_t empty_packet_queue_lock;
+
+ wait_queue_head_t wait_receive_queues;
+
+ /* Reassembly queue */
+ struct list_head reassembly_queue;
+ spinlock_t reassembly_queue_lock;
+ wait_queue_head_t wait_reassembly_queue;
+
+ /* total data length of reassembly queue */
+ int reassembly_data_length;
+ int reassembly_queue_length;
+ /* the offset to first buffer in reassembly queue */
+ int first_entry_offset;
+
+ bool send_immediate;
+
+ wait_queue_head_t wait_send_queue;
+
+ /*
+ * Indicate if we have received a full packet on the connection
+ * This is used to identify the first SMBD packet of a assembled
+ * payload (SMB packet) in reassembly queue so we can return a
+ * RFC1002 length to upper layer to indicate the length of the SMB
+ * packet received
+ */
+ bool full_packet_received;
+
+ struct workqueue_struct *workqueue;
+ struct delayed_work idle_timer_work;
+ struct delayed_work send_immediate_work;
+
+ /* Memory pool for preallocating buffers */
+ /* request pool for RDMA send */
+ struct kmem_cache *request_cache;
+ mempool_t *request_mempool;
+
+ /* response pool for RDMA receive */
+ struct kmem_cache *response_cache;
+ mempool_t *response_mempool;
+
+ /* for debug purposes */
+ unsigned int count_get_receive_buffer;
+ unsigned int count_put_receive_buffer;
+ unsigned int count_reassembly_queue;
+ unsigned int count_enqueue_reassembly_queue;
+ unsigned int count_dequeue_reassembly_queue;
+ unsigned int count_send_empty;
+
+#define SIZE_LONG_LONG 64
+ unsigned long long smbd_write_cycles[SIZE_LONG_LONG];
+ unsigned long long smbd_recv_cycles[SIZE_LONG_LONG];
+ unsigned long long recv_done_cycles[SIZE_LONG_LONG];
+};
+
+enum smbd_message_type {
+ SMBD_NEGOTIATE_RESP,
+ SMBD_TRANSFER_DATA,
+};
+
+#define SMB_DIRECT_RESPONSE_REQUESTED 0x0001
+
+/* SMBD negotiation request packet [MS-SMBD] 2.2.1 */
+struct smbd_negotiate_req {
+ __le16 min_version;
+ __le16 max_version;
+ __le16 reserved;
+ __le16 credits_requested;
+ __le32 preferred_send_size;
+ __le32 max_receive_size;
+ __le32 max_fragmented_size;
+} __packed;
+
+/* SMBD negotiation response packet [MS-SMBD] 2.2.2 */
+struct smbd_negotiate_resp {
+ __le16 min_version;
+ __le16 max_version;
+ __le16 negotiated_version;
+ __le16 reserved;
+ __le16 credits_requested;
+ __le16 credits_granted;
+ __le32 status;
+ __le32 max_readwrite_size;
+ __le32 preferred_send_size;
+ __le32 max_receive_size;
+ __le32 max_fragmented_size;
+} __packed;
+
+/* SMBD data transfer packet with payload [MS-SMBD] 2.2.3 */
+struct smbd_data_transfer {
+ __le16 credits_requested;
+ __le16 credits_granted;
+ __le16 flags;
+ __le16 reserved;
+ __le32 remaining_data_length;
+ __le32 data_offset;
+ __le32 data_length;
+ __le32 padding;
+ __u8 buffer[];
+} __packed;
+
+/* The packet fields for a registered RDMA buffer */
+struct smbd_buffer_descriptor_v1 {
+ __le64 offset;
+ __le32 token;
+ __le32 length;
+} __packed;
+
/* Default maximum number of SGEs in a RDMA send/recv */
#define SMBDIRECT_MAX_SGE 16
+/* The context for a SMBD request */
+struct smbd_request {
+ struct smbd_connection *info;
+ struct ib_cqe cqe;
+
+ /* true if this request carries upper layer payload */
+ bool has_payload;
+
+ /* the SGE entries for this packet */
+ struct ib_sge sge[SMBDIRECT_MAX_SGE];
+ int num_sge;
+
+ /* SMBD packet header follows this structure */
+ u8 packet[];
+};
+
+/* The context for a SMBD response */
+struct smbd_response {
+ struct smbd_connection *info;
+ struct ib_cqe cqe;
+ struct ib_sge sge;
+
+ enum smbd_message_type type;
+
+ /* Link to receive queue or reassembly queue */
+ struct list_head list;
+
+ /* Indicate if this is the 1st packet of a payload */
+ bool first_segment;
+
+ /* SMBD packet header and payload follows this structure */
+ u8 packet[];
+};
+
+void profiling_display_histogram(
+ struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 03/22] CIFS: SMBD: export protocol initial values
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
2017-10-02 2:30 ` [Patch v4 01/22] CIFS: SMBD: Add SMBDirect protocol initial values and constants Long Li
2017-10-02 2:30 ` [Patch v4 02/22] CIFS: SMBD: Establish SMBDirect connection Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 04/22] CIFS: SMBD: Add rdma mount option Long Li
` (12 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Those values can be configured by user. Export them to /proc/fs/cifs.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/cifs_debug.c | 70 ++++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 70 insertions(+)
diff --git a/fs/cifs/cifs_debug.c b/fs/cifs/cifs_debug.c
index 9727e1d..bdc2f38 100644
--- a/fs/cifs/cifs_debug.c
+++ b/fs/cifs/cifs_debug.c
@@ -369,6 +369,52 @@ static const struct file_operations cifs_stats_proc_fops = {
};
#endif /* STATS */
+#define PROC_FILE_DEFINE(name) \
+static ssize_t name##_write(struct file *file, const char __user *buffer, \
+ size_t count, loff_t *ppos) \
+{ \
+ int rc; \
+ rc = kstrtoint_from_user(buffer, count, 10, & name ); \
+ if (rc) \
+ return rc; \
+ return count; \
+} \
+static int name##_proc_show(struct seq_file *m, void *v) \
+{ \
+ seq_printf(m, "%d\n", name ); \
+ return 0; \
+} \
+static int name##_open(struct inode *inode, struct file *file) \
+{ \
+ return single_open(file, name##_proc_show, NULL); \
+} \
+\
+static const struct file_operations cifs_##name##_proc_fops = { \
+ .open = name##_open, \
+ .read = seq_read, \
+ .llseek = seq_lseek, \
+ .release = single_release, \
+ .write = name##_write, \
+}
+
+extern int rdma_readwrite_threshold;
+extern int smbd_max_frmr_depth;
+extern int smbd_keep_alive_interval;
+extern int smbd_max_receive_size;
+extern int smbd_max_fragmented_recv_size;
+extern int smbd_max_send_size;
+extern int smbd_send_credit_target;
+extern int smbd_receive_credit_max;
+
+PROC_FILE_DEFINE(rdma_readwrite_threshold);
+PROC_FILE_DEFINE(smbd_max_frmr_depth);
+PROC_FILE_DEFINE(smbd_keep_alive_interval);
+PROC_FILE_DEFINE(smbd_max_receive_size);
+PROC_FILE_DEFINE(smbd_max_fragmented_recv_size);
+PROC_FILE_DEFINE(smbd_max_send_size);
+PROC_FILE_DEFINE(smbd_send_credit_target);
+PROC_FILE_DEFINE(smbd_receive_credit_max);
+
static struct proc_dir_entry *proc_fs_cifs;
static const struct file_operations cifsFYI_proc_fops;
static const struct file_operations cifs_lookup_cache_proc_fops;
@@ -396,6 +442,22 @@ cifs_proc_init(void)
&cifs_security_flags_proc_fops);
proc_create("LookupCacheEnabled", 0, proc_fs_cifs,
&cifs_lookup_cache_proc_fops);
+ proc_create("rdma_readwrite_threshold", 0, proc_fs_cifs,
+ &cifs_rdma_readwrite_threshold_proc_fops);
+ proc_create("smbd_max_frmr_depth", 0, proc_fs_cifs,
+ &cifs_smbd_max_frmr_depth_proc_fops);
+ proc_create("smbd_keep_alive_interval", 0, proc_fs_cifs,
+ &cifs_smbd_keep_alive_interval_proc_fops);
+ proc_create("smbd_max_receive_size", 0, proc_fs_cifs,
+ &cifs_smbd_max_receive_size_proc_fops);
+ proc_create("smbd_max_fragmented_recv_size", 0, proc_fs_cifs,
+ &cifs_smbd_max_fragmented_recv_size_proc_fops);
+ proc_create("smbd_max_send_size", 0, proc_fs_cifs,
+ &cifs_smbd_max_send_size_proc_fops);
+ proc_create("smbd_send_credit_target", 0, proc_fs_cifs,
+ &cifs_smbd_send_credit_target_proc_fops);
+ proc_create("smbd_receive_credit_max", 0, proc_fs_cifs,
+ &cifs_smbd_receive_credit_max_proc_fops);
}
void
@@ -413,6 +475,14 @@ cifs_proc_clean(void)
remove_proc_entry("SecurityFlags", proc_fs_cifs);
remove_proc_entry("LinuxExtensionsEnabled", proc_fs_cifs);
remove_proc_entry("LookupCacheEnabled", proc_fs_cifs);
+ remove_proc_entry("rdma_readwrite_threshold", proc_fs_cifs);
+ remove_proc_entry("smbd_max_frmr_depth", proc_fs_cifs);
+ remove_proc_entry("smbd_keep_alive_interval", proc_fs_cifs);
+ remove_proc_entry("smbd_max_receive_size", proc_fs_cifs);
+ remove_proc_entry("smbd_max_fragmented_recv_size", proc_fs_cifs);
+ remove_proc_entry("smbd_max_send_size", proc_fs_cifs);
+ remove_proc_entry("smbd_send_credit_target", proc_fs_cifs);
+ remove_proc_entry("smbd_receive_credit_max", proc_fs_cifs);
remove_proc_entry("fs/cifs", NULL);
}
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 04/22] CIFS: SMBD: Add rdma mount option
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (2 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 03/22] CIFS: SMBD: export protocol initial values Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 08/22] CIFS: SMBD: Upper layer reconnects to SMBDirect session Long Li
` (11 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Add "rdma" to CIFS mount options to connect to SMB Direct.
Add checks to validate this is used on SMB 3.X dialects.
To connect to SMBDirect, use "mount.cifs -o rdma,vers=3.x".
At the time of this patch, 3.x can be 3.0, 3.02 or 3.1.1.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/cifs_debug.c | 2 ++
fs/cifs/cifsfs.c | 2 ++
fs/cifs/cifsglob.h | 5 +++++
fs/cifs/connect.c | 15 ++++++++++++++-
4 files changed, 23 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/cifs_debug.c b/fs/cifs/cifs_debug.c
index bdc2f38..9738026 100644
--- a/fs/cifs/cifs_debug.c
+++ b/fs/cifs/cifs_debug.c
@@ -171,6 +171,8 @@ static int cifs_debug_data_proc_show(struct seq_file *m, void *v)
ses->ses_count, ses->serverOS, ses->serverNOS,
ses->capabilities, ses->status);
}
+ if (server->rdma)
+ seq_printf(m, "RDMA\n\t");
seq_printf(m, "TCP status: %d\n\tLocal Users To "
"Server: %d SecMode: 0x%x Req On Wire: %d",
server->tcpStatus, server->srv_count,
diff --git a/fs/cifs/cifsfs.c b/fs/cifs/cifsfs.c
index 180b335..e15fbf1 100644
--- a/fs/cifs/cifsfs.c
+++ b/fs/cifs/cifsfs.c
@@ -327,6 +327,8 @@ cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server)
default:
seq_puts(s, "(unknown)");
}
+ if (server->rdma)
+ seq_puts(s, ",rdma");
}
static void
diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index 808486c..5585516 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -530,6 +530,7 @@ struct smb_vol {
bool nopersistent:1;
bool resilient:1; /* noresilient not required since not fored for CA */
bool domainauto:1;
+ bool rdma:1;
unsigned int rsize;
unsigned int wsize;
bool sockopt_tcp_nodelay:1;
@@ -646,6 +647,10 @@ struct TCP_Server_Info {
bool sec_kerberos; /* supports plain Kerberos */
bool sec_mskerberos; /* supports legacy MS Kerberos */
bool large_buf; /* is current buffer large? */
+ /* use SMBD connection instead of socket */
+ bool rdma;
+ /* point to the SMBD connection if RDMA is used instead of socket */
+ struct smbd_connection *smbd_conn;
struct delayed_work echo; /* echo ping workqueue job */
char *smallbuf; /* pointer to current "small" buffer */
char *bigbuf; /* pointer to current "big" buffer */
diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 59647eb..b5a575f 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -92,7 +92,7 @@ enum {
Opt_multiuser, Opt_sloppy, Opt_nosharesock,
Opt_persistent, Opt_nopersistent,
Opt_resilient, Opt_noresilient,
- Opt_domainauto,
+ Opt_domainauto, Opt_rdma,
/* Mount options which take numeric value */
Opt_backupuid, Opt_backupgid, Opt_uid,
@@ -183,6 +183,7 @@ static const match_table_t cifs_mount_option_tokens = {
{ Opt_resilient, "resilienthandles"},
{ Opt_noresilient, "noresilienthandles"},
{ Opt_domainauto, "domainauto"},
+ { Opt_rdma, "rdma"},
{ Opt_backupuid, "backupuid=%s" },
{ Opt_backupgid, "backupgid=%s" },
@@ -1538,6 +1539,9 @@ cifs_parse_mount_options(const char *mountdata, const char *devname,
case Opt_domainauto:
vol->domainauto = true;
break;
+ case Opt_rdma:
+ vol->rdma = true;
+ break;
/* Numeric Values */
case Opt_backupuid:
@@ -1928,6 +1932,11 @@ cifs_parse_mount_options(const char *mountdata, const char *devname,
goto cifs_parse_mount_err;
}
+ if (vol->rdma && vol->vals->protocol_id < SMB30_PROT_ID) {
+ cifs_dbg(VFS, "SMB Direct requires Version >=3.0\n");
+ goto cifs_parse_mount_err;
+ }
+
#ifndef CONFIG_KEYS
/* Muliuser mounts require CONFIG_KEYS support */
if (vol->multiuser) {
@@ -2131,6 +2140,9 @@ static int match_server(struct TCP_Server_Info *server, struct smb_vol *vol)
if (server->echo_interval != vol->echo_interval * HZ)
return 0;
+ if (server->rdma != vol->rdma)
+ return 0;
+
return 1;
}
@@ -2229,6 +2241,7 @@ cifs_get_tcp_session(struct smb_vol *volume_info)
tcp_ses->noblocksnd = volume_info->noblocksnd;
tcp_ses->noautotune = volume_info->noautotune;
tcp_ses->tcp_nodelay = volume_info->sockopt_tcp_nodelay;
+ tcp_ses->rdma = volume_info->rdma;
tcp_ses->in_flight = 0;
tcp_ses->credits = 1;
init_waitqueue_head(&tcp_ses->response_q);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 08/22] CIFS: SMBD: Upper layer reconnects to SMBDirect session
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (3 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 04/22] CIFS: SMBD: Add rdma mount option Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 09/22] CIFS: SMBD: Implement function to destroy a SMBDirect connection Long Li
` (10 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Do a reconnect on SMBDirect when it is used as the connection. Reconnect can
happen for many reasons and it's mostly the decision of upper layer SMB2.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/connect.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 94b6357..26ad706 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -405,7 +405,11 @@ cifs_reconnect(struct TCP_Server_Info *server)
/* we should try only the port we connected to before */
mutex_lock(&server->srv_mutex);
- rc = generic_ip_connect(server);
+ if (server->rdma)
+ rc = smbd_reconnect(server);
+ else
+ rc = generic_ip_connect(server);
+
if (rc) {
cifs_dbg(FYI, "reconnect error %d\n", rc);
mutex_unlock(&server->srv_mutex);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 09/22] CIFS: SMBD: Implement function to destroy a SMBDirect connection
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (4 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 08/22] CIFS: SMBD: Upper layer reconnects to SMBDirect session Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 11/22] CIFS: SMBD: Set SMBDirect maximum read or write size for I/O Long Li
` (9 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Add function to tear down a SMBDirect connection. This is used by upper layer
to free all SMBDirect connection and transport resources.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smbdirect.c | 16 ++++++++++++++++
fs/cifs/smbdirect.h | 3 +++
2 files changed, 19 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index 1f0f33c..cb129c2 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -1416,6 +1416,22 @@ static void idle_connection_timer(struct work_struct *work)
info->keep_alive_interval*HZ);
}
+/* Destroy this SMBD connection, called from upper layer */
+void smbd_destroy(struct smbd_connection *info)
+{
+ log_rdma_event(INFO, "destroying rdma session\n");
+
+ /* Kick off the disconnection process */
+ smbd_disconnect_rdma_connection(info);
+
+ log_rdma_event(INFO, "wait for transport being destroyed\n");
+ wait_event(info->wait_destroy,
+ info->transport_status == SMBD_DESTROYED);
+
+ destroy_workqueue(info->workqueue);
+ kfree(info);
+}
+
/*
* Reconnect this SMBD connection, called from upper layer
* return value: 0 on success, or actual error code
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index 9818852..d14a484 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -252,6 +252,9 @@ struct smbd_connection *smbd_get_connection(
/* Reconnect SMBDirect session */
int smbd_reconnect(struct TCP_Server_Info *server);
+/* Destroy SMBDirect session */
+void smbd_destroy(struct smbd_connection *info);
+
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 11/22] CIFS: SMBD: Set SMBDirect maximum read or write size for I/O
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (5 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 09/22] CIFS: SMBD: Implement function to destroy a SMBDirect connection Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 12/22] CIFS: SMBD: Implement function to receive data via RDMA receive Long Li
` (8 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
When connecting over SMBDirect, the transport negotiates its maximum I/O sizes
with the server and determines how to choose to do RDMA send/recv vs
read/write. Expose these maximum I/O sizes to upper layer so we will get the
correct sized payloads.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smb2ops.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c
index fb2934b..7ad35d6 100644
--- a/fs/cifs/smb2ops.c
+++ b/fs/cifs/smb2ops.c
@@ -32,6 +32,7 @@
#include "smb2status.h"
#include "smb2glob.h"
#include "cifs_ioctl.h"
+#include "smbdirect.h"
static int
change_conf(struct TCP_Server_Info *server)
@@ -249,7 +250,11 @@ smb2_negotiate_wsize(struct cifs_tcon *tcon, struct smb_vol *volume_info)
/* start with specified wsize, or default */
wsize = volume_info->wsize ? volume_info->wsize : CIFS_DEFAULT_IOSIZE;
- wsize = min_t(unsigned int, wsize, server->max_write);
+ if (server->rdma)
+ wsize = min_t(unsigned int,
+ wsize, server->smbd_conn->max_readwrite_size);
+ else
+ wsize = min_t(unsigned int, wsize, server->max_write);
if (!(server->capabilities & SMB2_GLOBAL_CAP_LARGE_MTU))
wsize = min_t(unsigned int, wsize, SMB2_MAX_BUFFER_SIZE);
@@ -265,7 +270,11 @@ smb2_negotiate_rsize(struct cifs_tcon *tcon, struct smb_vol *volume_info)
/* start with specified rsize, or default */
rsize = volume_info->rsize ? volume_info->rsize : CIFS_DEFAULT_IOSIZE;
- rsize = min_t(unsigned int, rsize, server->max_read);
+ if (server->rdma)
+ rsize = min_t(unsigned int,
+ rsize, server->smbd_conn->max_readwrite_size);
+ else
+ rsize = min_t(unsigned int, rsize, server->max_read);
if (!(server->capabilities & SMB2_GLOBAL_CAP_LARGE_MTU))
rsize = min_t(unsigned int, rsize, SMB2_MAX_BUFFER_SIZE);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 12/22] CIFS: SMBD: Implement function to receive data via RDMA receive
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (6 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 11/22] CIFS: SMBD: Set SMBDirect maximum read or write size for I/O Long Li
@ 2017-10-02 2:30 ` Long Li
[not found] ` <20171002023030.3582-13-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 14/22] CIFS: SMBD: Implement function to send data via RDMA send Long Li
` (7 subsequent siblings)
15 siblings, 1 reply; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
On the receive path, the transport maintains receive buffers and a reassembly
queue for transferring payload via RDMA recv. There is data copy in the
transport on recv when it copies the payload to upper layer.
The transport recognizes the RFC1002 header length use in the SMB
upper layer payloads in CIFS. Because this length is mainly used for TCP and
not applicable to RDMA, it is handled as a out-of-band information and is
never sent over the wire, and the trasnport behaves like TCP to upper layer
by processing and exposing the length correctly on data payloads.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smbdirect.c | 229 ++++++++++++++++++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 6 ++
2 files changed, 235 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index cb129c2..b9be9d6 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -200,6 +200,8 @@ static void smbd_destroy_rdma_work(struct work_struct *work)
log_rdma_event(INFO, "wait for all recv to finish\n");
wake_up_interruptible(&info->wait_reassembly_queue);
+ wait_event(info->wait_smbd_recv_pending,
+ info->smbd_recv_pending == 0);
log_rdma_event(INFO, "wait for all send posted to IB to finish\n");
wait_event(info->wait_send_pending,
@@ -1678,6 +1680,9 @@ struct smbd_connection *_smbd_get_connection(
queue_delayed_work(info->workqueue, &info->idle_timer_work,
info->keep_alive_interval*HZ);
+ init_waitqueue_head(&info->wait_smbd_recv_pending);
+ info->smbd_recv_pending = 0;
+
init_waitqueue_head(&info->wait_send_pending);
atomic_set(&info->send_pending, 0);
@@ -1744,3 +1749,227 @@ struct smbd_connection *smbd_get_connection(
}
return ret;
}
+
+/*
+ * Receive data from receive reassembly queue
+ * All the incoming data packets are placed in reassembly queue
+ * buf: 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
+ * 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.
+ */
+int smbd_recv_buf(struct smbd_connection *info, char *buf, unsigned int size)
+{
+ struct smbd_response *response;
+ struct smbd_data_transfer *data_transfer;
+ int to_copy, to_read, data_read, offset;
+ u32 data_length, remaining_data_length, data_offset;
+ int rc;
+ unsigned long flags;
+
+again:
+ if (info->transport_status != SMBD_CONNECTED) {
+ log_read(ERR, "disconnected\n");
+ return -ENODEV;
+ }
+
+ /*
+ * 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 queeu at the same time
+ */
+ log_read(INFO, "size=%d info->reassembly_data_length=%d\n", size,
+ info->reassembly_data_length);
+ if (info->reassembly_data_length >= size) {
+ unsigned long long t1 = rdtsc();
+ int queue_length;
+ int queue_removed = 0;
+
+ /*
+ * Need to make sure reassembly_data_length is read before
+ * reading reassembly_queue_length and calling
+ * _get_first_reassembly. This call is lock free
+ * as we never read at the end of the queue which are being
+ * updated in SOFTIRQ as more data is received
+ */
+ virt_rmb();
+ queue_length = info->reassembly_queue_length;
+ data_read = 0;
+ to_read = size;
+ offset = info->first_entry_offset;
+ while (data_read < size) {
+ response = _get_first_reassembly(info);
+ data_transfer = smbd_response_payload(response);
+ data_length = le32_to_cpu(data_transfer->data_length);
+ remaining_data_length =
+ le32_to_cpu(
+ data_transfer->remaining_data_length);
+ data_offset = le32_to_cpu(data_transfer->data_offset);
+
+ /*
+ * The upper layer expects RFC1002 length at the
+ * beginning of the payload. Return it to indicate
+ * the total length of the packet. This minimize the
+ * change to upper layer packet processing logic. This
+ * will be eventually remove when an intermediate
+ * transport layer is added
+ */
+ if (response->first_segment && size == 4) {
+ unsigned int rfc1002_len =
+ data_length + remaining_data_length;
+ *((__be32 *)buf) = cpu_to_be32(rfc1002_len);
+ data_read = 4;
+ response->first_segment = false;
+ log_read(INFO, "returning rfc1002 length %d\n",
+ rfc1002_len);
+ goto read_rfc1002_done;
+ }
+
+ to_copy = min_t(int, data_length - offset, to_read);
+ memcpy(
+ buf + data_read,
+ (char *)data_transfer + data_offset + offset,
+ to_copy);
+
+ /* move on to the next buffer? */
+ if (to_copy == data_length - offset) {
+ queue_length--;
+ /*
+ * No need to lock if we are not at the
+ * end of the queue
+ */
+ if (!queue_length)
+ spin_lock_irqsave(
+ &info->reassembly_queue_lock,
+ flags);
+ list_del(&response->list);
+ queue_removed++;
+ if (!queue_length)
+ spin_unlock_irqrestore(
+ &info->reassembly_queue_lock,
+ flags);
+
+ info->count_reassembly_queue--;
+ info->count_dequeue_reassembly_queue++;
+ put_receive_buffer(info, response, true);
+ offset = 0;
+ log_read(INFO, "put_receive_buffer offset=0\n");
+ } else
+ offset += to_copy;
+
+ to_read -= to_copy;
+ data_read += to_copy;
+
+ log_read(INFO, "_get_first_reassembly memcpy %d bytes "
+ "data_transfer_length-offset=%d after that "
+ "to_read=%d data_read=%d offset=%d\n",
+ to_copy, data_length - offset,
+ to_read, data_read, offset);
+ }
+
+ spin_lock_irqsave(&info->reassembly_queue_lock, flags);
+ info->reassembly_data_length -= data_read;
+ info->reassembly_queue_length -= queue_removed;
+ spin_unlock_irqrestore(&info->reassembly_queue_lock, flags);
+
+ info->first_entry_offset = offset;
+ log_read(INFO, "returning to thread data_read=%d "
+ "reassembly_data_length=%d first_entry_offset=%d\n",
+ data_read, info->reassembly_data_length,
+ info->first_entry_offset);
+read_rfc1002_done:
+ profiling_add_histogram(rdtsc() - t1, info->smbd_recv_cycles);
+ return data_read;
+ }
+
+ log_read(INFO, "wait_event on more data\n");
+ rc = wait_event_interruptible(
+ info->wait_reassembly_queue,
+ info->reassembly_data_length >= size ||
+ info->transport_status != SMBD_CONNECTED);
+ /* Don't return any data if interrupted */
+ if (rc)
+ return -ENODEV;
+
+ 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
+ */
+int smbd_recv_page(struct smbd_connection *info,
+ struct page *page, unsigned int to_read)
+{
+ int ret;
+ char *to_address;
+
+ /* make sure we have the page ready for read */
+ ret = wait_event_interruptible(
+ info->wait_reassembly_queue,
+ info->reassembly_data_length >= to_read ||
+ info->transport_status != SMBD_CONNECTED);
+ if (ret)
+ return 0;
+
+ /* now we can read from reassembly queue and not sleep */
+ to_address = kmap_atomic(page);
+
+ 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(to_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;
+ int rc;
+
+ info->smbd_recv_pending++;
+
+ switch (msg->msg_iter.type) {
+ case READ | 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 READ | ITER_BVEC:
+ page = msg->msg_iter.bvec->bv_page;
+ to_read = msg->msg_iter.bvec->bv_len;
+ rc = smbd_recv_page(info, page, to_read);
+ break;
+
+ default:
+ /* It's a bug in upper layer to get there */
+ cifs_dbg(VFS, "CIFS: invalid msg type %d\n",
+ msg->msg_iter.type);
+ rc = -EIO;
+ }
+
+ info->smbd_recv_pending--;
+ wake_up(&info->wait_smbd_recv_pending);
+
+ /* SMBDirect will read it all or nothing */
+ if (rc > 0)
+ msg->msg_iter.count = 0;
+ return rc;
+}
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index d14a484..26614fa 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -88,6 +88,9 @@ struct smbd_connection {
int fragment_reassembly_remaining;
/* Activity accoutning */
+ /* Pending reqeusts issued from upper layer */
+ int smbd_recv_pending;
+ wait_queue_head_t wait_smbd_recv_pending;
atomic_t send_pending;
wait_queue_head_t wait_send_pending;
@@ -255,6 +258,9 @@ int smbd_reconnect(struct TCP_Server_Info *server);
/* Destroy SMBDirect session */
void smbd_destroy(struct smbd_connection *info);
+/* Interface for carrying upper layer I/O through send/recv */
+int smbd_recv(struct smbd_connection *info, struct msghdr *msg);
+
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 14/22] CIFS: SMBD: Implement function to send data via RDMA send
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (7 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 12/22] CIFS: SMBD: Implement function to receive data via RDMA receive Long Li
@ 2017-10-02 2:30 ` Long Li
[not found] ` <20171002023030.3582-15-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 15/22] CIFS: SMBD: Upper layer sends " Long Li
` (6 subsequent siblings)
15 siblings, 1 reply; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
The transport doesn't maintain send buffers or send queue for transferring
payload via RDMA send. There is no data copy in the transport on send.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smbdirect.c | 248 ++++++++++++++++++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 4 +
2 files changed, 252 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index b9be9d6..90e2c94 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -42,6 +42,12 @@ static int smbd_post_recv(
struct smbd_response *response);
static int smbd_post_send_empty(struct smbd_connection *info);
+static int smbd_post_send_data(
+ struct smbd_connection *info,
+ struct kvec *iov, int n_vec, int remaining_data_length);
+static int smbd_post_send_page(struct smbd_connection *info,
+ struct page *page, unsigned long offset,
+ size_t size, int remaining_data_length);
/* SMBD version number */
#define SMBD_V1 0x0100
@@ -198,6 +204,10 @@ static void smbd_destroy_rdma_work(struct work_struct *work)
log_rdma_event(INFO, "cancelling send immediate work\n");
cancel_delayed_work_sync(&info->send_immediate_work);
+ log_rdma_event(INFO, "wait for all send to finish\n");
+ wait_event(info->wait_smbd_send_pending,
+ info->smbd_send_pending == 0);
+
log_rdma_event(INFO, "wait for all recv to finish\n");
wake_up_interruptible(&info->wait_reassembly_queue);
wait_event(info->wait_smbd_recv_pending,
@@ -1103,6 +1113,24 @@ static int smbd_post_send_sgl(struct smbd_connection *info,
}
/*
+ * Send a page
+ * page: the page to send
+ * offset: offset in the page to send
+ * size: length in the page to send
+ * remaining_data_length: remaining data to send in this payload
+ */
+static int smbd_post_send_page(struct smbd_connection *info, struct page *page,
+ unsigned long offset, size_t size, int remaining_data_length)
+{
+ struct scatterlist sgl;
+
+ sg_init_table(&sgl, 1);
+ sg_set_page(&sgl, page, size, offset);
+
+ return smbd_post_send_sgl(info, &sgl, size, remaining_data_length);
+}
+
+/*
* Send an empty message
* Empty message is used to extend credits to peer to for keep live
* while there is no upper layer payload to send at the time
@@ -1114,6 +1142,35 @@ static int smbd_post_send_empty(struct smbd_connection *info)
}
/*
+ * Send a data buffer
+ * iov: the iov array describing the data buffers
+ * n_vec: number of iov array
+ * remaining_data_length: remaining data to send following this packet
+ * in segmented SMBD packet
+ */
+static int smbd_post_send_data(
+ struct smbd_connection *info, struct kvec *iov, int n_vec,
+ int remaining_data_length)
+{
+ int i;
+ u32 data_length = 0;
+ struct scatterlist sgl[SMBDIRECT_MAX_SGE];
+
+ if (n_vec > SMBDIRECT_MAX_SGE) {
+ cifs_dbg(VFS, "Can't fit data to SGL, n_vec=%d\n", n_vec);
+ return -ENOMEM;
+ }
+
+ sg_init_table(sgl, n_vec);
+ for (i = 0; i < n_vec; i++) {
+ data_length += iov[i].iov_len;
+ sg_set_buf(&sgl[i], iov[i].iov_base, iov[i].iov_len);
+ }
+
+ return smbd_post_send_sgl(info, sgl, data_length, remaining_data_length);
+}
+
+/*
* Post a receive request to the transport
* The remote peer can only send data when a receive request is posted
* The interaction is controlled by send/receive credit system
@@ -1680,6 +1737,9 @@ struct smbd_connection *_smbd_get_connection(
queue_delayed_work(info->workqueue, &info->idle_timer_work,
info->keep_alive_interval*HZ);
+ init_waitqueue_head(&info->wait_smbd_send_pending);
+ info->smbd_send_pending = 0;
+
init_waitqueue_head(&info->wait_smbd_recv_pending);
info->smbd_recv_pending = 0;
@@ -1973,3 +2033,191 @@ int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
msg->msg_iter.count = 0;
return rc;
}
+
+/*
+ * Send data to transport
+ * Each rqst is transported as a SMBDirect payload
+ * rqst: the data to write
+ * return value: 0 if successfully write, otherwise error code
+ */
+int smbd_send(struct smbd_connection *info, struct smb_rqst *rqst)
+{
+ struct kvec vec;
+ int nvecs;
+ int size;
+ int buflen = 0, remaining_data_length;
+ int start, i, j;
+ int max_iov_size =
+ info->max_send_size - sizeof(struct smbd_data_transfer);
+ struct kvec iov[SMBDIRECT_MAX_SGE];
+ int rc;
+ unsigned long long t1 = rdtsc();
+
+ info->smbd_send_pending++;
+ if (info->transport_status != SMBD_CONNECTED) {
+ rc = -ENODEV;
+ goto done;
+ }
+
+ /*
+ * This usually means a configuration error
+ * We use RDMA read/write for packet size > rdma_readwrite_threshold
+ * as long as it's properly configured we should never get into this
+ * situation
+ */
+ if (rqst->rq_nvec + rqst->rq_npages > SMBDIRECT_MAX_SGE) {
+ log_write(ERR, "maximum send segment %x exceeding %x\n",
+ rqst->rq_nvec + rqst->rq_npages, SMBDIRECT_MAX_SGE);
+ rc = -EINVAL;
+ goto done;
+ }
+
+ /*
+ * Remove the RFC1002 length defined in MS-SMB2 section 2.1
+ * It is used only for TCP transport
+ * In future we may want to add a transport layer under protocol
+ * layer so this will only be issued to TCP transport
+ */
+ iov[0].iov_base = (char *)rqst->rq_iov[0].iov_base + 4;
+ iov[0].iov_len = rqst->rq_iov[0].iov_len - 4;
+ buflen += iov[0].iov_len;
+
+ /* total up iov array first */
+ for (i = 1; i < rqst->rq_nvec; i++) {
+ iov[i].iov_base = rqst->rq_iov[i].iov_base;
+ iov[i].iov_len = rqst->rq_iov[i].iov_len;
+ buflen += iov[i].iov_len;
+ }
+
+ /* add in the page array if there is one */
+ if (rqst->rq_npages) {
+ buflen += rqst->rq_pagesz * (rqst->rq_npages - 1);
+ buflen += rqst->rq_tailsz;
+ }
+
+ if (buflen + sizeof(struct smbd_data_transfer) >
+ info->max_fragmented_send_size) {
+ log_write(ERR, "payload size %d > max size %d\n",
+ buflen, info->max_fragmented_send_size);
+ rc = -EINVAL;
+ goto done;
+ }
+
+ remaining_data_length = buflen;
+
+ log_write(INFO, "rqst->rq_nvec=%d rqst->rq_npages=%d rq_pagesz=%d "
+ "rq_tailsz=%d buflen=%d\n",
+ rqst->rq_nvec, rqst->rq_npages, rqst->rq_pagesz,
+ rqst->rq_tailsz, buflen);
+
+ start = i = iov[0].iov_len ? 0 : 1;
+ buflen = 0;
+ while (true) {
+ buflen += iov[i].iov_len;
+ if (buflen > max_iov_size) {
+ if (i > start) {
+ remaining_data_length -=
+ (buflen-iov[i].iov_len);
+ log_write(INFO, "sending iov[] from start=%d "
+ "i=%d nvecs=%d "
+ "remaining_data_length=%d\n",
+ start, i, i-start,
+ remaining_data_length);
+ rc = smbd_post_send_data(
+ info, &iov[start], i-start,
+ remaining_data_length);
+ if (rc)
+ goto done;
+ } else {
+ /* iov[start] is too big, break it */
+ nvecs = (buflen+max_iov_size-1)/max_iov_size;
+ log_write(INFO, "iov[%d] iov_base=%p buflen=%d"
+ " break to %d vectors\n",
+ start, iov[start].iov_base,
+ buflen, nvecs);
+ for (j = 0; j < nvecs; j++) {
+ vec.iov_base =
+ (char *)iov[start].iov_base +
+ j*max_iov_size;
+ vec.iov_len = max_iov_size;
+ if (j == nvecs-1)
+ vec.iov_len =
+ buflen -
+ max_iov_size*(nvecs-1);
+ remaining_data_length -= vec.iov_len;
+ log_write(INFO,
+ "sending vec j=%d iov_base=%p"
+ " iov_len=%lu "
+ "remaining_data_length=%d\n",
+ j, vec.iov_base, vec.iov_len,
+ remaining_data_length);
+ rc = smbd_post_send_data(
+ info, &vec, 1,
+ remaining_data_length);
+ if (rc)
+ goto done;
+ }
+ i++;
+ }
+ start = i;
+ buflen = 0;
+ } else {
+ i++;
+ if (i == rqst->rq_nvec) {
+ /* send out all remaining vecs */
+ remaining_data_length -= buflen;
+ log_write(INFO,
+ "sending iov[] from start=%d i=%d "
+ "nvecs=%d remaining_data_length=%d\n",
+ start, i, i-start,
+ remaining_data_length);
+ rc = smbd_post_send_data(info, &iov[start],
+ i-start, remaining_data_length);
+ if (rc)
+ goto done;
+ break;
+ }
+ }
+ log_write(INFO, "looping i=%d buflen=%d\n", i, buflen);
+ }
+
+ /* now sending pages if there are any */
+ for (i = 0; i < rqst->rq_npages; i++) {
+ buflen = (i == rqst->rq_npages-1) ?
+ rqst->rq_tailsz : rqst->rq_pagesz;
+ nvecs = (buflen + max_iov_size - 1) / max_iov_size;
+ log_write(INFO, "sending pages buflen=%d nvecs=%d\n",
+ buflen, nvecs);
+ for (j = 0; j < nvecs; j++) {
+ size = max_iov_size;
+ if (j == nvecs-1)
+ size = buflen - j*max_iov_size;
+ remaining_data_length -= size;
+ log_write(INFO, "sending pages i=%d offset=%d size=%d"
+ " remaining_data_length=%d\n",
+ i, j*max_iov_size, size, remaining_data_length);
+ rc = smbd_post_send_page(
+ info, rqst->rq_pages[i], j*max_iov_size,
+ size, remaining_data_length);
+ if (rc)
+ goto done;
+ }
+ }
+
+done:
+ /*
+ * As an optimization, we don't wait for individual I/O to finish
+ * before sending the next one.
+ * Send them all and wait for pending send count to get to 0
+ * that means all the I/Os have been out and we are good to return
+ */
+
+ wait_event(info->wait_send_payload_pending,
+ atomic_read(&info->send_payload_pending) == 0);
+
+ info->smbd_send_pending--;
+ wake_up(&info->wait_smbd_send_pending);
+ profiling_add_histogram(rdtsc()-t1, info->smbd_write_cycles);
+
+ return rc;
+}
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index 26614fa..e9bd938 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -89,6 +89,9 @@ struct smbd_connection {
/* Activity accoutning */
/* Pending reqeusts issued from upper layer */
+ int smbd_send_pending;
+ wait_queue_head_t wait_smbd_send_pending;
+
int smbd_recv_pending;
wait_queue_head_t wait_smbd_recv_pending;
@@ -260,6 +263,7 @@ void smbd_destroy(struct smbd_connection *info);
/* Interface for carrying upper layer I/O through send/recv */
int smbd_recv(struct smbd_connection *info, struct msghdr *msg);
+int smbd_send(struct smbd_connection *info, struct smb_rqst *rqst);
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 15/22] CIFS: SMBD: Upper layer sends data via RDMA send
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (8 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 14/22] CIFS: SMBD: Implement function to send data via RDMA send Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 16/22] CIFS: SMBD: Fix the definition for SMB2_CHANNEL_RDMA_V1_INVALIDATE Long Li
` (5 subsequent siblings)
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
With SMBDirect connected, use it for sending data via RDMA send.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/transport.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
index 7efbab0..3a9b5a0 100644
--- a/fs/cifs/transport.c
+++ b/fs/cifs/transport.c
@@ -37,6 +37,7 @@
#include "cifsglob.h"
#include "cifsproto.h"
#include "cifs_debug.h"
+#include "smbdirect.h"
void
cifs_wake_up_task(struct mid_q_entry *mid)
@@ -230,6 +231,11 @@ __smb_send_rqst(struct TCP_Server_Info *server, struct smb_rqst *rqst)
struct msghdr smb_msg;
int val = 1;
+ if (server->smbd_conn) {
+ rc = smbd_send(server->smbd_conn, rqst);
+ goto done;
+ }
+
if (ssocket == NULL)
return -ENOTSOCK;
@@ -299,6 +305,7 @@ __smb_send_rqst(struct TCP_Server_Info *server, struct smb_rqst *rqst)
server->tcpStatus = CifsNeedReconnect;
}
+done:
if (rc < 0 && rc != -EINTR)
cifs_dbg(VFS, "Error %d sending data on socket to server\n",
rc);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 16/22] CIFS: SMBD: Fix the definition for SMB2_CHANNEL_RDMA_V1_INVALIDATE
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (9 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 15/22] CIFS: SMBD: Upper layer sends " Long Li
@ 2017-10-02 2:30 ` Long Li
[not found] ` <20171002023030.3582-17-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 17/22] CIFS: SMBD: Implement RDMA memory registration Long Li
` (4 subsequent siblings)
15 siblings, 1 reply; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
The channel value for requesting server remote invalidating local memory
registration should be 0x00000002
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smb2pdu.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/cifs/smb2pdu.h b/fs/cifs/smb2pdu.h
index 393ed5f..f783a08 100644
--- a/fs/cifs/smb2pdu.h
+++ b/fs/cifs/smb2pdu.h
@@ -832,7 +832,7 @@ struct smb2_flush_rsp {
/* Channel field for read and write: exactly one of following flags can be set*/
#define SMB2_CHANNEL_NONE 0x00000000
#define SMB2_CHANNEL_RDMA_V1 0x00000001 /* SMB3 or later */
-#define SMB2_CHANNEL_RDMA_V1_INVALIDATE 0x00000001 /* SMB3.02 or later */
+#define SMB2_CHANNEL_RDMA_V1_INVALIDATE 0x00000002 /* SMB3.02 or later */
/* SMB2 read request without RFC1001 length at the beginning */
struct smb2_read_plain_req {
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 17/22] CIFS: SMBD: Implement RDMA memory registration
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (10 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 16/22] CIFS: SMBD: Fix the definition for SMB2_CHANNEL_RDMA_V1_INVALIDATE Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-04 14:08 ` kbuild test robot
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
` (3 subsequent siblings)
15 siblings, 1 reply; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Memory registration is used for transferring payload via RDMA read or write.
After I/O is done, memory registrations are recovered and reused. This
process can be time consuming and is done in a work queue.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smbdirect.c | 428 ++++++++++++++++++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 55 +++++++
2 files changed, 483 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index 90e2c94..3f2de48 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -49,6 +49,9 @@ static int smbd_post_send_page(struct smbd_connection *info,
struct page *page, unsigned long offset,
size_t size, int remaining_data_length);
+static void destroy_mr_list(struct smbd_connection *info);
+static int allocate_mr_list(struct smbd_connection *info);
+
/* SMBD version number */
#define SMBD_V1 0x0100
@@ -219,6 +222,12 @@ static void smbd_destroy_rdma_work(struct work_struct *work)
wait_event(info->wait_send_payload_pending,
atomic_read(&info->send_payload_pending) == 0);
+ log_rdma_event(INFO, "freeing mr list\n");
+ wake_up_interruptible_all(&info->wait_mr);
+ wait_event(info->wait_for_mr_cleanup,
+ atomic_read(&info->mr_used_count) == 0);
+ destroy_mr_list(info);
+
/* It's not posssible for upper layer to get to reassembly */
log_rdma_event(INFO, "drain the reassembly queue\n");
do {
@@ -475,6 +484,16 @@ static bool process_negotiation_response(
}
info->max_fragmented_send_size =
le32_to_cpu(packet->max_fragmented_size);
+ info->rdma_readwrite_threshold =
+ rdma_readwrite_threshold > info->max_fragmented_send_size ?
+ info->max_fragmented_send_size :
+ rdma_readwrite_threshold;
+
+
+ info->max_readwrite_size = min_t(u32,
+ le32_to_cpu(packet->max_readwrite_size),
+ info->max_frmr_depth * PAGE_SIZE);
+ info->max_frmr_depth = info->max_readwrite_size / PAGE_SIZE;
return true;
}
@@ -773,6 +792,12 @@ static int smbd_ia_open(
rc = -EPROTONOSUPPORT;
goto out2;
}
+ info->max_frmr_depth = min_t(int,
+ smbd_max_frmr_depth,
+ info->id->device->attrs.max_fast_reg_page_list_len);
+ info->mr_type = IB_MR_TYPE_MEM_REG;
+ if (info->id->device->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
+ info->mr_type = IB_MR_TYPE_SG_GAPS;
info->pd = ib_alloc_pd(info->id->device, 0);
if (IS_ERR(info->pd)) {
@@ -1610,6 +1635,8 @@ struct smbd_connection *_smbd_get_connection(
struct rdma_conn_param conn_param;
struct ib_qp_init_attr qp_attr;
struct sockaddr_in *addr_in = (struct sockaddr_in *) dstaddr;
+ struct ib_port_immutable port_immutable;
+ u32 ird_ord_hdr[2];
info = kzalloc(sizeof(struct smbd_connection), GFP_KERNEL);
if (!info)
@@ -1698,6 +1725,28 @@ struct smbd_connection *_smbd_get_connection(
memset(&conn_param, 0, sizeof(conn_param));
conn_param.initiator_depth = 0;
+ conn_param.responder_resources =
+ info->id->device->attrs.max_qp_rd_atom
+ < SMBD_CM_RESPONDER_RESOURCES ?
+ info->id->device->attrs.max_qp_rd_atom :
+ SMBD_CM_RESPONDER_RESOURCES;
+ info->responder_resources = conn_param.responder_resources;
+ log_rdma_mr(INFO, "responder_resources=%d\n",
+ info->responder_resources);
+
+ /* Need to send IRD/ORD in private data for iWARP */
+ info->id->device->get_port_immutable(
+ info->id->device, info->id->port_num, &port_immutable);
+ if (port_immutable.core_cap_flags & RDMA_CORE_PORT_IWARP) {
+ ird_ord_hdr[0] = info->responder_resources;
+ ird_ord_hdr[1] = 1;
+ conn_param.private_data = ird_ord_hdr;
+ conn_param.private_data_len = sizeof(ird_ord_hdr);
+ } else {
+ conn_param.private_data = NULL;
+ conn_param.private_data_len = 0;
+ }
+
conn_param.retry_count = SMBD_CM_RETRY;
conn_param.rnr_retry_count = SMBD_CM_RNR_RETRY;
conn_param.flow_control = 0;
@@ -1762,8 +1811,19 @@ struct smbd_connection *_smbd_get_connection(
goto negotiation_failed;
}
+ rc = allocate_mr_list(info);
+ if (rc) {
+ log_rdma_mr(ERR, "memory registration allocation failed\n");
+ goto allocate_mr_failed;
+ }
+
return info;
+allocate_mr_failed:
+ /* At this point, need to a full transport shutdown */
+ smbd_destroy(info);
+ return NULL;
+
negotiation_failed:
cancel_delayed_work_sync(&info->idle_timer_work);
destroy_caches_and_workqueue(info);
@@ -2221,3 +2281,371 @@ int smbd_send(struct smbd_connection *info, struct smb_rqst *rqst)
return rc;
}
+
+static void register_mr_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct smbd_mr *mr;
+ struct ib_cqe *cqe;
+
+ if (wc->status) {
+ log_rdma_mr(ERR, "status=%d\n", wc->status);
+ cqe = wc->wr_cqe;
+ mr = container_of(cqe, struct smbd_mr, cqe);
+ smbd_disconnect_rdma_connection(mr->conn);
+ }
+}
+
+/*
+ * The work queue function that recovers MRs
+ * We need to call ib_dereg_mr() and ib_alloc_mr() before this MR can be used
+ * again. Both calls are slow, so finish them in a workqueue. This will not
+ * block I/O path.
+ * There is one workqueue that recovers MRs, there is no need to lock as the
+ * I/O requests calling smbd_register_mr will never update the links in the
+ * mr_list.
+ */
+static void smbd_mr_recovery_work(struct work_struct *work)
+{
+ struct smbd_connection *info =
+ container_of(work, struct smbd_connection, mr_recovery_work);
+ struct smbd_mr *smbdirect_mr;
+ int rc;
+
+ list_for_each_entry(smbdirect_mr, &info->mr_list, list) {
+ if (smbdirect_mr->state == MR_INVALIDATED ||
+ smbdirect_mr->state == MR_ERROR) {
+
+ if (smbdirect_mr->state == MR_INVALIDATED) {
+ ib_dma_unmap_sg(
+ info->id->device, smbdirect_mr->sgl,
+ smbdirect_mr->sgl_count,
+ smbdirect_mr->dir);
+ smbdirect_mr->state = MR_READY;
+ } else if (smbdirect_mr->state == MR_ERROR) {
+
+ /* recover this MR entry */
+ rc = ib_dereg_mr(smbdirect_mr->mr);
+ if (rc) {
+ log_rdma_mr(ERR,
+ "ib_dereg_mr faield rc=%x\n",
+ rc);
+ smbd_disconnect_rdma_connection(info);
+ }
+
+ smbdirect_mr->mr = ib_alloc_mr(
+ info->pd, info->mr_type,
+ info->max_frmr_depth);
+ if (IS_ERR(smbdirect_mr->mr)) {
+ log_rdma_mr(ERR,
+ "ib_alloc_mr failed mr_type=%x "
+ "max_frmr_depth=%x\n",
+ info->mr_type,
+ info->max_frmr_depth);
+ smbd_disconnect_rdma_connection(info);
+ }
+
+ smbdirect_mr->state = MR_READY;
+ }
+ /* smbdirect_mr->state is updated by this function
+ * and is read and updated by I/O issuing CPUs trying
+ * to get a MR, the call to atomic_inc_return
+ * implicates a memory barrier and guarantees this
+ * value is updated before waking up any calls to
+ * get_mr() from the I/O issuing CPUs
+ */
+ if (atomic_inc_return(&info->mr_ready_count) == 1)
+ wake_up_interruptible(&info->wait_mr);
+ }
+ }
+}
+
+static void destroy_mr_list(struct smbd_connection *info)
+{
+ struct smbd_mr *mr, *tmp;
+
+ cancel_work_sync(&info->mr_recovery_work);
+ list_for_each_entry_safe(mr, tmp, &info->mr_list, list) {
+ if (mr->state == MR_INVALIDATED)
+ ib_dma_unmap_sg(info->id->device, mr->sgl,
+ mr->sgl_count, mr->dir);
+ ib_dereg_mr(mr->mr);
+ kfree(mr->sgl);
+ kfree(mr);
+ }
+}
+
+/*
+ * Allocate MRs used for RDMA read/write
+ * The number of MRs will not exceed hardware capability in responder_resources
+ * All MRs are kept in mr_list. The MR can be recovered after it's used
+ * Recovery is done in smbd_mr_recovery_work. The content of list entry changes
+ * as MRs are used and recovered for I/O, but the list links will not change
+ */
+static int allocate_mr_list(struct smbd_connection *info)
+{
+ int i;
+ struct smbd_mr *smbdirect_mr, *tmp;
+
+ INIT_LIST_HEAD(&info->mr_list);
+ init_waitqueue_head(&info->wait_mr);
+ spin_lock_init(&info->mr_list_lock);
+ atomic_set(&info->mr_ready_count, 0);
+ atomic_set(&info->mr_used_count, 0);
+ init_waitqueue_head(&info->wait_for_mr_cleanup);
+ /* Allocate more MRs (2x) than hardware responder_resources */
+ for (i = 0; i < info->responder_resources * 2; i++) {
+ smbdirect_mr = kzalloc(sizeof(*smbdirect_mr), GFP_KERNEL);
+ if (!smbdirect_mr)
+ goto out;
+ smbdirect_mr->mr = ib_alloc_mr(info->pd, info->mr_type,
+ info->max_frmr_depth);
+ if (IS_ERR(smbdirect_mr->mr)) {
+ log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x "
+ "max_frmr_depth=%x\n",
+ info->mr_type, info->max_frmr_depth);
+ goto out;
+ }
+ smbdirect_mr->sgl = kcalloc(
+ info->max_frmr_depth,
+ sizeof(struct scatterlist),
+ GFP_KERNEL);
+ if (!smbdirect_mr->sgl) {
+ log_rdma_mr(ERR, "failed to allocate sgl\n");
+ ib_dereg_mr(smbdirect_mr->mr);
+ goto out;
+ }
+ smbdirect_mr->state = MR_READY;
+ smbdirect_mr->conn = info;
+
+ list_add_tail(&smbdirect_mr->list, &info->mr_list);
+ atomic_inc(&info->mr_ready_count);
+ }
+ INIT_WORK(&info->mr_recovery_work, smbd_mr_recovery_work);
+ return 0;
+
+out:
+ kfree(smbdirect_mr);
+
+ list_for_each_entry_safe(smbdirect_mr, tmp, &info->mr_list, list) {
+ ib_dereg_mr(smbdirect_mr->mr);
+ kfree(smbdirect_mr->sgl);
+ kfree(smbdirect_mr);
+ }
+ return -ENOMEM;
+}
+
+/*
+ * Get a MR from mr_list. This function waits until there is at least one
+ * MR available in the list. It may access the list while the
+ * smbd_mr_recovery_work is recovering the MR list. This doesn't need a lock
+ * as they never modify the same places. However, there may be several CPUs
+ * issueing I/O trying to get MR at the same time, mr_list_lock is used to
+ * protect this situation.
+ */
+static struct smbd_mr *get_mr(struct smbd_connection *info)
+{
+ struct smbd_mr *ret;
+ int rc;
+again:
+ rc = wait_event_interruptible(info->wait_mr,
+ atomic_read(&info->mr_ready_count) ||
+ info->transport_status != SMBD_CONNECTED);
+ if (rc) {
+ log_rdma_mr(ERR, "wait_event_interruptible rc=%x\n", rc);
+ return NULL;
+ }
+
+ if (info->transport_status != SMBD_CONNECTED) {
+ log_rdma_mr(ERR, "info->transport_status=%x\n",
+ info->transport_status);
+ return NULL;
+ }
+
+ spin_lock(&info->mr_list_lock);
+ list_for_each_entry(ret, &info->mr_list, list) {
+ if (ret->state == MR_READY) {
+ ret->state = MR_REGISTERED;
+ spin_unlock(&info->mr_list_lock);
+ atomic_dec(&info->mr_ready_count);
+ atomic_inc(&info->mr_used_count);
+ return ret;
+ }
+ }
+
+ spin_unlock(&info->mr_list_lock);
+ /*
+ * It is possible that we can get a MR because other processes may try
+ * to acquire a MR at the same time. If this is the case, retry it.
+ */
+ goto again;
+}
+
+/*
+ * Register memory for RDMA read/write
+ * pages[]: the list of pages to register memory with
+ * num_pages: the number of pages to register
+ * tailsz: if non-zero, the bytes to register in the last page
+ * writing: true if this is a RDMA write (SMB read), false for RDMA read
+ * need_invalidate: true if this MR needs to be locally invalidated after I/O
+ * return value: the MR registered, NULL if failed.
+ */
+struct smbd_mr *smbd_register_mr(
+ struct smbd_connection *info, struct page *pages[], int num_pages,
+ int tailsz, bool writing, bool need_invalidate)
+{
+ struct smbd_mr *smbdirect_mr;
+ int rc, i;
+ enum dma_data_direction dir;
+ struct ib_reg_wr *reg_wr;
+ struct ib_send_wr *bad_wr;
+ unsigned long long t1 = rdtsc();
+
+ if (num_pages > info->max_frmr_depth) {
+ log_rdma_mr(ERR, "num_pages=%d max_frmr_depth=%d\n",
+ num_pages, info->max_frmr_depth);
+ return NULL;
+ }
+
+ smbdirect_mr = get_mr(info);
+ if (!smbdirect_mr) {
+ log_rdma_mr(ERR, "get_mr returning NULL\n");
+ return NULL;
+ }
+ smbdirect_mr->need_invalidate = need_invalidate;
+ smbdirect_mr->sgl_count = num_pages;
+ sg_init_table(smbdirect_mr->sgl, num_pages);
+
+ for (i = 0; i < num_pages - 1; i++)
+ sg_set_page(&smbdirect_mr->sgl[i], pages[i], PAGE_SIZE, 0);
+
+ sg_set_page(&smbdirect_mr->sgl[i], pages[i],
+ tailsz ? tailsz : PAGE_SIZE, 0);
+
+ dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
+ smbdirect_mr->dir = dir;
+ rc = ib_dma_map_sg(info->id->device, smbdirect_mr->sgl, num_pages, dir);
+ if (!rc) {
+ log_rdma_mr(INFO, "ib_dma_map_sg num_pages=%x dir=%x rc=%x\n",
+ num_pages, dir, rc);
+ goto dma_map_error;
+ }
+
+ rc = ib_map_mr_sg(smbdirect_mr->mr, smbdirect_mr->sgl, num_pages,
+ NULL, PAGE_SIZE);
+ if (rc != num_pages) {
+ log_rdma_mr(INFO,
+ "ib_map_mr_sg failed rc = %x num_pages = %x\n",
+ rc, num_pages);
+ goto map_mr_error;
+ }
+
+ ib_update_fast_reg_key(smbdirect_mr->mr,
+ ib_inc_rkey(smbdirect_mr->mr->rkey));
+ reg_wr = &smbdirect_mr->wr;
+ reg_wr->wr.opcode = IB_WR_REG_MR;
+ smbdirect_mr->cqe.done = register_mr_done;
+ reg_wr->wr.wr_cqe = &smbdirect_mr->cqe;
+ reg_wr->wr.num_sge = 0;
+ reg_wr->wr.send_flags = IB_SEND_SIGNALED;
+ reg_wr->mr = smbdirect_mr->mr;
+ reg_wr->key = smbdirect_mr->mr->rkey;
+ reg_wr->access = writing ?
+ IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
+ IB_ACCESS_REMOTE_READ;
+
+ /*
+ * There is no need for waiting for complemtion on ib_post_send
+ * on IB_WR_REG_MR. Hardware enforces a barrier and order of execution
+ * on the next ib_post_send when we actaully send I/O to remote peer
+ */
+ rc = ib_post_send(info->id->qp, ®_wr->wr, &bad_wr);
+ if (!rc) {
+ profiling_add_histogram(
+ rdtsc()-t1, info->smbd_register_mr_cycles);
+ return smbdirect_mr;
+ }
+
+ log_rdma_mr(ERR, "ib_post_send failed rc=%x reg_wr->key=%x\n",
+ rc, reg_wr->key);
+
+ /* If all failed, attempt to recover this MR by setting it MR_ERROR*/
+map_mr_error:
+ ib_dma_unmap_sg(info->id->device, smbdirect_mr->sgl,
+ smbdirect_mr->sgl_count, smbdirect_mr->dir);
+
+dma_map_error:
+ smbdirect_mr->state = MR_ERROR;
+ if (atomic_dec_and_test(&info->mr_used_count))
+ wake_up(&info->wait_for_mr_cleanup);
+
+ return NULL;
+}
+
+static void local_inv_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct smbd_mr *smbdirect_mr;
+ struct ib_cqe *cqe;
+
+ cqe = wc->wr_cqe;
+ smbdirect_mr = container_of(cqe, struct smbd_mr, cqe);
+ smbdirect_mr->state = MR_INVALIDATED;
+ if (wc->status != IB_WC_SUCCESS) {
+ log_rdma_mr(ERR, "invalidate failed status=%x\n", wc->status);
+ smbdirect_mr->state = MR_ERROR;
+ }
+ complete(&smbdirect_mr->invalidate_done);
+}
+
+/*
+ * Deregister a MR after I/O is done
+ * This function may wait if remote invalidation is not used
+ * and we have to locally invalidate the buffer to prevent data is being
+ * modified by remote peer after upper layer consumes it
+ */
+int smbd_deregister_mr(struct smbd_mr *smbdirect_mr)
+{
+ struct ib_send_wr *wr, *bad_wr;
+ struct smbd_connection *info = smbdirect_mr->conn;
+ int rc = 0;
+ unsigned long long t1 = rdtsc();
+
+ if (smbdirect_mr->need_invalidate) {
+ /* Need to finish local invalidation before returning */
+ wr = &smbdirect_mr->inv_wr;
+ wr->opcode = IB_WR_LOCAL_INV;
+ smbdirect_mr->cqe.done = local_inv_done;
+ wr->wr_cqe = &smbdirect_mr->cqe;
+ wr->num_sge = 0;
+ wr->ex.invalidate_rkey = smbdirect_mr->mr->rkey;
+ wr->send_flags = IB_SEND_SIGNALED;
+
+ init_completion(&smbdirect_mr->invalidate_done);
+ rc = ib_post_send(info->id->qp, wr, &bad_wr);
+ if (rc) {
+ log_rdma_mr(ERR, "ib_post_send failed rc=%x\n", rc);
+ smbd_disconnect_rdma_connection(info);
+ goto done;
+ }
+ wait_for_completion(&smbdirect_mr->invalidate_done);
+ smbdirect_mr->need_invalidate = false;
+ } else
+ /*
+ * For remote invalidation, just set it to MR_INVALIDATED
+ * and defer to mr_recovery_work to recover the MR for next use
+ */
+ smbdirect_mr->state = MR_INVALIDATED;
+
+ /*
+ * Schedule the work to do MR recovery for future I/Os
+ * MR recovery is slow and we don't want it to block the current I/O
+ */
+ queue_work(info->workqueue, &info->mr_recovery_work);
+
+ profiling_add_histogram(rdtsc()-t1, info->smbd_deregister_mr_cycles);
+
+done:
+ if (atomic_dec_and_test(&info->mr_used_count))
+ wake_up(&info->wait_for_mr_cleanup);
+
+ return rc;
+}
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index e9bd938..b8ddfe9 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -87,6 +87,29 @@ struct smbd_connection {
int receive_credit_target;
int fragment_reassembly_remaining;
+ /* Memory registrations */
+ /* Maximum number of RDMA read/write outstanding on this connection */
+ int responder_resources;
+ /* Maximum number of SGEs in a RDMA write/read */
+ int max_frmr_depth;
+ /*
+ * If payload is less than or equal to the threshold,
+ * use RDMA send/recv to send upper layer I/O.
+ * If payload is more than the threshold,
+ * use RDMA read/write through memory registration for I/O.
+ */
+ int rdma_readwrite_threshold;
+ enum ib_mr_type mr_type;
+ struct list_head mr_list;
+ spinlock_t mr_list_lock;
+ /* The number of available MRs ready for memory registration */
+ atomic_t mr_ready_count;
+ atomic_t mr_used_count;
+ wait_queue_head_t wait_mr;
+ struct work_struct mr_recovery_work;
+ /* Used by transport to wait until all MRs are returned */
+ wait_queue_head_t wait_for_mr_cleanup;
+
/* Activity accoutning */
/* Pending reqeusts issued from upper layer */
int smbd_send_pending;
@@ -157,6 +180,8 @@ struct smbd_connection {
unsigned int count_send_empty;
#define SIZE_LONG_LONG 64
+ unsigned long long smbd_register_mr_cycles[SIZE_LONG_LONG];
+ unsigned long long smbd_deregister_mr_cycles[SIZE_LONG_LONG];
unsigned long long smbd_write_cycles[SIZE_LONG_LONG];
unsigned long long smbd_recv_cycles[SIZE_LONG_LONG];
unsigned long long recv_done_cycles[SIZE_LONG_LONG];
@@ -265,6 +290,36 @@ void smbd_destroy(struct smbd_connection *info);
int smbd_recv(struct smbd_connection *info, struct msghdr *msg);
int smbd_send(struct smbd_connection *info, struct smb_rqst *rqst);
+enum mr_state {
+ MR_READY,
+ MR_REGISTERED,
+ MR_INVALIDATED,
+ MR_ERROR
+};
+
+struct smbd_mr {
+ struct smbd_connection *conn;
+ struct list_head list;
+ enum mr_state state;
+ struct ib_mr *mr;
+ struct scatterlist *sgl;
+ int sgl_count;
+ enum dma_data_direction dir;
+ union {
+ struct ib_reg_wr wr;
+ struct ib_send_wr inv_wr;
+ };
+ struct ib_cqe cqe;
+ bool need_invalidate;
+ struct completion invalidate_done;
+};
+
+/* Interfaces to register and deregister MR for RDMA read/write */
+struct smbd_mr *smbd_register_mr(
+ struct smbd_connection *info, struct page *pages[], int num_pages,
+ int tailsz, bool writing, bool need_invalidate);
+int smbd_deregister_mr(struct smbd_mr *mr);
+
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* Re: [Patch v4 17/22] CIFS: SMBD: Implement RDMA memory registration
2017-10-02 2:30 ` [Patch v4 17/22] CIFS: SMBD: Implement RDMA memory registration Long Li
@ 2017-10-04 14:08 ` kbuild test robot
0 siblings, 0 replies; 29+ messages in thread
From: kbuild test robot @ 2017-10-04 14:08 UTC (permalink / raw)
To: Long Li
Cc: kbuild-all, Steve French, linux-cifs, samba-technical,
linux-kernel, linux-rdma, Christoph Hellwig, Tom Talpey,
Matthew Wilcox, Long Li
[-- Attachment #1: Type: text/plain, Size: 3278 bytes --]
Hi Long,
[auto build test ERROR on cifs/for-next]
[also build test ERROR on v4.14-rc3 next-20170929]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Long-Li/CIFS-Implement-SMBDirect/20171004-165915
base: git://git.samba.org/sfrench/cifs-2.6.git for-next
config: x86_64-acpi-redef (attached as .config)
compiler: gcc-6 (Debian 6.2.0-3) 6.2.0 20160901
reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64
All errors (new ones prefixed by >>):
fs/cifs/smbdirect.o: In function `smbd_qp_async_error_upcall':
smbdirect.c:(.text+0xb2f): undefined reference to `ib_event_msg'
fs/cifs/smbdirect.o: In function `smbd_mr_recovery_work':
>> smbdirect.c:(.text+0xca4): undefined reference to `ib_dereg_mr'
>> smbdirect.c:(.text+0xce7): undefined reference to `ib_alloc_mr'
fs/cifs/smbdirect.o: In function `smbd_disconnect_rdma_work':
smbdirect.c:(.text+0xed0): undefined reference to `rdma_disconnect'
fs/cifs/smbdirect.o: In function `smbd_destroy_rdma_work':
smbdirect.c:(.text+0x1625): undefined reference to `ib_drain_qp'
smbdirect.c:(.text+0x1631): undefined reference to `rdma_destroy_qp'
smbdirect.c:(.text+0x19de): undefined reference to `ib_dereg_mr'
smbdirect.c:(.text+0x1b65): undefined reference to `ib_free_cq'
smbdirect.c:(.text+0x1b71): undefined reference to `ib_free_cq'
smbdirect.c:(.text+0x1b7d): undefined reference to `ib_dealloc_pd'
smbdirect.c:(.text+0x1b89): undefined reference to `rdma_destroy_id'
fs/cifs/smbdirect.o: In function `_smbd_get_connection':
smbdirect.c:(.text+0x1f86): undefined reference to `rdma_create_id'
smbdirect.c:(.text+0x2008): undefined reference to `rdma_resolve_addr'
smbdirect.c:(.text+0x2075): undefined reference to `rdma_resolve_route'
smbdirect.c:(.text+0x20d6): undefined reference to `rdma_destroy_id'
smbdirect.c:(.text+0x214c): undefined reference to `__ib_alloc_pd'
smbdirect.c:(.text+0x217e): undefined reference to `rdma_destroy_id'
smbdirect.c:(.text+0x22d7): undefined reference to `ib_alloc_cq'
smbdirect.c:(.text+0x2313): undefined reference to `ib_alloc_cq'
smbdirect.c:(.text+0x23a8): undefined reference to `rdma_create_qp'
smbdirect.c:(.text+0x2521): undefined reference to `rdma_connect'
smbdirect.c:(.text+0x3003): undefined reference to `rdma_destroy_qp'
smbdirect.c:(.text+0x3014): undefined reference to `ib_free_cq'
smbdirect.c:(.text+0x3025): undefined reference to `ib_free_cq'
smbdirect.c:(.text+0x302e): undefined reference to `ib_dealloc_pd'
smbdirect.c:(.text+0x3037): undefined reference to `rdma_destroy_id'
smbdirect.c:(.text+0x31b4): undefined reference to `rdma_disconnect'
smbdirect.c:(.text+0x32c6): undefined reference to `ib_alloc_mr'
smbdirect.c:(.text+0x3341): undefined reference to `ib_dereg_mr'
smbdirect.c:(.text+0x3410): undefined reference to `ib_dereg_mr'
fs/cifs/smbdirect.o: In function `smbd_register_mr':
>> smbdirect.c:(.text+0x45e8): undefined reference to `ib_map_mr_sg'
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 30455 bytes --]
^ permalink raw reply [flat|nested] 29+ messages in thread
[parent not found: <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>]
* [Patch v4 05/22] CIFS: SMBD: Implement function to create a SMBDirect connection
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 06/22] CIFS: SMBD: Upper layer connects to SMBDirect session Long Li
` (5 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
The upper layer calls this function to connect to peer through SMBDirect.
Each SMBDirect connection is based on a RC Queue Pair.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/smbdirect.c | 17 +++++++++++++++++
fs/cifs/smbdirect.h | 4 ++++
2 files changed, 21 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index e8f976f..34f73e2 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -1675,3 +1675,20 @@ struct smbd_connection *_smbd_get_connection(
kfree(info);
return NULL;
}
+
+struct smbd_connection *smbd_get_connection(
+ struct TCP_Server_Info *server, struct sockaddr *dstaddr)
+{
+ struct smbd_connection *ret;
+ int port = SMBD_PORT;
+
+try_again:
+ ret = _smbd_get_connection(server, dstaddr, port);
+
+ /* Try SMB_PORT if SMBD_PORT doesn't work */
+ if (!ret && port == SMBD_PORT) {
+ port = SMB_PORT;
+ goto try_again;
+ }
+ return ret;
+}
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index ca60700..42a9338 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -245,6 +245,10 @@ struct smbd_response {
u8 packet[];
};
+/* Create a SMBDirect session */
+struct smbd_connection *smbd_get_connection(
+ struct TCP_Server_Info *server, struct sockaddr *dstaddr);
+
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 06/22] CIFS: SMBD: Upper layer connects to SMBDirect session
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 05/22] CIFS: SMBD: Implement function to create a SMBDirect connection Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 07/22] CIFS: SMBD: Implement function to reconnect to a SMBDirect transport Long Li
` (4 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
When "rdma" is specified in the mount option, CIFS attempts to connect to
SMBDirect instead of TCP socket.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/connect.c | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index b5a575f..94b6357 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -45,6 +45,7 @@
#include <linux/parser.h>
#include <linux/bvec.h>
+#include "smbdirect.h"
#include "cifspdu.h"
#include "cifsglob.h"
#include "cifsproto.h"
@@ -2280,12 +2281,26 @@ cifs_get_tcp_session(struct smb_vol *volume_info)
else
tcp_ses->echo_interval = SMB_ECHO_INTERVAL_DEFAULT * HZ;
+ if (tcp_ses->rdma) {
+ tcp_ses->smbd_conn = smbd_get_connection(
+ tcp_ses, (struct sockaddr *)&volume_info->dstaddr);
+ if (tcp_ses->smbd_conn) {
+ cifs_dbg(VFS, "RDMA transport established\n");
+ rc = 0;
+ goto connected;
+ } else {
+ rc = -ENOENT;
+ goto out_err_crypto_release;
+ }
+ }
+
rc = ip_connect(tcp_ses);
if (rc < 0) {
cifs_dbg(VFS, "Error connecting to socket. Aborting operation.\n");
goto out_err_crypto_release;
}
+connected:
/*
* since we're in a cifs function already, we know that
* this will succeed. No need for try_module_get().
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 07/22] CIFS: SMBD: Implement function to reconnect to a SMBDirect transport
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
2017-10-02 2:30 ` [Patch v4 05/22] CIFS: SMBD: Implement function to create a SMBDirect connection Long Li
2017-10-02 2:30 ` [Patch v4 06/22] CIFS: SMBD: Upper layer connects to SMBDirect session Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 10/22] CIFS: SMBD: Upper layer destroys SMBDirect session on shutdown or umount Long Li
` (3 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
Add function to implement a reconnect to SMBDirect. This involves tearing down
the current connection and establishing/negotiating a new connection.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/smbdirect.c | 36 ++++++++++++++++++++++++++++++++++++
fs/cifs/smbdirect.h | 3 +++
2 files changed, 39 insertions(+)
diff --git a/fs/cifs/smbdirect.c b/fs/cifs/smbdirect.c
index 34f73e2..1f0f33c 100644
--- a/fs/cifs/smbdirect.c
+++ b/fs/cifs/smbdirect.c
@@ -1416,6 +1416,42 @@ static void idle_connection_timer(struct work_struct *work)
info->keep_alive_interval*HZ);
}
+/*
+ * Reconnect this SMBD connection, called from upper layer
+ * return value: 0 on success, or actual error code
+ */
+int smbd_reconnect(struct TCP_Server_Info *server)
+{
+ log_rdma_event(INFO, "reconnecting rdma session\n");
+
+ if (!server->smbd_conn) {
+ log_rdma_event(ERR, "rdma session already destroyed\n");
+ return -EINVAL;
+ }
+
+ /*
+ * This is possible if transport is disconnected and we haven't received
+ * notification from RDMA, but upper layer has detected timeout
+ */
+ if (server->smbd_conn->transport_status == SMBD_CONNECTED) {
+ log_rdma_event(INFO, "disconnecting transport\n");
+ smbd_disconnect_rdma_connection(server->smbd_conn);
+ }
+
+ /* wait until the transport is destroyed */
+ wait_event(server->smbd_conn->wait_destroy,
+ server->smbd_conn->transport_status == SMBD_DESTROYED);
+
+ destroy_workqueue(server->smbd_conn->workqueue);
+ kfree(server->smbd_conn);
+
+ log_rdma_event(INFO, "creating rdma session\n");
+ server->smbd_conn = smbd_get_connection(
+ server, (struct sockaddr *) &server->dstaddr);
+
+ return server->smbd_conn ? 0 : -ENOENT;
+}
+
static void destroy_caches_and_workqueue(struct smbd_connection *info)
{
destroy_receive_buffers(info);
diff --git a/fs/cifs/smbdirect.h b/fs/cifs/smbdirect.h
index 42a9338..9818852 100644
--- a/fs/cifs/smbdirect.h
+++ b/fs/cifs/smbdirect.h
@@ -249,6 +249,9 @@ struct smbd_response {
struct smbd_connection *smbd_get_connection(
struct TCP_Server_Info *server, struct sockaddr *dstaddr);
+/* Reconnect SMBDirect session */
+int smbd_reconnect(struct TCP_Server_Info *server);
+
void profiling_display_histogram(
struct seq_file *m, unsigned long long array[]);
#endif
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 10/22] CIFS: SMBD: Upper layer destroys SMBDirect session on shutdown or umount
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
` (2 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 07/22] CIFS: SMBD: Implement function to reconnect to a SMBDirect transport Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 13/22] CIFS: SMBD: Upper layer receives data via RDMA receive Long Li
` (2 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
When CIFS wants to umount, call shutdown on transport when SMBDirect is used.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/connect.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 26ad706..1a9f22f 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -704,6 +704,11 @@ static void clean_demultiplex_info(struct TCP_Server_Info *server)
/* give those requests time to exit */
msleep(125);
+ if (server->smbd_conn) {
+ smbd_destroy(server->smbd_conn);
+ server->smbd_conn = NULL;
+ }
+
if (server->ssocket) {
sock_release(server->ssocket);
server->ssocket = NULL;
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 13/22] CIFS: SMBD: Upper layer receives data via RDMA receive
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
` (3 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 10/22] CIFS: SMBD: Upper layer destroys SMBDirect session on shutdown or umount Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 18/22] CIFS: SMBD: Upper layer performs SMB write via RDMA read through memory registration Long Li
2017-10-02 2:30 ` [Patch v4 21/22] CIFS: SMBD: Upper layer performs SMB read via RDMA write " Long Li
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
With SMBDirect connected, use it for receiving data via RDMA receive.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/connect.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 1a9f22f..8026682 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -542,7 +542,10 @@ cifs_readv_from_socket(struct TCP_Server_Info *server, struct msghdr *smb_msg)
if (server_unresponsive(server))
return -ECONNABORTED;
- length = sock_recvmsg(server->ssocket, smb_msg, 0);
+ if (server->smbd_conn)
+ length = smbd_recv(server->smbd_conn, smb_msg);
+ else
+ length = sock_recvmsg(server->ssocket, smb_msg, 0);
if (server->tcpStatus == CifsExiting)
return -ESHUTDOWN;
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 18/22] CIFS: SMBD: Upper layer performs SMB write via RDMA read through memory registration
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
` (4 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 13/22] CIFS: SMBD: Upper layer receives data via RDMA receive Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 21/22] CIFS: SMBD: Upper layer performs SMB read via RDMA write " Long Li
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
When sending I/O, if size is larger than rdma_readwrite_threshold we prepare
to send SMB write packet for a RDMA read via memory registration. The actual
I/O is done by remote peer through local RDMA hardware. Modify the relevant
fields in the packet accordingly, and append a smbd_buffer_descriptor_v1 to
the end of the SMB write packet.
On write I/O finish, deregister the memory region if this was for a RDMA read.
If remote invalidation is not used, the call to smbd_deregister_mr will do
local invalidation and possibly wait. Memory region is normally deregistered
in MID callback as soon as it's used. There are situations where the MID may
not be created on I/O failure, under which memory region is deregistered when
write data context is released.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/cifsglob.h | 1 +
fs/cifs/cifssmb.c | 6 ++++++
fs/cifs/smb2pdu.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
3 files changed, 63 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index 5585516..bcb6df1 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -1168,6 +1168,7 @@ struct cifs_writedata {
pid_t pid;
unsigned int bytes;
int result;
+ struct smbd_mr *mr;
unsigned int pagesz;
unsigned int tailsz;
unsigned int credits;
diff --git a/fs/cifs/cifssmb.c b/fs/cifs/cifssmb.c
index 5857009..0e29ecf 100644
--- a/fs/cifs/cifssmb.c
+++ b/fs/cifs/cifssmb.c
@@ -43,6 +43,7 @@
#include "cifs_unicode.h"
#include "cifs_debug.h"
#include "fscache.h"
+#include "smbdirect.h"
#ifdef CONFIG_CIFS_POSIX
static struct {
@@ -1912,6 +1913,11 @@ cifs_writedata_release(struct kref *refcount)
struct cifs_writedata *wdata = container_of(refcount,
struct cifs_writedata, refcount);
+ if (wdata->mr) {
+ smbd_deregister_mr(wdata->mr);
+ wdata->mr = NULL;
+ }
+
if (wdata->cfile)
cifsFileInfo_put(wdata->cfile);
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
index bab3da6..6089957 100644
--- a/fs/cifs/smb2pdu.c
+++ b/fs/cifs/smb2pdu.c
@@ -48,6 +48,7 @@
#include "smb2glob.h"
#include "cifspdu.h"
#include "cifs_spnego.h"
+#include "smbdirect.h"
/*
* The following table defines the expected "StructureSize" of SMB2 requests
@@ -2653,6 +2654,18 @@ smb2_writev_callback(struct mid_q_entry *mid)
break;
}
+ /*
+ * If this wdata has a memory registered, the MR can be freed
+ * The number of MRs available is limited, it's important to recover
+ * used MR as soon as I/O is finished. Hold MR longer in the later
+ * I/O process can possibly result in I/O deadlock due to lack of MR
+ * to send request on I/O retry
+ */
+ if (wdata->mr) {
+ smbd_deregister_mr(wdata->mr);
+ wdata->mr = NULL;
+ }
+
if (wdata->result)
cifs_stats_fail_inc(tcon, SMB2_WRITE_HE);
@@ -2704,6 +2717,41 @@ smb2_async_writev(struct cifs_writedata *wdata,
offsetof(struct smb2_write_req, Buffer) - 4);
req->RemainingBytes = 0;
+ /*
+ * If we want to do a server RDMA read, fill in and append
+ * smbd_buffer_descriptor_v1 to the end of write request
+ */
+ if (server->rdma && wdata->bytes >=
+ server->smbd_conn->rdma_readwrite_threshold) {
+
+ struct smbd_buffer_descriptor_v1 *v1;
+ bool need_invalidate = server->dialect == SMB30_PROT_ID;
+
+ wdata->mr = smbd_register_mr(
+ server->smbd_conn, wdata->pages,
+ wdata->nr_pages, wdata->tailsz,
+ false, need_invalidate);
+ if (!wdata->mr) {
+ rc = -ENOBUFS;
+ goto async_writev_out;
+ }
+ req->Length = 0;
+ req->DataOffset = 0;
+ req->RemainingBytes =
+ (wdata->nr_pages-1)*PAGE_SIZE + wdata->tailsz;
+ req->Channel = SMB2_CHANNEL_RDMA_V1_INVALIDATE;
+ if (need_invalidate)
+ req->Channel = SMB2_CHANNEL_RDMA_V1;
+ req->WriteChannelInfoOffset =
+ offsetof(struct smb2_write_req, Buffer) - 4;
+ req->WriteChannelInfoLength =
+ sizeof(struct smbd_buffer_descriptor_v1);
+ v1 = (struct smbd_buffer_descriptor_v1 *) &req->Buffer[0];
+ v1->offset = wdata->mr->mr->iova;
+ v1->token = wdata->mr->mr->rkey;
+ v1->length = wdata->mr->mr->length;
+ }
+
/* 4 for rfc1002 length field and 1 for Buffer */
iov[0].iov_len = 4;
iov[0].iov_base = req;
@@ -2717,10 +2765,17 @@ smb2_async_writev(struct cifs_writedata *wdata,
rqst.rq_pagesz = wdata->pagesz;
rqst.rq_tailsz = wdata->tailsz;
+ if (wdata->mr) {
+ iov[1].iov_len += sizeof(struct smbd_buffer_descriptor_v1);
+ rqst.rq_npages = 0;
+ }
+
cifs_dbg(FYI, "async write at %llu %u bytes\n",
wdata->offset, wdata->bytes);
- req->Length = cpu_to_le32(wdata->bytes);
+ /* For RDMA read, I/O size is in RemainingBytes not in Length */
+ if (!wdata->mr)
+ req->Length = cpu_to_le32(wdata->bytes);
inc_rfc1001_len(&req->hdr, wdata->bytes - 1 /* Buffer */);
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 21/22] CIFS: SMBD: Upper layer performs SMB read via RDMA write through memory registration
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
` (5 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 18/22] CIFS: SMBD: Upper layer performs SMB write via RDMA read through memory registration Long Li
@ 2017-10-02 2:30 ` Long Li
6 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs-u79uwXL29TY76Z2rM5mHXA,
samba-technical-w/Ol4Ecudpl8XjKLYN78aQ,
linux-kernel-u79uwXL29TY76Z2rM5mHXA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, Christoph Hellwig, Tom Talpey,
Matthew Wilcox
Cc: Long Li
From: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
If I/O size is larger than rdma_readwrite_threshold, use RDMA write for
SMB read by specifying channel SMB2_CHANNEL_RDMA_V1 or
SMB2_CHANNEL_RDMA_V1_INVALIDATE in the SMB packet, depending on SMB dialect
used. Append a smbd_buffer_descriptor_v1 to the end of the SMB packet and fill
in other values to indicate this SMB read uses RDMA write.
There is no need to read from the transport for incoming payload. At the time
SMB read response comes back, the data is already transfered and placed in the
pages by RDMA hardware.
When SMB read is finished, deregister the memory regions if RDMA write is used
for this SMB read. smbd_deregister_mr may need to do local invalidation and
sleep, if server remote invalidation is not used.
There are situations where the MID may not be created on I/O failure, under
which memory region is deregistered when read data context is released.
Signed-off-by: Long Li <longli-0li6OtcxBFHby3iVrkZq2A@public.gmane.org>
---
fs/cifs/cifsglob.h | 1 +
fs/cifs/file.c | 10 ++++++++++
fs/cifs/smb2pdu.c | 43 +++++++++++++++++++++++++++++++++++++++++++
3 files changed, 54 insertions(+)
diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index f851b50..30b99a5 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -1152,6 +1152,7 @@ struct cifs_readdata {
struct cifs_readdata *rdata,
struct iov_iter *iter);
struct kvec iov[2];
+ struct smbd_mr *mr;
unsigned int pagesz;
unsigned int tailsz;
unsigned int credits;
diff --git a/fs/cifs/file.c b/fs/cifs/file.c
index 0786f19..8396f1e 100644
--- a/fs/cifs/file.c
+++ b/fs/cifs/file.c
@@ -42,6 +42,7 @@
#include "cifs_debug.h"
#include "cifs_fs_sb.h"
#include "fscache.h"
+#include "smbdirect.h"
static inline int cifs_convert_flags(unsigned int flags)
@@ -2909,6 +2910,11 @@ cifs_readdata_release(struct kref *refcount)
struct cifs_readdata *rdata = container_of(refcount,
struct cifs_readdata, refcount);
+ if (rdata->mr) {
+ smbd_deregister_mr(rdata->mr);
+ rdata->mr = NULL;
+ }
+
if (rdata->cfile)
cifsFileInfo_put(rdata->cfile);
@@ -3037,6 +3043,8 @@ uncached_fill_pages(struct TCP_Server_Info *server,
}
if (iter)
result = copy_page_from_iter(page, 0, n, iter);
+ else if (rdata->mr)
+ result = n;
else
result = cifs_read_page_from_socket(server, page, n);
if (result < 0)
@@ -3606,6 +3614,8 @@ readpages_fill_pages(struct TCP_Server_Info *server,
if (iter)
result = copy_page_from_iter(page, 0, n, iter);
+ else if (rdata->mr)
+ result = n;
else
result = cifs_read_page_from_socket(server, page, n);
if (result < 0)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
index 7053db9..31dcee0 100644
--- a/fs/cifs/smb2pdu.c
+++ b/fs/cifs/smb2pdu.c
@@ -2380,6 +2380,39 @@ smb2_new_read_req(void **buf, unsigned int *total_len,
req->Length = cpu_to_le32(io_parms->length);
req->Offset = cpu_to_le64(io_parms->offset);
+ /*
+ * If we want to do a RDMA write, fill in and append
+ * smbd_buffer_descriptor_v1 to the end of read request
+ */
+ if (server->rdma && rdata &&
+ rdata->bytes >= server->smbd_conn->rdma_readwrite_threshold) {
+
+ struct smbd_buffer_descriptor_v1 *v1;
+ bool need_invalidate =
+ io_parms->tcon->ses->server->dialect == SMB30_PROT_ID;
+
+ rdata->mr = smbd_register_mr(
+ server->smbd_conn, rdata->pages,
+ rdata->nr_pages, rdata->tailsz,
+ true, need_invalidate);
+ if (!rdata->mr)
+ return -ENOBUFS;
+
+ req->Channel = SMB2_CHANNEL_RDMA_V1_INVALIDATE;
+ if (need_invalidate)
+ req->Channel = SMB2_CHANNEL_RDMA_V1;
+ req->ReadChannelInfoOffset =
+ offsetof(struct smb2_read_plain_req, Buffer);
+ req->ReadChannelInfoLength =
+ sizeof(struct smbd_buffer_descriptor_v1);
+ v1 = (struct smbd_buffer_descriptor_v1 *) &req->Buffer[0];
+ v1->offset = rdata->mr->mr->iova;
+ v1->token = rdata->mr->mr->rkey;
+ v1->length = rdata->mr->mr->length;
+
+ *total_len += sizeof(*v1) - 1;
+ }
+
if (request_type & CHAINED_REQUEST) {
if (!(request_type & END_OF_CHAIN)) {
/* next 8-byte aligned request */
@@ -2459,6 +2492,16 @@ smb2_readv_callback(struct mid_q_entry *mid)
rdata->result = -EIO;
}
+ /*
+ * If this rdata has a memmory registered, the MR can be freed
+ * MR needs to be freed as soon as I/O finishes to prevent deadlock
+ * because they have limited number and are used for future I/Os
+ */
+ if (rdata->mr) {
+ smbd_deregister_mr(rdata->mr);
+ rdata->mr = NULL;
+ }
+
if (rdata->result)
cifs_stats_fail_inc(tcon, SMB2_READ_HE);
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 29+ messages in thread
* [Patch v4 19/22] CIFS: SMBD: Add parameter rdata to smb2_new_read_req
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (12 preceding siblings ...)
[not found] ` <20171002023030.3582-1-longli-Lp/cVzEoVyZiJJESP9tAQJZ3qXmFLfmx@public.gmane.org>
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 20/22] CIFS: SMBD: Read correct returned data length for RDMA write (SMB read) I/O Long Li
2017-10-02 2:30 ` [Patch v4 22/22] CIFS: SMBD: Add SMBDirect debug counters Long Li
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
This patch is for preparing upper layer for doing SMB read via RDMA write.
When we assemble the SMB read packet header, we need to know the I/O layout
if this request is to use a RDMA write. rdata has all the information we need
for memory registration. Add rdata to smb2_new_read_req.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/smb2pdu.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
index 6089957..7053db9 100644
--- a/fs/cifs/smb2pdu.c
+++ b/fs/cifs/smb2pdu.c
@@ -2351,18 +2351,21 @@ SMB2_flush(const unsigned int xid, struct cifs_tcon *tcon, u64 persistent_fid,
*/
static int
smb2_new_read_req(void **buf, unsigned int *total_len,
- struct cifs_io_parms *io_parms, unsigned int remaining_bytes,
- int request_type)
+ struct cifs_io_parms *io_parms, struct cifs_readdata *rdata,
+ unsigned int remaining_bytes, int request_type)
{
int rc = -EACCES;
struct smb2_read_plain_req *req = NULL;
struct smb2_sync_hdr *shdr;
+ struct TCP_Server_Info *server;
rc = smb2_plain_req_init(SMB2_READ, io_parms->tcon, (void **) &req,
total_len);
if (rc)
return rc;
- if (io_parms->tcon->ses->server == NULL)
+
+ server = io_parms->tcon->ses->server;
+ if (server == NULL)
return -ECONNABORTED;
shdr = &req->sync_hdr;
@@ -2490,7 +2493,8 @@ smb2_async_readv(struct cifs_readdata *rdata)
server = io_parms.tcon->ses->server;
- rc = smb2_new_read_req((void **) &buf, &total_len, &io_parms, 0, 0);
+ rc = smb2_new_read_req(
+ (void **) &buf, &total_len, &io_parms, rdata, 0, 0);
if (rc) {
if (rc == -EAGAIN && rdata->credits) {
/* credits was reset by reconnect */
@@ -2558,7 +2562,7 @@ SMB2_read(const unsigned int xid, struct cifs_io_parms *io_parms,
struct cifs_ses *ses = io_parms->tcon->ses;
*nbytes = 0;
- rc = smb2_new_read_req((void **)&req, &total_len, io_parms, 0, 0);
+ rc = smb2_new_read_req((void **)&req, &total_len, io_parms, NULL, 0, 0);
if (rc)
return rc;
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 20/22] CIFS: SMBD: Read correct returned data length for RDMA write (SMB read) I/O
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (13 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 19/22] CIFS: SMBD: Add parameter rdata to smb2_new_read_req Long Li
@ 2017-10-02 2:30 ` Long Li
2017-10-02 2:30 ` [Patch v4 22/22] CIFS: SMBD: Add SMBDirect debug counters Long Li
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
This patch is for preparing upper layer for doing SMB read via RDMA write.
When RDMA write is used for SMB read, the returned data length is in
DataRemaining in the response packet. Reading it properly by adding a
parameter to specifiy where the returned data length is.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/cifsglob.h | 10 ++++++++--
fs/cifs/cifssmb.c | 4 ++--
fs/cifs/smb1ops.c | 2 +-
fs/cifs/smb2ops.c | 8 ++++++--
4 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index bcb6df1..f851b50 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -228,8 +228,14 @@ struct smb_version_operations {
__u64 (*get_next_mid)(struct TCP_Server_Info *);
/* data offset from read response message */
unsigned int (*read_data_offset)(char *);
- /* data length from read response message */
- unsigned int (*read_data_length)(char *);
+ /*
+ * Data length from read response message
+ * When in_remaining is true, the returned data length is in
+ * message field DataRemaining for out-of-band data read (e.g through
+ * Memory Registration RDMA write in SMBD).
+ * Otherwise, the returned data length is in message field DataLength.
+ */
+ unsigned int (*read_data_length)(char *, bool in_remaining);
/* map smb to linux error */
int (*map_error)(char *, bool);
/* find mid corresponding to the response message */
diff --git a/fs/cifs/cifssmb.c b/fs/cifs/cifssmb.c
index 0e29ecf..b9410e1 100644
--- a/fs/cifs/cifssmb.c
+++ b/fs/cifs/cifssmb.c
@@ -1531,8 +1531,8 @@ cifs_readv_receive(struct TCP_Server_Info *server, struct mid_q_entry *mid)
rdata->iov[0].iov_base, server->total_read);
/* how much data is in the response? */
- data_len = server->ops->read_data_length(buf);
- if (data_offset + data_len > buflen) {
+ data_len = server->ops->read_data_length(buf, rdata->mr);
+ if (!rdata->mr && (data_offset + data_len > buflen)) {
/* data_len is corrupt -- discard frame */
rdata->result = -EIO;
return cifs_readv_discard(server, mid);
diff --git a/fs/cifs/smb1ops.c b/fs/cifs/smb1ops.c
index a723df3..27a8280 100644
--- a/fs/cifs/smb1ops.c
+++ b/fs/cifs/smb1ops.c
@@ -87,7 +87,7 @@ cifs_read_data_offset(char *buf)
}
static unsigned int
-cifs_read_data_length(char *buf)
+cifs_read_data_length(char *buf, bool in_remaining)
{
READ_RSP *rsp = (READ_RSP *)buf;
return (le16_to_cpu(rsp->DataLengthHigh) << 16) +
diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c
index 7ad35d6..a765877 100644
--- a/fs/cifs/smb2ops.c
+++ b/fs/cifs/smb2ops.c
@@ -935,9 +935,13 @@ smb2_read_data_offset(char *buf)
}
static unsigned int
-smb2_read_data_length(char *buf)
+smb2_read_data_length(char *buf, bool in_remaining)
{
struct smb2_read_rsp *rsp = (struct smb2_read_rsp *)buf;
+
+ if (in_remaining)
+ return le32_to_cpu(rsp->DataRemaining);
+
return le32_to_cpu(rsp->DataLength);
}
@@ -2446,7 +2450,7 @@ handle_read_data(struct TCP_Server_Info *server, struct mid_q_entry *mid,
}
data_offset = server->ops->read_data_offset(buf) + 4;
- data_len = server->ops->read_data_length(buf);
+ data_len = server->ops->read_data_length(buf, rdata->mr);
if (data_offset < server->vals->read_rsp_size) {
/*
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread* [Patch v4 22/22] CIFS: SMBD: Add SMBDirect debug counters
2017-10-02 2:30 [Patch v4 00/22] CIFS: Implement SMBDirect Long Li
` (14 preceding siblings ...)
2017-10-02 2:30 ` [Patch v4 20/22] CIFS: SMBD: Read correct returned data length for RDMA write (SMB read) I/O Long Li
@ 2017-10-02 2:30 ` Long Li
15 siblings, 0 replies; 29+ messages in thread
From: Long Li @ 2017-10-02 2:30 UTC (permalink / raw)
To: Steve French, linux-cifs, samba-technical, linux-kernel,
linux-rdma, Christoph Hellwig, Tom Talpey, Matthew Wilcox
Cc: Long Li
From: Long Li <longli@microsoft.com>
Export SMBDirect debug counters to /proc/fs/cifs/DebugData.
Those are used for debugging, troubleshooting and profiling.
Signed-off-by: Long Li <longli@microsoft.com>
---
fs/cifs/cifs_debug.c | 87 ++++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 87 insertions(+)
diff --git a/fs/cifs/cifs_debug.c b/fs/cifs/cifs_debug.c
index 9738026..1ea78d5 100644
--- a/fs/cifs/cifs_debug.c
+++ b/fs/cifs/cifs_debug.c
@@ -30,6 +30,7 @@
#include "cifsproto.h"
#include "cifs_debug.h"
#include "cifsfs.h"
+#include "smbdirect.h"
void
cifs_dump_mem(char *label, void *data, int length)
@@ -152,6 +153,92 @@ static int cifs_debug_data_proc_show(struct seq_file *m, void *v)
list_for_each(tmp1, &cifs_tcp_ses_list) {
server = list_entry(tmp1, struct TCP_Server_Info,
tcp_ses_list);
+
+ if (!server->rdma)
+ goto skip_rdma;
+
+ seq_printf(m, "\nSMBDirect (in hex) protocol version: %x "
+ "transport status: %x",
+ server->smbd_conn->protocol,
+ server->smbd_conn->transport_status);
+ seq_printf(m, "\nConn receive_credit_max: %x "
+ "send_credit_target: %x max_send_size: %x",
+ server->smbd_conn->receive_credit_max,
+ server->smbd_conn->send_credit_target,
+ server->smbd_conn->max_send_size);
+ seq_printf(m, "\nConn max_fragmented_recv_size: %x "
+ "max_fragmented_send_size: %x max_receive_size:%x",
+ server->smbd_conn->max_fragmented_recv_size,
+ server->smbd_conn->max_fragmented_send_size,
+ server->smbd_conn->max_receive_size);
+ seq_printf(m, "\nConn keep_alive_interval: %x "
+ "max_readwrite_size: %x rdma_readwrite_threshold: %x",
+ server->smbd_conn->keep_alive_interval,
+ server->smbd_conn->max_readwrite_size,
+ server->smbd_conn->rdma_readwrite_threshold);
+ seq_printf(m, "\nDebug count_get_receive_buffer: %x "
+ "count_put_receive_buffer: %x count_send_empty: %x",
+ server->smbd_conn->count_get_receive_buffer,
+ server->smbd_conn->count_put_receive_buffer,
+ server->smbd_conn->count_send_empty);
+ seq_printf(m, "\nRead Queue count_reassembly_queue: %x "
+ "count_enqueue_reassembly_queue: %x "
+ "count_dequeue_reassembly_queue: %x "
+ "fragment_reassembly_remaining: %x "
+ "reassembly_data_length: %x "
+ "reassembly_queue_length: %x",
+ server->smbd_conn->count_reassembly_queue,
+ server->smbd_conn->count_enqueue_reassembly_queue,
+ server->smbd_conn->count_dequeue_reassembly_queue,
+ server->smbd_conn->fragment_reassembly_remaining,
+ server->smbd_conn->reassembly_data_length,
+ server->smbd_conn->reassembly_queue_length);
+ seq_printf(m, "\nCurrent Credits send_credits: %x "
+ "receive_credits: %x receive_credit_target: %x",
+ atomic_read(&server->smbd_conn->send_credits),
+ atomic_read(&server->smbd_conn->receive_credits),
+ server->smbd_conn->receive_credit_target);
+ seq_printf(m, "\nPending send_pending: %x send_payload_pending:"
+ " %x smbd_send_pending: %x smbd_recv_pending: %x",
+ atomic_read(&server->smbd_conn->send_pending),
+ atomic_read(&server->smbd_conn->send_payload_pending),
+ server->smbd_conn->smbd_send_pending,
+ server->smbd_conn->smbd_recv_pending);
+ seq_printf(m, "\nReceive buffers count_receive_queue: %x "
+ "count_empty_packet_queue: %x",
+ server->smbd_conn->count_receive_queue,
+ server->smbd_conn->count_empty_packet_queue);
+ seq_printf(m, "\nMR responder_resources: %x "
+ "max_frmr_depth: %x mr_type: %x",
+ server->smbd_conn->responder_resources,
+ server->smbd_conn->max_frmr_depth,
+ server->smbd_conn->mr_type);
+ seq_printf(m, "\nMR mr_ready_count: %x mr_used_count: %x",
+ atomic_read(&server->smbd_conn->mr_ready_count),
+ atomic_read(&server->smbd_conn->mr_used_count));
+
+ seq_printf(m, "\nTSC cycle histogram in I/O path: "
+ "(the number of most significant bits)");
+ seq_printf(m, "\nsmbd_register_mr cycles [1/10000]: ");
+ profiling_display_histogram(
+ m, server->smbd_conn->smbd_register_mr_cycles);
+
+ seq_printf(m, "\nsmbd_deregister_mr cycles [1/10000]: ");
+ profiling_display_histogram(
+ m, server->smbd_conn->smbd_deregister_mr_cycles);
+
+ seq_printf(m, "\nsmbd_send cycles [1/10000]: ");
+ profiling_display_histogram(
+ m, server->smbd_conn->smbd_write_cycles);
+
+ seq_printf(m, "\nsmbd_recv cycles [1/10000]: ");
+ profiling_display_histogram(
+ m, server->smbd_conn->smbd_recv_cycles);
+
+ seq_printf(m, "\nrecv_done cycles [1/10000]: ");
+ profiling_display_histogram(
+ m, server->smbd_conn->recv_done_cycles);
+skip_rdma:
seq_printf(m, "\nNumber of credits: %d", server->credits);
i++;
list_for_each(tmp2, &server->smb_ses_list) {
--
2.7.4
^ permalink raw reply related [flat|nested] 29+ messages in thread