1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2016 Namjae Jeon <namjae.jeon@protocolfreedom.org>
4 * Copyright (C) 2018 Samsung Electronics Co., Ltd.
5 */
6
7 #include <linux/mutex.h>
8 #include <linux/freezer.h>
9 #include <linux/module.h>
10
11 #include "server.h"
12 #include "smb_common.h"
13 #include "mgmt/ksmbd_ida.h"
14 #include "connection.h"
15 #include "transport_tcp.h"
16 #include "transport_rdma.h"
17
18 static DEFINE_MUTEX(init_lock);
19
20 static struct ksmbd_conn_ops default_conn_ops;
21
22 LIST_HEAD(conn_list);
23 DEFINE_RWLOCK(conn_list_lock);
24
25 /**
26 * ksmbd_conn_free() - free resources of the connection instance
27 *
28 * @conn: connection instance to be cleand up
29 *
30 * During the thread termination, the corresponding conn instance
31 * resources(sock/memory) are released and finally the conn object is freed.
32 */
ksmbd_conn_free(struct ksmbd_conn * conn)33 void ksmbd_conn_free(struct ksmbd_conn *conn)
34 {
35 write_lock(&conn_list_lock);
36 list_del(&conn->conns_list);
37 write_unlock(&conn_list_lock);
38
39 kvfree(conn->request_buf);
40 kfree(conn->preauth_info);
41 kfree(conn);
42 }
43
44 /**
45 * ksmbd_conn_alloc() - initialize a new connection instance
46 *
47 * Return: ksmbd_conn struct on success, otherwise NULL
48 */
ksmbd_conn_alloc(void)49 struct ksmbd_conn *ksmbd_conn_alloc(void)
50 {
51 struct ksmbd_conn *conn;
52
53 conn = kzalloc(sizeof(struct ksmbd_conn), GFP_KERNEL);
54 if (!conn)
55 return NULL;
56
57 conn->need_neg = true;
58 conn->status = KSMBD_SESS_NEW;
59 conn->local_nls = load_nls("utf8");
60 if (!conn->local_nls)
61 conn->local_nls = load_nls_default();
62 atomic_set(&conn->req_running, 0);
63 atomic_set(&conn->r_count, 0);
64 conn->total_credits = 1;
65
66 init_waitqueue_head(&conn->req_running_q);
67 INIT_LIST_HEAD(&conn->conns_list);
68 INIT_LIST_HEAD(&conn->sessions);
69 INIT_LIST_HEAD(&conn->requests);
70 INIT_LIST_HEAD(&conn->async_requests);
71 spin_lock_init(&conn->request_lock);
72 spin_lock_init(&conn->credits_lock);
73 ida_init(&conn->async_ida);
74
75 spin_lock_init(&conn->llist_lock);
76 INIT_LIST_HEAD(&conn->lock_list);
77
78 write_lock(&conn_list_lock);
79 list_add(&conn->conns_list, &conn_list);
80 write_unlock(&conn_list_lock);
81 return conn;
82 }
83
ksmbd_conn_lookup_dialect(struct ksmbd_conn * c)84 bool ksmbd_conn_lookup_dialect(struct ksmbd_conn *c)
85 {
86 struct ksmbd_conn *t;
87 bool ret = false;
88
89 read_lock(&conn_list_lock);
90 list_for_each_entry(t, &conn_list, conns_list) {
91 if (memcmp(t->ClientGUID, c->ClientGUID, SMB2_CLIENT_GUID_SIZE))
92 continue;
93
94 ret = true;
95 break;
96 }
97 read_unlock(&conn_list_lock);
98 return ret;
99 }
100
ksmbd_conn_enqueue_request(struct ksmbd_work * work)101 void ksmbd_conn_enqueue_request(struct ksmbd_work *work)
102 {
103 struct ksmbd_conn *conn = work->conn;
104 struct list_head *requests_queue = NULL;
105
106 if (conn->ops->get_cmd_val(work) != SMB2_CANCEL_HE) {
107 requests_queue = &conn->requests;
108 work->syncronous = true;
109 }
110
111 if (requests_queue) {
112 atomic_inc(&conn->req_running);
113 spin_lock(&conn->request_lock);
114 list_add_tail(&work->request_entry, requests_queue);
115 spin_unlock(&conn->request_lock);
116 }
117 }
118
ksmbd_conn_try_dequeue_request(struct ksmbd_work * work)119 int ksmbd_conn_try_dequeue_request(struct ksmbd_work *work)
120 {
121 struct ksmbd_conn *conn = work->conn;
122 int ret = 1;
123
124 if (list_empty(&work->request_entry) &&
125 list_empty(&work->async_request_entry))
126 return 0;
127
128 if (!work->multiRsp)
129 atomic_dec(&conn->req_running);
130 spin_lock(&conn->request_lock);
131 if (!work->multiRsp) {
132 list_del_init(&work->request_entry);
133 if (work->syncronous == false)
134 list_del_init(&work->async_request_entry);
135 ret = 0;
136 }
137 spin_unlock(&conn->request_lock);
138
139 wake_up_all(&conn->req_running_q);
140 return ret;
141 }
142
ksmbd_conn_lock(struct ksmbd_conn * conn)143 static void ksmbd_conn_lock(struct ksmbd_conn *conn)
144 {
145 mutex_lock(&conn->srv_mutex);
146 }
147
ksmbd_conn_unlock(struct ksmbd_conn * conn)148 static void ksmbd_conn_unlock(struct ksmbd_conn *conn)
149 {
150 mutex_unlock(&conn->srv_mutex);
151 }
152
ksmbd_conn_wait_idle(struct ksmbd_conn * conn)153 void ksmbd_conn_wait_idle(struct ksmbd_conn *conn)
154 {
155 wait_event(conn->req_running_q, atomic_read(&conn->req_running) < 2);
156 }
157
ksmbd_conn_write(struct ksmbd_work * work)158 int ksmbd_conn_write(struct ksmbd_work *work)
159 {
160 struct ksmbd_conn *conn = work->conn;
161 size_t len = 0;
162 int sent;
163 struct kvec iov[3];
164 int iov_idx = 0;
165
166 ksmbd_conn_try_dequeue_request(work);
167 if (!work->response_buf) {
168 pr_err("NULL response header\n");
169 return -EINVAL;
170 }
171
172 if (work->tr_buf) {
173 iov[iov_idx] = (struct kvec) { work->tr_buf,
174 sizeof(struct smb2_transform_hdr) + 4 };
175 len += iov[iov_idx++].iov_len;
176 }
177
178 if (work->aux_payload_sz) {
179 iov[iov_idx] = (struct kvec) { work->response_buf, work->resp_hdr_sz };
180 len += iov[iov_idx++].iov_len;
181 iov[iov_idx] = (struct kvec) { work->aux_payload_buf, work->aux_payload_sz };
182 len += iov[iov_idx++].iov_len;
183 } else {
184 if (work->tr_buf)
185 iov[iov_idx].iov_len = work->resp_hdr_sz;
186 else
187 iov[iov_idx].iov_len = get_rfc1002_len(work->response_buf) + 4;
188 iov[iov_idx].iov_base = work->response_buf;
189 len += iov[iov_idx++].iov_len;
190 }
191
192 ksmbd_conn_lock(conn);
193 sent = conn->transport->ops->writev(conn->transport, &iov[0],
194 iov_idx, len,
195 work->need_invalidate_rkey,
196 work->remote_key);
197 ksmbd_conn_unlock(conn);
198
199 if (sent < 0) {
200 pr_err("Failed to send message: %d\n", sent);
201 return sent;
202 }
203
204 return 0;
205 }
206
ksmbd_conn_rdma_read(struct ksmbd_conn * conn,void * buf,unsigned int buflen,u32 remote_key,u64 remote_offset,u32 remote_len)207 int ksmbd_conn_rdma_read(struct ksmbd_conn *conn, void *buf,
208 unsigned int buflen, u32 remote_key, u64 remote_offset,
209 u32 remote_len)
210 {
211 int ret = -EINVAL;
212
213 if (conn->transport->ops->rdma_read)
214 ret = conn->transport->ops->rdma_read(conn->transport,
215 buf, buflen,
216 remote_key, remote_offset,
217 remote_len);
218 return ret;
219 }
220
ksmbd_conn_rdma_write(struct ksmbd_conn * conn,void * buf,unsigned int buflen,u32 remote_key,u64 remote_offset,u32 remote_len)221 int ksmbd_conn_rdma_write(struct ksmbd_conn *conn, void *buf,
222 unsigned int buflen, u32 remote_key,
223 u64 remote_offset, u32 remote_len)
224 {
225 int ret = -EINVAL;
226
227 if (conn->transport->ops->rdma_write)
228 ret = conn->transport->ops->rdma_write(conn->transport,
229 buf, buflen,
230 remote_key, remote_offset,
231 remote_len);
232 return ret;
233 }
234
ksmbd_conn_alive(struct ksmbd_conn * conn)235 bool ksmbd_conn_alive(struct ksmbd_conn *conn)
236 {
237 if (!ksmbd_server_running())
238 return false;
239
240 if (conn->status == KSMBD_SESS_EXITING)
241 return false;
242
243 if (kthread_should_stop())
244 return false;
245
246 if (atomic_read(&conn->stats.open_files_count) > 0)
247 return true;
248
249 /*
250 * Stop current session if the time that get last request from client
251 * is bigger than deadtime user configured and opening file count is
252 * zero.
253 */
254 if (server_conf.deadtime > 0 &&
255 time_after(jiffies, conn->last_active + server_conf.deadtime)) {
256 ksmbd_debug(CONN, "No response from client in %lu minutes\n",
257 server_conf.deadtime / SMB_ECHO_INTERVAL);
258 return false;
259 }
260 return true;
261 }
262
263 /**
264 * ksmbd_conn_handler_loop() - session thread to listen on new smb requests
265 * @p: connection instance
266 *
267 * One thread each per connection
268 *
269 * Return: 0 on success
270 */
ksmbd_conn_handler_loop(void * p)271 int ksmbd_conn_handler_loop(void *p)
272 {
273 struct ksmbd_conn *conn = (struct ksmbd_conn *)p;
274 struct ksmbd_transport *t = conn->transport;
275 unsigned int pdu_size;
276 char hdr_buf[4] = {0,};
277 int size;
278
279 mutex_init(&conn->srv_mutex);
280 __module_get(THIS_MODULE);
281
282 if (t->ops->prepare && t->ops->prepare(t))
283 goto out;
284
285 conn->last_active = jiffies;
286 while (ksmbd_conn_alive(conn)) {
287 if (try_to_freeze())
288 continue;
289
290 kvfree(conn->request_buf);
291 conn->request_buf = NULL;
292
293 size = t->ops->read(t, hdr_buf, sizeof(hdr_buf));
294 if (size != sizeof(hdr_buf))
295 break;
296
297 pdu_size = get_rfc1002_len(hdr_buf);
298 ksmbd_debug(CONN, "RFC1002 header %u bytes\n", pdu_size);
299
300 /*
301 * Check if pdu size is valid (min : smb header size,
302 * max : 0x00FFFFFF).
303 */
304 if (pdu_size < __SMB2_HEADER_STRUCTURE_SIZE ||
305 pdu_size > MAX_STREAM_PROT_LEN) {
306 continue;
307 }
308
309 /* 4 for rfc1002 length field */
310 size = pdu_size + 4;
311 conn->request_buf = kvmalloc(size, GFP_KERNEL);
312 if (!conn->request_buf)
313 continue;
314
315 memcpy(conn->request_buf, hdr_buf, sizeof(hdr_buf));
316 if (!ksmbd_smb_request(conn))
317 break;
318
319 /*
320 * We already read 4 bytes to find out PDU size, now
321 * read in PDU
322 */
323 size = t->ops->read(t, conn->request_buf + 4, pdu_size);
324 if (size < 0) {
325 pr_err("sock_read failed: %d\n", size);
326 break;
327 }
328
329 if (size != pdu_size) {
330 pr_err("PDU error. Read: %d, Expected: %d\n",
331 size, pdu_size);
332 continue;
333 }
334
335 if (!default_conn_ops.process_fn) {
336 pr_err("No connection request callback\n");
337 break;
338 }
339
340 if (default_conn_ops.process_fn(conn)) {
341 pr_err("Cannot handle request\n");
342 break;
343 }
344 }
345
346 out:
347 /* Wait till all reference dropped to the Server object*/
348 while (atomic_read(&conn->r_count) > 0)
349 schedule_timeout(HZ);
350
351 unload_nls(conn->local_nls);
352 if (default_conn_ops.terminate_fn)
353 default_conn_ops.terminate_fn(conn);
354 t->ops->disconnect(t);
355 module_put(THIS_MODULE);
356 return 0;
357 }
358
ksmbd_conn_init_server_callbacks(struct ksmbd_conn_ops * ops)359 void ksmbd_conn_init_server_callbacks(struct ksmbd_conn_ops *ops)
360 {
361 default_conn_ops.process_fn = ops->process_fn;
362 default_conn_ops.terminate_fn = ops->terminate_fn;
363 }
364
ksmbd_conn_transport_init(void)365 int ksmbd_conn_transport_init(void)
366 {
367 int ret;
368
369 mutex_lock(&init_lock);
370 ret = ksmbd_tcp_init();
371 if (ret) {
372 pr_err("Failed to init TCP subsystem: %d\n", ret);
373 goto out;
374 }
375
376 ret = ksmbd_rdma_init();
377 if (ret) {
378 pr_err("Failed to init RDMA subsystem: %d\n", ret);
379 goto out;
380 }
381 out:
382 mutex_unlock(&init_lock);
383 return ret;
384 }
385
stop_sessions(void)386 static void stop_sessions(void)
387 {
388 struct ksmbd_conn *conn;
389
390 again:
391 read_lock(&conn_list_lock);
392 list_for_each_entry(conn, &conn_list, conns_list) {
393 struct task_struct *task;
394
395 task = conn->transport->handler;
396 if (task)
397 ksmbd_debug(CONN, "Stop session handler %s/%d\n",
398 task->comm, task_pid_nr(task));
399 conn->status = KSMBD_SESS_EXITING;
400 }
401 read_unlock(&conn_list_lock);
402
403 if (!list_empty(&conn_list)) {
404 schedule_timeout_interruptible(HZ / 10); /* 100ms */
405 goto again;
406 }
407 }
408
ksmbd_conn_transport_destroy(void)409 void ksmbd_conn_transport_destroy(void)
410 {
411 mutex_lock(&init_lock);
412 ksmbd_tcp_destroy();
413 ksmbd_rdma_destroy();
414 stop_sessions();
415 mutex_unlock(&init_lock);
416 }
417