xref: /openbmc/linux/net/ceph/messenger_v2.c (revision 2ea88716)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Ceph msgr2 protocol implementation
4  *
5  * Copyright (C) 2020 Ilya Dryomov <idryomov@gmail.com>
6  */
7 
8 #include <linux/ceph/ceph_debug.h>
9 
10 #include <crypto/aead.h>
11 #include <crypto/algapi.h>  /* for crypto_memneq() */
12 #include <crypto/hash.h>
13 #include <crypto/sha2.h>
14 #include <linux/bvec.h>
15 #include <linux/crc32c.h>
16 #include <linux/net.h>
17 #include <linux/scatterlist.h>
18 #include <linux/socket.h>
19 #include <linux/sched/mm.h>
20 #include <net/sock.h>
21 #include <net/tcp.h>
22 
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/libceph.h>
26 #include <linux/ceph/messenger.h>
27 
28 #include "crypto.h"  /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */
29 
30 #define FRAME_TAG_HELLO			1
31 #define FRAME_TAG_AUTH_REQUEST		2
32 #define FRAME_TAG_AUTH_BAD_METHOD	3
33 #define FRAME_TAG_AUTH_REPLY_MORE	4
34 #define FRAME_TAG_AUTH_REQUEST_MORE	5
35 #define FRAME_TAG_AUTH_DONE		6
36 #define FRAME_TAG_AUTH_SIGNATURE	7
37 #define FRAME_TAG_CLIENT_IDENT		8
38 #define FRAME_TAG_SERVER_IDENT		9
39 #define FRAME_TAG_IDENT_MISSING_FEATURES 10
40 #define FRAME_TAG_SESSION_RECONNECT	11
41 #define FRAME_TAG_SESSION_RESET		12
42 #define FRAME_TAG_SESSION_RETRY		13
43 #define FRAME_TAG_SESSION_RETRY_GLOBAL	14
44 #define FRAME_TAG_SESSION_RECONNECT_OK	15
45 #define FRAME_TAG_WAIT			16
46 #define FRAME_TAG_MESSAGE		17
47 #define FRAME_TAG_KEEPALIVE2		18
48 #define FRAME_TAG_KEEPALIVE2_ACK	19
49 #define FRAME_TAG_ACK			20
50 
51 #define FRAME_LATE_STATUS_ABORTED	0x1
52 #define FRAME_LATE_STATUS_COMPLETE	0xe
53 #define FRAME_LATE_STATUS_ABORTED_MASK	0xf
54 
55 #define IN_S_HANDLE_PREAMBLE		1
56 #define IN_S_HANDLE_CONTROL		2
57 #define IN_S_HANDLE_CONTROL_REMAINDER	3
58 #define IN_S_PREPARE_READ_DATA		4
59 #define IN_S_PREPARE_READ_DATA_CONT	5
60 #define IN_S_PREPARE_READ_ENC_PAGE	6
61 #define IN_S_HANDLE_EPILOGUE		7
62 #define IN_S_FINISH_SKIP		8
63 
64 #define OUT_S_QUEUE_DATA		1
65 #define OUT_S_QUEUE_DATA_CONT		2
66 #define OUT_S_QUEUE_ENC_PAGE		3
67 #define OUT_S_QUEUE_ZEROS		4
68 #define OUT_S_FINISH_MESSAGE		5
69 #define OUT_S_GET_NEXT			6
70 
71 #define CTRL_BODY(p)	((void *)(p) + CEPH_PREAMBLE_LEN)
72 #define FRONT_PAD(p)	((void *)(p) + CEPH_EPILOGUE_SECURE_LEN)
73 #define MIDDLE_PAD(p)	(FRONT_PAD(p) + CEPH_GCM_BLOCK_LEN)
74 #define DATA_PAD(p)	(MIDDLE_PAD(p) + CEPH_GCM_BLOCK_LEN)
75 
76 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
77 
78 static int do_recvmsg(struct socket *sock, struct iov_iter *it)
79 {
80 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
81 	int ret;
82 
83 	msg.msg_iter = *it;
84 	while (iov_iter_count(it)) {
85 		ret = sock_recvmsg(sock, &msg, msg.msg_flags);
86 		if (ret <= 0) {
87 			if (ret == -EAGAIN)
88 				ret = 0;
89 			return ret;
90 		}
91 
92 		iov_iter_advance(it, ret);
93 	}
94 
95 	WARN_ON(msg_data_left(&msg));
96 	return 1;
97 }
98 
99 /*
100  * Read as much as possible.
101  *
102  * Return:
103  *   1 - done, nothing (else) to read
104  *   0 - socket is empty, need to wait
105  *  <0 - error
106  */
107 static int ceph_tcp_recv(struct ceph_connection *con)
108 {
109 	int ret;
110 
111 	dout("%s con %p %s %zu\n", __func__, con,
112 	     iov_iter_is_discard(&con->v2.in_iter) ? "discard" : "need",
113 	     iov_iter_count(&con->v2.in_iter));
114 	ret = do_recvmsg(con->sock, &con->v2.in_iter);
115 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
116 	     iov_iter_count(&con->v2.in_iter));
117 	return ret;
118 }
119 
120 static int do_sendmsg(struct socket *sock, struct iov_iter *it)
121 {
122 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
123 	int ret;
124 
125 	msg.msg_iter = *it;
126 	while (iov_iter_count(it)) {
127 		ret = sock_sendmsg(sock, &msg);
128 		if (ret <= 0) {
129 			if (ret == -EAGAIN)
130 				ret = 0;
131 			return ret;
132 		}
133 
134 		iov_iter_advance(it, ret);
135 	}
136 
137 	WARN_ON(msg_data_left(&msg));
138 	return 1;
139 }
140 
141 static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
142 {
143 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
144 	struct bio_vec bv;
145 	int ret;
146 
147 	if (WARN_ON(!iov_iter_is_bvec(it)))
148 		return -EINVAL;
149 
150 	while (iov_iter_count(it)) {
151 		/* iov_iter_iovec() for ITER_BVEC */
152 		bv.bv_page = it->bvec->bv_page;
153 		bv.bv_offset = it->bvec->bv_offset + it->iov_offset;
154 		bv.bv_len = min(iov_iter_count(it),
155 				it->bvec->bv_len - it->iov_offset);
156 
157 		/*
158 		 * sendpage cannot properly handle pages with
159 		 * page_count == 0, we need to fall back to sendmsg if
160 		 * that's the case.
161 		 *
162 		 * Same goes for slab pages: skb_can_coalesce() allows
163 		 * coalescing neighboring slab objects into a single frag
164 		 * which triggers one of hardened usercopy checks.
165 		 */
166 		if (sendpage_ok(bv.bv_page)) {
167 			ret = sock->ops->sendpage(sock, bv.bv_page,
168 						  bv.bv_offset, bv.bv_len,
169 						  CEPH_MSG_FLAGS);
170 		} else {
171 			iov_iter_bvec(&msg.msg_iter, WRITE, &bv, 1, bv.bv_len);
172 			ret = sock_sendmsg(sock, &msg);
173 		}
174 		if (ret <= 0) {
175 			if (ret == -EAGAIN)
176 				ret = 0;
177 			return ret;
178 		}
179 
180 		iov_iter_advance(it, ret);
181 	}
182 
183 	return 1;
184 }
185 
186 /*
187  * Write as much as possible.  The socket is expected to be corked,
188  * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here.
189  *
190  * Return:
191  *   1 - done, nothing (else) to write
192  *   0 - socket is full, need to wait
193  *  <0 - error
194  */
195 static int ceph_tcp_send(struct ceph_connection *con)
196 {
197 	int ret;
198 
199 	dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
200 	     iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
201 	if (con->v2.out_iter_sendpage)
202 		ret = do_try_sendpage(con->sock, &con->v2.out_iter);
203 	else
204 		ret = do_sendmsg(con->sock, &con->v2.out_iter);
205 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
206 	     iov_iter_count(&con->v2.out_iter));
207 	return ret;
208 }
209 
210 static void add_in_kvec(struct ceph_connection *con, void *buf, int len)
211 {
212 	BUG_ON(con->v2.in_kvec_cnt >= ARRAY_SIZE(con->v2.in_kvecs));
213 	WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
214 
215 	con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_base = buf;
216 	con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_len = len;
217 	con->v2.in_kvec_cnt++;
218 
219 	con->v2.in_iter.nr_segs++;
220 	con->v2.in_iter.count += len;
221 }
222 
223 static void reset_in_kvecs(struct ceph_connection *con)
224 {
225 	WARN_ON(iov_iter_count(&con->v2.in_iter));
226 
227 	con->v2.in_kvec_cnt = 0;
228 	iov_iter_kvec(&con->v2.in_iter, READ, con->v2.in_kvecs, 0, 0);
229 }
230 
231 static void set_in_bvec(struct ceph_connection *con, const struct bio_vec *bv)
232 {
233 	WARN_ON(iov_iter_count(&con->v2.in_iter));
234 
235 	con->v2.in_bvec = *bv;
236 	iov_iter_bvec(&con->v2.in_iter, READ, &con->v2.in_bvec, 1, bv->bv_len);
237 }
238 
239 static void set_in_skip(struct ceph_connection *con, int len)
240 {
241 	WARN_ON(iov_iter_count(&con->v2.in_iter));
242 
243 	dout("%s con %p len %d\n", __func__, con, len);
244 	iov_iter_discard(&con->v2.in_iter, READ, len);
245 }
246 
247 static void add_out_kvec(struct ceph_connection *con, void *buf, int len)
248 {
249 	BUG_ON(con->v2.out_kvec_cnt >= ARRAY_SIZE(con->v2.out_kvecs));
250 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
251 	WARN_ON(con->v2.out_zero);
252 
253 	con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_base = buf;
254 	con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_len = len;
255 	con->v2.out_kvec_cnt++;
256 
257 	con->v2.out_iter.nr_segs++;
258 	con->v2.out_iter.count += len;
259 }
260 
261 static void reset_out_kvecs(struct ceph_connection *con)
262 {
263 	WARN_ON(iov_iter_count(&con->v2.out_iter));
264 	WARN_ON(con->v2.out_zero);
265 
266 	con->v2.out_kvec_cnt = 0;
267 
268 	iov_iter_kvec(&con->v2.out_iter, WRITE, con->v2.out_kvecs, 0, 0);
269 	con->v2.out_iter_sendpage = false;
270 }
271 
272 static void set_out_bvec(struct ceph_connection *con, const struct bio_vec *bv,
273 			 bool zerocopy)
274 {
275 	WARN_ON(iov_iter_count(&con->v2.out_iter));
276 	WARN_ON(con->v2.out_zero);
277 
278 	con->v2.out_bvec = *bv;
279 	con->v2.out_iter_sendpage = zerocopy;
280 	iov_iter_bvec(&con->v2.out_iter, WRITE, &con->v2.out_bvec, 1,
281 		      con->v2.out_bvec.bv_len);
282 }
283 
284 static void set_out_bvec_zero(struct ceph_connection *con)
285 {
286 	WARN_ON(iov_iter_count(&con->v2.out_iter));
287 	WARN_ON(!con->v2.out_zero);
288 
289 	con->v2.out_bvec.bv_page = ceph_zero_page;
290 	con->v2.out_bvec.bv_offset = 0;
291 	con->v2.out_bvec.bv_len = min(con->v2.out_zero, (int)PAGE_SIZE);
292 	con->v2.out_iter_sendpage = true;
293 	iov_iter_bvec(&con->v2.out_iter, WRITE, &con->v2.out_bvec, 1,
294 		      con->v2.out_bvec.bv_len);
295 }
296 
297 static void out_zero_add(struct ceph_connection *con, int len)
298 {
299 	dout("%s con %p len %d\n", __func__, con, len);
300 	con->v2.out_zero += len;
301 }
302 
303 static void *alloc_conn_buf(struct ceph_connection *con, int len)
304 {
305 	void *buf;
306 
307 	dout("%s con %p len %d\n", __func__, con, len);
308 
309 	if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
310 		return NULL;
311 
312 	buf = kvmalloc(len, GFP_NOIO);
313 	if (!buf)
314 		return NULL;
315 
316 	con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf;
317 	return buf;
318 }
319 
320 static void free_conn_bufs(struct ceph_connection *con)
321 {
322 	while (con->v2.conn_buf_cnt)
323 		kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]);
324 }
325 
326 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len)
327 {
328 	BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs));
329 
330 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf;
331 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len;
332 	con->v2.in_sign_kvec_cnt++;
333 }
334 
335 static void clear_in_sign_kvecs(struct ceph_connection *con)
336 {
337 	con->v2.in_sign_kvec_cnt = 0;
338 }
339 
340 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len)
341 {
342 	BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs));
343 
344 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf;
345 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len;
346 	con->v2.out_sign_kvec_cnt++;
347 }
348 
349 static void clear_out_sign_kvecs(struct ceph_connection *con)
350 {
351 	con->v2.out_sign_kvec_cnt = 0;
352 }
353 
354 static bool con_secure(struct ceph_connection *con)
355 {
356 	return con->v2.con_mode == CEPH_CON_MODE_SECURE;
357 }
358 
359 static int front_len(const struct ceph_msg *msg)
360 {
361 	return le32_to_cpu(msg->hdr.front_len);
362 }
363 
364 static int middle_len(const struct ceph_msg *msg)
365 {
366 	return le32_to_cpu(msg->hdr.middle_len);
367 }
368 
369 static int data_len(const struct ceph_msg *msg)
370 {
371 	return le32_to_cpu(msg->hdr.data_len);
372 }
373 
374 static bool need_padding(int len)
375 {
376 	return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN);
377 }
378 
379 static int padded_len(int len)
380 {
381 	return ALIGN(len, CEPH_GCM_BLOCK_LEN);
382 }
383 
384 static int padding_len(int len)
385 {
386 	return padded_len(len) - len;
387 }
388 
389 /* preamble + control segment */
390 static int head_onwire_len(int ctrl_len, bool secure)
391 {
392 	int head_len;
393 	int rem_len;
394 
395 	if (secure) {
396 		head_len = CEPH_PREAMBLE_SECURE_LEN;
397 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
398 			rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
399 			head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN;
400 		}
401 	} else {
402 		head_len = CEPH_PREAMBLE_PLAIN_LEN;
403 		if (ctrl_len)
404 			head_len += ctrl_len + CEPH_CRC_LEN;
405 	}
406 	return head_len;
407 }
408 
409 /* front, middle and data segments + epilogue */
410 static int __tail_onwire_len(int front_len, int middle_len, int data_len,
411 			     bool secure)
412 {
413 	if (!front_len && !middle_len && !data_len)
414 		return 0;
415 
416 	if (!secure)
417 		return front_len + middle_len + data_len +
418 		       CEPH_EPILOGUE_PLAIN_LEN;
419 
420 	return padded_len(front_len) + padded_len(middle_len) +
421 	       padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN;
422 }
423 
424 static int tail_onwire_len(const struct ceph_msg *msg, bool secure)
425 {
426 	return __tail_onwire_len(front_len(msg), middle_len(msg),
427 				 data_len(msg), secure);
428 }
429 
430 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */
431 #define MESSAGE_HEAD_PLAIN_LEN	(CEPH_PREAMBLE_PLAIN_LEN +		\
432 				 sizeof(struct ceph_msg_header2) +	\
433 				 CEPH_CRC_LEN)
434 
435 static const int frame_aligns[] = {
436 	sizeof(void *),
437 	sizeof(void *),
438 	sizeof(void *),
439 	PAGE_SIZE
440 };
441 
442 /*
443  * Discards trailing empty segments, unless there is just one segment.
444  * A frame always has at least one (possibly empty) segment.
445  */
446 static int calc_segment_count(const int *lens, int len_cnt)
447 {
448 	int i;
449 
450 	for (i = len_cnt - 1; i >= 0; i--) {
451 		if (lens[i])
452 			return i + 1;
453 	}
454 
455 	return 1;
456 }
457 
458 static void init_frame_desc(struct ceph_frame_desc *desc, int tag,
459 			    const int *lens, int len_cnt)
460 {
461 	int i;
462 
463 	memset(desc, 0, sizeof(*desc));
464 
465 	desc->fd_tag = tag;
466 	desc->fd_seg_cnt = calc_segment_count(lens, len_cnt);
467 	BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT);
468 	for (i = 0; i < desc->fd_seg_cnt; i++) {
469 		desc->fd_lens[i] = lens[i];
470 		desc->fd_aligns[i] = frame_aligns[i];
471 	}
472 }
473 
474 /*
475  * Preamble crc covers everything up to itself (28 bytes) and
476  * is calculated and verified irrespective of the connection mode
477  * (i.e. even if the frame is encrypted).
478  */
479 static void encode_preamble(const struct ceph_frame_desc *desc, void *p)
480 {
481 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
482 	void *start = p;
483 	int i;
484 
485 	memset(p, 0, CEPH_PREAMBLE_LEN);
486 
487 	ceph_encode_8(&p, desc->fd_tag);
488 	ceph_encode_8(&p, desc->fd_seg_cnt);
489 	for (i = 0; i < desc->fd_seg_cnt; i++) {
490 		ceph_encode_32(&p, desc->fd_lens[i]);
491 		ceph_encode_16(&p, desc->fd_aligns[i]);
492 	}
493 
494 	put_unaligned_le32(crc32c(0, start, crcp - start), crcp);
495 }
496 
497 static int decode_preamble(void *p, struct ceph_frame_desc *desc)
498 {
499 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
500 	u32 crc, expected_crc;
501 	int i;
502 
503 	crc = crc32c(0, p, crcp - p);
504 	expected_crc = get_unaligned_le32(crcp);
505 	if (crc != expected_crc) {
506 		pr_err("bad preamble crc, calculated %u, expected %u\n",
507 		       crc, expected_crc);
508 		return -EBADMSG;
509 	}
510 
511 	memset(desc, 0, sizeof(*desc));
512 
513 	desc->fd_tag = ceph_decode_8(&p);
514 	desc->fd_seg_cnt = ceph_decode_8(&p);
515 	if (desc->fd_seg_cnt < 1 ||
516 	    desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) {
517 		pr_err("bad segment count %d\n", desc->fd_seg_cnt);
518 		return -EINVAL;
519 	}
520 	for (i = 0; i < desc->fd_seg_cnt; i++) {
521 		desc->fd_lens[i] = ceph_decode_32(&p);
522 		desc->fd_aligns[i] = ceph_decode_16(&p);
523 	}
524 
525 	/*
526 	 * This would fire for FRAME_TAG_WAIT (it has one empty
527 	 * segment), but we should never get it as client.
528 	 */
529 	if (!desc->fd_lens[desc->fd_seg_cnt - 1]) {
530 		pr_err("last segment empty\n");
531 		return -EINVAL;
532 	}
533 
534 	if (desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) {
535 		pr_err("control segment too big %d\n", desc->fd_lens[0]);
536 		return -EINVAL;
537 	}
538 	if (desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) {
539 		pr_err("front segment too big %d\n", desc->fd_lens[1]);
540 		return -EINVAL;
541 	}
542 	if (desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) {
543 		pr_err("middle segment too big %d\n", desc->fd_lens[2]);
544 		return -EINVAL;
545 	}
546 	if (desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) {
547 		pr_err("data segment too big %d\n", desc->fd_lens[3]);
548 		return -EINVAL;
549 	}
550 
551 	return 0;
552 }
553 
554 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted)
555 {
556 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
557 						 FRAME_LATE_STATUS_COMPLETE;
558 	cpu_to_le32s(&con->v2.out_epil.front_crc);
559 	cpu_to_le32s(&con->v2.out_epil.middle_crc);
560 	cpu_to_le32s(&con->v2.out_epil.data_crc);
561 }
562 
563 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted)
564 {
565 	memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil));
566 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
567 						 FRAME_LATE_STATUS_COMPLETE;
568 }
569 
570 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc,
571 			   u32 *data_crc)
572 {
573 	u8 late_status;
574 
575 	late_status = ceph_decode_8(&p);
576 	if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) !=
577 			FRAME_LATE_STATUS_COMPLETE) {
578 		/* we should never get an aborted message as client */
579 		pr_err("bad late_status 0x%x\n", late_status);
580 		return -EINVAL;
581 	}
582 
583 	if (front_crc && middle_crc && data_crc) {
584 		*front_crc = ceph_decode_32(&p);
585 		*middle_crc = ceph_decode_32(&p);
586 		*data_crc = ceph_decode_32(&p);
587 	}
588 
589 	return 0;
590 }
591 
592 static void fill_header(struct ceph_msg_header *hdr,
593 			const struct ceph_msg_header2 *hdr2,
594 			int front_len, int middle_len, int data_len,
595 			const struct ceph_entity_name *peer_name)
596 {
597 	hdr->seq = hdr2->seq;
598 	hdr->tid = hdr2->tid;
599 	hdr->type = hdr2->type;
600 	hdr->priority = hdr2->priority;
601 	hdr->version = hdr2->version;
602 	hdr->front_len = cpu_to_le32(front_len);
603 	hdr->middle_len = cpu_to_le32(middle_len);
604 	hdr->data_len = cpu_to_le32(data_len);
605 	hdr->data_off = hdr2->data_off;
606 	hdr->src = *peer_name;
607 	hdr->compat_version = hdr2->compat_version;
608 	hdr->reserved = 0;
609 	hdr->crc = 0;
610 }
611 
612 static void fill_header2(struct ceph_msg_header2 *hdr2,
613 			 const struct ceph_msg_header *hdr, u64 ack_seq)
614 {
615 	hdr2->seq = hdr->seq;
616 	hdr2->tid = hdr->tid;
617 	hdr2->type = hdr->type;
618 	hdr2->priority = hdr->priority;
619 	hdr2->version = hdr->version;
620 	hdr2->data_pre_padding_len = 0;
621 	hdr2->data_off = hdr->data_off;
622 	hdr2->ack_seq = cpu_to_le64(ack_seq);
623 	hdr2->flags = 0;
624 	hdr2->compat_version = hdr->compat_version;
625 	hdr2->reserved = 0;
626 }
627 
628 static int verify_control_crc(struct ceph_connection *con)
629 {
630 	int ctrl_len = con->v2.in_desc.fd_lens[0];
631 	u32 crc, expected_crc;
632 
633 	WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len);
634 	WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN);
635 
636 	crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len);
637 	expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base);
638 	if (crc != expected_crc) {
639 		pr_err("bad control crc, calculated %u, expected %u\n",
640 		       crc, expected_crc);
641 		return -EBADMSG;
642 	}
643 
644 	return 0;
645 }
646 
647 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc,
648 				u32 middle_crc, u32 data_crc)
649 {
650 	if (front_len(con->in_msg)) {
651 		con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base,
652 					   front_len(con->in_msg));
653 	} else {
654 		WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg));
655 		con->in_front_crc = -1;
656 	}
657 
658 	if (middle_len(con->in_msg))
659 		con->in_middle_crc = crc32c(-1,
660 					    con->in_msg->middle->vec.iov_base,
661 					    middle_len(con->in_msg));
662 	else if (data_len(con->in_msg))
663 		con->in_middle_crc = -1;
664 	else
665 		con->in_middle_crc = 0;
666 
667 	if (!data_len(con->in_msg))
668 		con->in_data_crc = 0;
669 
670 	dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg,
671 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
672 
673 	if (con->in_front_crc != front_crc) {
674 		pr_err("bad front crc, calculated %u, expected %u\n",
675 		       con->in_front_crc, front_crc);
676 		return -EBADMSG;
677 	}
678 	if (con->in_middle_crc != middle_crc) {
679 		pr_err("bad middle crc, calculated %u, expected %u\n",
680 		       con->in_middle_crc, middle_crc);
681 		return -EBADMSG;
682 	}
683 	if (con->in_data_crc != data_crc) {
684 		pr_err("bad data crc, calculated %u, expected %u\n",
685 		       con->in_data_crc, data_crc);
686 		return -EBADMSG;
687 	}
688 
689 	return 0;
690 }
691 
692 static int setup_crypto(struct ceph_connection *con,
693 			const u8 *session_key, int session_key_len,
694 			const u8 *con_secret, int con_secret_len)
695 {
696 	unsigned int noio_flag;
697 	int ret;
698 
699 	dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n",
700 	     __func__, con, con->v2.con_mode, session_key_len, con_secret_len);
701 	WARN_ON(con->v2.hmac_tfm || con->v2.gcm_tfm || con->v2.gcm_req);
702 
703 	if (con->v2.con_mode != CEPH_CON_MODE_CRC &&
704 	    con->v2.con_mode != CEPH_CON_MODE_SECURE) {
705 		pr_err("bad con_mode %d\n", con->v2.con_mode);
706 		return -EINVAL;
707 	}
708 
709 	if (!session_key_len) {
710 		WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC);
711 		WARN_ON(con_secret_len);
712 		return 0;  /* auth_none */
713 	}
714 
715 	noio_flag = memalloc_noio_save();
716 	con->v2.hmac_tfm = crypto_alloc_shash("hmac(sha256)", 0, 0);
717 	memalloc_noio_restore(noio_flag);
718 	if (IS_ERR(con->v2.hmac_tfm)) {
719 		ret = PTR_ERR(con->v2.hmac_tfm);
720 		con->v2.hmac_tfm = NULL;
721 		pr_err("failed to allocate hmac tfm context: %d\n", ret);
722 		return ret;
723 	}
724 
725 	WARN_ON((unsigned long)session_key &
726 		crypto_shash_alignmask(con->v2.hmac_tfm));
727 	ret = crypto_shash_setkey(con->v2.hmac_tfm, session_key,
728 				  session_key_len);
729 	if (ret) {
730 		pr_err("failed to set hmac key: %d\n", ret);
731 		return ret;
732 	}
733 
734 	if (con->v2.con_mode == CEPH_CON_MODE_CRC) {
735 		WARN_ON(con_secret_len);
736 		return 0;  /* auth_x, plain mode */
737 	}
738 
739 	if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) {
740 		pr_err("con_secret too small %d\n", con_secret_len);
741 		return -EINVAL;
742 	}
743 
744 	noio_flag = memalloc_noio_save();
745 	con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0);
746 	memalloc_noio_restore(noio_flag);
747 	if (IS_ERR(con->v2.gcm_tfm)) {
748 		ret = PTR_ERR(con->v2.gcm_tfm);
749 		con->v2.gcm_tfm = NULL;
750 		pr_err("failed to allocate gcm tfm context: %d\n", ret);
751 		return ret;
752 	}
753 
754 	WARN_ON((unsigned long)con_secret &
755 		crypto_aead_alignmask(con->v2.gcm_tfm));
756 	ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN);
757 	if (ret) {
758 		pr_err("failed to set gcm key: %d\n", ret);
759 		return ret;
760 	}
761 
762 	WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN);
763 	ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN);
764 	if (ret) {
765 		pr_err("failed to set gcm tag size: %d\n", ret);
766 		return ret;
767 	}
768 
769 	con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO);
770 	if (!con->v2.gcm_req) {
771 		pr_err("failed to allocate gcm request\n");
772 		return -ENOMEM;
773 	}
774 
775 	crypto_init_wait(&con->v2.gcm_wait);
776 	aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
777 				  crypto_req_done, &con->v2.gcm_wait);
778 
779 	memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN,
780 	       CEPH_GCM_IV_LEN);
781 	memcpy(&con->v2.out_gcm_nonce,
782 	       con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN,
783 	       CEPH_GCM_IV_LEN);
784 	return 0;  /* auth_x, secure mode */
785 }
786 
787 static int hmac_sha256(struct ceph_connection *con, const struct kvec *kvecs,
788 		       int kvec_cnt, u8 *hmac)
789 {
790 	SHASH_DESC_ON_STACK(desc, con->v2.hmac_tfm);  /* tfm arg is ignored */
791 	int ret;
792 	int i;
793 
794 	dout("%s con %p hmac_tfm %p kvec_cnt %d\n", __func__, con,
795 	     con->v2.hmac_tfm, kvec_cnt);
796 
797 	if (!con->v2.hmac_tfm) {
798 		memset(hmac, 0, SHA256_DIGEST_SIZE);
799 		return 0;  /* auth_none */
800 	}
801 
802 	desc->tfm = con->v2.hmac_tfm;
803 	ret = crypto_shash_init(desc);
804 	if (ret)
805 		goto out;
806 
807 	for (i = 0; i < kvec_cnt; i++) {
808 		WARN_ON((unsigned long)kvecs[i].iov_base &
809 			crypto_shash_alignmask(con->v2.hmac_tfm));
810 		ret = crypto_shash_update(desc, kvecs[i].iov_base,
811 					  kvecs[i].iov_len);
812 		if (ret)
813 			goto out;
814 	}
815 
816 	ret = crypto_shash_final(desc, hmac);
817 
818 out:
819 	shash_desc_zero(desc);
820 	return ret;  /* auth_x, both plain and secure modes */
821 }
822 
823 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce)
824 {
825 	u64 counter;
826 
827 	counter = le64_to_cpu(nonce->counter);
828 	nonce->counter = cpu_to_le64(counter + 1);
829 }
830 
831 static int gcm_crypt(struct ceph_connection *con, bool encrypt,
832 		     struct scatterlist *src, struct scatterlist *dst,
833 		     int src_len)
834 {
835 	struct ceph_gcm_nonce *nonce;
836 	int ret;
837 
838 	nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce;
839 
840 	aead_request_set_ad(con->v2.gcm_req, 0);  /* no AAD */
841 	aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce);
842 	ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) :
843 					crypto_aead_decrypt(con->v2.gcm_req),
844 			      &con->v2.gcm_wait);
845 	if (ret)
846 		return ret;
847 
848 	gcm_inc_nonce(nonce);
849 	return 0;
850 }
851 
852 static void get_bvec_at(struct ceph_msg_data_cursor *cursor,
853 			struct bio_vec *bv)
854 {
855 	struct page *page;
856 	size_t off, len;
857 
858 	WARN_ON(!cursor->total_resid);
859 
860 	/* skip zero-length data items */
861 	while (!cursor->resid)
862 		ceph_msg_data_advance(cursor, 0);
863 
864 	/* get a piece of data, cursor isn't advanced */
865 	page = ceph_msg_data_next(cursor, &off, &len, NULL);
866 
867 	bv->bv_page = page;
868 	bv->bv_offset = off;
869 	bv->bv_len = len;
870 }
871 
872 static int calc_sg_cnt(void *buf, int buf_len)
873 {
874 	int sg_cnt;
875 
876 	if (!buf_len)
877 		return 0;
878 
879 	sg_cnt = need_padding(buf_len) ? 1 : 0;
880 	if (is_vmalloc_addr(buf)) {
881 		WARN_ON(offset_in_page(buf));
882 		sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT;
883 	} else {
884 		sg_cnt++;
885 	}
886 
887 	return sg_cnt;
888 }
889 
890 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor)
891 {
892 	int data_len = cursor->total_resid;
893 	struct bio_vec bv;
894 	int sg_cnt;
895 
896 	if (!data_len)
897 		return 0;
898 
899 	sg_cnt = need_padding(data_len) ? 1 : 0;
900 	do {
901 		get_bvec_at(cursor, &bv);
902 		sg_cnt++;
903 
904 		ceph_msg_data_advance(cursor, bv.bv_len);
905 	} while (cursor->total_resid);
906 
907 	return sg_cnt;
908 }
909 
910 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad)
911 {
912 	void *end = buf + buf_len;
913 	struct page *page;
914 	int len;
915 	void *p;
916 
917 	if (!buf_len)
918 		return;
919 
920 	if (is_vmalloc_addr(buf)) {
921 		p = buf;
922 		do {
923 			page = vmalloc_to_page(p);
924 			len = min_t(int, end - p, PAGE_SIZE);
925 			WARN_ON(!page || !len || offset_in_page(p));
926 			sg_set_page(*sg, page, len, 0);
927 			*sg = sg_next(*sg);
928 			p += len;
929 		} while (p != end);
930 	} else {
931 		sg_set_buf(*sg, buf, buf_len);
932 		*sg = sg_next(*sg);
933 	}
934 
935 	if (need_padding(buf_len)) {
936 		sg_set_buf(*sg, pad, padding_len(buf_len));
937 		*sg = sg_next(*sg);
938 	}
939 }
940 
941 static void init_sgs_cursor(struct scatterlist **sg,
942 			    struct ceph_msg_data_cursor *cursor, u8 *pad)
943 {
944 	int data_len = cursor->total_resid;
945 	struct bio_vec bv;
946 
947 	if (!data_len)
948 		return;
949 
950 	do {
951 		get_bvec_at(cursor, &bv);
952 		sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
953 		*sg = sg_next(*sg);
954 
955 		ceph_msg_data_advance(cursor, bv.bv_len);
956 	} while (cursor->total_resid);
957 
958 	if (need_padding(data_len)) {
959 		sg_set_buf(*sg, pad, padding_len(data_len));
960 		*sg = sg_next(*sg);
961 	}
962 }
963 
964 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
965 			     u8 *front_pad, u8 *middle_pad, u8 *data_pad,
966 			     void *epilogue, bool add_tag)
967 {
968 	struct ceph_msg_data_cursor cursor;
969 	struct scatterlist *cur_sg;
970 	int sg_cnt;
971 	int ret;
972 
973 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
974 		return 0;
975 
976 	sg_cnt = 1;  /* epilogue + [auth tag] */
977 	if (front_len(msg))
978 		sg_cnt += calc_sg_cnt(msg->front.iov_base,
979 				      front_len(msg));
980 	if (middle_len(msg))
981 		sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
982 				      middle_len(msg));
983 	if (data_len(msg)) {
984 		ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
985 		sg_cnt += calc_sg_cnt_cursor(&cursor);
986 	}
987 
988 	ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
989 	if (ret)
990 		return ret;
991 
992 	cur_sg = sgt->sgl;
993 	if (front_len(msg))
994 		init_sgs(&cur_sg, msg->front.iov_base, front_len(msg),
995 			 front_pad);
996 	if (middle_len(msg))
997 		init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
998 			 middle_pad);
999 	if (data_len(msg)) {
1000 		ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
1001 		init_sgs_cursor(&cur_sg, &cursor, data_pad);
1002 	}
1003 
1004 	WARN_ON(!sg_is_last(cur_sg));
1005 	sg_set_buf(cur_sg, epilogue,
1006 		   CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0));
1007 	return 0;
1008 }
1009 
1010 static int decrypt_preamble(struct ceph_connection *con)
1011 {
1012 	struct scatterlist sg;
1013 
1014 	sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN);
1015 	return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN);
1016 }
1017 
1018 static int decrypt_control_remainder(struct ceph_connection *con)
1019 {
1020 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1021 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1022 	int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN;
1023 	struct scatterlist sgs[2];
1024 
1025 	WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len);
1026 	WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len);
1027 
1028 	sg_init_table(sgs, 2);
1029 	sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len);
1030 	sg_set_buf(&sgs[1], con->v2.in_buf, pt_len);
1031 
1032 	return gcm_crypt(con, false, sgs, sgs,
1033 			 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
1034 }
1035 
1036 static int decrypt_tail(struct ceph_connection *con)
1037 {
1038 	struct sg_table enc_sgt = {};
1039 	struct sg_table sgt = {};
1040 	int tail_len;
1041 	int ret;
1042 
1043 	tail_len = tail_onwire_len(con->in_msg, true);
1044 	ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages,
1045 					con->v2.in_enc_page_cnt, 0, tail_len,
1046 					GFP_NOIO);
1047 	if (ret)
1048 		goto out;
1049 
1050 	ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
1051 			MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
1052 			con->v2.in_buf, true);
1053 	if (ret)
1054 		goto out;
1055 
1056 	dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con,
1057 	     con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents);
1058 	ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len);
1059 	if (ret)
1060 		goto out;
1061 
1062 	WARN_ON(!con->v2.in_enc_page_cnt);
1063 	ceph_release_page_vector(con->v2.in_enc_pages,
1064 				 con->v2.in_enc_page_cnt);
1065 	con->v2.in_enc_pages = NULL;
1066 	con->v2.in_enc_page_cnt = 0;
1067 
1068 out:
1069 	sg_free_table(&sgt);
1070 	sg_free_table(&enc_sgt);
1071 	return ret;
1072 }
1073 
1074 static int prepare_banner(struct ceph_connection *con)
1075 {
1076 	int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8;
1077 	void *buf, *p;
1078 
1079 	buf = alloc_conn_buf(con, buf_len);
1080 	if (!buf)
1081 		return -ENOMEM;
1082 
1083 	p = buf;
1084 	ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN);
1085 	ceph_encode_16(&p, sizeof(u64) + sizeof(u64));
1086 	ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES);
1087 	ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES);
1088 	WARN_ON(p != buf + buf_len);
1089 
1090 	add_out_kvec(con, buf, buf_len);
1091 	add_out_sign_kvec(con, buf, buf_len);
1092 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1093 	return 0;
1094 }
1095 
1096 /*
1097  * base:
1098  *   preamble
1099  *   control body (ctrl_len bytes)
1100  *   space for control crc
1101  *
1102  * extdata (optional):
1103  *   control body (extdata_len bytes)
1104  *
1105  * Compute control crc and gather base and extdata into:
1106  *
1107  *   preamble
1108  *   control body (ctrl_len + extdata_len bytes)
1109  *   control crc
1110  *
1111  * Preamble should already be encoded at the start of base.
1112  */
1113 static void prepare_head_plain(struct ceph_connection *con, void *base,
1114 			       int ctrl_len, void *extdata, int extdata_len,
1115 			       bool to_be_signed)
1116 {
1117 	int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN;
1118 	void *crcp = base + base_len - CEPH_CRC_LEN;
1119 	u32 crc;
1120 
1121 	crc = crc32c(-1, CTRL_BODY(base), ctrl_len);
1122 	if (extdata_len)
1123 		crc = crc32c(crc, extdata, extdata_len);
1124 	put_unaligned_le32(crc, crcp);
1125 
1126 	if (!extdata_len) {
1127 		add_out_kvec(con, base, base_len);
1128 		if (to_be_signed)
1129 			add_out_sign_kvec(con, base, base_len);
1130 		return;
1131 	}
1132 
1133 	add_out_kvec(con, base, crcp - base);
1134 	add_out_kvec(con, extdata, extdata_len);
1135 	add_out_kvec(con, crcp, CEPH_CRC_LEN);
1136 	if (to_be_signed) {
1137 		add_out_sign_kvec(con, base, crcp - base);
1138 		add_out_sign_kvec(con, extdata, extdata_len);
1139 		add_out_sign_kvec(con, crcp, CEPH_CRC_LEN);
1140 	}
1141 }
1142 
1143 static int prepare_head_secure_small(struct ceph_connection *con,
1144 				     void *base, int ctrl_len)
1145 {
1146 	struct scatterlist sg;
1147 	int ret;
1148 
1149 	/* inline buffer padding? */
1150 	if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN)
1151 		memset(CTRL_BODY(base) + ctrl_len, 0,
1152 		       CEPH_PREAMBLE_INLINE_LEN - ctrl_len);
1153 
1154 	sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN);
1155 	ret = gcm_crypt(con, true, &sg, &sg,
1156 			CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN);
1157 	if (ret)
1158 		return ret;
1159 
1160 	add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN);
1161 	return 0;
1162 }
1163 
1164 /*
1165  * base:
1166  *   preamble
1167  *   control body (ctrl_len bytes)
1168  *   space for padding, if needed
1169  *   space for control remainder auth tag
1170  *   space for preamble auth tag
1171  *
1172  * Encrypt preamble and the inline portion, then encrypt the remainder
1173  * and gather into:
1174  *
1175  *   preamble
1176  *   control body (48 bytes)
1177  *   preamble auth tag
1178  *   control body (ctrl_len - 48 bytes)
1179  *   zero padding, if needed
1180  *   control remainder auth tag
1181  *
1182  * Preamble should already be encoded at the start of base.
1183  */
1184 static int prepare_head_secure_big(struct ceph_connection *con,
1185 				   void *base, int ctrl_len)
1186 {
1187 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1188 	void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN;
1189 	void *rem_tag = rem + padded_len(rem_len);
1190 	void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN;
1191 	struct scatterlist sgs[2];
1192 	int ret;
1193 
1194 	sg_init_table(sgs, 2);
1195 	sg_set_buf(&sgs[0], base, rem - base);
1196 	sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN);
1197 	ret = gcm_crypt(con, true, sgs, sgs, rem - base);
1198 	if (ret)
1199 		return ret;
1200 
1201 	/* control remainder padding? */
1202 	if (need_padding(rem_len))
1203 		memset(rem + rem_len, 0, padding_len(rem_len));
1204 
1205 	sg_init_one(&sgs[0], rem, pmbl_tag - rem);
1206 	ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem);
1207 	if (ret)
1208 		return ret;
1209 
1210 	add_out_kvec(con, base, rem - base);
1211 	add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN);
1212 	add_out_kvec(con, rem, pmbl_tag - rem);
1213 	return 0;
1214 }
1215 
1216 static int __prepare_control(struct ceph_connection *con, int tag,
1217 			     void *base, int ctrl_len, void *extdata,
1218 			     int extdata_len, bool to_be_signed)
1219 {
1220 	int total_len = ctrl_len + extdata_len;
1221 	struct ceph_frame_desc desc;
1222 	int ret;
1223 
1224 	dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag,
1225 	     total_len, ctrl_len, extdata_len);
1226 
1227 	/* extdata may be vmalloc'ed but not base */
1228 	if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len))
1229 		return -EINVAL;
1230 
1231 	init_frame_desc(&desc, tag, &total_len, 1);
1232 	encode_preamble(&desc, base);
1233 
1234 	if (con_secure(con)) {
1235 		if (WARN_ON(extdata_len || to_be_signed))
1236 			return -EINVAL;
1237 
1238 		if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN)
1239 			/* fully inlined, inline buffer may need padding */
1240 			ret = prepare_head_secure_small(con, base, ctrl_len);
1241 		else
1242 			/* partially inlined, inline buffer is full */
1243 			ret = prepare_head_secure_big(con, base, ctrl_len);
1244 		if (ret)
1245 			return ret;
1246 	} else {
1247 		prepare_head_plain(con, base, ctrl_len, extdata, extdata_len,
1248 				   to_be_signed);
1249 	}
1250 
1251 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1252 	return 0;
1253 }
1254 
1255 static int prepare_control(struct ceph_connection *con, int tag,
1256 			   void *base, int ctrl_len)
1257 {
1258 	return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false);
1259 }
1260 
1261 static int prepare_hello(struct ceph_connection *con)
1262 {
1263 	void *buf, *p;
1264 	int ctrl_len;
1265 
1266 	ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr);
1267 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1268 	if (!buf)
1269 		return -ENOMEM;
1270 
1271 	p = CTRL_BODY(buf);
1272 	ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT);
1273 	ceph_encode_entity_addr(&p, &con->peer_addr);
1274 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1275 
1276 	return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len,
1277 				 NULL, 0, true);
1278 }
1279 
1280 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */
1281 #define AUTH_BUF_LEN	(512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN)
1282 
1283 static int prepare_auth_request(struct ceph_connection *con)
1284 {
1285 	void *authorizer, *authorizer_copy;
1286 	int ctrl_len, authorizer_len;
1287 	void *buf;
1288 	int ret;
1289 
1290 	ctrl_len = AUTH_BUF_LEN;
1291 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1292 	if (!buf)
1293 		return -ENOMEM;
1294 
1295 	mutex_unlock(&con->mutex);
1296 	ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len,
1297 					 &authorizer, &authorizer_len);
1298 	mutex_lock(&con->mutex);
1299 	if (con->state != CEPH_CON_S_V2_HELLO) {
1300 		dout("%s con %p state changed to %d\n", __func__, con,
1301 		     con->state);
1302 		return -EAGAIN;
1303 	}
1304 
1305 	dout("%s con %p get_auth_request ret %d\n", __func__, con, ret);
1306 	if (ret)
1307 		return ret;
1308 
1309 	authorizer_copy = alloc_conn_buf(con, authorizer_len);
1310 	if (!authorizer_copy)
1311 		return -ENOMEM;
1312 
1313 	memcpy(authorizer_copy, authorizer, authorizer_len);
1314 
1315 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len,
1316 				 authorizer_copy, authorizer_len, true);
1317 }
1318 
1319 static int prepare_auth_request_more(struct ceph_connection *con,
1320 				     void *reply, int reply_len)
1321 {
1322 	int ctrl_len, authorizer_len;
1323 	void *authorizer;
1324 	void *buf;
1325 	int ret;
1326 
1327 	ctrl_len = AUTH_BUF_LEN;
1328 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1329 	if (!buf)
1330 		return -ENOMEM;
1331 
1332 	mutex_unlock(&con->mutex);
1333 	ret = con->ops->handle_auth_reply_more(con, reply, reply_len,
1334 					       CTRL_BODY(buf), &ctrl_len,
1335 					       &authorizer, &authorizer_len);
1336 	mutex_lock(&con->mutex);
1337 	if (con->state != CEPH_CON_S_V2_AUTH) {
1338 		dout("%s con %p state changed to %d\n", __func__, con,
1339 		     con->state);
1340 		return -EAGAIN;
1341 	}
1342 
1343 	dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret);
1344 	if (ret)
1345 		return ret;
1346 
1347 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf,
1348 				 ctrl_len, authorizer, authorizer_len, true);
1349 }
1350 
1351 static int prepare_auth_signature(struct ceph_connection *con)
1352 {
1353 	void *buf;
1354 	int ret;
1355 
1356 	buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE,
1357 						  con_secure(con)));
1358 	if (!buf)
1359 		return -ENOMEM;
1360 
1361 	ret = hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt,
1362 			  CTRL_BODY(buf));
1363 	if (ret)
1364 		return ret;
1365 
1366 	return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf,
1367 			       SHA256_DIGEST_SIZE);
1368 }
1369 
1370 static int prepare_client_ident(struct ceph_connection *con)
1371 {
1372 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1373 	struct ceph_client *client = from_msgr(con->msgr);
1374 	u64 global_id = ceph_client_gid(client);
1375 	void *buf, *p;
1376 	int ctrl_len;
1377 
1378 	WARN_ON(con->v2.server_cookie);
1379 	WARN_ON(con->v2.connect_seq);
1380 	WARN_ON(con->v2.peer_global_seq);
1381 
1382 	if (!con->v2.client_cookie) {
1383 		do {
1384 			get_random_bytes(&con->v2.client_cookie,
1385 					 sizeof(con->v2.client_cookie));
1386 		} while (!con->v2.client_cookie);
1387 		dout("%s con %p generated cookie 0x%llx\n", __func__, con,
1388 		     con->v2.client_cookie);
1389 	} else {
1390 		dout("%s con %p cookie already set 0x%llx\n", __func__, con,
1391 		     con->v2.client_cookie);
1392 	}
1393 
1394 	dout("%s con %p my_addr %s/%u peer_addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx cookie 0x%llx\n",
1395 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1396 	     ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce),
1397 	     global_id, con->v2.global_seq, client->supported_features,
1398 	     client->required_features, con->v2.client_cookie);
1399 
1400 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) +
1401 		   ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8;
1402 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1403 	if (!buf)
1404 		return -ENOMEM;
1405 
1406 	p = CTRL_BODY(buf);
1407 	ceph_encode_8(&p, 2);  /* addrvec marker */
1408 	ceph_encode_32(&p, 1);  /* addr_cnt */
1409 	ceph_encode_entity_addr(&p, my_addr);
1410 	ceph_encode_entity_addr(&p, &con->peer_addr);
1411 	ceph_encode_64(&p, global_id);
1412 	ceph_encode_64(&p, con->v2.global_seq);
1413 	ceph_encode_64(&p, client->supported_features);
1414 	ceph_encode_64(&p, client->required_features);
1415 	ceph_encode_64(&p, 0);  /* flags */
1416 	ceph_encode_64(&p, con->v2.client_cookie);
1417 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1418 
1419 	return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len);
1420 }
1421 
1422 static int prepare_session_reconnect(struct ceph_connection *con)
1423 {
1424 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1425 	void *buf, *p;
1426 	int ctrl_len;
1427 
1428 	WARN_ON(!con->v2.client_cookie);
1429 	WARN_ON(!con->v2.server_cookie);
1430 	WARN_ON(!con->v2.connect_seq);
1431 	WARN_ON(!con->v2.peer_global_seq);
1432 
1433 	dout("%s con %p my_addr %s/%u client_cookie 0x%llx server_cookie 0x%llx global_seq %llu connect_seq %llu in_seq %llu\n",
1434 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1435 	     con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq,
1436 	     con->v2.connect_seq, con->in_seq);
1437 
1438 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8;
1439 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1440 	if (!buf)
1441 		return -ENOMEM;
1442 
1443 	p = CTRL_BODY(buf);
1444 	ceph_encode_8(&p, 2);  /* entity_addrvec_t marker */
1445 	ceph_encode_32(&p, 1);  /* my_addrs len */
1446 	ceph_encode_entity_addr(&p, my_addr);
1447 	ceph_encode_64(&p, con->v2.client_cookie);
1448 	ceph_encode_64(&p, con->v2.server_cookie);
1449 	ceph_encode_64(&p, con->v2.global_seq);
1450 	ceph_encode_64(&p, con->v2.connect_seq);
1451 	ceph_encode_64(&p, con->in_seq);
1452 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1453 
1454 	return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len);
1455 }
1456 
1457 static int prepare_keepalive2(struct ceph_connection *con)
1458 {
1459 	struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
1460 	struct timespec64 now;
1461 
1462 	ktime_get_real_ts64(&now);
1463 	dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec,
1464 	     now.tv_nsec);
1465 
1466 	ceph_encode_timespec64(ts, &now);
1467 
1468 	reset_out_kvecs(con);
1469 	return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf,
1470 			       sizeof(struct ceph_timespec));
1471 }
1472 
1473 static int prepare_ack(struct ceph_connection *con)
1474 {
1475 	void *p;
1476 
1477 	dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1478 	     con->in_seq_acked, con->in_seq);
1479 	con->in_seq_acked = con->in_seq;
1480 
1481 	p = CTRL_BODY(con->v2.out_buf);
1482 	ceph_encode_64(&p, con->in_seq_acked);
1483 
1484 	reset_out_kvecs(con);
1485 	return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8);
1486 }
1487 
1488 static void prepare_epilogue_plain(struct ceph_connection *con, bool aborted)
1489 {
1490 	dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con,
1491 	     con->out_msg, aborted, con->v2.out_epil.front_crc,
1492 	     con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc);
1493 
1494 	encode_epilogue_plain(con, aborted);
1495 	add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN);
1496 }
1497 
1498 /*
1499  * For "used" empty segments, crc is -1.  For unused (trailing)
1500  * segments, crc is 0.
1501  */
1502 static void prepare_message_plain(struct ceph_connection *con)
1503 {
1504 	struct ceph_msg *msg = con->out_msg;
1505 
1506 	prepare_head_plain(con, con->v2.out_buf,
1507 			   sizeof(struct ceph_msg_header2), NULL, 0, false);
1508 
1509 	if (!front_len(msg) && !middle_len(msg)) {
1510 		if (!data_len(msg)) {
1511 			/*
1512 			 * Empty message: once the head is written,
1513 			 * we are done -- there is no epilogue.
1514 			 */
1515 			con->v2.out_state = OUT_S_FINISH_MESSAGE;
1516 			return;
1517 		}
1518 
1519 		con->v2.out_epil.front_crc = -1;
1520 		con->v2.out_epil.middle_crc = -1;
1521 		con->v2.out_state = OUT_S_QUEUE_DATA;
1522 		return;
1523 	}
1524 
1525 	if (front_len(msg)) {
1526 		con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base,
1527 						    front_len(msg));
1528 		add_out_kvec(con, msg->front.iov_base, front_len(msg));
1529 	} else {
1530 		/* middle (at least) is there, checked above */
1531 		con->v2.out_epil.front_crc = -1;
1532 	}
1533 
1534 	if (middle_len(msg)) {
1535 		con->v2.out_epil.middle_crc =
1536 			crc32c(-1, msg->middle->vec.iov_base, middle_len(msg));
1537 		add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1538 	} else {
1539 		con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0;
1540 	}
1541 
1542 	if (data_len(msg)) {
1543 		con->v2.out_state = OUT_S_QUEUE_DATA;
1544 	} else {
1545 		con->v2.out_epil.data_crc = 0;
1546 		prepare_epilogue_plain(con, false);
1547 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1548 	}
1549 }
1550 
1551 /*
1552  * Unfortunately the kernel crypto API doesn't support streaming
1553  * (piecewise) operation for AEAD algorithms, so we can't get away
1554  * with a fixed size buffer and a couple sgs.  Instead, we have to
1555  * allocate pages for the entire tail of the message (currently up
1556  * to ~32M) and two sgs arrays (up to ~256K each)...
1557  */
1558 static int prepare_message_secure(struct ceph_connection *con)
1559 {
1560 	void *zerop = page_address(ceph_zero_page);
1561 	struct sg_table enc_sgt = {};
1562 	struct sg_table sgt = {};
1563 	struct page **enc_pages;
1564 	int enc_page_cnt;
1565 	int tail_len;
1566 	int ret;
1567 
1568 	ret = prepare_head_secure_small(con, con->v2.out_buf,
1569 					sizeof(struct ceph_msg_header2));
1570 	if (ret)
1571 		return ret;
1572 
1573 	tail_len = tail_onwire_len(con->out_msg, true);
1574 	if (!tail_len) {
1575 		/*
1576 		 * Empty message: once the head is written,
1577 		 * we are done -- there is no epilogue.
1578 		 */
1579 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1580 		return 0;
1581 	}
1582 
1583 	encode_epilogue_secure(con, false);
1584 	ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop,
1585 				&con->v2.out_epil, false);
1586 	if (ret)
1587 		goto out;
1588 
1589 	enc_page_cnt = calc_pages_for(0, tail_len);
1590 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1591 	if (IS_ERR(enc_pages)) {
1592 		ret = PTR_ERR(enc_pages);
1593 		goto out;
1594 	}
1595 
1596 	WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt);
1597 	con->v2.out_enc_pages = enc_pages;
1598 	con->v2.out_enc_page_cnt = enc_page_cnt;
1599 	con->v2.out_enc_resid = tail_len;
1600 	con->v2.out_enc_i = 0;
1601 
1602 	ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt,
1603 					0, tail_len, GFP_NOIO);
1604 	if (ret)
1605 		goto out;
1606 
1607 	ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl,
1608 			tail_len - CEPH_GCM_TAG_LEN);
1609 	if (ret)
1610 		goto out;
1611 
1612 	dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con,
1613 	     con->out_msg, sgt.orig_nents, enc_page_cnt);
1614 	con->v2.out_state = OUT_S_QUEUE_ENC_PAGE;
1615 
1616 out:
1617 	sg_free_table(&sgt);
1618 	sg_free_table(&enc_sgt);
1619 	return ret;
1620 }
1621 
1622 static int prepare_message(struct ceph_connection *con)
1623 {
1624 	int lens[] = {
1625 		sizeof(struct ceph_msg_header2),
1626 		front_len(con->out_msg),
1627 		middle_len(con->out_msg),
1628 		data_len(con->out_msg)
1629 	};
1630 	struct ceph_frame_desc desc;
1631 	int ret;
1632 
1633 	dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con,
1634 	     con->out_msg, lens[0], lens[1], lens[2], lens[3]);
1635 
1636 	if (con->in_seq > con->in_seq_acked) {
1637 		dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1638 		     con->in_seq_acked, con->in_seq);
1639 		con->in_seq_acked = con->in_seq;
1640 	}
1641 
1642 	reset_out_kvecs(con);
1643 	init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4);
1644 	encode_preamble(&desc, con->v2.out_buf);
1645 	fill_header2(CTRL_BODY(con->v2.out_buf), &con->out_msg->hdr,
1646 		     con->in_seq_acked);
1647 
1648 	if (con_secure(con)) {
1649 		ret = prepare_message_secure(con);
1650 		if (ret)
1651 			return ret;
1652 	} else {
1653 		prepare_message_plain(con);
1654 	}
1655 
1656 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1657 	return 0;
1658 }
1659 
1660 static int prepare_read_banner_prefix(struct ceph_connection *con)
1661 {
1662 	void *buf;
1663 
1664 	buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN);
1665 	if (!buf)
1666 		return -ENOMEM;
1667 
1668 	reset_in_kvecs(con);
1669 	add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1670 	add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1671 	con->state = CEPH_CON_S_V2_BANNER_PREFIX;
1672 	return 0;
1673 }
1674 
1675 static int prepare_read_banner_payload(struct ceph_connection *con,
1676 				       int payload_len)
1677 {
1678 	void *buf;
1679 
1680 	buf = alloc_conn_buf(con, payload_len);
1681 	if (!buf)
1682 		return -ENOMEM;
1683 
1684 	reset_in_kvecs(con);
1685 	add_in_kvec(con, buf, payload_len);
1686 	add_in_sign_kvec(con, buf, payload_len);
1687 	con->state = CEPH_CON_S_V2_BANNER_PAYLOAD;
1688 	return 0;
1689 }
1690 
1691 static void prepare_read_preamble(struct ceph_connection *con)
1692 {
1693 	reset_in_kvecs(con);
1694 	add_in_kvec(con, con->v2.in_buf,
1695 		    con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN :
1696 				      CEPH_PREAMBLE_PLAIN_LEN);
1697 	con->v2.in_state = IN_S_HANDLE_PREAMBLE;
1698 }
1699 
1700 static int prepare_read_control(struct ceph_connection *con)
1701 {
1702 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1703 	int head_len;
1704 	void *buf;
1705 
1706 	reset_in_kvecs(con);
1707 	if (con->state == CEPH_CON_S_V2_HELLO ||
1708 	    con->state == CEPH_CON_S_V2_AUTH) {
1709 		head_len = head_onwire_len(ctrl_len, false);
1710 		buf = alloc_conn_buf(con, head_len);
1711 		if (!buf)
1712 			return -ENOMEM;
1713 
1714 		/* preserve preamble */
1715 		memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN);
1716 
1717 		add_in_kvec(con, CTRL_BODY(buf), ctrl_len);
1718 		add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN);
1719 		add_in_sign_kvec(con, buf, head_len);
1720 	} else {
1721 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
1722 			buf = alloc_conn_buf(con, ctrl_len);
1723 			if (!buf)
1724 				return -ENOMEM;
1725 
1726 			add_in_kvec(con, buf, ctrl_len);
1727 		} else {
1728 			add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len);
1729 		}
1730 		add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN);
1731 	}
1732 	con->v2.in_state = IN_S_HANDLE_CONTROL;
1733 	return 0;
1734 }
1735 
1736 static int prepare_read_control_remainder(struct ceph_connection *con)
1737 {
1738 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1739 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1740 	void *buf;
1741 
1742 	buf = alloc_conn_buf(con, ctrl_len);
1743 	if (!buf)
1744 		return -ENOMEM;
1745 
1746 	memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN);
1747 
1748 	reset_in_kvecs(con);
1749 	add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len);
1750 	add_in_kvec(con, con->v2.in_buf,
1751 		    padding_len(rem_len) + CEPH_GCM_TAG_LEN);
1752 	con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER;
1753 	return 0;
1754 }
1755 
1756 static void prepare_read_data(struct ceph_connection *con)
1757 {
1758 	struct bio_vec bv;
1759 
1760 	con->in_data_crc = -1;
1761 	ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg,
1762 				  data_len(con->in_msg));
1763 
1764 	get_bvec_at(&con->v2.in_cursor, &bv);
1765 	set_in_bvec(con, &bv);
1766 	con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT;
1767 }
1768 
1769 static void prepare_read_data_cont(struct ceph_connection *con)
1770 {
1771 	struct bio_vec bv;
1772 
1773 	con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1774 					    con->v2.in_bvec.bv_page,
1775 					    con->v2.in_bvec.bv_offset,
1776 					    con->v2.in_bvec.bv_len);
1777 
1778 	ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len);
1779 	if (con->v2.in_cursor.total_resid) {
1780 		get_bvec_at(&con->v2.in_cursor, &bv);
1781 		set_in_bvec(con, &bv);
1782 		WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT);
1783 		return;
1784 	}
1785 
1786 	/*
1787 	 * We've read all data.  Prepare to read epilogue.
1788 	 */
1789 	reset_in_kvecs(con);
1790 	add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1791 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1792 }
1793 
1794 static void prepare_read_tail_plain(struct ceph_connection *con)
1795 {
1796 	struct ceph_msg *msg = con->in_msg;
1797 
1798 	if (!front_len(msg) && !middle_len(msg)) {
1799 		WARN_ON(!data_len(msg));
1800 		prepare_read_data(con);
1801 		return;
1802 	}
1803 
1804 	reset_in_kvecs(con);
1805 	if (front_len(msg)) {
1806 		add_in_kvec(con, msg->front.iov_base, front_len(msg));
1807 		WARN_ON(msg->front.iov_len != front_len(msg));
1808 	}
1809 	if (middle_len(msg)) {
1810 		add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1811 		WARN_ON(msg->middle->vec.iov_len != middle_len(msg));
1812 	}
1813 
1814 	if (data_len(msg)) {
1815 		con->v2.in_state = IN_S_PREPARE_READ_DATA;
1816 	} else {
1817 		add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1818 		con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1819 	}
1820 }
1821 
1822 static void prepare_read_enc_page(struct ceph_connection *con)
1823 {
1824 	struct bio_vec bv;
1825 
1826 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i,
1827 	     con->v2.in_enc_resid);
1828 	WARN_ON(!con->v2.in_enc_resid);
1829 
1830 	bv.bv_page = con->v2.in_enc_pages[con->v2.in_enc_i];
1831 	bv.bv_offset = 0;
1832 	bv.bv_len = min(con->v2.in_enc_resid, (int)PAGE_SIZE);
1833 
1834 	set_in_bvec(con, &bv);
1835 	con->v2.in_enc_i++;
1836 	con->v2.in_enc_resid -= bv.bv_len;
1837 
1838 	if (con->v2.in_enc_resid) {
1839 		con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE;
1840 		return;
1841 	}
1842 
1843 	/*
1844 	 * We are set to read the last piece of ciphertext (ending
1845 	 * with epilogue) + auth tag.
1846 	 */
1847 	WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt);
1848 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1849 }
1850 
1851 static int prepare_read_tail_secure(struct ceph_connection *con)
1852 {
1853 	struct page **enc_pages;
1854 	int enc_page_cnt;
1855 	int tail_len;
1856 
1857 	tail_len = tail_onwire_len(con->in_msg, true);
1858 	WARN_ON(!tail_len);
1859 
1860 	enc_page_cnt = calc_pages_for(0, tail_len);
1861 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1862 	if (IS_ERR(enc_pages))
1863 		return PTR_ERR(enc_pages);
1864 
1865 	WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt);
1866 	con->v2.in_enc_pages = enc_pages;
1867 	con->v2.in_enc_page_cnt = enc_page_cnt;
1868 	con->v2.in_enc_resid = tail_len;
1869 	con->v2.in_enc_i = 0;
1870 
1871 	prepare_read_enc_page(con);
1872 	return 0;
1873 }
1874 
1875 static void __finish_skip(struct ceph_connection *con)
1876 {
1877 	con->in_seq++;
1878 	prepare_read_preamble(con);
1879 }
1880 
1881 static void prepare_skip_message(struct ceph_connection *con)
1882 {
1883 	struct ceph_frame_desc *desc = &con->v2.in_desc;
1884 	int tail_len;
1885 
1886 	dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1],
1887 	     desc->fd_lens[2], desc->fd_lens[3]);
1888 
1889 	tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2],
1890 				     desc->fd_lens[3], con_secure(con));
1891 	if (!tail_len) {
1892 		__finish_skip(con);
1893 	} else {
1894 		set_in_skip(con, tail_len);
1895 		con->v2.in_state = IN_S_FINISH_SKIP;
1896 	}
1897 }
1898 
1899 static int process_banner_prefix(struct ceph_connection *con)
1900 {
1901 	int payload_len;
1902 	void *p;
1903 
1904 	WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN);
1905 
1906 	p = con->v2.in_kvecs[0].iov_base;
1907 	if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) {
1908 		if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN))
1909 			con->error_msg = "server is speaking msgr1 protocol";
1910 		else
1911 			con->error_msg = "protocol error, bad banner";
1912 		return -EINVAL;
1913 	}
1914 
1915 	p += CEPH_BANNER_V2_LEN;
1916 	payload_len = ceph_decode_16(&p);
1917 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
1918 
1919 	return prepare_read_banner_payload(con, payload_len);
1920 }
1921 
1922 static int process_banner_payload(struct ceph_connection *con)
1923 {
1924 	void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len;
1925 	u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES;
1926 	u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES;
1927 	u64 server_feat, server_req_feat;
1928 	void *p;
1929 	int ret;
1930 
1931 	p = con->v2.in_kvecs[0].iov_base;
1932 	ceph_decode_64_safe(&p, end, server_feat, bad);
1933 	ceph_decode_64_safe(&p, end, server_req_feat, bad);
1934 
1935 	dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n",
1936 	     __func__, con, server_feat, server_req_feat);
1937 
1938 	if (req_feat & ~server_feat) {
1939 		pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
1940 		       server_feat, req_feat & ~server_feat);
1941 		con->error_msg = "missing required protocol features";
1942 		return -EINVAL;
1943 	}
1944 	if (server_req_feat & ~feat) {
1945 		pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
1946 		       feat, server_req_feat & ~feat);
1947 		con->error_msg = "missing required protocol features";
1948 		return -EINVAL;
1949 	}
1950 
1951 	/* no reset_out_kvecs() as our banner may still be pending */
1952 	ret = prepare_hello(con);
1953 	if (ret) {
1954 		pr_err("prepare_hello failed: %d\n", ret);
1955 		return ret;
1956 	}
1957 
1958 	con->state = CEPH_CON_S_V2_HELLO;
1959 	prepare_read_preamble(con);
1960 	return 0;
1961 
1962 bad:
1963 	pr_err("failed to decode banner payload\n");
1964 	return -EINVAL;
1965 }
1966 
1967 static int process_hello(struct ceph_connection *con, void *p, void *end)
1968 {
1969 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1970 	struct ceph_entity_addr addr_for_me;
1971 	u8 entity_type;
1972 	int ret;
1973 
1974 	if (con->state != CEPH_CON_S_V2_HELLO) {
1975 		con->error_msg = "protocol error, unexpected hello";
1976 		return -EINVAL;
1977 	}
1978 
1979 	ceph_decode_8_safe(&p, end, entity_type, bad);
1980 	ret = ceph_decode_entity_addr(&p, end, &addr_for_me);
1981 	if (ret) {
1982 		pr_err("failed to decode addr_for_me: %d\n", ret);
1983 		return ret;
1984 	}
1985 
1986 	dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con,
1987 	     entity_type, ceph_pr_addr(&addr_for_me));
1988 
1989 	if (entity_type != con->peer_name.type) {
1990 		pr_err("bad peer type, want %d, got %d\n",
1991 		       con->peer_name.type, entity_type);
1992 		con->error_msg = "wrong peer at address";
1993 		return -EINVAL;
1994 	}
1995 
1996 	/*
1997 	 * Set our address to the address our first peer (i.e. monitor)
1998 	 * sees that we are connecting from.  If we are behind some sort
1999 	 * of NAT and want to be identified by some private (not NATed)
2000 	 * address, ip option should be used.
2001 	 */
2002 	if (ceph_addr_is_blank(my_addr)) {
2003 		memcpy(&my_addr->in_addr, &addr_for_me.in_addr,
2004 		       sizeof(my_addr->in_addr));
2005 		ceph_addr_set_port(my_addr, 0);
2006 		dout("%s con %p set my addr %s, as seen by peer %s\n",
2007 		     __func__, con, ceph_pr_addr(my_addr),
2008 		     ceph_pr_addr(&con->peer_addr));
2009 	} else {
2010 		dout("%s con %p my addr already set %s\n",
2011 		     __func__, con, ceph_pr_addr(my_addr));
2012 	}
2013 
2014 	WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr));
2015 	WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY);
2016 	WARN_ON(!my_addr->nonce);
2017 
2018 	/* no reset_out_kvecs() as our hello may still be pending */
2019 	ret = prepare_auth_request(con);
2020 	if (ret) {
2021 		if (ret != -EAGAIN)
2022 			pr_err("prepare_auth_request failed: %d\n", ret);
2023 		return ret;
2024 	}
2025 
2026 	con->state = CEPH_CON_S_V2_AUTH;
2027 	return 0;
2028 
2029 bad:
2030 	pr_err("failed to decode hello\n");
2031 	return -EINVAL;
2032 }
2033 
2034 static int process_auth_bad_method(struct ceph_connection *con,
2035 				   void *p, void *end)
2036 {
2037 	int allowed_protos[8], allowed_modes[8];
2038 	int allowed_proto_cnt, allowed_mode_cnt;
2039 	int used_proto, result;
2040 	int ret;
2041 	int i;
2042 
2043 	if (con->state != CEPH_CON_S_V2_AUTH) {
2044 		con->error_msg = "protocol error, unexpected auth_bad_method";
2045 		return -EINVAL;
2046 	}
2047 
2048 	ceph_decode_32_safe(&p, end, used_proto, bad);
2049 	ceph_decode_32_safe(&p, end, result, bad);
2050 	dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto,
2051 	     result);
2052 
2053 	ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad);
2054 	if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) {
2055 		pr_err("allowed_protos too big %d\n", allowed_proto_cnt);
2056 		return -EINVAL;
2057 	}
2058 	for (i = 0; i < allowed_proto_cnt; i++) {
2059 		ceph_decode_32_safe(&p, end, allowed_protos[i], bad);
2060 		dout("%s con %p allowed_protos[%d] %d\n", __func__, con,
2061 		     i, allowed_protos[i]);
2062 	}
2063 
2064 	ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad);
2065 	if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) {
2066 		pr_err("allowed_modes too big %d\n", allowed_mode_cnt);
2067 		return -EINVAL;
2068 	}
2069 	for (i = 0; i < allowed_mode_cnt; i++) {
2070 		ceph_decode_32_safe(&p, end, allowed_modes[i], bad);
2071 		dout("%s con %p allowed_modes[%d] %d\n", __func__, con,
2072 		     i, allowed_modes[i]);
2073 	}
2074 
2075 	mutex_unlock(&con->mutex);
2076 	ret = con->ops->handle_auth_bad_method(con, used_proto, result,
2077 					       allowed_protos,
2078 					       allowed_proto_cnt,
2079 					       allowed_modes,
2080 					       allowed_mode_cnt);
2081 	mutex_lock(&con->mutex);
2082 	if (con->state != CEPH_CON_S_V2_AUTH) {
2083 		dout("%s con %p state changed to %d\n", __func__, con,
2084 		     con->state);
2085 		return -EAGAIN;
2086 	}
2087 
2088 	dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret);
2089 	return ret;
2090 
2091 bad:
2092 	pr_err("failed to decode auth_bad_method\n");
2093 	return -EINVAL;
2094 }
2095 
2096 static int process_auth_reply_more(struct ceph_connection *con,
2097 				   void *p, void *end)
2098 {
2099 	int payload_len;
2100 	int ret;
2101 
2102 	if (con->state != CEPH_CON_S_V2_AUTH) {
2103 		con->error_msg = "protocol error, unexpected auth_reply_more";
2104 		return -EINVAL;
2105 	}
2106 
2107 	ceph_decode_32_safe(&p, end, payload_len, bad);
2108 	ceph_decode_need(&p, end, payload_len, bad);
2109 
2110 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2111 
2112 	reset_out_kvecs(con);
2113 	ret = prepare_auth_request_more(con, p, payload_len);
2114 	if (ret) {
2115 		if (ret != -EAGAIN)
2116 			pr_err("prepare_auth_request_more failed: %d\n", ret);
2117 		return ret;
2118 	}
2119 
2120 	return 0;
2121 
2122 bad:
2123 	pr_err("failed to decode auth_reply_more\n");
2124 	return -EINVAL;
2125 }
2126 
2127 /*
2128  * Align session_key and con_secret to avoid GFP_ATOMIC allocation
2129  * inside crypto_shash_setkey() and crypto_aead_setkey() called from
2130  * setup_crypto().  __aligned(16) isn't guaranteed to work for stack
2131  * objects, so do it by hand.
2132  */
2133 static int process_auth_done(struct ceph_connection *con, void *p, void *end)
2134 {
2135 	u8 session_key_buf[CEPH_KEY_LEN + 16];
2136 	u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16];
2137 	u8 *session_key = PTR_ALIGN(&session_key_buf[0], 16);
2138 	u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16);
2139 	int session_key_len, con_secret_len;
2140 	int payload_len;
2141 	u64 global_id;
2142 	int ret;
2143 
2144 	if (con->state != CEPH_CON_S_V2_AUTH) {
2145 		con->error_msg = "protocol error, unexpected auth_done";
2146 		return -EINVAL;
2147 	}
2148 
2149 	ceph_decode_64_safe(&p, end, global_id, bad);
2150 	ceph_decode_32_safe(&p, end, con->v2.con_mode, bad);
2151 	ceph_decode_32_safe(&p, end, payload_len, bad);
2152 
2153 	dout("%s con %p global_id %llu con_mode %d payload_len %d\n",
2154 	     __func__, con, global_id, con->v2.con_mode, payload_len);
2155 
2156 	mutex_unlock(&con->mutex);
2157 	session_key_len = 0;
2158 	con_secret_len = 0;
2159 	ret = con->ops->handle_auth_done(con, global_id, p, payload_len,
2160 					 session_key, &session_key_len,
2161 					 con_secret, &con_secret_len);
2162 	mutex_lock(&con->mutex);
2163 	if (con->state != CEPH_CON_S_V2_AUTH) {
2164 		dout("%s con %p state changed to %d\n", __func__, con,
2165 		     con->state);
2166 		ret = -EAGAIN;
2167 		goto out;
2168 	}
2169 
2170 	dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret);
2171 	if (ret)
2172 		goto out;
2173 
2174 	ret = setup_crypto(con, session_key, session_key_len, con_secret,
2175 			   con_secret_len);
2176 	if (ret)
2177 		goto out;
2178 
2179 	reset_out_kvecs(con);
2180 	ret = prepare_auth_signature(con);
2181 	if (ret) {
2182 		pr_err("prepare_auth_signature failed: %d\n", ret);
2183 		goto out;
2184 	}
2185 
2186 	con->state = CEPH_CON_S_V2_AUTH_SIGNATURE;
2187 
2188 out:
2189 	memzero_explicit(session_key_buf, sizeof(session_key_buf));
2190 	memzero_explicit(con_secret_buf, sizeof(con_secret_buf));
2191 	return ret;
2192 
2193 bad:
2194 	pr_err("failed to decode auth_done\n");
2195 	return -EINVAL;
2196 }
2197 
2198 static int process_auth_signature(struct ceph_connection *con,
2199 				  void *p, void *end)
2200 {
2201 	u8 hmac[SHA256_DIGEST_SIZE];
2202 	int ret;
2203 
2204 	if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) {
2205 		con->error_msg = "protocol error, unexpected auth_signature";
2206 		return -EINVAL;
2207 	}
2208 
2209 	ret = hmac_sha256(con, con->v2.out_sign_kvecs,
2210 			  con->v2.out_sign_kvec_cnt, hmac);
2211 	if (ret)
2212 		return ret;
2213 
2214 	ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad);
2215 	if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) {
2216 		con->error_msg = "integrity error, bad auth signature";
2217 		return -EBADMSG;
2218 	}
2219 
2220 	dout("%s con %p auth signature ok\n", __func__, con);
2221 
2222 	/* no reset_out_kvecs() as our auth_signature may still be pending */
2223 	if (!con->v2.server_cookie) {
2224 		ret = prepare_client_ident(con);
2225 		if (ret) {
2226 			pr_err("prepare_client_ident failed: %d\n", ret);
2227 			return ret;
2228 		}
2229 
2230 		con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2231 	} else {
2232 		ret = prepare_session_reconnect(con);
2233 		if (ret) {
2234 			pr_err("prepare_session_reconnect failed: %d\n", ret);
2235 			return ret;
2236 		}
2237 
2238 		con->state = CEPH_CON_S_V2_SESSION_RECONNECT;
2239 	}
2240 
2241 	return 0;
2242 
2243 bad:
2244 	pr_err("failed to decode auth_signature\n");
2245 	return -EINVAL;
2246 }
2247 
2248 static int process_server_ident(struct ceph_connection *con,
2249 				void *p, void *end)
2250 {
2251 	struct ceph_client *client = from_msgr(con->msgr);
2252 	u64 features, required_features;
2253 	struct ceph_entity_addr addr;
2254 	u64 global_seq;
2255 	u64 global_id;
2256 	u64 cookie;
2257 	u64 flags;
2258 	int ret;
2259 
2260 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2261 		con->error_msg = "protocol error, unexpected server_ident";
2262 		return -EINVAL;
2263 	}
2264 
2265 	ret = ceph_decode_entity_addrvec(&p, end, true, &addr);
2266 	if (ret) {
2267 		pr_err("failed to decode server addrs: %d\n", ret);
2268 		return ret;
2269 	}
2270 
2271 	ceph_decode_64_safe(&p, end, global_id, bad);
2272 	ceph_decode_64_safe(&p, end, global_seq, bad);
2273 	ceph_decode_64_safe(&p, end, features, bad);
2274 	ceph_decode_64_safe(&p, end, required_features, bad);
2275 	ceph_decode_64_safe(&p, end, flags, bad);
2276 	ceph_decode_64_safe(&p, end, cookie, bad);
2277 
2278 	dout("%s con %p addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx flags 0x%llx cookie 0x%llx\n",
2279 	     __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce),
2280 	     global_id, global_seq, features, required_features, flags, cookie);
2281 
2282 	/* is this who we intended to talk to? */
2283 	if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
2284 		pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
2285 		       ceph_pr_addr(&con->peer_addr),
2286 		       le32_to_cpu(con->peer_addr.nonce),
2287 		       ceph_pr_addr(&addr), le32_to_cpu(addr.nonce));
2288 		con->error_msg = "wrong peer at address";
2289 		return -EINVAL;
2290 	}
2291 
2292 	if (client->required_features & ~features) {
2293 		pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2294 		       features, client->required_features & ~features);
2295 		con->error_msg = "missing required protocol features";
2296 		return -EINVAL;
2297 	}
2298 
2299 	/*
2300 	 * Both name->type and name->num are set in ceph_con_open() but
2301 	 * name->num may be bogus in the initial monmap.  name->type is
2302 	 * verified in handle_hello().
2303 	 */
2304 	WARN_ON(!con->peer_name.type);
2305 	con->peer_name.num = cpu_to_le64(global_id);
2306 	con->v2.peer_global_seq = global_seq;
2307 	con->peer_features = features;
2308 	WARN_ON(required_features & ~client->supported_features);
2309 	con->v2.server_cookie = cookie;
2310 
2311 	if (flags & CEPH_MSG_CONNECT_LOSSY) {
2312 		ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
2313 		WARN_ON(con->v2.server_cookie);
2314 	} else {
2315 		WARN_ON(!con->v2.server_cookie);
2316 	}
2317 
2318 	clear_in_sign_kvecs(con);
2319 	clear_out_sign_kvecs(con);
2320 	free_conn_bufs(con);
2321 	con->delay = 0;  /* reset backoff memory */
2322 
2323 	con->state = CEPH_CON_S_OPEN;
2324 	con->v2.out_state = OUT_S_GET_NEXT;
2325 	return 0;
2326 
2327 bad:
2328 	pr_err("failed to decode server_ident\n");
2329 	return -EINVAL;
2330 }
2331 
2332 static int process_ident_missing_features(struct ceph_connection *con,
2333 					  void *p, void *end)
2334 {
2335 	struct ceph_client *client = from_msgr(con->msgr);
2336 	u64 missing_features;
2337 
2338 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2339 		con->error_msg = "protocol error, unexpected ident_missing_features";
2340 		return -EINVAL;
2341 	}
2342 
2343 	ceph_decode_64_safe(&p, end, missing_features, bad);
2344 	pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2345 	       client->supported_features, missing_features);
2346 	con->error_msg = "missing required protocol features";
2347 	return -EINVAL;
2348 
2349 bad:
2350 	pr_err("failed to decode ident_missing_features\n");
2351 	return -EINVAL;
2352 }
2353 
2354 static int process_session_reconnect_ok(struct ceph_connection *con,
2355 					void *p, void *end)
2356 {
2357 	u64 seq;
2358 
2359 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2360 		con->error_msg = "protocol error, unexpected session_reconnect_ok";
2361 		return -EINVAL;
2362 	}
2363 
2364 	ceph_decode_64_safe(&p, end, seq, bad);
2365 
2366 	dout("%s con %p seq %llu\n", __func__, con, seq);
2367 	ceph_con_discard_requeued(con, seq);
2368 
2369 	clear_in_sign_kvecs(con);
2370 	clear_out_sign_kvecs(con);
2371 	free_conn_bufs(con);
2372 	con->delay = 0;  /* reset backoff memory */
2373 
2374 	con->state = CEPH_CON_S_OPEN;
2375 	con->v2.out_state = OUT_S_GET_NEXT;
2376 	return 0;
2377 
2378 bad:
2379 	pr_err("failed to decode session_reconnect_ok\n");
2380 	return -EINVAL;
2381 }
2382 
2383 static int process_session_retry(struct ceph_connection *con,
2384 				 void *p, void *end)
2385 {
2386 	u64 connect_seq;
2387 	int ret;
2388 
2389 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2390 		con->error_msg = "protocol error, unexpected session_retry";
2391 		return -EINVAL;
2392 	}
2393 
2394 	ceph_decode_64_safe(&p, end, connect_seq, bad);
2395 
2396 	dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq);
2397 	WARN_ON(connect_seq <= con->v2.connect_seq);
2398 	con->v2.connect_seq = connect_seq + 1;
2399 
2400 	free_conn_bufs(con);
2401 
2402 	reset_out_kvecs(con);
2403 	ret = prepare_session_reconnect(con);
2404 	if (ret) {
2405 		pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret);
2406 		return ret;
2407 	}
2408 
2409 	return 0;
2410 
2411 bad:
2412 	pr_err("failed to decode session_retry\n");
2413 	return -EINVAL;
2414 }
2415 
2416 static int process_session_retry_global(struct ceph_connection *con,
2417 					void *p, void *end)
2418 {
2419 	u64 global_seq;
2420 	int ret;
2421 
2422 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2423 		con->error_msg = "protocol error, unexpected session_retry_global";
2424 		return -EINVAL;
2425 	}
2426 
2427 	ceph_decode_64_safe(&p, end, global_seq, bad);
2428 
2429 	dout("%s con %p global_seq %llu\n", __func__, con, global_seq);
2430 	WARN_ON(global_seq <= con->v2.global_seq);
2431 	con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq);
2432 
2433 	free_conn_bufs(con);
2434 
2435 	reset_out_kvecs(con);
2436 	ret = prepare_session_reconnect(con);
2437 	if (ret) {
2438 		pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret);
2439 		return ret;
2440 	}
2441 
2442 	return 0;
2443 
2444 bad:
2445 	pr_err("failed to decode session_retry_global\n");
2446 	return -EINVAL;
2447 }
2448 
2449 static int process_session_reset(struct ceph_connection *con,
2450 				 void *p, void *end)
2451 {
2452 	bool full;
2453 	int ret;
2454 
2455 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2456 		con->error_msg = "protocol error, unexpected session_reset";
2457 		return -EINVAL;
2458 	}
2459 
2460 	ceph_decode_8_safe(&p, end, full, bad);
2461 	if (!full) {
2462 		con->error_msg = "protocol error, bad session_reset";
2463 		return -EINVAL;
2464 	}
2465 
2466 	pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name),
2467 		ceph_pr_addr(&con->peer_addr));
2468 	ceph_con_reset_session(con);
2469 
2470 	mutex_unlock(&con->mutex);
2471 	if (con->ops->peer_reset)
2472 		con->ops->peer_reset(con);
2473 	mutex_lock(&con->mutex);
2474 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2475 		dout("%s con %p state changed to %d\n", __func__, con,
2476 		     con->state);
2477 		return -EAGAIN;
2478 	}
2479 
2480 	free_conn_bufs(con);
2481 
2482 	reset_out_kvecs(con);
2483 	ret = prepare_client_ident(con);
2484 	if (ret) {
2485 		pr_err("prepare_client_ident (rst) failed: %d\n", ret);
2486 		return ret;
2487 	}
2488 
2489 	con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2490 	return 0;
2491 
2492 bad:
2493 	pr_err("failed to decode session_reset\n");
2494 	return -EINVAL;
2495 }
2496 
2497 static int process_keepalive2_ack(struct ceph_connection *con,
2498 				  void *p, void *end)
2499 {
2500 	if (con->state != CEPH_CON_S_OPEN) {
2501 		con->error_msg = "protocol error, unexpected keepalive2_ack";
2502 		return -EINVAL;
2503 	}
2504 
2505 	ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad);
2506 	ceph_decode_timespec64(&con->last_keepalive_ack, p);
2507 
2508 	dout("%s con %p timestamp %lld.%09ld\n", __func__, con,
2509 	     con->last_keepalive_ack.tv_sec, con->last_keepalive_ack.tv_nsec);
2510 
2511 	return 0;
2512 
2513 bad:
2514 	pr_err("failed to decode keepalive2_ack\n");
2515 	return -EINVAL;
2516 }
2517 
2518 static int process_ack(struct ceph_connection *con, void *p, void *end)
2519 {
2520 	u64 seq;
2521 
2522 	if (con->state != CEPH_CON_S_OPEN) {
2523 		con->error_msg = "protocol error, unexpected ack";
2524 		return -EINVAL;
2525 	}
2526 
2527 	ceph_decode_64_safe(&p, end, seq, bad);
2528 
2529 	dout("%s con %p seq %llu\n", __func__, con, seq);
2530 	ceph_con_discard_sent(con, seq);
2531 	return 0;
2532 
2533 bad:
2534 	pr_err("failed to decode ack\n");
2535 	return -EINVAL;
2536 }
2537 
2538 static int process_control(struct ceph_connection *con, void *p, void *end)
2539 {
2540 	int tag = con->v2.in_desc.fd_tag;
2541 	int ret;
2542 
2543 	dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p));
2544 
2545 	switch (tag) {
2546 	case FRAME_TAG_HELLO:
2547 		ret = process_hello(con, p, end);
2548 		break;
2549 	case FRAME_TAG_AUTH_BAD_METHOD:
2550 		ret = process_auth_bad_method(con, p, end);
2551 		break;
2552 	case FRAME_TAG_AUTH_REPLY_MORE:
2553 		ret = process_auth_reply_more(con, p, end);
2554 		break;
2555 	case FRAME_TAG_AUTH_DONE:
2556 		ret = process_auth_done(con, p, end);
2557 		break;
2558 	case FRAME_TAG_AUTH_SIGNATURE:
2559 		ret = process_auth_signature(con, p, end);
2560 		break;
2561 	case FRAME_TAG_SERVER_IDENT:
2562 		ret = process_server_ident(con, p, end);
2563 		break;
2564 	case FRAME_TAG_IDENT_MISSING_FEATURES:
2565 		ret = process_ident_missing_features(con, p, end);
2566 		break;
2567 	case FRAME_TAG_SESSION_RECONNECT_OK:
2568 		ret = process_session_reconnect_ok(con, p, end);
2569 		break;
2570 	case FRAME_TAG_SESSION_RETRY:
2571 		ret = process_session_retry(con, p, end);
2572 		break;
2573 	case FRAME_TAG_SESSION_RETRY_GLOBAL:
2574 		ret = process_session_retry_global(con, p, end);
2575 		break;
2576 	case FRAME_TAG_SESSION_RESET:
2577 		ret = process_session_reset(con, p, end);
2578 		break;
2579 	case FRAME_TAG_KEEPALIVE2_ACK:
2580 		ret = process_keepalive2_ack(con, p, end);
2581 		break;
2582 	case FRAME_TAG_ACK:
2583 		ret = process_ack(con, p, end);
2584 		break;
2585 	default:
2586 		pr_err("bad tag %d\n", tag);
2587 		con->error_msg = "protocol error, bad tag";
2588 		return -EINVAL;
2589 	}
2590 	if (ret) {
2591 		dout("%s con %p error %d\n", __func__, con, ret);
2592 		return ret;
2593 	}
2594 
2595 	prepare_read_preamble(con);
2596 	return 0;
2597 }
2598 
2599 /*
2600  * Return:
2601  *   1 - con->in_msg set, read message
2602  *   0 - skip message
2603  *  <0 - error
2604  */
2605 static int process_message_header(struct ceph_connection *con,
2606 				  void *p, void *end)
2607 {
2608 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2609 	struct ceph_msg_header2 *hdr2 = p;
2610 	struct ceph_msg_header hdr;
2611 	int skip;
2612 	int ret;
2613 	u64 seq;
2614 
2615 	/* verify seq# */
2616 	seq = le64_to_cpu(hdr2->seq);
2617 	if ((s64)seq - (s64)con->in_seq < 1) {
2618 		pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n",
2619 			ENTITY_NAME(con->peer_name),
2620 			ceph_pr_addr(&con->peer_addr),
2621 			seq, con->in_seq + 1);
2622 		return 0;
2623 	}
2624 	if ((s64)seq - (s64)con->in_seq > 1) {
2625 		pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1);
2626 		con->error_msg = "bad message sequence # for incoming message";
2627 		return -EBADE;
2628 	}
2629 
2630 	ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq));
2631 
2632 	fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2],
2633 		    desc->fd_lens[3], &con->peer_name);
2634 	ret = ceph_con_in_msg_alloc(con, &hdr, &skip);
2635 	if (ret)
2636 		return ret;
2637 
2638 	WARN_ON(!con->in_msg ^ skip);
2639 	if (skip)
2640 		return 0;
2641 
2642 	WARN_ON(!con->in_msg);
2643 	WARN_ON(con->in_msg->con != con);
2644 	return 1;
2645 }
2646 
2647 static int process_message(struct ceph_connection *con)
2648 {
2649 	ceph_con_process_message(con);
2650 
2651 	/*
2652 	 * We could have been closed by ceph_con_close() because
2653 	 * ceph_con_process_message() temporarily drops con->mutex.
2654 	 */
2655 	if (con->state != CEPH_CON_S_OPEN) {
2656 		dout("%s con %p state changed to %d\n", __func__, con,
2657 		     con->state);
2658 		return -EAGAIN;
2659 	}
2660 
2661 	prepare_read_preamble(con);
2662 	return 0;
2663 }
2664 
2665 static int __handle_control(struct ceph_connection *con, void *p)
2666 {
2667 	void *end = p + con->v2.in_desc.fd_lens[0];
2668 	struct ceph_msg *msg;
2669 	int ret;
2670 
2671 	if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE)
2672 		return process_control(con, p, end);
2673 
2674 	ret = process_message_header(con, p, end);
2675 	if (ret < 0)
2676 		return ret;
2677 	if (ret == 0) {
2678 		prepare_skip_message(con);
2679 		return 0;
2680 	}
2681 
2682 	msg = con->in_msg;  /* set in process_message_header() */
2683 	if (front_len(msg)) {
2684 		WARN_ON(front_len(msg) > msg->front_alloc_len);
2685 		msg->front.iov_len = front_len(msg);
2686 	} else {
2687 		msg->front.iov_len = 0;
2688 	}
2689 	if (middle_len(msg)) {
2690 		WARN_ON(middle_len(msg) > msg->middle->alloc_len);
2691 		msg->middle->vec.iov_len = middle_len(msg);
2692 	} else if (msg->middle) {
2693 		msg->middle->vec.iov_len = 0;
2694 	}
2695 
2696 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
2697 		return process_message(con);
2698 
2699 	if (con_secure(con))
2700 		return prepare_read_tail_secure(con);
2701 
2702 	prepare_read_tail_plain(con);
2703 	return 0;
2704 }
2705 
2706 static int handle_preamble(struct ceph_connection *con)
2707 {
2708 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2709 	int ret;
2710 
2711 	if (con_secure(con)) {
2712 		ret = decrypt_preamble(con);
2713 		if (ret) {
2714 			if (ret == -EBADMSG)
2715 				con->error_msg = "integrity error, bad preamble auth tag";
2716 			return ret;
2717 		}
2718 	}
2719 
2720 	ret = decode_preamble(con->v2.in_buf, desc);
2721 	if (ret) {
2722 		if (ret == -EBADMSG)
2723 			con->error_msg = "integrity error, bad crc";
2724 		else
2725 			con->error_msg = "protocol error, bad preamble";
2726 		return ret;
2727 	}
2728 
2729 	dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__,
2730 	     con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0],
2731 	     desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]);
2732 
2733 	if (!con_secure(con))
2734 		return prepare_read_control(con);
2735 
2736 	if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN)
2737 		return prepare_read_control_remainder(con);
2738 
2739 	return __handle_control(con, CTRL_BODY(con->v2.in_buf));
2740 }
2741 
2742 static int handle_control(struct ceph_connection *con)
2743 {
2744 	int ctrl_len = con->v2.in_desc.fd_lens[0];
2745 	void *buf;
2746 	int ret;
2747 
2748 	WARN_ON(con_secure(con));
2749 
2750 	ret = verify_control_crc(con);
2751 	if (ret) {
2752 		con->error_msg = "integrity error, bad crc";
2753 		return ret;
2754 	}
2755 
2756 	if (con->state == CEPH_CON_S_V2_AUTH) {
2757 		buf = alloc_conn_buf(con, ctrl_len);
2758 		if (!buf)
2759 			return -ENOMEM;
2760 
2761 		memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len);
2762 		return __handle_control(con, buf);
2763 	}
2764 
2765 	return __handle_control(con, con->v2.in_kvecs[0].iov_base);
2766 }
2767 
2768 static int handle_control_remainder(struct ceph_connection *con)
2769 {
2770 	int ret;
2771 
2772 	WARN_ON(!con_secure(con));
2773 
2774 	ret = decrypt_control_remainder(con);
2775 	if (ret) {
2776 		if (ret == -EBADMSG)
2777 			con->error_msg = "integrity error, bad control remainder auth tag";
2778 		return ret;
2779 	}
2780 
2781 	return __handle_control(con, con->v2.in_kvecs[0].iov_base -
2782 				     CEPH_PREAMBLE_INLINE_LEN);
2783 }
2784 
2785 static int handle_epilogue(struct ceph_connection *con)
2786 {
2787 	u32 front_crc, middle_crc, data_crc;
2788 	int ret;
2789 
2790 	if (con_secure(con)) {
2791 		ret = decrypt_tail(con);
2792 		if (ret) {
2793 			if (ret == -EBADMSG)
2794 				con->error_msg = "integrity error, bad epilogue auth tag";
2795 			return ret;
2796 		}
2797 
2798 		/* just late_status */
2799 		ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL);
2800 		if (ret) {
2801 			con->error_msg = "protocol error, bad epilogue";
2802 			return ret;
2803 		}
2804 	} else {
2805 		ret = decode_epilogue(con->v2.in_buf, &front_crc,
2806 				      &middle_crc, &data_crc);
2807 		if (ret) {
2808 			con->error_msg = "protocol error, bad epilogue";
2809 			return ret;
2810 		}
2811 
2812 		ret = verify_epilogue_crcs(con, front_crc, middle_crc,
2813 					   data_crc);
2814 		if (ret) {
2815 			con->error_msg = "integrity error, bad crc";
2816 			return ret;
2817 		}
2818 	}
2819 
2820 	return process_message(con);
2821 }
2822 
2823 static void finish_skip(struct ceph_connection *con)
2824 {
2825 	dout("%s con %p\n", __func__, con);
2826 
2827 	if (con_secure(con))
2828 		gcm_inc_nonce(&con->v2.in_gcm_nonce);
2829 
2830 	__finish_skip(con);
2831 }
2832 
2833 static int populate_in_iter(struct ceph_connection *con)
2834 {
2835 	int ret;
2836 
2837 	dout("%s con %p state %d in_state %d\n", __func__, con, con->state,
2838 	     con->v2.in_state);
2839 	WARN_ON(iov_iter_count(&con->v2.in_iter));
2840 
2841 	if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) {
2842 		ret = process_banner_prefix(con);
2843 	} else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) {
2844 		ret = process_banner_payload(con);
2845 	} else if ((con->state >= CEPH_CON_S_V2_HELLO &&
2846 		    con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) ||
2847 		   con->state == CEPH_CON_S_OPEN) {
2848 		switch (con->v2.in_state) {
2849 		case IN_S_HANDLE_PREAMBLE:
2850 			ret = handle_preamble(con);
2851 			break;
2852 		case IN_S_HANDLE_CONTROL:
2853 			ret = handle_control(con);
2854 			break;
2855 		case IN_S_HANDLE_CONTROL_REMAINDER:
2856 			ret = handle_control_remainder(con);
2857 			break;
2858 		case IN_S_PREPARE_READ_DATA:
2859 			prepare_read_data(con);
2860 			ret = 0;
2861 			break;
2862 		case IN_S_PREPARE_READ_DATA_CONT:
2863 			prepare_read_data_cont(con);
2864 			ret = 0;
2865 			break;
2866 		case IN_S_PREPARE_READ_ENC_PAGE:
2867 			prepare_read_enc_page(con);
2868 			ret = 0;
2869 			break;
2870 		case IN_S_HANDLE_EPILOGUE:
2871 			ret = handle_epilogue(con);
2872 			break;
2873 		case IN_S_FINISH_SKIP:
2874 			finish_skip(con);
2875 			ret = 0;
2876 			break;
2877 		default:
2878 			WARN(1, "bad in_state %d", con->v2.in_state);
2879 			return -EINVAL;
2880 		}
2881 	} else {
2882 		WARN(1, "bad state %d", con->state);
2883 		return -EINVAL;
2884 	}
2885 	if (ret) {
2886 		dout("%s con %p error %d\n", __func__, con, ret);
2887 		return ret;
2888 	}
2889 
2890 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
2891 		return -ENODATA;
2892 	dout("%s con %p populated %zu\n", __func__, con,
2893 	     iov_iter_count(&con->v2.in_iter));
2894 	return 1;
2895 }
2896 
2897 int ceph_con_v2_try_read(struct ceph_connection *con)
2898 {
2899 	int ret;
2900 
2901 	dout("%s con %p state %d need %zu\n", __func__, con, con->state,
2902 	     iov_iter_count(&con->v2.in_iter));
2903 
2904 	if (con->state == CEPH_CON_S_PREOPEN)
2905 		return 0;
2906 
2907 	/*
2908 	 * We should always have something pending here.  If not,
2909 	 * avoid calling populate_in_iter() as if we read something
2910 	 * (ceph_tcp_recv() would immediately return 1).
2911 	 */
2912 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
2913 		return -ENODATA;
2914 
2915 	for (;;) {
2916 		ret = ceph_tcp_recv(con);
2917 		if (ret <= 0)
2918 			return ret;
2919 
2920 		ret = populate_in_iter(con);
2921 		if (ret <= 0) {
2922 			if (ret && ret != -EAGAIN && !con->error_msg)
2923 				con->error_msg = "read processing error";
2924 			return ret;
2925 		}
2926 	}
2927 }
2928 
2929 static void queue_data(struct ceph_connection *con)
2930 {
2931 	struct bio_vec bv;
2932 
2933 	con->v2.out_epil.data_crc = -1;
2934 	ceph_msg_data_cursor_init(&con->v2.out_cursor, con->out_msg,
2935 				  data_len(con->out_msg));
2936 
2937 	get_bvec_at(&con->v2.out_cursor, &bv);
2938 	set_out_bvec(con, &bv, true);
2939 	con->v2.out_state = OUT_S_QUEUE_DATA_CONT;
2940 }
2941 
2942 static void queue_data_cont(struct ceph_connection *con)
2943 {
2944 	struct bio_vec bv;
2945 
2946 	con->v2.out_epil.data_crc = ceph_crc32c_page(
2947 		con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
2948 		con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len);
2949 
2950 	ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len);
2951 	if (con->v2.out_cursor.total_resid) {
2952 		get_bvec_at(&con->v2.out_cursor, &bv);
2953 		set_out_bvec(con, &bv, true);
2954 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT);
2955 		return;
2956 	}
2957 
2958 	/*
2959 	 * We've written all data.  Queue epilogue.  Once it's written,
2960 	 * we are done.
2961 	 */
2962 	reset_out_kvecs(con);
2963 	prepare_epilogue_plain(con, false);
2964 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
2965 }
2966 
2967 static void queue_enc_page(struct ceph_connection *con)
2968 {
2969 	struct bio_vec bv;
2970 
2971 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i,
2972 	     con->v2.out_enc_resid);
2973 	WARN_ON(!con->v2.out_enc_resid);
2974 
2975 	bv.bv_page = con->v2.out_enc_pages[con->v2.out_enc_i];
2976 	bv.bv_offset = 0;
2977 	bv.bv_len = min(con->v2.out_enc_resid, (int)PAGE_SIZE);
2978 
2979 	set_out_bvec(con, &bv, false);
2980 	con->v2.out_enc_i++;
2981 	con->v2.out_enc_resid -= bv.bv_len;
2982 
2983 	if (con->v2.out_enc_resid) {
2984 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE);
2985 		return;
2986 	}
2987 
2988 	/*
2989 	 * We've queued the last piece of ciphertext (ending with
2990 	 * epilogue) + auth tag.  Once it's written, we are done.
2991 	 */
2992 	WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt);
2993 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
2994 }
2995 
2996 static void queue_zeros(struct ceph_connection *con)
2997 {
2998 	dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero);
2999 
3000 	if (con->v2.out_zero) {
3001 		set_out_bvec_zero(con);
3002 		con->v2.out_zero -= con->v2.out_bvec.bv_len;
3003 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3004 		return;
3005 	}
3006 
3007 	/*
3008 	 * We've zero-filled everything up to epilogue.  Queue epilogue
3009 	 * with late_status set to ABORTED and crcs adjusted for zeros.
3010 	 * Once it's written, we are done patching up for the revoke.
3011 	 */
3012 	reset_out_kvecs(con);
3013 	prepare_epilogue_plain(con, true);
3014 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3015 }
3016 
3017 static void finish_message(struct ceph_connection *con)
3018 {
3019 	dout("%s con %p msg %p\n", __func__, con, con->out_msg);
3020 
3021 	/* we end up here both plain and secure modes */
3022 	if (con->v2.out_enc_pages) {
3023 		WARN_ON(!con->v2.out_enc_page_cnt);
3024 		ceph_release_page_vector(con->v2.out_enc_pages,
3025 					 con->v2.out_enc_page_cnt);
3026 		con->v2.out_enc_pages = NULL;
3027 		con->v2.out_enc_page_cnt = 0;
3028 	}
3029 	/* message may have been revoked */
3030 	if (con->out_msg) {
3031 		ceph_msg_put(con->out_msg);
3032 		con->out_msg = NULL;
3033 	}
3034 
3035 	con->v2.out_state = OUT_S_GET_NEXT;
3036 }
3037 
3038 static int populate_out_iter(struct ceph_connection *con)
3039 {
3040 	int ret;
3041 
3042 	dout("%s con %p state %d out_state %d\n", __func__, con, con->state,
3043 	     con->v2.out_state);
3044 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3045 
3046 	if (con->state != CEPH_CON_S_OPEN) {
3047 		WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX ||
3048 			con->state > CEPH_CON_S_V2_SESSION_RECONNECT);
3049 		goto nothing_pending;
3050 	}
3051 
3052 	switch (con->v2.out_state) {
3053 	case OUT_S_QUEUE_DATA:
3054 		WARN_ON(!con->out_msg);
3055 		queue_data(con);
3056 		goto populated;
3057 	case OUT_S_QUEUE_DATA_CONT:
3058 		WARN_ON(!con->out_msg);
3059 		queue_data_cont(con);
3060 		goto populated;
3061 	case OUT_S_QUEUE_ENC_PAGE:
3062 		queue_enc_page(con);
3063 		goto populated;
3064 	case OUT_S_QUEUE_ZEROS:
3065 		WARN_ON(con->out_msg);  /* revoked */
3066 		queue_zeros(con);
3067 		goto populated;
3068 	case OUT_S_FINISH_MESSAGE:
3069 		finish_message(con);
3070 		break;
3071 	case OUT_S_GET_NEXT:
3072 		break;
3073 	default:
3074 		WARN(1, "bad out_state %d", con->v2.out_state);
3075 		return -EINVAL;
3076 	}
3077 
3078 	WARN_ON(con->v2.out_state != OUT_S_GET_NEXT);
3079 	if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
3080 		ret = prepare_keepalive2(con);
3081 		if (ret) {
3082 			pr_err("prepare_keepalive2 failed: %d\n", ret);
3083 			return ret;
3084 		}
3085 	} else if (!list_empty(&con->out_queue)) {
3086 		ceph_con_get_out_msg(con);
3087 		ret = prepare_message(con);
3088 		if (ret) {
3089 			pr_err("prepare_message failed: %d\n", ret);
3090 			return ret;
3091 		}
3092 	} else if (con->in_seq > con->in_seq_acked) {
3093 		ret = prepare_ack(con);
3094 		if (ret) {
3095 			pr_err("prepare_ack failed: %d\n", ret);
3096 			return ret;
3097 		}
3098 	} else {
3099 		goto nothing_pending;
3100 	}
3101 
3102 populated:
3103 	if (WARN_ON(!iov_iter_count(&con->v2.out_iter)))
3104 		return -ENODATA;
3105 	dout("%s con %p populated %zu\n", __func__, con,
3106 	     iov_iter_count(&con->v2.out_iter));
3107 	return 1;
3108 
3109 nothing_pending:
3110 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3111 	dout("%s con %p nothing pending\n", __func__, con);
3112 	ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
3113 	return 0;
3114 }
3115 
3116 int ceph_con_v2_try_write(struct ceph_connection *con)
3117 {
3118 	int ret;
3119 
3120 	dout("%s con %p state %d have %zu\n", __func__, con, con->state,
3121 	     iov_iter_count(&con->v2.out_iter));
3122 
3123 	/* open the socket first? */
3124 	if (con->state == CEPH_CON_S_PREOPEN) {
3125 		WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2);
3126 
3127 		/*
3128 		 * Always bump global_seq.  Bump connect_seq only if
3129 		 * there is a session (i.e. we are reconnecting and will
3130 		 * send session_reconnect instead of client_ident).
3131 		 */
3132 		con->v2.global_seq = ceph_get_global_seq(con->msgr, 0);
3133 		if (con->v2.server_cookie)
3134 			con->v2.connect_seq++;
3135 
3136 		ret = prepare_read_banner_prefix(con);
3137 		if (ret) {
3138 			pr_err("prepare_read_banner_prefix failed: %d\n", ret);
3139 			con->error_msg = "connect error";
3140 			return ret;
3141 		}
3142 
3143 		reset_out_kvecs(con);
3144 		ret = prepare_banner(con);
3145 		if (ret) {
3146 			pr_err("prepare_banner failed: %d\n", ret);
3147 			con->error_msg = "connect error";
3148 			return ret;
3149 		}
3150 
3151 		ret = ceph_tcp_connect(con);
3152 		if (ret) {
3153 			pr_err("ceph_tcp_connect failed: %d\n", ret);
3154 			con->error_msg = "connect error";
3155 			return ret;
3156 		}
3157 	}
3158 
3159 	if (!iov_iter_count(&con->v2.out_iter)) {
3160 		ret = populate_out_iter(con);
3161 		if (ret <= 0) {
3162 			if (ret && ret != -EAGAIN && !con->error_msg)
3163 				con->error_msg = "write processing error";
3164 			return ret;
3165 		}
3166 	}
3167 
3168 	tcp_sock_set_cork(con->sock->sk, true);
3169 	for (;;) {
3170 		ret = ceph_tcp_send(con);
3171 		if (ret <= 0)
3172 			break;
3173 
3174 		ret = populate_out_iter(con);
3175 		if (ret <= 0) {
3176 			if (ret && ret != -EAGAIN && !con->error_msg)
3177 				con->error_msg = "write processing error";
3178 			break;
3179 		}
3180 	}
3181 
3182 	tcp_sock_set_cork(con->sock->sk, false);
3183 	return ret;
3184 }
3185 
3186 static u32 crc32c_zeros(u32 crc, int zero_len)
3187 {
3188 	int len;
3189 
3190 	while (zero_len) {
3191 		len = min(zero_len, (int)PAGE_SIZE);
3192 		crc = crc32c(crc, page_address(ceph_zero_page), len);
3193 		zero_len -= len;
3194 	}
3195 
3196 	return crc;
3197 }
3198 
3199 static void prepare_zero_front(struct ceph_connection *con, int resid)
3200 {
3201 	int sent;
3202 
3203 	WARN_ON(!resid || resid > front_len(con->out_msg));
3204 	sent = front_len(con->out_msg) - resid;
3205 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3206 
3207 	if (sent) {
3208 		con->v2.out_epil.front_crc =
3209 			crc32c(-1, con->out_msg->front.iov_base, sent);
3210 		con->v2.out_epil.front_crc =
3211 			crc32c_zeros(con->v2.out_epil.front_crc, resid);
3212 	} else {
3213 		con->v2.out_epil.front_crc = crc32c_zeros(-1, resid);
3214 	}
3215 
3216 	con->v2.out_iter.count -= resid;
3217 	out_zero_add(con, resid);
3218 }
3219 
3220 static void prepare_zero_middle(struct ceph_connection *con, int resid)
3221 {
3222 	int sent;
3223 
3224 	WARN_ON(!resid || resid > middle_len(con->out_msg));
3225 	sent = middle_len(con->out_msg) - resid;
3226 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3227 
3228 	if (sent) {
3229 		con->v2.out_epil.middle_crc =
3230 			crc32c(-1, con->out_msg->middle->vec.iov_base, sent);
3231 		con->v2.out_epil.middle_crc =
3232 			crc32c_zeros(con->v2.out_epil.middle_crc, resid);
3233 	} else {
3234 		con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid);
3235 	}
3236 
3237 	con->v2.out_iter.count -= resid;
3238 	out_zero_add(con, resid);
3239 }
3240 
3241 static void prepare_zero_data(struct ceph_connection *con)
3242 {
3243 	dout("%s con %p\n", __func__, con);
3244 	con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(con->out_msg));
3245 	out_zero_add(con, data_len(con->out_msg));
3246 }
3247 
3248 static void revoke_at_queue_data(struct ceph_connection *con)
3249 {
3250 	int boundary;
3251 	int resid;
3252 
3253 	WARN_ON(!data_len(con->out_msg));
3254 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3255 	resid = iov_iter_count(&con->v2.out_iter);
3256 
3257 	boundary = front_len(con->out_msg) + middle_len(con->out_msg);
3258 	if (resid > boundary) {
3259 		resid -= boundary;
3260 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3261 		dout("%s con %p was sending head\n", __func__, con);
3262 		if (front_len(con->out_msg))
3263 			prepare_zero_front(con, front_len(con->out_msg));
3264 		if (middle_len(con->out_msg))
3265 			prepare_zero_middle(con, middle_len(con->out_msg));
3266 		prepare_zero_data(con);
3267 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3268 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3269 		return;
3270 	}
3271 
3272 	boundary = middle_len(con->out_msg);
3273 	if (resid > boundary) {
3274 		resid -= boundary;
3275 		dout("%s con %p was sending front\n", __func__, con);
3276 		prepare_zero_front(con, resid);
3277 		if (middle_len(con->out_msg))
3278 			prepare_zero_middle(con, middle_len(con->out_msg));
3279 		prepare_zero_data(con);
3280 		queue_zeros(con);
3281 		return;
3282 	}
3283 
3284 	WARN_ON(!resid);
3285 	dout("%s con %p was sending middle\n", __func__, con);
3286 	prepare_zero_middle(con, resid);
3287 	prepare_zero_data(con);
3288 	queue_zeros(con);
3289 }
3290 
3291 static void revoke_at_queue_data_cont(struct ceph_connection *con)
3292 {
3293 	int sent, resid;  /* current piece of data */
3294 
3295 	WARN_ON(!data_len(con->out_msg));
3296 	WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter));
3297 	resid = iov_iter_count(&con->v2.out_iter);
3298 	WARN_ON(!resid || resid > con->v2.out_bvec.bv_len);
3299 	sent = con->v2.out_bvec.bv_len - resid;
3300 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3301 
3302 	if (sent) {
3303 		con->v2.out_epil.data_crc = ceph_crc32c_page(
3304 			con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3305 			con->v2.out_bvec.bv_offset, sent);
3306 		ceph_msg_data_advance(&con->v2.out_cursor, sent);
3307 	}
3308 	WARN_ON(resid > con->v2.out_cursor.total_resid);
3309 	con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc,
3310 						con->v2.out_cursor.total_resid);
3311 
3312 	con->v2.out_iter.count -= resid;
3313 	out_zero_add(con, con->v2.out_cursor.total_resid);
3314 	queue_zeros(con);
3315 }
3316 
3317 static void revoke_at_finish_message(struct ceph_connection *con)
3318 {
3319 	int boundary;
3320 	int resid;
3321 
3322 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3323 	resid = iov_iter_count(&con->v2.out_iter);
3324 
3325 	if (!front_len(con->out_msg) && !middle_len(con->out_msg) &&
3326 	    !data_len(con->out_msg)) {
3327 		WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN);
3328 		dout("%s con %p was sending head (empty message) - noop\n",
3329 		     __func__, con);
3330 		return;
3331 	}
3332 
3333 	boundary = front_len(con->out_msg) + middle_len(con->out_msg) +
3334 		   CEPH_EPILOGUE_PLAIN_LEN;
3335 	if (resid > boundary) {
3336 		resid -= boundary;
3337 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3338 		dout("%s con %p was sending head\n", __func__, con);
3339 		if (front_len(con->out_msg))
3340 			prepare_zero_front(con, front_len(con->out_msg));
3341 		if (middle_len(con->out_msg))
3342 			prepare_zero_middle(con, middle_len(con->out_msg));
3343 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3344 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3345 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3346 		return;
3347 	}
3348 
3349 	boundary = middle_len(con->out_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3350 	if (resid > boundary) {
3351 		resid -= boundary;
3352 		dout("%s con %p was sending front\n", __func__, con);
3353 		prepare_zero_front(con, resid);
3354 		if (middle_len(con->out_msg))
3355 			prepare_zero_middle(con, middle_len(con->out_msg));
3356 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3357 		queue_zeros(con);
3358 		return;
3359 	}
3360 
3361 	boundary = CEPH_EPILOGUE_PLAIN_LEN;
3362 	if (resid > boundary) {
3363 		resid -= boundary;
3364 		dout("%s con %p was sending middle\n", __func__, con);
3365 		prepare_zero_middle(con, resid);
3366 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3367 		queue_zeros(con);
3368 		return;
3369 	}
3370 
3371 	WARN_ON(!resid);
3372 	dout("%s con %p was sending epilogue - noop\n", __func__, con);
3373 }
3374 
3375 void ceph_con_v2_revoke(struct ceph_connection *con)
3376 {
3377 	WARN_ON(con->v2.out_zero);
3378 
3379 	if (con_secure(con)) {
3380 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE &&
3381 			con->v2.out_state != OUT_S_FINISH_MESSAGE);
3382 		dout("%s con %p secure - noop\n", __func__, con);
3383 		return;
3384 	}
3385 
3386 	switch (con->v2.out_state) {
3387 	case OUT_S_QUEUE_DATA:
3388 		revoke_at_queue_data(con);
3389 		break;
3390 	case OUT_S_QUEUE_DATA_CONT:
3391 		revoke_at_queue_data_cont(con);
3392 		break;
3393 	case OUT_S_FINISH_MESSAGE:
3394 		revoke_at_finish_message(con);
3395 		break;
3396 	default:
3397 		WARN(1, "bad out_state %d", con->v2.out_state);
3398 		break;
3399 	}
3400 }
3401 
3402 static void revoke_at_prepare_read_data(struct ceph_connection *con)
3403 {
3404 	int remaining;
3405 	int resid;
3406 
3407 	WARN_ON(con_secure(con));
3408 	WARN_ON(!data_len(con->in_msg));
3409 	WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
3410 	resid = iov_iter_count(&con->v2.in_iter);
3411 	WARN_ON(!resid);
3412 
3413 	remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3414 	dout("%s con %p resid %d remaining %d\n", __func__, con, resid,
3415 	     remaining);
3416 	con->v2.in_iter.count -= resid;
3417 	set_in_skip(con, resid + remaining);
3418 	con->v2.in_state = IN_S_FINISH_SKIP;
3419 }
3420 
3421 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con)
3422 {
3423 	int recved, resid;  /* current piece of data */
3424 	int remaining;
3425 
3426 	WARN_ON(con_secure(con));
3427 	WARN_ON(!data_len(con->in_msg));
3428 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3429 	resid = iov_iter_count(&con->v2.in_iter);
3430 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3431 	recved = con->v2.in_bvec.bv_len - resid;
3432 	dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid);
3433 
3434 	if (recved)
3435 		ceph_msg_data_advance(&con->v2.in_cursor, recved);
3436 	WARN_ON(resid > con->v2.in_cursor.total_resid);
3437 
3438 	remaining = CEPH_EPILOGUE_PLAIN_LEN;
3439 	dout("%s con %p total_resid %zu remaining %d\n", __func__, con,
3440 	     con->v2.in_cursor.total_resid, remaining);
3441 	con->v2.in_iter.count -= resid;
3442 	set_in_skip(con, con->v2.in_cursor.total_resid + remaining);
3443 	con->v2.in_state = IN_S_FINISH_SKIP;
3444 }
3445 
3446 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con)
3447 {
3448 	int resid;  /* current enc page (not necessarily data) */
3449 
3450 	WARN_ON(!con_secure(con));
3451 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3452 	resid = iov_iter_count(&con->v2.in_iter);
3453 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3454 
3455 	dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid,
3456 	     con->v2.in_enc_resid);
3457 	con->v2.in_iter.count -= resid;
3458 	set_in_skip(con, resid + con->v2.in_enc_resid);
3459 	con->v2.in_state = IN_S_FINISH_SKIP;
3460 }
3461 
3462 static void revoke_at_handle_epilogue(struct ceph_connection *con)
3463 {
3464 	int resid;
3465 
3466 	resid = iov_iter_count(&con->v2.in_iter);
3467 	WARN_ON(!resid);
3468 
3469 	dout("%s con %p resid %d\n", __func__, con, resid);
3470 	con->v2.in_iter.count -= resid;
3471 	set_in_skip(con, resid);
3472 	con->v2.in_state = IN_S_FINISH_SKIP;
3473 }
3474 
3475 void ceph_con_v2_revoke_incoming(struct ceph_connection *con)
3476 {
3477 	switch (con->v2.in_state) {
3478 	case IN_S_PREPARE_READ_DATA:
3479 		revoke_at_prepare_read_data(con);
3480 		break;
3481 	case IN_S_PREPARE_READ_DATA_CONT:
3482 		revoke_at_prepare_read_data_cont(con);
3483 		break;
3484 	case IN_S_PREPARE_READ_ENC_PAGE:
3485 		revoke_at_prepare_read_enc_page(con);
3486 		break;
3487 	case IN_S_HANDLE_EPILOGUE:
3488 		revoke_at_handle_epilogue(con);
3489 		break;
3490 	default:
3491 		WARN(1, "bad in_state %d", con->v2.in_state);
3492 		break;
3493 	}
3494 }
3495 
3496 bool ceph_con_v2_opened(struct ceph_connection *con)
3497 {
3498 	return con->v2.peer_global_seq;
3499 }
3500 
3501 void ceph_con_v2_reset_session(struct ceph_connection *con)
3502 {
3503 	con->v2.client_cookie = 0;
3504 	con->v2.server_cookie = 0;
3505 	con->v2.global_seq = 0;
3506 	con->v2.connect_seq = 0;
3507 	con->v2.peer_global_seq = 0;
3508 }
3509 
3510 void ceph_con_v2_reset_protocol(struct ceph_connection *con)
3511 {
3512 	iov_iter_truncate(&con->v2.in_iter, 0);
3513 	iov_iter_truncate(&con->v2.out_iter, 0);
3514 	con->v2.out_zero = 0;
3515 
3516 	clear_in_sign_kvecs(con);
3517 	clear_out_sign_kvecs(con);
3518 	free_conn_bufs(con);
3519 
3520 	if (con->v2.in_enc_pages) {
3521 		WARN_ON(!con->v2.in_enc_page_cnt);
3522 		ceph_release_page_vector(con->v2.in_enc_pages,
3523 					 con->v2.in_enc_page_cnt);
3524 		con->v2.in_enc_pages = NULL;
3525 		con->v2.in_enc_page_cnt = 0;
3526 	}
3527 	if (con->v2.out_enc_pages) {
3528 		WARN_ON(!con->v2.out_enc_page_cnt);
3529 		ceph_release_page_vector(con->v2.out_enc_pages,
3530 					 con->v2.out_enc_page_cnt);
3531 		con->v2.out_enc_pages = NULL;
3532 		con->v2.out_enc_page_cnt = 0;
3533 	}
3534 
3535 	con->v2.con_mode = CEPH_CON_MODE_UNKNOWN;
3536 	memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN);
3537 	memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN);
3538 
3539 	if (con->v2.hmac_tfm) {
3540 		crypto_free_shash(con->v2.hmac_tfm);
3541 		con->v2.hmac_tfm = NULL;
3542 	}
3543 	if (con->v2.gcm_req) {
3544 		aead_request_free(con->v2.gcm_req);
3545 		con->v2.gcm_req = NULL;
3546 	}
3547 	if (con->v2.gcm_tfm) {
3548 		crypto_free_aead(con->v2.gcm_tfm);
3549 		con->v2.gcm_tfm = NULL;
3550 	}
3551 }
3552