xref: /linux/fs/smb/common/smbdirect/smbdirect_socket.h (revision a9401710a5f5681abd2a6f21f9e76bc9f2e81891)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  *   Copyright (c) 2025 Stefan Metzmacher
4  */
5 
6 #ifndef __FS_SMB_COMMON_SMBDIRECT_SMBDIRECT_SOCKET_H__
7 #define __FS_SMB_COMMON_SMBDIRECT_SMBDIRECT_SOCKET_H__
8 
9 #include <rdma/rw.h>
10 
11 enum smbdirect_socket_status {
12 	SMBDIRECT_SOCKET_CREATED,
13 	SMBDIRECT_SOCKET_RESOLVE_ADDR_NEEDED,
14 	SMBDIRECT_SOCKET_RESOLVE_ADDR_RUNNING,
15 	SMBDIRECT_SOCKET_RESOLVE_ADDR_FAILED,
16 	SMBDIRECT_SOCKET_RESOLVE_ROUTE_NEEDED,
17 	SMBDIRECT_SOCKET_RESOLVE_ROUTE_RUNNING,
18 	SMBDIRECT_SOCKET_RESOLVE_ROUTE_FAILED,
19 	SMBDIRECT_SOCKET_RDMA_CONNECT_NEEDED,
20 	SMBDIRECT_SOCKET_RDMA_CONNECT_RUNNING,
21 	SMBDIRECT_SOCKET_RDMA_CONNECT_FAILED,
22 	SMBDIRECT_SOCKET_NEGOTIATE_NEEDED,
23 	SMBDIRECT_SOCKET_NEGOTIATE_RUNNING,
24 	SMBDIRECT_SOCKET_NEGOTIATE_FAILED,
25 	SMBDIRECT_SOCKET_CONNECTED,
26 	SMBDIRECT_SOCKET_ERROR,
27 	SMBDIRECT_SOCKET_DISCONNECTING,
28 	SMBDIRECT_SOCKET_DISCONNECTED,
29 	SMBDIRECT_SOCKET_DESTROYED
30 };
31 
32 static __always_inline
smbdirect_socket_status_string(enum smbdirect_socket_status status)33 const char *smbdirect_socket_status_string(enum smbdirect_socket_status status)
34 {
35 	switch (status) {
36 	case SMBDIRECT_SOCKET_CREATED:
37 		return "CREATED";
38 	case SMBDIRECT_SOCKET_RESOLVE_ADDR_NEEDED:
39 		return "RESOLVE_ADDR_NEEDED";
40 	case SMBDIRECT_SOCKET_RESOLVE_ADDR_RUNNING:
41 		return "RESOLVE_ADDR_RUNNING";
42 	case SMBDIRECT_SOCKET_RESOLVE_ADDR_FAILED:
43 		return "RESOLVE_ADDR_FAILED";
44 	case SMBDIRECT_SOCKET_RESOLVE_ROUTE_NEEDED:
45 		return "RESOLVE_ROUTE_NEEDED";
46 	case SMBDIRECT_SOCKET_RESOLVE_ROUTE_RUNNING:
47 		return "RESOLVE_ROUTE_RUNNING";
48 	case SMBDIRECT_SOCKET_RESOLVE_ROUTE_FAILED:
49 		return "RESOLVE_ROUTE_FAILED";
50 	case SMBDIRECT_SOCKET_RDMA_CONNECT_NEEDED:
51 		return "RDMA_CONNECT_NEEDED";
52 	case SMBDIRECT_SOCKET_RDMA_CONNECT_RUNNING:
53 		return "RDMA_CONNECT_RUNNING";
54 	case SMBDIRECT_SOCKET_RDMA_CONNECT_FAILED:
55 		return "RDMA_CONNECT_FAILED";
56 	case SMBDIRECT_SOCKET_NEGOTIATE_NEEDED:
57 		return "NEGOTIATE_NEEDED";
58 	case SMBDIRECT_SOCKET_NEGOTIATE_RUNNING:
59 		return "NEGOTIATE_RUNNING";
60 	case SMBDIRECT_SOCKET_NEGOTIATE_FAILED:
61 		return "NEGOTIATE_FAILED";
62 	case SMBDIRECT_SOCKET_CONNECTED:
63 		return "CONNECTED";
64 	case SMBDIRECT_SOCKET_ERROR:
65 		return "ERROR";
66 	case SMBDIRECT_SOCKET_DISCONNECTING:
67 		return "DISCONNECTING";
68 	case SMBDIRECT_SOCKET_DISCONNECTED:
69 		return "DISCONNECTED";
70 	case SMBDIRECT_SOCKET_DESTROYED:
71 		return "DESTROYED";
72 	}
73 
74 	return "<unknown>";
75 }
76 
77 enum smbdirect_keepalive_status {
78 	SMBDIRECT_KEEPALIVE_NONE,
79 	SMBDIRECT_KEEPALIVE_PENDING,
80 	SMBDIRECT_KEEPALIVE_SENT
81 };
82 
83 struct smbdirect_socket {
84 	enum smbdirect_socket_status status;
85 	wait_queue_head_t status_wait;
86 	int first_error;
87 
88 	/*
89 	 * This points to the workqueue to
90 	 * be used for this socket.
91 	 * It can be per socket (on the client)
92 	 * or point to a global workqueue (on the server)
93 	 */
94 	struct workqueue_struct *workqueue;
95 
96 	struct work_struct disconnect_work;
97 
98 	/* RDMA related */
99 	struct {
100 		struct rdma_cm_id *cm_id;
101 		/*
102 		 * This is for iWarp MPA v1
103 		 */
104 		bool legacy_iwarp;
105 	} rdma;
106 
107 	/* IB verbs related */
108 	struct {
109 		struct ib_pd *pd;
110 		struct ib_cq *send_cq;
111 		struct ib_cq *recv_cq;
112 
113 		/*
114 		 * shortcuts for rdma.cm_id->{qp,device};
115 		 */
116 		struct ib_qp *qp;
117 		struct ib_device *dev;
118 	} ib;
119 
120 	struct smbdirect_socket_parameters parameters;
121 
122 	/*
123 	 * The state for keepalive and timeout handling
124 	 */
125 	struct {
126 		enum smbdirect_keepalive_status keepalive;
127 		struct work_struct immediate_work;
128 		struct delayed_work timer_work;
129 	} idle;
130 
131 	/*
132 	 * The state for posted send buffers
133 	 */
134 	struct {
135 		/*
136 		 * Memory pools for preallocating
137 		 * smbdirect_send_io buffers
138 		 */
139 		struct {
140 			struct kmem_cache	*cache;
141 			mempool_t		*pool;
142 		} mem;
143 
144 		/*
145 		 * The credit state for the send side
146 		 */
147 		struct {
148 			atomic_t count;
149 			wait_queue_head_t wait_queue;
150 		} credits;
151 
152 		/*
153 		 * The state about posted/pending sends
154 		 */
155 		struct {
156 			atomic_t count;
157 			/*
158 			 * woken when count is decremented
159 			 */
160 			wait_queue_head_t dec_wait_queue;
161 			/*
162 			 * woken when count reached zero
163 			 */
164 			wait_queue_head_t zero_wait_queue;
165 		} pending;
166 	} send_io;
167 
168 	/*
169 	 * The state for posted receive buffers
170 	 */
171 	struct {
172 		/*
173 		 * The type of PDU we are expecting
174 		 */
175 		enum {
176 			SMBDIRECT_EXPECT_NEGOTIATE_REQ = 1,
177 			SMBDIRECT_EXPECT_NEGOTIATE_REP = 2,
178 			SMBDIRECT_EXPECT_DATA_TRANSFER = 3,
179 		} expected;
180 
181 		/*
182 		 * Memory pools for preallocating
183 		 * smbdirect_recv_io buffers
184 		 */
185 		struct {
186 			struct kmem_cache	*cache;
187 			mempool_t		*pool;
188 		} mem;
189 
190 		/*
191 		 * The list of free smbdirect_recv_io
192 		 * structures
193 		 */
194 		struct {
195 			struct list_head list;
196 			spinlock_t lock;
197 		} free;
198 
199 		/*
200 		 * The state for posted recv_io messages
201 		 * and the refill work struct.
202 		 */
203 		struct {
204 			atomic_t count;
205 			struct work_struct refill_work;
206 		} posted;
207 
208 		/*
209 		 * The credit state for the recv side
210 		 */
211 		struct {
212 			u16 target;
213 			atomic_t count;
214 		} credits;
215 
216 		/*
217 		 * The list of arrived non-empty smbdirect_recv_io
218 		 * structures
219 		 *
220 		 * This represents the reassembly queue.
221 		 */
222 		struct {
223 			struct list_head list;
224 			spinlock_t lock;
225 			wait_queue_head_t wait_queue;
226 			/* total data length of reassembly queue */
227 			int data_length;
228 			int queue_length;
229 			/* the offset to first buffer in reassembly queue */
230 			int first_entry_offset;
231 			/*
232 			 * Indicate if we have received a full packet on the
233 			 * connection This is used to identify the first SMBD
234 			 * packet of a assembled payload (SMB packet) in
235 			 * reassembly queue so we can return a RFC1002 length to
236 			 * upper layer to indicate the length of the SMB packet
237 			 * received
238 			 */
239 			bool full_packet_received;
240 		} reassembly;
241 	} recv_io;
242 
243 	/*
244 	 * The state for Memory registrations on the client
245 	 */
246 	struct {
247 		enum ib_mr_type type;
248 
249 		/*
250 		 * The list of free smbdirect_mr_io
251 		 * structures
252 		 */
253 		struct {
254 			struct list_head list;
255 			spinlock_t lock;
256 		} all;
257 
258 		/*
259 		 * The number of available MRs ready for memory registration
260 		 */
261 		struct {
262 			atomic_t count;
263 			wait_queue_head_t wait_queue;
264 		} ready;
265 
266 		/*
267 		 * The number of used MRs
268 		 */
269 		struct {
270 			atomic_t count;
271 		} used;
272 
273 		struct work_struct recovery_work;
274 
275 		/* Used by transport to wait until all MRs are returned */
276 		struct {
277 			wait_queue_head_t wait_queue;
278 		} cleanup;
279 	} mr_io;
280 
281 	/*
282 	 * The state for RDMA read/write requests on the server
283 	 */
284 	struct {
285 		/*
286 		 * The credit state for the send side
287 		 */
288 		struct {
289 			/*
290 			 * The maximum number of rw credits
291 			 */
292 			size_t max;
293 			/*
294 			 * The number of pages per credit
295 			 */
296 			size_t num_pages;
297 			atomic_t count;
298 			wait_queue_head_t wait_queue;
299 		} credits;
300 	} rw_io;
301 
302 	/*
303 	 * For debug purposes
304 	 */
305 	struct {
306 		u64 get_receive_buffer;
307 		u64 put_receive_buffer;
308 		u64 enqueue_reassembly_queue;
309 		u64 dequeue_reassembly_queue;
310 		u64 send_empty;
311 	} statistics;
312 };
313 
__smbdirect_socket_disabled_work(struct work_struct * work)314 static void __smbdirect_socket_disabled_work(struct work_struct *work)
315 {
316 	/*
317 	 * Should never be called as disable_[delayed_]work_sync() was used.
318 	 */
319 	WARN_ON_ONCE(1);
320 }
321 
smbdirect_socket_init(struct smbdirect_socket * sc)322 static __always_inline void smbdirect_socket_init(struct smbdirect_socket *sc)
323 {
324 	/*
325 	 * This also sets status = SMBDIRECT_SOCKET_CREATED
326 	 */
327 	BUILD_BUG_ON(SMBDIRECT_SOCKET_CREATED != 0);
328 	memset(sc, 0, sizeof(*sc));
329 
330 	init_waitqueue_head(&sc->status_wait);
331 
332 	INIT_WORK(&sc->disconnect_work, __smbdirect_socket_disabled_work);
333 	disable_work_sync(&sc->disconnect_work);
334 
335 	INIT_WORK(&sc->idle.immediate_work, __smbdirect_socket_disabled_work);
336 	disable_work_sync(&sc->idle.immediate_work);
337 	INIT_DELAYED_WORK(&sc->idle.timer_work, __smbdirect_socket_disabled_work);
338 	disable_delayed_work_sync(&sc->idle.timer_work);
339 
340 	atomic_set(&sc->send_io.credits.count, 0);
341 	init_waitqueue_head(&sc->send_io.credits.wait_queue);
342 
343 	atomic_set(&sc->send_io.pending.count, 0);
344 	init_waitqueue_head(&sc->send_io.pending.dec_wait_queue);
345 	init_waitqueue_head(&sc->send_io.pending.zero_wait_queue);
346 
347 	INIT_LIST_HEAD(&sc->recv_io.free.list);
348 	spin_lock_init(&sc->recv_io.free.lock);
349 
350 	atomic_set(&sc->recv_io.posted.count, 0);
351 	INIT_WORK(&sc->recv_io.posted.refill_work, __smbdirect_socket_disabled_work);
352 	disable_work_sync(&sc->recv_io.posted.refill_work);
353 
354 	atomic_set(&sc->recv_io.credits.count, 0);
355 
356 	INIT_LIST_HEAD(&sc->recv_io.reassembly.list);
357 	spin_lock_init(&sc->recv_io.reassembly.lock);
358 	init_waitqueue_head(&sc->recv_io.reassembly.wait_queue);
359 
360 	atomic_set(&sc->rw_io.credits.count, 0);
361 	init_waitqueue_head(&sc->rw_io.credits.wait_queue);
362 
363 	spin_lock_init(&sc->mr_io.all.lock);
364 	INIT_LIST_HEAD(&sc->mr_io.all.list);
365 	atomic_set(&sc->mr_io.ready.count, 0);
366 	init_waitqueue_head(&sc->mr_io.ready.wait_queue);
367 	atomic_set(&sc->mr_io.used.count, 0);
368 	INIT_WORK(&sc->mr_io.recovery_work, __smbdirect_socket_disabled_work);
369 	disable_work_sync(&sc->mr_io.recovery_work);
370 	init_waitqueue_head(&sc->mr_io.cleanup.wait_queue);
371 }
372 
373 struct smbdirect_send_io {
374 	struct smbdirect_socket *socket;
375 	struct ib_cqe cqe;
376 
377 	/*
378 	 * The SGE entries for this work request
379 	 *
380 	 * The first points to the packet header
381 	 */
382 #define SMBDIRECT_SEND_IO_MAX_SGE 6
383 	size_t num_sge;
384 	struct ib_sge sge[SMBDIRECT_SEND_IO_MAX_SGE];
385 
386 	/*
387 	 * Link to the list of sibling smbdirect_send_io
388 	 * messages.
389 	 */
390 	struct list_head sibling_list;
391 	struct ib_send_wr wr;
392 
393 	/* SMBD packet header follows this structure */
394 	u8 packet[];
395 };
396 
397 struct smbdirect_send_batch {
398 	/*
399 	 * List of smbdirect_send_io messages
400 	 */
401 	struct list_head msg_list;
402 	/*
403 	 * Number of list entries
404 	 */
405 	size_t wr_cnt;
406 
407 	/*
408 	 * Possible remote key invalidation state
409 	 */
410 	bool need_invalidate_rkey;
411 	u32 remote_key;
412 };
413 
414 struct smbdirect_recv_io {
415 	struct smbdirect_socket *socket;
416 	struct ib_cqe cqe;
417 
418 	/*
419 	 * For now we only use a single SGE
420 	 * as we have just one large buffer
421 	 * per posted recv.
422 	 */
423 #define SMBDIRECT_RECV_IO_MAX_SGE 1
424 	struct ib_sge sge;
425 
426 	/* Link to free or reassembly list */
427 	struct list_head list;
428 
429 	/* Indicate if this is the 1st packet of a payload */
430 	bool first_segment;
431 
432 	/* SMBD packet header and payload follows this structure */
433 	u8 packet[];
434 };
435 
436 enum smbdirect_mr_state {
437 	SMBDIRECT_MR_READY,
438 	SMBDIRECT_MR_REGISTERED,
439 	SMBDIRECT_MR_INVALIDATED,
440 	SMBDIRECT_MR_ERROR
441 };
442 
443 struct smbdirect_mr_io {
444 	struct smbdirect_socket *socket;
445 	struct ib_cqe cqe;
446 
447 	struct list_head list;
448 
449 	enum smbdirect_mr_state state;
450 	struct ib_mr *mr;
451 	struct sg_table sgt;
452 	enum dma_data_direction dir;
453 	union {
454 		struct ib_reg_wr wr;
455 		struct ib_send_wr inv_wr;
456 	};
457 
458 	bool need_invalidate;
459 	struct completion invalidate_done;
460 };
461 
462 struct smbdirect_rw_io {
463 	struct smbdirect_socket *socket;
464 	struct ib_cqe cqe;
465 
466 	struct list_head list;
467 
468 	int error;
469 	struct completion *completion;
470 
471 	struct rdma_rw_ctx rdma_ctx;
472 	struct sg_table sgt;
473 	struct scatterlist sg_list[];
474 };
475 
476 #endif /* __FS_SMB_COMMON_SMBDIRECT_SMBDIRECT_SOCKET_H__ */
477