xref: /linux/net/ceph/messenger_v2.c (revision ac2dc6d57425ffa9629941d7c9d7c0e51082cb5a)
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/sha2.h>
12 #include <crypto/utils.h>
13 #include <linux/bvec.h>
14 #include <linux/crc32c.h>
15 #include <linux/net.h>
16 #include <linux/scatterlist.h>
17 #include <linux/socket.h>
18 #include <linux/sched/mm.h>
19 #include <net/sock.h>
20 #include <net/tcp.h>
21 
22 #include <linux/ceph/ceph_features.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/libceph.h>
25 #include <linux/ceph/messenger.h>
26 
27 #include "crypto.h"  /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */
28 
29 #define FRAME_TAG_HELLO			1
30 #define FRAME_TAG_AUTH_REQUEST		2
31 #define FRAME_TAG_AUTH_BAD_METHOD	3
32 #define FRAME_TAG_AUTH_REPLY_MORE	4
33 #define FRAME_TAG_AUTH_REQUEST_MORE	5
34 #define FRAME_TAG_AUTH_DONE		6
35 #define FRAME_TAG_AUTH_SIGNATURE	7
36 #define FRAME_TAG_CLIENT_IDENT		8
37 #define FRAME_TAG_SERVER_IDENT		9
38 #define FRAME_TAG_IDENT_MISSING_FEATURES 10
39 #define FRAME_TAG_SESSION_RECONNECT	11
40 #define FRAME_TAG_SESSION_RESET		12
41 #define FRAME_TAG_SESSION_RETRY		13
42 #define FRAME_TAG_SESSION_RETRY_GLOBAL	14
43 #define FRAME_TAG_SESSION_RECONNECT_OK	15
44 #define FRAME_TAG_WAIT			16
45 #define FRAME_TAG_MESSAGE		17
46 #define FRAME_TAG_KEEPALIVE2		18
47 #define FRAME_TAG_KEEPALIVE2_ACK	19
48 #define FRAME_TAG_ACK			20
49 
50 #define FRAME_LATE_STATUS_ABORTED	0x1
51 #define FRAME_LATE_STATUS_COMPLETE	0xe
52 #define FRAME_LATE_STATUS_ABORTED_MASK	0xf
53 
54 #define IN_S_HANDLE_PREAMBLE			1
55 #define IN_S_HANDLE_CONTROL			2
56 #define IN_S_HANDLE_CONTROL_REMAINDER		3
57 #define IN_S_PREPARE_READ_DATA			4
58 #define IN_S_PREPARE_READ_DATA_CONT		5
59 #define IN_S_PREPARE_READ_ENC_PAGE		6
60 #define IN_S_PREPARE_SPARSE_DATA		7
61 #define IN_S_PREPARE_SPARSE_DATA_CONT		8
62 #define IN_S_HANDLE_EPILOGUE			9
63 #define IN_S_FINISH_SKIP			10
64 
65 #define OUT_S_QUEUE_DATA		1
66 #define OUT_S_QUEUE_DATA_CONT		2
67 #define OUT_S_QUEUE_ENC_PAGE		3
68 #define OUT_S_QUEUE_ZEROS		4
69 #define OUT_S_FINISH_MESSAGE		5
70 #define OUT_S_GET_NEXT			6
71 
72 #define CTRL_BODY(p)	((void *)(p) + CEPH_PREAMBLE_LEN)
73 #define FRONT_PAD(p)	((void *)(p) + CEPH_EPILOGUE_SECURE_LEN)
74 #define MIDDLE_PAD(p)	(FRONT_PAD(p) + CEPH_GCM_BLOCK_LEN)
75 #define DATA_PAD(p)	(MIDDLE_PAD(p) + CEPH_GCM_BLOCK_LEN)
76 
77 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
78 
do_recvmsg(struct socket * sock,struct iov_iter * it)79 static int do_recvmsg(struct socket *sock, struct iov_iter *it)
80 {
81 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
82 	int ret;
83 
84 	msg.msg_iter = *it;
85 	while (iov_iter_count(it)) {
86 		ret = sock_recvmsg(sock, &msg, msg.msg_flags);
87 		if (ret <= 0) {
88 			if (ret == -EAGAIN)
89 				ret = 0;
90 			return ret;
91 		}
92 
93 		iov_iter_advance(it, ret);
94 	}
95 
96 	WARN_ON(msg_data_left(&msg));
97 	return 1;
98 }
99 
100 /*
101  * Read as much as possible.
102  *
103  * Return:
104  *   1 - done, nothing (else) to read
105  *   0 - socket is empty, need to wait
106  *  <0 - error
107  */
ceph_tcp_recv(struct ceph_connection * con)108 static int ceph_tcp_recv(struct ceph_connection *con)
109 {
110 	int ret;
111 
112 	dout("%s con %p %s %zu\n", __func__, con,
113 	     iov_iter_is_discard(&con->v2.in_iter) ? "discard" : "need",
114 	     iov_iter_count(&con->v2.in_iter));
115 	ret = do_recvmsg(con->sock, &con->v2.in_iter);
116 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
117 	     iov_iter_count(&con->v2.in_iter));
118 	return ret;
119 }
120 
do_sendmsg(struct socket * sock,struct iov_iter * it)121 static int do_sendmsg(struct socket *sock, struct iov_iter *it)
122 {
123 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
124 	int ret;
125 
126 	msg.msg_iter = *it;
127 	while (iov_iter_count(it)) {
128 		ret = sock_sendmsg(sock, &msg);
129 		if (ret <= 0) {
130 			if (ret == -EAGAIN)
131 				ret = 0;
132 			return ret;
133 		}
134 
135 		iov_iter_advance(it, ret);
136 	}
137 
138 	WARN_ON(msg_data_left(&msg));
139 	return 1;
140 }
141 
do_try_sendpage(struct socket * sock,struct iov_iter * it)142 static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
143 {
144 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
145 	struct bio_vec bv;
146 	int ret;
147 
148 	if (WARN_ON(!iov_iter_is_bvec(it)))
149 		return -EINVAL;
150 
151 	while (iov_iter_count(it)) {
152 		/* iov_iter_iovec() for ITER_BVEC */
153 		bvec_set_page(&bv, it->bvec->bv_page,
154 			      min(iov_iter_count(it),
155 				  it->bvec->bv_len - it->iov_offset),
156 			      it->bvec->bv_offset + it->iov_offset);
157 
158 		/*
159 		 * MSG_SPLICE_PAGES cannot properly handle pages with
160 		 * page_count == 0, we need to fall back to sendmsg if
161 		 * that's the case.
162 		 *
163 		 * Same goes for slab pages: skb_can_coalesce() allows
164 		 * coalescing neighboring slab objects into a single frag
165 		 * which triggers one of hardened usercopy checks.
166 		 */
167 		if (sendpage_ok(bv.bv_page))
168 			msg.msg_flags |= MSG_SPLICE_PAGES;
169 		else
170 			msg.msg_flags &= ~MSG_SPLICE_PAGES;
171 
172 		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
173 		ret = sock_sendmsg(sock, &msg);
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 here.
189  *
190  * Return:
191  *   1 - done, nothing (else) to write
192  *   0 - socket is full, need to wait
193  *  <0 - error
194  */
ceph_tcp_send(struct ceph_connection * con)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 
add_in_kvec(struct ceph_connection * con,void * buf,int len)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 
reset_in_kvecs(struct ceph_connection * con)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, ITER_DEST, con->v2.in_kvecs, 0, 0);
229 }
230 
set_in_bvec(struct ceph_connection * con,const struct bio_vec * bv)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, ITER_DEST, &con->v2.in_bvec, 1, bv->bv_len);
237 }
238 
set_in_skip(struct ceph_connection * con,int len)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, ITER_DEST, len);
245 }
246 
add_out_kvec(struct ceph_connection * con,void * buf,int len)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 
reset_out_kvecs(struct ceph_connection * con)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, ITER_SOURCE, con->v2.out_kvecs, 0, 0);
269 	con->v2.out_iter_sendpage = false;
270 }
271 
set_out_bvec(struct ceph_connection * con,const struct bio_vec * bv,bool zerocopy)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, ITER_SOURCE, &con->v2.out_bvec, 1,
281 		      con->v2.out_bvec.bv_len);
282 }
283 
set_out_bvec_zero(struct ceph_connection * con)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 	bvec_set_page(&con->v2.out_bvec, ceph_zero_page,
290 		      min(con->v2.out_zero, (int)PAGE_SIZE), 0);
291 	con->v2.out_iter_sendpage = true;
292 	iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
293 		      con->v2.out_bvec.bv_len);
294 }
295 
out_zero_add(struct ceph_connection * con,int len)296 static void out_zero_add(struct ceph_connection *con, int len)
297 {
298 	dout("%s con %p len %d\n", __func__, con, len);
299 	con->v2.out_zero += len;
300 }
301 
alloc_conn_buf(struct ceph_connection * con,int len)302 static void *alloc_conn_buf(struct ceph_connection *con, int len)
303 {
304 	void *buf;
305 
306 	dout("%s con %p len %d\n", __func__, con, len);
307 
308 	if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
309 		return NULL;
310 
311 	buf = kvmalloc(len, GFP_NOIO);
312 	if (!buf)
313 		return NULL;
314 
315 	con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf;
316 	return buf;
317 }
318 
free_conn_bufs(struct ceph_connection * con)319 static void free_conn_bufs(struct ceph_connection *con)
320 {
321 	while (con->v2.conn_buf_cnt)
322 		kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]);
323 }
324 
add_in_sign_kvec(struct ceph_connection * con,void * buf,int len)325 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len)
326 {
327 	BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs));
328 
329 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf;
330 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len;
331 	con->v2.in_sign_kvec_cnt++;
332 }
333 
clear_in_sign_kvecs(struct ceph_connection * con)334 static void clear_in_sign_kvecs(struct ceph_connection *con)
335 {
336 	con->v2.in_sign_kvec_cnt = 0;
337 }
338 
add_out_sign_kvec(struct ceph_connection * con,void * buf,int len)339 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len)
340 {
341 	BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs));
342 
343 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf;
344 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len;
345 	con->v2.out_sign_kvec_cnt++;
346 }
347 
clear_out_sign_kvecs(struct ceph_connection * con)348 static void clear_out_sign_kvecs(struct ceph_connection *con)
349 {
350 	con->v2.out_sign_kvec_cnt = 0;
351 }
352 
con_secure(struct ceph_connection * con)353 static bool con_secure(struct ceph_connection *con)
354 {
355 	return con->v2.con_mode == CEPH_CON_MODE_SECURE;
356 }
357 
front_len(const struct ceph_msg * msg)358 static int front_len(const struct ceph_msg *msg)
359 {
360 	return le32_to_cpu(msg->hdr.front_len);
361 }
362 
middle_len(const struct ceph_msg * msg)363 static int middle_len(const struct ceph_msg *msg)
364 {
365 	return le32_to_cpu(msg->hdr.middle_len);
366 }
367 
data_len(const struct ceph_msg * msg)368 static int data_len(const struct ceph_msg *msg)
369 {
370 	return le32_to_cpu(msg->hdr.data_len);
371 }
372 
need_padding(int len)373 static bool need_padding(int len)
374 {
375 	return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN);
376 }
377 
padded_len(int len)378 static int padded_len(int len)
379 {
380 	return ALIGN(len, CEPH_GCM_BLOCK_LEN);
381 }
382 
padding_len(int len)383 static int padding_len(int len)
384 {
385 	return padded_len(len) - len;
386 }
387 
388 /* preamble + control segment */
head_onwire_len(int ctrl_len,bool secure)389 static int head_onwire_len(int ctrl_len, bool secure)
390 {
391 	int head_len;
392 	int rem_len;
393 
394 	BUG_ON(ctrl_len < 1 || ctrl_len > CEPH_MSG_MAX_CONTROL_LEN);
395 
396 	if (secure) {
397 		head_len = CEPH_PREAMBLE_SECURE_LEN;
398 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
399 			rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
400 			head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN;
401 		}
402 	} else {
403 		head_len = CEPH_PREAMBLE_PLAIN_LEN + ctrl_len + CEPH_CRC_LEN;
404 	}
405 	return head_len;
406 }
407 
408 /* front, middle and data segments + epilogue */
__tail_onwire_len(int front_len,int middle_len,int data_len,bool secure)409 static int __tail_onwire_len(int front_len, int middle_len, int data_len,
410 			     bool secure)
411 {
412 	BUG_ON(front_len < 0 || front_len > CEPH_MSG_MAX_FRONT_LEN ||
413 	       middle_len < 0 || middle_len > CEPH_MSG_MAX_MIDDLE_LEN ||
414 	       data_len < 0 || data_len > CEPH_MSG_MAX_DATA_LEN);
415 
416 	if (!front_len && !middle_len && !data_len)
417 		return 0;
418 
419 	if (!secure)
420 		return front_len + middle_len + data_len +
421 		       CEPH_EPILOGUE_PLAIN_LEN;
422 
423 	return padded_len(front_len) + padded_len(middle_len) +
424 	       padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN;
425 }
426 
tail_onwire_len(const struct ceph_msg * msg,bool secure)427 static int tail_onwire_len(const struct ceph_msg *msg, bool secure)
428 {
429 	return __tail_onwire_len(front_len(msg), middle_len(msg),
430 				 data_len(msg), secure);
431 }
432 
433 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */
434 #define MESSAGE_HEAD_PLAIN_LEN	(CEPH_PREAMBLE_PLAIN_LEN +		\
435 				 sizeof(struct ceph_msg_header2) +	\
436 				 CEPH_CRC_LEN)
437 
438 static const int frame_aligns[] = {
439 	sizeof(void *),
440 	sizeof(void *),
441 	sizeof(void *),
442 	PAGE_SIZE
443 };
444 
445 /*
446  * Discards trailing empty segments, unless there is just one segment.
447  * A frame always has at least one (possibly empty) segment.
448  */
calc_segment_count(const int * lens,int len_cnt)449 static int calc_segment_count(const int *lens, int len_cnt)
450 {
451 	int i;
452 
453 	for (i = len_cnt - 1; i >= 0; i--) {
454 		if (lens[i])
455 			return i + 1;
456 	}
457 
458 	return 1;
459 }
460 
init_frame_desc(struct ceph_frame_desc * desc,int tag,const int * lens,int len_cnt)461 static void init_frame_desc(struct ceph_frame_desc *desc, int tag,
462 			    const int *lens, int len_cnt)
463 {
464 	int i;
465 
466 	memset(desc, 0, sizeof(*desc));
467 
468 	desc->fd_tag = tag;
469 	desc->fd_seg_cnt = calc_segment_count(lens, len_cnt);
470 	BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT);
471 	for (i = 0; i < desc->fd_seg_cnt; i++) {
472 		desc->fd_lens[i] = lens[i];
473 		desc->fd_aligns[i] = frame_aligns[i];
474 	}
475 }
476 
477 /*
478  * Preamble crc covers everything up to itself (28 bytes) and
479  * is calculated and verified irrespective of the connection mode
480  * (i.e. even if the frame is encrypted).
481  */
encode_preamble(const struct ceph_frame_desc * desc,void * p)482 static void encode_preamble(const struct ceph_frame_desc *desc, void *p)
483 {
484 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
485 	void *start = p;
486 	int i;
487 
488 	memset(p, 0, CEPH_PREAMBLE_LEN);
489 
490 	ceph_encode_8(&p, desc->fd_tag);
491 	ceph_encode_8(&p, desc->fd_seg_cnt);
492 	for (i = 0; i < desc->fd_seg_cnt; i++) {
493 		ceph_encode_32(&p, desc->fd_lens[i]);
494 		ceph_encode_16(&p, desc->fd_aligns[i]);
495 	}
496 
497 	put_unaligned_le32(crc32c(0, start, crcp - start), crcp);
498 }
499 
decode_preamble(void * p,struct ceph_frame_desc * desc)500 static int decode_preamble(void *p, struct ceph_frame_desc *desc)
501 {
502 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
503 	u32 crc, expected_crc;
504 	int i;
505 
506 	crc = crc32c(0, p, crcp - p);
507 	expected_crc = get_unaligned_le32(crcp);
508 	if (crc != expected_crc) {
509 		pr_err("bad preamble crc, calculated %u, expected %u\n",
510 		       crc, expected_crc);
511 		return -EBADMSG;
512 	}
513 
514 	memset(desc, 0, sizeof(*desc));
515 
516 	desc->fd_tag = ceph_decode_8(&p);
517 	desc->fd_seg_cnt = ceph_decode_8(&p);
518 	if (desc->fd_seg_cnt < 1 ||
519 	    desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) {
520 		pr_err("bad segment count %d\n", desc->fd_seg_cnt);
521 		return -EINVAL;
522 	}
523 	for (i = 0; i < desc->fd_seg_cnt; i++) {
524 		desc->fd_lens[i] = ceph_decode_32(&p);
525 		desc->fd_aligns[i] = ceph_decode_16(&p);
526 	}
527 
528 	/*
529 	 * This would fire for FRAME_TAG_WAIT (it has one empty
530 	 * segment), but we should never get it as client.
531 	 */
532 	if (desc->fd_lens[0] < 1 ||
533 	    desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) {
534 		pr_err("bad control segment length %d\n", desc->fd_lens[0]);
535 		return -EINVAL;
536 	}
537 
538 	if (desc->fd_lens[1] < 0 ||
539 	    desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) {
540 		pr_err("bad front segment length %d\n", desc->fd_lens[1]);
541 		return -EINVAL;
542 	}
543 	if (desc->fd_lens[2] < 0 ||
544 	    desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) {
545 		pr_err("bad middle segment length %d\n", desc->fd_lens[2]);
546 		return -EINVAL;
547 	}
548 	if (desc->fd_lens[3] < 0 ||
549 	    desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) {
550 		pr_err("bad data segment length %d\n", desc->fd_lens[3]);
551 		return -EINVAL;
552 	}
553 
554 	if (!desc->fd_lens[desc->fd_seg_cnt - 1]) {
555 		pr_err("last segment empty, segment count %d\n",
556 		       desc->fd_seg_cnt);
557 		return -EINVAL;
558 	}
559 
560 	return 0;
561 }
562 
encode_epilogue_plain(struct ceph_connection * con,bool aborted)563 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted)
564 {
565 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
566 						 FRAME_LATE_STATUS_COMPLETE;
567 	cpu_to_le32s(&con->v2.out_epil.front_crc);
568 	cpu_to_le32s(&con->v2.out_epil.middle_crc);
569 	cpu_to_le32s(&con->v2.out_epil.data_crc);
570 }
571 
encode_epilogue_secure(struct ceph_connection * con,bool aborted)572 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted)
573 {
574 	memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil));
575 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
576 						 FRAME_LATE_STATUS_COMPLETE;
577 }
578 
decode_epilogue(void * p,u32 * front_crc,u32 * middle_crc,u32 * data_crc)579 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc,
580 			   u32 *data_crc)
581 {
582 	u8 late_status;
583 
584 	late_status = ceph_decode_8(&p);
585 	if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) !=
586 			FRAME_LATE_STATUS_COMPLETE) {
587 		/* we should never get an aborted message as client */
588 		pr_err("bad late_status 0x%x\n", late_status);
589 		return -EINVAL;
590 	}
591 
592 	if (front_crc && middle_crc && data_crc) {
593 		*front_crc = ceph_decode_32(&p);
594 		*middle_crc = ceph_decode_32(&p);
595 		*data_crc = ceph_decode_32(&p);
596 	}
597 
598 	return 0;
599 }
600 
fill_header(struct ceph_msg_header * hdr,const struct ceph_msg_header2 * hdr2,int front_len,int middle_len,int data_len,const struct ceph_entity_name * peer_name)601 static void fill_header(struct ceph_msg_header *hdr,
602 			const struct ceph_msg_header2 *hdr2,
603 			int front_len, int middle_len, int data_len,
604 			const struct ceph_entity_name *peer_name)
605 {
606 	hdr->seq = hdr2->seq;
607 	hdr->tid = hdr2->tid;
608 	hdr->type = hdr2->type;
609 	hdr->priority = hdr2->priority;
610 	hdr->version = hdr2->version;
611 	hdr->front_len = cpu_to_le32(front_len);
612 	hdr->middle_len = cpu_to_le32(middle_len);
613 	hdr->data_len = cpu_to_le32(data_len);
614 	hdr->data_off = hdr2->data_off;
615 	hdr->src = *peer_name;
616 	hdr->compat_version = hdr2->compat_version;
617 	hdr->reserved = 0;
618 	hdr->crc = 0;
619 }
620 
fill_header2(struct ceph_msg_header2 * hdr2,const struct ceph_msg_header * hdr,u64 ack_seq)621 static void fill_header2(struct ceph_msg_header2 *hdr2,
622 			 const struct ceph_msg_header *hdr, u64 ack_seq)
623 {
624 	hdr2->seq = hdr->seq;
625 	hdr2->tid = hdr->tid;
626 	hdr2->type = hdr->type;
627 	hdr2->priority = hdr->priority;
628 	hdr2->version = hdr->version;
629 	hdr2->data_pre_padding_len = 0;
630 	hdr2->data_off = hdr->data_off;
631 	hdr2->ack_seq = cpu_to_le64(ack_seq);
632 	hdr2->flags = 0;
633 	hdr2->compat_version = hdr->compat_version;
634 	hdr2->reserved = 0;
635 }
636 
verify_control_crc(struct ceph_connection * con)637 static int verify_control_crc(struct ceph_connection *con)
638 {
639 	int ctrl_len = con->v2.in_desc.fd_lens[0];
640 	u32 crc, expected_crc;
641 
642 	WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len);
643 	WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN);
644 
645 	crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len);
646 	expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base);
647 	if (crc != expected_crc) {
648 		pr_err("bad control crc, calculated %u, expected %u\n",
649 		       crc, expected_crc);
650 		return -EBADMSG;
651 	}
652 
653 	return 0;
654 }
655 
verify_epilogue_crcs(struct ceph_connection * con,u32 front_crc,u32 middle_crc,u32 data_crc)656 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc,
657 				u32 middle_crc, u32 data_crc)
658 {
659 	if (front_len(con->in_msg)) {
660 		con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base,
661 					   front_len(con->in_msg));
662 	} else {
663 		WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg));
664 		con->in_front_crc = -1;
665 	}
666 
667 	if (middle_len(con->in_msg))
668 		con->in_middle_crc = crc32c(-1,
669 					    con->in_msg->middle->vec.iov_base,
670 					    middle_len(con->in_msg));
671 	else if (data_len(con->in_msg))
672 		con->in_middle_crc = -1;
673 	else
674 		con->in_middle_crc = 0;
675 
676 	if (!data_len(con->in_msg))
677 		con->in_data_crc = 0;
678 
679 	dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg,
680 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
681 
682 	if (con->in_front_crc != front_crc) {
683 		pr_err("bad front crc, calculated %u, expected %u\n",
684 		       con->in_front_crc, front_crc);
685 		return -EBADMSG;
686 	}
687 	if (con->in_middle_crc != middle_crc) {
688 		pr_err("bad middle crc, calculated %u, expected %u\n",
689 		       con->in_middle_crc, middle_crc);
690 		return -EBADMSG;
691 	}
692 	if (con->in_data_crc != data_crc) {
693 		pr_err("bad data crc, calculated %u, expected %u\n",
694 		       con->in_data_crc, data_crc);
695 		return -EBADMSG;
696 	}
697 
698 	return 0;
699 }
700 
setup_crypto(struct ceph_connection * con,const u8 * session_key,int session_key_len,const u8 * con_secret,int con_secret_len)701 static int setup_crypto(struct ceph_connection *con,
702 			const u8 *session_key, int session_key_len,
703 			const u8 *con_secret, int con_secret_len)
704 {
705 	unsigned int noio_flag;
706 	int ret;
707 
708 	dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n",
709 	     __func__, con, con->v2.con_mode, session_key_len, con_secret_len);
710 	WARN_ON(con->v2.hmac_key_set || con->v2.gcm_tfm || con->v2.gcm_req);
711 
712 	if (con->v2.con_mode != CEPH_CON_MODE_CRC &&
713 	    con->v2.con_mode != CEPH_CON_MODE_SECURE) {
714 		pr_err("bad con_mode %d\n", con->v2.con_mode);
715 		return -EINVAL;
716 	}
717 
718 	if (!session_key_len) {
719 		WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC);
720 		WARN_ON(con_secret_len);
721 		return 0;  /* auth_none */
722 	}
723 
724 	hmac_sha256_preparekey(&con->v2.hmac_key, session_key, session_key_len);
725 	con->v2.hmac_key_set = true;
726 
727 	if (con->v2.con_mode == CEPH_CON_MODE_CRC) {
728 		WARN_ON(con_secret_len);
729 		return 0;  /* auth_x, plain mode */
730 	}
731 
732 	if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) {
733 		pr_err("con_secret too small %d\n", con_secret_len);
734 		return -EINVAL;
735 	}
736 
737 	noio_flag = memalloc_noio_save();
738 	con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0);
739 	memalloc_noio_restore(noio_flag);
740 	if (IS_ERR(con->v2.gcm_tfm)) {
741 		ret = PTR_ERR(con->v2.gcm_tfm);
742 		con->v2.gcm_tfm = NULL;
743 		pr_err("failed to allocate gcm tfm context: %d\n", ret);
744 		return ret;
745 	}
746 
747 	WARN_ON((unsigned long)con_secret &
748 		crypto_aead_alignmask(con->v2.gcm_tfm));
749 	ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN);
750 	if (ret) {
751 		pr_err("failed to set gcm key: %d\n", ret);
752 		return ret;
753 	}
754 
755 	WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN);
756 	ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN);
757 	if (ret) {
758 		pr_err("failed to set gcm tag size: %d\n", ret);
759 		return ret;
760 	}
761 
762 	con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO);
763 	if (!con->v2.gcm_req) {
764 		pr_err("failed to allocate gcm request\n");
765 		return -ENOMEM;
766 	}
767 
768 	crypto_init_wait(&con->v2.gcm_wait);
769 	aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
770 				  crypto_req_done, &con->v2.gcm_wait);
771 
772 	memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN,
773 	       CEPH_GCM_IV_LEN);
774 	memcpy(&con->v2.out_gcm_nonce,
775 	       con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN,
776 	       CEPH_GCM_IV_LEN);
777 	return 0;  /* auth_x, secure mode */
778 }
779 
con_hmac_sha256(struct ceph_connection * con,const struct kvec * kvecs,int kvec_cnt,u8 hmac[SHA256_DIGEST_SIZE])780 static void con_hmac_sha256(struct ceph_connection *con,
781 			    const struct kvec *kvecs, int kvec_cnt,
782 			    u8 hmac[SHA256_DIGEST_SIZE])
783 {
784 	struct hmac_sha256_ctx ctx;
785 	int i;
786 
787 	dout("%s con %p hmac_key_set %d kvec_cnt %d\n", __func__, con,
788 	     con->v2.hmac_key_set, kvec_cnt);
789 
790 	if (!con->v2.hmac_key_set) {
791 		memset(hmac, 0, SHA256_DIGEST_SIZE);
792 		return;  /* auth_none */
793 	}
794 
795 	/* auth_x, both plain and secure modes */
796 	hmac_sha256_init(&ctx, &con->v2.hmac_key);
797 	for (i = 0; i < kvec_cnt; i++)
798 		hmac_sha256_update(&ctx, kvecs[i].iov_base, kvecs[i].iov_len);
799 	hmac_sha256_final(&ctx, hmac);
800 }
801 
gcm_inc_nonce(struct ceph_gcm_nonce * nonce)802 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce)
803 {
804 	u64 counter;
805 
806 	counter = le64_to_cpu(nonce->counter);
807 	nonce->counter = cpu_to_le64(counter + 1);
808 }
809 
gcm_crypt(struct ceph_connection * con,bool encrypt,struct scatterlist * src,struct scatterlist * dst,int src_len)810 static int gcm_crypt(struct ceph_connection *con, bool encrypt,
811 		     struct scatterlist *src, struct scatterlist *dst,
812 		     int src_len)
813 {
814 	struct ceph_gcm_nonce *nonce;
815 	int ret;
816 
817 	nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce;
818 
819 	aead_request_set_ad(con->v2.gcm_req, 0);  /* no AAD */
820 	aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce);
821 	ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) :
822 					crypto_aead_decrypt(con->v2.gcm_req),
823 			      &con->v2.gcm_wait);
824 	if (ret)
825 		return ret;
826 
827 	gcm_inc_nonce(nonce);
828 	return 0;
829 }
830 
get_bvec_at(struct ceph_msg_data_cursor * cursor,struct bio_vec * bv)831 static void get_bvec_at(struct ceph_msg_data_cursor *cursor,
832 			struct bio_vec *bv)
833 {
834 	struct page *page;
835 	size_t off, len;
836 
837 	WARN_ON(!cursor->total_resid);
838 
839 	/* skip zero-length data items */
840 	while (!cursor->resid)
841 		ceph_msg_data_advance(cursor, 0);
842 
843 	/* get a piece of data, cursor isn't advanced */
844 	page = ceph_msg_data_next(cursor, &off, &len);
845 	bvec_set_page(bv, page, len, off);
846 }
847 
calc_sg_cnt(void * buf,int buf_len)848 static int calc_sg_cnt(void *buf, int buf_len)
849 {
850 	int sg_cnt;
851 
852 	if (!buf_len)
853 		return 0;
854 
855 	sg_cnt = need_padding(buf_len) ? 1 : 0;
856 	if (is_vmalloc_addr(buf)) {
857 		WARN_ON(offset_in_page(buf));
858 		sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT;
859 	} else {
860 		sg_cnt++;
861 	}
862 
863 	return sg_cnt;
864 }
865 
calc_sg_cnt_cursor(struct ceph_msg_data_cursor * cursor)866 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor)
867 {
868 	int data_len = cursor->total_resid;
869 	struct bio_vec bv;
870 	int sg_cnt;
871 
872 	if (!data_len)
873 		return 0;
874 
875 	sg_cnt = need_padding(data_len) ? 1 : 0;
876 	do {
877 		get_bvec_at(cursor, &bv);
878 		sg_cnt++;
879 
880 		ceph_msg_data_advance(cursor, bv.bv_len);
881 	} while (cursor->total_resid);
882 
883 	return sg_cnt;
884 }
885 
init_sgs(struct scatterlist ** sg,void * buf,int buf_len,u8 * pad)886 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad)
887 {
888 	void *end = buf + buf_len;
889 	struct page *page;
890 	int len;
891 	void *p;
892 
893 	if (!buf_len)
894 		return;
895 
896 	if (is_vmalloc_addr(buf)) {
897 		p = buf;
898 		do {
899 			page = vmalloc_to_page(p);
900 			len = min_t(int, end - p, PAGE_SIZE);
901 			WARN_ON(!page || !len || offset_in_page(p));
902 			sg_set_page(*sg, page, len, 0);
903 			*sg = sg_next(*sg);
904 			p += len;
905 		} while (p != end);
906 	} else {
907 		sg_set_buf(*sg, buf, buf_len);
908 		*sg = sg_next(*sg);
909 	}
910 
911 	if (need_padding(buf_len)) {
912 		sg_set_buf(*sg, pad, padding_len(buf_len));
913 		*sg = sg_next(*sg);
914 	}
915 }
916 
init_sgs_cursor(struct scatterlist ** sg,struct ceph_msg_data_cursor * cursor,u8 * pad)917 static void init_sgs_cursor(struct scatterlist **sg,
918 			    struct ceph_msg_data_cursor *cursor, u8 *pad)
919 {
920 	int data_len = cursor->total_resid;
921 	struct bio_vec bv;
922 
923 	if (!data_len)
924 		return;
925 
926 	do {
927 		get_bvec_at(cursor, &bv);
928 		sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
929 		*sg = sg_next(*sg);
930 
931 		ceph_msg_data_advance(cursor, bv.bv_len);
932 	} while (cursor->total_resid);
933 
934 	if (need_padding(data_len)) {
935 		sg_set_buf(*sg, pad, padding_len(data_len));
936 		*sg = sg_next(*sg);
937 	}
938 }
939 
940 /**
941  * init_sgs_pages: set up scatterlist on an array of page pointers
942  * @sg:		scatterlist to populate
943  * @pages:	pointer to page array
944  * @dpos:	position in the array to start (bytes)
945  * @dlen:	len to add to sg (bytes)
946  * @pad:	pointer to pad destination (if any)
947  *
948  * Populate the scatterlist from the page array, starting at an arbitrary
949  * byte in the array and running for a specified length.
950  */
init_sgs_pages(struct scatterlist ** sg,struct page ** pages,int dpos,int dlen,u8 * pad)951 static void init_sgs_pages(struct scatterlist **sg, struct page **pages,
952 			   int dpos, int dlen, u8 *pad)
953 {
954 	int idx = dpos >> PAGE_SHIFT;
955 	int off = offset_in_page(dpos);
956 	int resid = dlen;
957 
958 	do {
959 		int len = min(resid, (int)PAGE_SIZE - off);
960 
961 		sg_set_page(*sg, pages[idx], len, off);
962 		*sg = sg_next(*sg);
963 		off = 0;
964 		++idx;
965 		resid -= len;
966 	} while (resid);
967 
968 	if (need_padding(dlen)) {
969 		sg_set_buf(*sg, pad, padding_len(dlen));
970 		*sg = sg_next(*sg);
971 	}
972 }
973 
setup_message_sgs(struct sg_table * sgt,struct ceph_msg * msg,u8 * front_pad,u8 * middle_pad,u8 * data_pad,void * epilogue,struct page ** pages,int dpos,bool add_tag)974 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
975 			     u8 *front_pad, u8 *middle_pad, u8 *data_pad,
976 			     void *epilogue, struct page **pages, int dpos,
977 			     bool add_tag)
978 {
979 	struct ceph_msg_data_cursor cursor;
980 	struct scatterlist *cur_sg;
981 	int dlen = data_len(msg);
982 	int sg_cnt;
983 	int ret;
984 
985 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
986 		return 0;
987 
988 	sg_cnt = 1;  /* epilogue + [auth tag] */
989 	if (front_len(msg))
990 		sg_cnt += calc_sg_cnt(msg->front.iov_base,
991 				      front_len(msg));
992 	if (middle_len(msg))
993 		sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
994 				      middle_len(msg));
995 	if (dlen) {
996 		if (pages) {
997 			sg_cnt += calc_pages_for(dpos, dlen);
998 			if (need_padding(dlen))
999 				sg_cnt++;
1000 		} else {
1001 			ceph_msg_data_cursor_init(&cursor, msg, dlen);
1002 			sg_cnt += calc_sg_cnt_cursor(&cursor);
1003 		}
1004 	}
1005 
1006 	ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
1007 	if (ret)
1008 		return ret;
1009 
1010 	cur_sg = sgt->sgl;
1011 	if (front_len(msg))
1012 		init_sgs(&cur_sg, msg->front.iov_base, front_len(msg),
1013 			 front_pad);
1014 	if (middle_len(msg))
1015 		init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
1016 			 middle_pad);
1017 	if (dlen) {
1018 		if (pages) {
1019 			init_sgs_pages(&cur_sg, pages, dpos, dlen, data_pad);
1020 		} else {
1021 			ceph_msg_data_cursor_init(&cursor, msg, dlen);
1022 			init_sgs_cursor(&cur_sg, &cursor, data_pad);
1023 		}
1024 	}
1025 
1026 	WARN_ON(!sg_is_last(cur_sg));
1027 	sg_set_buf(cur_sg, epilogue,
1028 		   CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0));
1029 	return 0;
1030 }
1031 
decrypt_preamble(struct ceph_connection * con)1032 static int decrypt_preamble(struct ceph_connection *con)
1033 {
1034 	struct scatterlist sg;
1035 
1036 	sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN);
1037 	return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN);
1038 }
1039 
decrypt_control_remainder(struct ceph_connection * con)1040 static int decrypt_control_remainder(struct ceph_connection *con)
1041 {
1042 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1043 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1044 	int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN;
1045 	struct scatterlist sgs[2];
1046 
1047 	WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len);
1048 	WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len);
1049 
1050 	sg_init_table(sgs, 2);
1051 	sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len);
1052 	sg_set_buf(&sgs[1], con->v2.in_buf, pt_len);
1053 
1054 	return gcm_crypt(con, false, sgs, sgs,
1055 			 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
1056 }
1057 
1058 /* Process sparse read data that lives in a buffer */
process_v2_sparse_read(struct ceph_connection * con,struct page ** pages,int spos)1059 static int process_v2_sparse_read(struct ceph_connection *con,
1060 				  struct page **pages, int spos)
1061 {
1062 	struct ceph_msg_data_cursor cursor;
1063 	int ret;
1064 
1065 	ceph_msg_data_cursor_init(&cursor, con->in_msg,
1066 				  con->in_msg->sparse_read_total);
1067 
1068 	for (;;) {
1069 		char *buf = NULL;
1070 
1071 		ret = con->ops->sparse_read(con, &cursor, &buf);
1072 		if (ret <= 0)
1073 			return ret;
1074 
1075 		dout("%s: sparse_read return %x buf %p\n", __func__, ret, buf);
1076 
1077 		do {
1078 			int idx = spos >> PAGE_SHIFT;
1079 			int soff = offset_in_page(spos);
1080 			struct page *spage = con->v2.in_enc_pages[idx];
1081 			int len = min_t(int, ret, PAGE_SIZE - soff);
1082 
1083 			if (buf) {
1084 				memcpy_from_page(buf, spage, soff, len);
1085 				buf += len;
1086 			} else {
1087 				struct bio_vec bv;
1088 
1089 				get_bvec_at(&cursor, &bv);
1090 				len = min_t(int, len, bv.bv_len);
1091 				memcpy_page(bv.bv_page, bv.bv_offset,
1092 					    spage, soff, len);
1093 				ceph_msg_data_advance(&cursor, len);
1094 			}
1095 			spos += len;
1096 			ret -= len;
1097 		} while (ret);
1098 	}
1099 }
1100 
decrypt_tail(struct ceph_connection * con)1101 static int decrypt_tail(struct ceph_connection *con)
1102 {
1103 	struct sg_table enc_sgt = {};
1104 	struct sg_table sgt = {};
1105 	struct page **pages = NULL;
1106 	bool sparse = !!con->in_msg->sparse_read_total;
1107 	int dpos = 0;
1108 	int tail_len;
1109 	int ret;
1110 
1111 	tail_len = tail_onwire_len(con->in_msg, true);
1112 	ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages,
1113 					con->v2.in_enc_page_cnt, 0, tail_len,
1114 					GFP_NOIO);
1115 	if (ret)
1116 		goto out;
1117 
1118 	if (sparse) {
1119 		dpos = padded_len(front_len(con->in_msg) + padded_len(middle_len(con->in_msg)));
1120 		pages = con->v2.in_enc_pages;
1121 	}
1122 
1123 	ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
1124 				MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
1125 				con->v2.in_buf, pages, dpos, true);
1126 	if (ret)
1127 		goto out;
1128 
1129 	dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con,
1130 	     con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents);
1131 	ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len);
1132 	if (ret)
1133 		goto out;
1134 
1135 	if (sparse && data_len(con->in_msg)) {
1136 		ret = process_v2_sparse_read(con, con->v2.in_enc_pages, dpos);
1137 		if (ret)
1138 			goto out;
1139 	}
1140 
1141 	WARN_ON(!con->v2.in_enc_page_cnt);
1142 	ceph_release_page_vector(con->v2.in_enc_pages,
1143 				 con->v2.in_enc_page_cnt);
1144 	con->v2.in_enc_pages = NULL;
1145 	con->v2.in_enc_page_cnt = 0;
1146 
1147 out:
1148 	sg_free_table(&sgt);
1149 	sg_free_table(&enc_sgt);
1150 	return ret;
1151 }
1152 
prepare_banner(struct ceph_connection * con)1153 static int prepare_banner(struct ceph_connection *con)
1154 {
1155 	int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8;
1156 	void *buf, *p;
1157 
1158 	buf = alloc_conn_buf(con, buf_len);
1159 	if (!buf)
1160 		return -ENOMEM;
1161 
1162 	p = buf;
1163 	ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN);
1164 	ceph_encode_16(&p, sizeof(u64) + sizeof(u64));
1165 	ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES);
1166 	ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES);
1167 	WARN_ON(p != buf + buf_len);
1168 
1169 	add_out_kvec(con, buf, buf_len);
1170 	add_out_sign_kvec(con, buf, buf_len);
1171 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1172 	return 0;
1173 }
1174 
1175 /*
1176  * base:
1177  *   preamble
1178  *   control body (ctrl_len bytes)
1179  *   space for control crc
1180  *
1181  * extdata (optional):
1182  *   control body (extdata_len bytes)
1183  *
1184  * Compute control crc and gather base and extdata into:
1185  *
1186  *   preamble
1187  *   control body (ctrl_len + extdata_len bytes)
1188  *   control crc
1189  *
1190  * Preamble should already be encoded at the start of base.
1191  */
prepare_head_plain(struct ceph_connection * con,void * base,int ctrl_len,void * extdata,int extdata_len,bool to_be_signed)1192 static void prepare_head_plain(struct ceph_connection *con, void *base,
1193 			       int ctrl_len, void *extdata, int extdata_len,
1194 			       bool to_be_signed)
1195 {
1196 	int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN;
1197 	void *crcp = base + base_len - CEPH_CRC_LEN;
1198 	u32 crc;
1199 
1200 	crc = crc32c(-1, CTRL_BODY(base), ctrl_len);
1201 	if (extdata_len)
1202 		crc = crc32c(crc, extdata, extdata_len);
1203 	put_unaligned_le32(crc, crcp);
1204 
1205 	if (!extdata_len) {
1206 		add_out_kvec(con, base, base_len);
1207 		if (to_be_signed)
1208 			add_out_sign_kvec(con, base, base_len);
1209 		return;
1210 	}
1211 
1212 	add_out_kvec(con, base, crcp - base);
1213 	add_out_kvec(con, extdata, extdata_len);
1214 	add_out_kvec(con, crcp, CEPH_CRC_LEN);
1215 	if (to_be_signed) {
1216 		add_out_sign_kvec(con, base, crcp - base);
1217 		add_out_sign_kvec(con, extdata, extdata_len);
1218 		add_out_sign_kvec(con, crcp, CEPH_CRC_LEN);
1219 	}
1220 }
1221 
prepare_head_secure_small(struct ceph_connection * con,void * base,int ctrl_len)1222 static int prepare_head_secure_small(struct ceph_connection *con,
1223 				     void *base, int ctrl_len)
1224 {
1225 	struct scatterlist sg;
1226 	int ret;
1227 
1228 	/* inline buffer padding? */
1229 	if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN)
1230 		memset(CTRL_BODY(base) + ctrl_len, 0,
1231 		       CEPH_PREAMBLE_INLINE_LEN - ctrl_len);
1232 
1233 	sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN);
1234 	ret = gcm_crypt(con, true, &sg, &sg,
1235 			CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN);
1236 	if (ret)
1237 		return ret;
1238 
1239 	add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN);
1240 	return 0;
1241 }
1242 
1243 /*
1244  * base:
1245  *   preamble
1246  *   control body (ctrl_len bytes)
1247  *   space for padding, if needed
1248  *   space for control remainder auth tag
1249  *   space for preamble auth tag
1250  *
1251  * Encrypt preamble and the inline portion, then encrypt the remainder
1252  * and gather into:
1253  *
1254  *   preamble
1255  *   control body (48 bytes)
1256  *   preamble auth tag
1257  *   control body (ctrl_len - 48 bytes)
1258  *   zero padding, if needed
1259  *   control remainder auth tag
1260  *
1261  * Preamble should already be encoded at the start of base.
1262  */
prepare_head_secure_big(struct ceph_connection * con,void * base,int ctrl_len)1263 static int prepare_head_secure_big(struct ceph_connection *con,
1264 				   void *base, int ctrl_len)
1265 {
1266 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1267 	void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN;
1268 	void *rem_tag = rem + padded_len(rem_len);
1269 	void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN;
1270 	struct scatterlist sgs[2];
1271 	int ret;
1272 
1273 	sg_init_table(sgs, 2);
1274 	sg_set_buf(&sgs[0], base, rem - base);
1275 	sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN);
1276 	ret = gcm_crypt(con, true, sgs, sgs, rem - base);
1277 	if (ret)
1278 		return ret;
1279 
1280 	/* control remainder padding? */
1281 	if (need_padding(rem_len))
1282 		memset(rem + rem_len, 0, padding_len(rem_len));
1283 
1284 	sg_init_one(&sgs[0], rem, pmbl_tag - rem);
1285 	ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem);
1286 	if (ret)
1287 		return ret;
1288 
1289 	add_out_kvec(con, base, rem - base);
1290 	add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN);
1291 	add_out_kvec(con, rem, pmbl_tag - rem);
1292 	return 0;
1293 }
1294 
__prepare_control(struct ceph_connection * con,int tag,void * base,int ctrl_len,void * extdata,int extdata_len,bool to_be_signed)1295 static int __prepare_control(struct ceph_connection *con, int tag,
1296 			     void *base, int ctrl_len, void *extdata,
1297 			     int extdata_len, bool to_be_signed)
1298 {
1299 	int total_len = ctrl_len + extdata_len;
1300 	struct ceph_frame_desc desc;
1301 	int ret;
1302 
1303 	dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag,
1304 	     total_len, ctrl_len, extdata_len);
1305 
1306 	/* extdata may be vmalloc'ed but not base */
1307 	if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len))
1308 		return -EINVAL;
1309 
1310 	init_frame_desc(&desc, tag, &total_len, 1);
1311 	encode_preamble(&desc, base);
1312 
1313 	if (con_secure(con)) {
1314 		if (WARN_ON(extdata_len || to_be_signed))
1315 			return -EINVAL;
1316 
1317 		if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN)
1318 			/* fully inlined, inline buffer may need padding */
1319 			ret = prepare_head_secure_small(con, base, ctrl_len);
1320 		else
1321 			/* partially inlined, inline buffer is full */
1322 			ret = prepare_head_secure_big(con, base, ctrl_len);
1323 		if (ret)
1324 			return ret;
1325 	} else {
1326 		prepare_head_plain(con, base, ctrl_len, extdata, extdata_len,
1327 				   to_be_signed);
1328 	}
1329 
1330 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1331 	return 0;
1332 }
1333 
prepare_control(struct ceph_connection * con,int tag,void * base,int ctrl_len)1334 static int prepare_control(struct ceph_connection *con, int tag,
1335 			   void *base, int ctrl_len)
1336 {
1337 	return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false);
1338 }
1339 
prepare_hello(struct ceph_connection * con)1340 static int prepare_hello(struct ceph_connection *con)
1341 {
1342 	void *buf, *p;
1343 	int ctrl_len;
1344 
1345 	ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr);
1346 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1347 	if (!buf)
1348 		return -ENOMEM;
1349 
1350 	p = CTRL_BODY(buf);
1351 	ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT);
1352 	ceph_encode_entity_addr(&p, &con->peer_addr);
1353 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1354 
1355 	return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len,
1356 				 NULL, 0, true);
1357 }
1358 
1359 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */
1360 #define AUTH_BUF_LEN	(512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN)
1361 
prepare_auth_request(struct ceph_connection * con)1362 static int prepare_auth_request(struct ceph_connection *con)
1363 {
1364 	void *authorizer, *authorizer_copy;
1365 	int ctrl_len, authorizer_len;
1366 	void *buf;
1367 	int ret;
1368 
1369 	ctrl_len = AUTH_BUF_LEN;
1370 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1371 	if (!buf)
1372 		return -ENOMEM;
1373 
1374 	mutex_unlock(&con->mutex);
1375 	ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len,
1376 					 &authorizer, &authorizer_len);
1377 	mutex_lock(&con->mutex);
1378 	if (con->state != CEPH_CON_S_V2_HELLO) {
1379 		dout("%s con %p state changed to %d\n", __func__, con,
1380 		     con->state);
1381 		return -EAGAIN;
1382 	}
1383 
1384 	dout("%s con %p get_auth_request ret %d\n", __func__, con, ret);
1385 	if (ret)
1386 		return ret;
1387 
1388 	authorizer_copy = alloc_conn_buf(con, authorizer_len);
1389 	if (!authorizer_copy)
1390 		return -ENOMEM;
1391 
1392 	memcpy(authorizer_copy, authorizer, authorizer_len);
1393 
1394 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len,
1395 				 authorizer_copy, authorizer_len, true);
1396 }
1397 
prepare_auth_request_more(struct ceph_connection * con,void * reply,int reply_len)1398 static int prepare_auth_request_more(struct ceph_connection *con,
1399 				     void *reply, int reply_len)
1400 {
1401 	int ctrl_len, authorizer_len;
1402 	void *authorizer;
1403 	void *buf;
1404 	int ret;
1405 
1406 	ctrl_len = AUTH_BUF_LEN;
1407 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1408 	if (!buf)
1409 		return -ENOMEM;
1410 
1411 	mutex_unlock(&con->mutex);
1412 	ret = con->ops->handle_auth_reply_more(con, reply, reply_len,
1413 					       CTRL_BODY(buf), &ctrl_len,
1414 					       &authorizer, &authorizer_len);
1415 	mutex_lock(&con->mutex);
1416 	if (con->state != CEPH_CON_S_V2_AUTH) {
1417 		dout("%s con %p state changed to %d\n", __func__, con,
1418 		     con->state);
1419 		return -EAGAIN;
1420 	}
1421 
1422 	dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret);
1423 	if (ret)
1424 		return ret;
1425 
1426 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf,
1427 				 ctrl_len, authorizer, authorizer_len, true);
1428 }
1429 
prepare_auth_signature(struct ceph_connection * con)1430 static int prepare_auth_signature(struct ceph_connection *con)
1431 {
1432 	void *buf;
1433 
1434 	buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE,
1435 						  con_secure(con)));
1436 	if (!buf)
1437 		return -ENOMEM;
1438 
1439 	con_hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt,
1440 			CTRL_BODY(buf));
1441 
1442 	return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf,
1443 			       SHA256_DIGEST_SIZE);
1444 }
1445 
prepare_client_ident(struct ceph_connection * con)1446 static int prepare_client_ident(struct ceph_connection *con)
1447 {
1448 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1449 	struct ceph_client *client = from_msgr(con->msgr);
1450 	u64 global_id = ceph_client_gid(client);
1451 	void *buf, *p;
1452 	int ctrl_len;
1453 
1454 	WARN_ON(con->v2.server_cookie);
1455 	WARN_ON(con->v2.connect_seq);
1456 	WARN_ON(con->v2.peer_global_seq);
1457 
1458 	if (!con->v2.client_cookie) {
1459 		do {
1460 			get_random_bytes(&con->v2.client_cookie,
1461 					 sizeof(con->v2.client_cookie));
1462 		} while (!con->v2.client_cookie);
1463 		dout("%s con %p generated cookie 0x%llx\n", __func__, con,
1464 		     con->v2.client_cookie);
1465 	} else {
1466 		dout("%s con %p cookie already set 0x%llx\n", __func__, con,
1467 		     con->v2.client_cookie);
1468 	}
1469 
1470 	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",
1471 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1472 	     ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce),
1473 	     global_id, con->v2.global_seq, client->supported_features,
1474 	     client->required_features, con->v2.client_cookie);
1475 
1476 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) +
1477 		   ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8;
1478 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1479 	if (!buf)
1480 		return -ENOMEM;
1481 
1482 	p = CTRL_BODY(buf);
1483 	ceph_encode_8(&p, 2);  /* addrvec marker */
1484 	ceph_encode_32(&p, 1);  /* addr_cnt */
1485 	ceph_encode_entity_addr(&p, my_addr);
1486 	ceph_encode_entity_addr(&p, &con->peer_addr);
1487 	ceph_encode_64(&p, global_id);
1488 	ceph_encode_64(&p, con->v2.global_seq);
1489 	ceph_encode_64(&p, client->supported_features);
1490 	ceph_encode_64(&p, client->required_features);
1491 	ceph_encode_64(&p, 0);  /* flags */
1492 	ceph_encode_64(&p, con->v2.client_cookie);
1493 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1494 
1495 	return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len);
1496 }
1497 
prepare_session_reconnect(struct ceph_connection * con)1498 static int prepare_session_reconnect(struct ceph_connection *con)
1499 {
1500 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1501 	void *buf, *p;
1502 	int ctrl_len;
1503 
1504 	WARN_ON(!con->v2.client_cookie);
1505 	WARN_ON(!con->v2.server_cookie);
1506 	WARN_ON(!con->v2.connect_seq);
1507 	WARN_ON(!con->v2.peer_global_seq);
1508 
1509 	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",
1510 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1511 	     con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq,
1512 	     con->v2.connect_seq, con->in_seq);
1513 
1514 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8;
1515 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1516 	if (!buf)
1517 		return -ENOMEM;
1518 
1519 	p = CTRL_BODY(buf);
1520 	ceph_encode_8(&p, 2);  /* entity_addrvec_t marker */
1521 	ceph_encode_32(&p, 1);  /* my_addrs len */
1522 	ceph_encode_entity_addr(&p, my_addr);
1523 	ceph_encode_64(&p, con->v2.client_cookie);
1524 	ceph_encode_64(&p, con->v2.server_cookie);
1525 	ceph_encode_64(&p, con->v2.global_seq);
1526 	ceph_encode_64(&p, con->v2.connect_seq);
1527 	ceph_encode_64(&p, con->in_seq);
1528 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1529 
1530 	return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len);
1531 }
1532 
prepare_keepalive2(struct ceph_connection * con)1533 static int prepare_keepalive2(struct ceph_connection *con)
1534 {
1535 	struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
1536 	struct timespec64 now;
1537 
1538 	ktime_get_real_ts64(&now);
1539 	dout("%s con %p timestamp %ptSp\n", __func__, con, &now);
1540 
1541 	ceph_encode_timespec64(ts, &now);
1542 
1543 	reset_out_kvecs(con);
1544 	return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf,
1545 			       sizeof(struct ceph_timespec));
1546 }
1547 
prepare_ack(struct ceph_connection * con)1548 static int prepare_ack(struct ceph_connection *con)
1549 {
1550 	void *p;
1551 
1552 	dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1553 	     con->in_seq_acked, con->in_seq);
1554 	con->in_seq_acked = con->in_seq;
1555 
1556 	p = CTRL_BODY(con->v2.out_buf);
1557 	ceph_encode_64(&p, con->in_seq_acked);
1558 
1559 	reset_out_kvecs(con);
1560 	return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8);
1561 }
1562 
prepare_epilogue_plain(struct ceph_connection * con,struct ceph_msg * msg,bool aborted)1563 static void prepare_epilogue_plain(struct ceph_connection *con,
1564 				   struct ceph_msg *msg, bool aborted)
1565 {
1566 	dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con,
1567 	     msg, aborted, con->v2.out_epil.front_crc,
1568 	     con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc);
1569 
1570 	encode_epilogue_plain(con, aborted);
1571 	add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN);
1572 }
1573 
1574 /*
1575  * For "used" empty segments, crc is -1.  For unused (trailing)
1576  * segments, crc is 0.
1577  */
prepare_message_plain(struct ceph_connection * con,struct ceph_msg * msg)1578 static void prepare_message_plain(struct ceph_connection *con,
1579 				  struct ceph_msg *msg)
1580 {
1581 	prepare_head_plain(con, con->v2.out_buf,
1582 			   sizeof(struct ceph_msg_header2), NULL, 0, false);
1583 
1584 	if (!front_len(msg) && !middle_len(msg)) {
1585 		if (!data_len(msg)) {
1586 			/*
1587 			 * Empty message: once the head is written,
1588 			 * we are done -- there is no epilogue.
1589 			 */
1590 			con->v2.out_state = OUT_S_FINISH_MESSAGE;
1591 			return;
1592 		}
1593 
1594 		con->v2.out_epil.front_crc = -1;
1595 		con->v2.out_epil.middle_crc = -1;
1596 		con->v2.out_state = OUT_S_QUEUE_DATA;
1597 		return;
1598 	}
1599 
1600 	if (front_len(msg)) {
1601 		con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base,
1602 						    front_len(msg));
1603 		add_out_kvec(con, msg->front.iov_base, front_len(msg));
1604 	} else {
1605 		/* middle (at least) is there, checked above */
1606 		con->v2.out_epil.front_crc = -1;
1607 	}
1608 
1609 	if (middle_len(msg)) {
1610 		con->v2.out_epil.middle_crc =
1611 			crc32c(-1, msg->middle->vec.iov_base, middle_len(msg));
1612 		add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1613 	} else {
1614 		con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0;
1615 	}
1616 
1617 	if (data_len(msg)) {
1618 		con->v2.out_state = OUT_S_QUEUE_DATA;
1619 	} else {
1620 		con->v2.out_epil.data_crc = 0;
1621 		prepare_epilogue_plain(con, msg, false);
1622 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1623 	}
1624 }
1625 
1626 /*
1627  * Unfortunately the kernel crypto API doesn't support streaming
1628  * (piecewise) operation for AEAD algorithms, so we can't get away
1629  * with a fixed size buffer and a couple sgs.  Instead, we have to
1630  * allocate pages for the entire tail of the message (currently up
1631  * to ~32M) and two sgs arrays (up to ~256K each)...
1632  */
prepare_message_secure(struct ceph_connection * con,struct ceph_msg * msg)1633 static int prepare_message_secure(struct ceph_connection *con,
1634 				  struct ceph_msg *msg)
1635 {
1636 	void *zerop = page_address(ceph_zero_page);
1637 	struct sg_table enc_sgt = {};
1638 	struct sg_table sgt = {};
1639 	struct page **enc_pages;
1640 	int enc_page_cnt;
1641 	int tail_len;
1642 	int ret;
1643 
1644 	ret = prepare_head_secure_small(con, con->v2.out_buf,
1645 					sizeof(struct ceph_msg_header2));
1646 	if (ret)
1647 		return ret;
1648 
1649 	tail_len = tail_onwire_len(msg, true);
1650 	if (!tail_len) {
1651 		/*
1652 		 * Empty message: once the head is written,
1653 		 * we are done -- there is no epilogue.
1654 		 */
1655 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1656 		return 0;
1657 	}
1658 
1659 	encode_epilogue_secure(con, false);
1660 	ret = setup_message_sgs(&sgt, msg, zerop, zerop, zerop,
1661 				&con->v2.out_epil, NULL, 0, false);
1662 	if (ret)
1663 		goto out;
1664 
1665 	enc_page_cnt = calc_pages_for(0, tail_len);
1666 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1667 	if (IS_ERR(enc_pages)) {
1668 		ret = PTR_ERR(enc_pages);
1669 		goto out;
1670 	}
1671 
1672 	WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt);
1673 	con->v2.out_enc_pages = enc_pages;
1674 	con->v2.out_enc_page_cnt = enc_page_cnt;
1675 	con->v2.out_enc_resid = tail_len;
1676 	con->v2.out_enc_i = 0;
1677 
1678 	ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt,
1679 					0, tail_len, GFP_NOIO);
1680 	if (ret)
1681 		goto out;
1682 
1683 	ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl,
1684 			tail_len - CEPH_GCM_TAG_LEN);
1685 	if (ret)
1686 		goto out;
1687 
1688 	dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con,
1689 	     msg, sgt.orig_nents, enc_page_cnt);
1690 	con->v2.out_state = OUT_S_QUEUE_ENC_PAGE;
1691 
1692 out:
1693 	sg_free_table(&sgt);
1694 	sg_free_table(&enc_sgt);
1695 	return ret;
1696 }
1697 
prepare_message(struct ceph_connection * con,struct ceph_msg * msg)1698 static int prepare_message(struct ceph_connection *con, struct ceph_msg *msg)
1699 {
1700 	int lens[] = {
1701 		sizeof(struct ceph_msg_header2),
1702 		front_len(msg),
1703 		middle_len(msg),
1704 		data_len(msg)
1705 	};
1706 	struct ceph_frame_desc desc;
1707 	int ret;
1708 
1709 	dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con,
1710 	     msg, lens[0], lens[1], lens[2], lens[3]);
1711 
1712 	if (con->in_seq > con->in_seq_acked) {
1713 		dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1714 		     con->in_seq_acked, con->in_seq);
1715 		con->in_seq_acked = con->in_seq;
1716 	}
1717 
1718 	reset_out_kvecs(con);
1719 	init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4);
1720 	encode_preamble(&desc, con->v2.out_buf);
1721 	fill_header2(CTRL_BODY(con->v2.out_buf), &msg->hdr,
1722 		     con->in_seq_acked);
1723 
1724 	if (con_secure(con)) {
1725 		ret = prepare_message_secure(con, msg);
1726 		if (ret)
1727 			return ret;
1728 	} else {
1729 		prepare_message_plain(con, msg);
1730 	}
1731 
1732 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1733 	return 0;
1734 }
1735 
prepare_read_banner_prefix(struct ceph_connection * con)1736 static int prepare_read_banner_prefix(struct ceph_connection *con)
1737 {
1738 	void *buf;
1739 
1740 	buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN);
1741 	if (!buf)
1742 		return -ENOMEM;
1743 
1744 	reset_in_kvecs(con);
1745 	add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1746 	add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1747 	con->state = CEPH_CON_S_V2_BANNER_PREFIX;
1748 	return 0;
1749 }
1750 
prepare_read_banner_payload(struct ceph_connection * con,int payload_len)1751 static int prepare_read_banner_payload(struct ceph_connection *con,
1752 				       int payload_len)
1753 {
1754 	void *buf;
1755 
1756 	buf = alloc_conn_buf(con, payload_len);
1757 	if (!buf)
1758 		return -ENOMEM;
1759 
1760 	reset_in_kvecs(con);
1761 	add_in_kvec(con, buf, payload_len);
1762 	add_in_sign_kvec(con, buf, payload_len);
1763 	con->state = CEPH_CON_S_V2_BANNER_PAYLOAD;
1764 	return 0;
1765 }
1766 
prepare_read_preamble(struct ceph_connection * con)1767 static void prepare_read_preamble(struct ceph_connection *con)
1768 {
1769 	reset_in_kvecs(con);
1770 	add_in_kvec(con, con->v2.in_buf,
1771 		    con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN :
1772 				      CEPH_PREAMBLE_PLAIN_LEN);
1773 	con->v2.in_state = IN_S_HANDLE_PREAMBLE;
1774 }
1775 
prepare_read_control(struct ceph_connection * con)1776 static int prepare_read_control(struct ceph_connection *con)
1777 {
1778 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1779 	int head_len;
1780 	void *buf;
1781 
1782 	reset_in_kvecs(con);
1783 	if (con->state == CEPH_CON_S_V2_HELLO ||
1784 	    con->state == CEPH_CON_S_V2_AUTH) {
1785 		head_len = head_onwire_len(ctrl_len, false);
1786 		buf = alloc_conn_buf(con, head_len);
1787 		if (!buf)
1788 			return -ENOMEM;
1789 
1790 		/* preserve preamble */
1791 		memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN);
1792 
1793 		add_in_kvec(con, CTRL_BODY(buf), ctrl_len);
1794 		add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN);
1795 		add_in_sign_kvec(con, buf, head_len);
1796 	} else {
1797 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
1798 			buf = alloc_conn_buf(con, ctrl_len);
1799 			if (!buf)
1800 				return -ENOMEM;
1801 
1802 			add_in_kvec(con, buf, ctrl_len);
1803 		} else {
1804 			add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len);
1805 		}
1806 		add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN);
1807 	}
1808 	con->v2.in_state = IN_S_HANDLE_CONTROL;
1809 	return 0;
1810 }
1811 
prepare_read_control_remainder(struct ceph_connection * con)1812 static int prepare_read_control_remainder(struct ceph_connection *con)
1813 {
1814 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1815 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1816 	void *buf;
1817 
1818 	buf = alloc_conn_buf(con, ctrl_len);
1819 	if (!buf)
1820 		return -ENOMEM;
1821 
1822 	memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN);
1823 
1824 	reset_in_kvecs(con);
1825 	add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len);
1826 	add_in_kvec(con, con->v2.in_buf,
1827 		    padding_len(rem_len) + CEPH_GCM_TAG_LEN);
1828 	con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER;
1829 	return 0;
1830 }
1831 
prepare_read_data(struct ceph_connection * con)1832 static int prepare_read_data(struct ceph_connection *con)
1833 {
1834 	struct bio_vec bv;
1835 
1836 	con->in_data_crc = -1;
1837 	ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg,
1838 				  data_len(con->in_msg));
1839 
1840 	get_bvec_at(&con->v2.in_cursor, &bv);
1841 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1842 		if (unlikely(!con->bounce_page)) {
1843 			con->bounce_page = alloc_page(GFP_NOIO);
1844 			if (!con->bounce_page) {
1845 				pr_err("failed to allocate bounce page\n");
1846 				return -ENOMEM;
1847 			}
1848 		}
1849 
1850 		bv.bv_page = con->bounce_page;
1851 		bv.bv_offset = 0;
1852 	}
1853 	set_in_bvec(con, &bv);
1854 	con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT;
1855 	return 0;
1856 }
1857 
prepare_read_data_cont(struct ceph_connection * con)1858 static void prepare_read_data_cont(struct ceph_connection *con)
1859 {
1860 	struct bio_vec bv;
1861 
1862 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1863 		con->in_data_crc = crc32c(con->in_data_crc,
1864 					  page_address(con->bounce_page),
1865 					  con->v2.in_bvec.bv_len);
1866 
1867 		get_bvec_at(&con->v2.in_cursor, &bv);
1868 		memcpy_to_page(bv.bv_page, bv.bv_offset,
1869 			       page_address(con->bounce_page),
1870 			       con->v2.in_bvec.bv_len);
1871 	} else {
1872 		con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1873 						    con->v2.in_bvec.bv_page,
1874 						    con->v2.in_bvec.bv_offset,
1875 						    con->v2.in_bvec.bv_len);
1876 	}
1877 
1878 	ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len);
1879 	if (con->v2.in_cursor.total_resid) {
1880 		get_bvec_at(&con->v2.in_cursor, &bv);
1881 		if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1882 			bv.bv_page = con->bounce_page;
1883 			bv.bv_offset = 0;
1884 		}
1885 		set_in_bvec(con, &bv);
1886 		WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT);
1887 		return;
1888 	}
1889 
1890 	/*
1891 	 * We've read all data.  Prepare to read epilogue.
1892 	 */
1893 	reset_in_kvecs(con);
1894 	add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1895 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1896 }
1897 
prepare_sparse_read_cont(struct ceph_connection * con)1898 static int prepare_sparse_read_cont(struct ceph_connection *con)
1899 {
1900 	int ret;
1901 	struct bio_vec bv;
1902 	char *buf = NULL;
1903 	struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
1904 
1905 	WARN_ON(con->v2.in_state != IN_S_PREPARE_SPARSE_DATA_CONT);
1906 
1907 	if (iov_iter_is_bvec(&con->v2.in_iter)) {
1908 		if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1909 			con->in_data_crc = crc32c(con->in_data_crc,
1910 						  page_address(con->bounce_page),
1911 						  con->v2.in_bvec.bv_len);
1912 			get_bvec_at(cursor, &bv);
1913 			memcpy_to_page(bv.bv_page, bv.bv_offset,
1914 				       page_address(con->bounce_page),
1915 				       con->v2.in_bvec.bv_len);
1916 		} else {
1917 			con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1918 							    con->v2.in_bvec.bv_page,
1919 							    con->v2.in_bvec.bv_offset,
1920 							    con->v2.in_bvec.bv_len);
1921 		}
1922 
1923 		ceph_msg_data_advance(cursor, con->v2.in_bvec.bv_len);
1924 		cursor->sr_resid -= con->v2.in_bvec.bv_len;
1925 		dout("%s: advance by 0x%x sr_resid 0x%x\n", __func__,
1926 		     con->v2.in_bvec.bv_len, cursor->sr_resid);
1927 		WARN_ON_ONCE(cursor->sr_resid > cursor->total_resid);
1928 		if (cursor->sr_resid) {
1929 			get_bvec_at(cursor, &bv);
1930 			if (bv.bv_len > cursor->sr_resid)
1931 				bv.bv_len = cursor->sr_resid;
1932 			if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1933 				bv.bv_page = con->bounce_page;
1934 				bv.bv_offset = 0;
1935 			}
1936 			set_in_bvec(con, &bv);
1937 			con->v2.data_len_remain -= bv.bv_len;
1938 			return 0;
1939 		}
1940 	} else if (iov_iter_is_kvec(&con->v2.in_iter)) {
1941 		/* On first call, we have no kvec so don't compute crc */
1942 		if (con->v2.in_kvec_cnt) {
1943 			WARN_ON_ONCE(con->v2.in_kvec_cnt > 1);
1944 			con->in_data_crc = crc32c(con->in_data_crc,
1945 						  con->v2.in_kvecs[0].iov_base,
1946 						  con->v2.in_kvecs[0].iov_len);
1947 		}
1948 	} else {
1949 		return -EIO;
1950 	}
1951 
1952 	/* get next extent */
1953 	ret = con->ops->sparse_read(con, cursor, &buf);
1954 	if (ret <= 0) {
1955 		if (ret < 0)
1956 			return ret;
1957 
1958 		reset_in_kvecs(con);
1959 		add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1960 		con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1961 		return 0;
1962 	}
1963 
1964 	if (buf) {
1965 		/* receive into buffer */
1966 		reset_in_kvecs(con);
1967 		add_in_kvec(con, buf, ret);
1968 		con->v2.data_len_remain -= ret;
1969 		return 0;
1970 	}
1971 
1972 	if (ret > cursor->total_resid) {
1973 		pr_warn("%s: ret 0x%x total_resid 0x%zx resid 0x%zx\n",
1974 			__func__, ret, cursor->total_resid, cursor->resid);
1975 		return -EIO;
1976 	}
1977 	get_bvec_at(cursor, &bv);
1978 	if (bv.bv_len > cursor->sr_resid)
1979 		bv.bv_len = cursor->sr_resid;
1980 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1981 		if (unlikely(!con->bounce_page)) {
1982 			con->bounce_page = alloc_page(GFP_NOIO);
1983 			if (!con->bounce_page) {
1984 				pr_err("failed to allocate bounce page\n");
1985 				return -ENOMEM;
1986 			}
1987 		}
1988 
1989 		bv.bv_page = con->bounce_page;
1990 		bv.bv_offset = 0;
1991 	}
1992 	set_in_bvec(con, &bv);
1993 	con->v2.data_len_remain -= ret;
1994 	return ret;
1995 }
1996 
prepare_sparse_read_data(struct ceph_connection * con)1997 static int prepare_sparse_read_data(struct ceph_connection *con)
1998 {
1999 	struct ceph_msg *msg = con->in_msg;
2000 
2001 	dout("%s: starting sparse read\n", __func__);
2002 
2003 	if (WARN_ON_ONCE(!con->ops->sparse_read))
2004 		return -EOPNOTSUPP;
2005 
2006 	if (!con_secure(con))
2007 		con->in_data_crc = -1;
2008 
2009 	ceph_msg_data_cursor_init(&con->v2.in_cursor, msg,
2010 				  msg->sparse_read_total);
2011 
2012 	reset_in_kvecs(con);
2013 	con->v2.in_state = IN_S_PREPARE_SPARSE_DATA_CONT;
2014 	con->v2.data_len_remain = data_len(msg);
2015 	return prepare_sparse_read_cont(con);
2016 }
2017 
prepare_read_tail_plain(struct ceph_connection * con)2018 static int prepare_read_tail_plain(struct ceph_connection *con)
2019 {
2020 	struct ceph_msg *msg = con->in_msg;
2021 
2022 	if (!front_len(msg) && !middle_len(msg)) {
2023 		WARN_ON(!data_len(msg));
2024 		return prepare_read_data(con);
2025 	}
2026 
2027 	reset_in_kvecs(con);
2028 	if (front_len(msg)) {
2029 		add_in_kvec(con, msg->front.iov_base, front_len(msg));
2030 		WARN_ON(msg->front.iov_len != front_len(msg));
2031 	}
2032 	if (middle_len(msg)) {
2033 		add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
2034 		WARN_ON(msg->middle->vec.iov_len != middle_len(msg));
2035 	}
2036 
2037 	if (data_len(msg)) {
2038 		if (msg->sparse_read_total)
2039 			con->v2.in_state = IN_S_PREPARE_SPARSE_DATA;
2040 		else
2041 			con->v2.in_state = IN_S_PREPARE_READ_DATA;
2042 	} else {
2043 		add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
2044 		con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2045 	}
2046 	return 0;
2047 }
2048 
prepare_read_enc_page(struct ceph_connection * con)2049 static void prepare_read_enc_page(struct ceph_connection *con)
2050 {
2051 	struct bio_vec bv;
2052 
2053 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i,
2054 	     con->v2.in_enc_resid);
2055 	WARN_ON(!con->v2.in_enc_resid);
2056 
2057 	bvec_set_page(&bv, con->v2.in_enc_pages[con->v2.in_enc_i],
2058 		      min(con->v2.in_enc_resid, (int)PAGE_SIZE), 0);
2059 
2060 	set_in_bvec(con, &bv);
2061 	con->v2.in_enc_i++;
2062 	con->v2.in_enc_resid -= bv.bv_len;
2063 
2064 	if (con->v2.in_enc_resid) {
2065 		con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE;
2066 		return;
2067 	}
2068 
2069 	/*
2070 	 * We are set to read the last piece of ciphertext (ending
2071 	 * with epilogue) + auth tag.
2072 	 */
2073 	WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt);
2074 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2075 }
2076 
prepare_read_tail_secure(struct ceph_connection * con)2077 static int prepare_read_tail_secure(struct ceph_connection *con)
2078 {
2079 	struct page **enc_pages;
2080 	int enc_page_cnt;
2081 	int tail_len;
2082 
2083 	tail_len = tail_onwire_len(con->in_msg, true);
2084 	WARN_ON(!tail_len);
2085 
2086 	enc_page_cnt = calc_pages_for(0, tail_len);
2087 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
2088 	if (IS_ERR(enc_pages))
2089 		return PTR_ERR(enc_pages);
2090 
2091 	WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt);
2092 	con->v2.in_enc_pages = enc_pages;
2093 	con->v2.in_enc_page_cnt = enc_page_cnt;
2094 	con->v2.in_enc_resid = tail_len;
2095 	con->v2.in_enc_i = 0;
2096 
2097 	prepare_read_enc_page(con);
2098 	return 0;
2099 }
2100 
__finish_skip(struct ceph_connection * con)2101 static void __finish_skip(struct ceph_connection *con)
2102 {
2103 	con->in_seq++;
2104 	prepare_read_preamble(con);
2105 }
2106 
prepare_skip_message(struct ceph_connection * con)2107 static void prepare_skip_message(struct ceph_connection *con)
2108 {
2109 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2110 	int tail_len;
2111 
2112 	dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1],
2113 	     desc->fd_lens[2], desc->fd_lens[3]);
2114 
2115 	tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2],
2116 				     desc->fd_lens[3], con_secure(con));
2117 	if (!tail_len) {
2118 		__finish_skip(con);
2119 	} else {
2120 		set_in_skip(con, tail_len);
2121 		con->v2.in_state = IN_S_FINISH_SKIP;
2122 	}
2123 }
2124 
process_banner_prefix(struct ceph_connection * con)2125 static int process_banner_prefix(struct ceph_connection *con)
2126 {
2127 	int payload_len;
2128 	void *p;
2129 
2130 	WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN);
2131 
2132 	p = con->v2.in_kvecs[0].iov_base;
2133 	if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) {
2134 		if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN))
2135 			con->error_msg = "server is speaking msgr1 protocol";
2136 		else
2137 			con->error_msg = "protocol error, bad banner";
2138 		return -EINVAL;
2139 	}
2140 
2141 	p += CEPH_BANNER_V2_LEN;
2142 	payload_len = ceph_decode_16(&p);
2143 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2144 
2145 	return prepare_read_banner_payload(con, payload_len);
2146 }
2147 
process_banner_payload(struct ceph_connection * con)2148 static int process_banner_payload(struct ceph_connection *con)
2149 {
2150 	void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len;
2151 	u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES;
2152 	u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES;
2153 	u64 server_feat, server_req_feat;
2154 	void *p;
2155 	int ret;
2156 
2157 	p = con->v2.in_kvecs[0].iov_base;
2158 	ceph_decode_64_safe(&p, end, server_feat, bad);
2159 	ceph_decode_64_safe(&p, end, server_req_feat, bad);
2160 
2161 	dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n",
2162 	     __func__, con, server_feat, server_req_feat);
2163 
2164 	if (req_feat & ~server_feat) {
2165 		pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2166 		       server_feat, req_feat & ~server_feat);
2167 		con->error_msg = "missing required protocol features";
2168 		return -EINVAL;
2169 	}
2170 	if (server_req_feat & ~feat) {
2171 		pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2172 		       feat, server_req_feat & ~feat);
2173 		con->error_msg = "missing required protocol features";
2174 		return -EINVAL;
2175 	}
2176 
2177 	/* no reset_out_kvecs() as our banner may still be pending */
2178 	ret = prepare_hello(con);
2179 	if (ret) {
2180 		pr_err("prepare_hello failed: %d\n", ret);
2181 		return ret;
2182 	}
2183 
2184 	con->state = CEPH_CON_S_V2_HELLO;
2185 	prepare_read_preamble(con);
2186 	return 0;
2187 
2188 bad:
2189 	pr_err("failed to decode banner payload\n");
2190 	return -EINVAL;
2191 }
2192 
process_hello(struct ceph_connection * con,void * p,void * end)2193 static int process_hello(struct ceph_connection *con, void *p, void *end)
2194 {
2195 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
2196 	struct ceph_entity_addr addr_for_me;
2197 	u8 entity_type;
2198 	int ret;
2199 
2200 	if (con->state != CEPH_CON_S_V2_HELLO) {
2201 		con->error_msg = "protocol error, unexpected hello";
2202 		return -EINVAL;
2203 	}
2204 
2205 	ceph_decode_8_safe(&p, end, entity_type, bad);
2206 	ret = ceph_decode_entity_addr(&p, end, &addr_for_me);
2207 	if (ret) {
2208 		pr_err("failed to decode addr_for_me: %d\n", ret);
2209 		return ret;
2210 	}
2211 
2212 	dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con,
2213 	     entity_type, ceph_pr_addr(&addr_for_me));
2214 
2215 	if (entity_type != con->peer_name.type) {
2216 		pr_err("bad peer type, want %d, got %d\n",
2217 		       con->peer_name.type, entity_type);
2218 		con->error_msg = "wrong peer at address";
2219 		return -EINVAL;
2220 	}
2221 
2222 	/*
2223 	 * Set our address to the address our first peer (i.e. monitor)
2224 	 * sees that we are connecting from.  If we are behind some sort
2225 	 * of NAT and want to be identified by some private (not NATed)
2226 	 * address, ip option should be used.
2227 	 */
2228 	if (ceph_addr_is_blank(my_addr)) {
2229 		memcpy(&my_addr->in_addr, &addr_for_me.in_addr,
2230 		       sizeof(my_addr->in_addr));
2231 		ceph_addr_set_port(my_addr, 0);
2232 		dout("%s con %p set my addr %s, as seen by peer %s\n",
2233 		     __func__, con, ceph_pr_addr(my_addr),
2234 		     ceph_pr_addr(&con->peer_addr));
2235 	} else {
2236 		dout("%s con %p my addr already set %s\n",
2237 		     __func__, con, ceph_pr_addr(my_addr));
2238 	}
2239 
2240 	WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr));
2241 	WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY);
2242 	WARN_ON(!my_addr->nonce);
2243 
2244 	/* no reset_out_kvecs() as our hello may still be pending */
2245 	ret = prepare_auth_request(con);
2246 	if (ret) {
2247 		if (ret != -EAGAIN)
2248 			pr_err("prepare_auth_request failed: %d\n", ret);
2249 		return ret;
2250 	}
2251 
2252 	con->state = CEPH_CON_S_V2_AUTH;
2253 	return 0;
2254 
2255 bad:
2256 	pr_err("failed to decode hello\n");
2257 	return -EINVAL;
2258 }
2259 
process_auth_bad_method(struct ceph_connection * con,void * p,void * end)2260 static int process_auth_bad_method(struct ceph_connection *con,
2261 				   void *p, void *end)
2262 {
2263 	int allowed_protos[8], allowed_modes[8];
2264 	int allowed_proto_cnt, allowed_mode_cnt;
2265 	int used_proto, result;
2266 	int ret;
2267 	int i;
2268 
2269 	if (con->state != CEPH_CON_S_V2_AUTH) {
2270 		con->error_msg = "protocol error, unexpected auth_bad_method";
2271 		return -EINVAL;
2272 	}
2273 
2274 	ceph_decode_32_safe(&p, end, used_proto, bad);
2275 	ceph_decode_32_safe(&p, end, result, bad);
2276 	dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto,
2277 	     result);
2278 
2279 	ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad);
2280 	if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) {
2281 		pr_err("allowed_protos too big %d\n", allowed_proto_cnt);
2282 		return -EINVAL;
2283 	}
2284 	for (i = 0; i < allowed_proto_cnt; i++) {
2285 		ceph_decode_32_safe(&p, end, allowed_protos[i], bad);
2286 		dout("%s con %p allowed_protos[%d] %d\n", __func__, con,
2287 		     i, allowed_protos[i]);
2288 	}
2289 
2290 	ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad);
2291 	if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) {
2292 		pr_err("allowed_modes too big %d\n", allowed_mode_cnt);
2293 		return -EINVAL;
2294 	}
2295 	for (i = 0; i < allowed_mode_cnt; i++) {
2296 		ceph_decode_32_safe(&p, end, allowed_modes[i], bad);
2297 		dout("%s con %p allowed_modes[%d] %d\n", __func__, con,
2298 		     i, allowed_modes[i]);
2299 	}
2300 
2301 	mutex_unlock(&con->mutex);
2302 	ret = con->ops->handle_auth_bad_method(con, used_proto, result,
2303 					       allowed_protos,
2304 					       allowed_proto_cnt,
2305 					       allowed_modes,
2306 					       allowed_mode_cnt);
2307 	mutex_lock(&con->mutex);
2308 	if (con->state != CEPH_CON_S_V2_AUTH) {
2309 		dout("%s con %p state changed to %d\n", __func__, con,
2310 		     con->state);
2311 		return -EAGAIN;
2312 	}
2313 
2314 	dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret);
2315 	return ret;
2316 
2317 bad:
2318 	pr_err("failed to decode auth_bad_method\n");
2319 	return -EINVAL;
2320 }
2321 
process_auth_reply_more(struct ceph_connection * con,void * p,void * end)2322 static int process_auth_reply_more(struct ceph_connection *con,
2323 				   void *p, void *end)
2324 {
2325 	int payload_len;
2326 	int ret;
2327 
2328 	if (con->state != CEPH_CON_S_V2_AUTH) {
2329 		con->error_msg = "protocol error, unexpected auth_reply_more";
2330 		return -EINVAL;
2331 	}
2332 
2333 	ceph_decode_32_safe(&p, end, payload_len, bad);
2334 	ceph_decode_need(&p, end, payload_len, bad);
2335 
2336 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2337 
2338 	reset_out_kvecs(con);
2339 	ret = prepare_auth_request_more(con, p, payload_len);
2340 	if (ret) {
2341 		if (ret != -EAGAIN)
2342 			pr_err("prepare_auth_request_more failed: %d\n", ret);
2343 		return ret;
2344 	}
2345 
2346 	return 0;
2347 
2348 bad:
2349 	pr_err("failed to decode auth_reply_more\n");
2350 	return -EINVAL;
2351 }
2352 
2353 /*
2354  * Align con_secret to avoid GFP_ATOMIC allocation inside
2355  * crypto_aead_setkey() called from setup_crypto().  __aligned(16)
2356  * isn't guaranteed to work for stack objects, so do it by hand.
2357  */
process_auth_done(struct ceph_connection * con,void * p,void * end)2358 static int process_auth_done(struct ceph_connection *con, void *p, void *end)
2359 {
2360 	u8 session_key[CEPH_MAX_KEY_LEN];
2361 	u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16];
2362 	u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16);
2363 	int session_key_len, con_secret_len;
2364 	int payload_len;
2365 	u64 global_id;
2366 	int ret;
2367 
2368 	if (con->state != CEPH_CON_S_V2_AUTH) {
2369 		con->error_msg = "protocol error, unexpected auth_done";
2370 		return -EINVAL;
2371 	}
2372 
2373 	ceph_decode_64_safe(&p, end, global_id, bad);
2374 	ceph_decode_32_safe(&p, end, con->v2.con_mode, bad);
2375 
2376 	ceph_decode_32_safe(&p, end, payload_len, bad);
2377 	ceph_decode_need(&p, end, payload_len, bad);
2378 
2379 	dout("%s con %p global_id %llu con_mode %d payload_len %d\n",
2380 	     __func__, con, global_id, con->v2.con_mode, payload_len);
2381 
2382 	mutex_unlock(&con->mutex);
2383 	session_key_len = 0;
2384 	con_secret_len = 0;
2385 	ret = con->ops->handle_auth_done(con, global_id, p, payload_len,
2386 					 session_key, &session_key_len,
2387 					 con_secret, &con_secret_len);
2388 	mutex_lock(&con->mutex);
2389 	if (con->state != CEPH_CON_S_V2_AUTH) {
2390 		dout("%s con %p state changed to %d\n", __func__, con,
2391 		     con->state);
2392 		ret = -EAGAIN;
2393 		goto out;
2394 	}
2395 
2396 	dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret);
2397 	if (ret)
2398 		goto out;
2399 
2400 	ret = setup_crypto(con, session_key, session_key_len, con_secret,
2401 			   con_secret_len);
2402 	if (ret)
2403 		goto out;
2404 
2405 	reset_out_kvecs(con);
2406 	ret = prepare_auth_signature(con);
2407 	if (ret) {
2408 		pr_err("prepare_auth_signature failed: %d\n", ret);
2409 		goto out;
2410 	}
2411 
2412 	con->state = CEPH_CON_S_V2_AUTH_SIGNATURE;
2413 
2414 out:
2415 	memzero_explicit(session_key, sizeof(session_key));
2416 	memzero_explicit(con_secret_buf, sizeof(con_secret_buf));
2417 	return ret;
2418 
2419 bad:
2420 	pr_err("failed to decode auth_done\n");
2421 	return -EINVAL;
2422 }
2423 
process_auth_signature(struct ceph_connection * con,void * p,void * end)2424 static int process_auth_signature(struct ceph_connection *con,
2425 				  void *p, void *end)
2426 {
2427 	u8 hmac[SHA256_DIGEST_SIZE];
2428 	int ret;
2429 
2430 	if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) {
2431 		con->error_msg = "protocol error, unexpected auth_signature";
2432 		return -EINVAL;
2433 	}
2434 
2435 	con_hmac_sha256(con, con->v2.out_sign_kvecs, con->v2.out_sign_kvec_cnt,
2436 			hmac);
2437 
2438 	ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad);
2439 	if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) {
2440 		con->error_msg = "integrity error, bad auth signature";
2441 		return -EBADMSG;
2442 	}
2443 
2444 	dout("%s con %p auth signature ok\n", __func__, con);
2445 
2446 	/* no reset_out_kvecs() as our auth_signature may still be pending */
2447 	if (!con->v2.server_cookie) {
2448 		ret = prepare_client_ident(con);
2449 		if (ret) {
2450 			pr_err("prepare_client_ident failed: %d\n", ret);
2451 			return ret;
2452 		}
2453 
2454 		con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2455 	} else {
2456 		ret = prepare_session_reconnect(con);
2457 		if (ret) {
2458 			pr_err("prepare_session_reconnect failed: %d\n", ret);
2459 			return ret;
2460 		}
2461 
2462 		con->state = CEPH_CON_S_V2_SESSION_RECONNECT;
2463 	}
2464 
2465 	return 0;
2466 
2467 bad:
2468 	pr_err("failed to decode auth_signature\n");
2469 	return -EINVAL;
2470 }
2471 
process_server_ident(struct ceph_connection * con,void * p,void * end)2472 static int process_server_ident(struct ceph_connection *con,
2473 				void *p, void *end)
2474 {
2475 	struct ceph_client *client = from_msgr(con->msgr);
2476 	u64 features, required_features;
2477 	struct ceph_entity_addr addr;
2478 	u64 global_seq;
2479 	u64 global_id;
2480 	u64 cookie;
2481 	u64 flags;
2482 	int ret;
2483 
2484 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2485 		con->error_msg = "protocol error, unexpected server_ident";
2486 		return -EINVAL;
2487 	}
2488 
2489 	ret = ceph_decode_entity_addrvec(&p, end, true, &addr);
2490 	if (ret) {
2491 		pr_err("failed to decode server addrs: %d\n", ret);
2492 		return ret;
2493 	}
2494 
2495 	ceph_decode_64_safe(&p, end, global_id, bad);
2496 	ceph_decode_64_safe(&p, end, global_seq, bad);
2497 	ceph_decode_64_safe(&p, end, features, bad);
2498 	ceph_decode_64_safe(&p, end, required_features, bad);
2499 	ceph_decode_64_safe(&p, end, flags, bad);
2500 	ceph_decode_64_safe(&p, end, cookie, bad);
2501 
2502 	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",
2503 	     __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce),
2504 	     global_id, global_seq, features, required_features, flags, cookie);
2505 
2506 	/* is this who we intended to talk to? */
2507 	if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
2508 		pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
2509 		       ceph_pr_addr(&con->peer_addr),
2510 		       le32_to_cpu(con->peer_addr.nonce),
2511 		       ceph_pr_addr(&addr), le32_to_cpu(addr.nonce));
2512 		con->error_msg = "wrong peer at address";
2513 		return -EINVAL;
2514 	}
2515 
2516 	if (client->required_features & ~features) {
2517 		pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2518 		       features, client->required_features & ~features);
2519 		con->error_msg = "missing required protocol features";
2520 		return -EINVAL;
2521 	}
2522 
2523 	/*
2524 	 * Both name->type and name->num are set in ceph_con_open() but
2525 	 * name->num may be bogus in the initial monmap.  name->type is
2526 	 * verified in handle_hello().
2527 	 */
2528 	WARN_ON(!con->peer_name.type);
2529 	con->peer_name.num = cpu_to_le64(global_id);
2530 	con->v2.peer_global_seq = global_seq;
2531 	con->peer_features = features;
2532 	WARN_ON(required_features & ~client->supported_features);
2533 	con->v2.server_cookie = cookie;
2534 
2535 	if (flags & CEPH_MSG_CONNECT_LOSSY) {
2536 		ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
2537 		WARN_ON(con->v2.server_cookie);
2538 	} else {
2539 		WARN_ON(!con->v2.server_cookie);
2540 	}
2541 
2542 	clear_in_sign_kvecs(con);
2543 	clear_out_sign_kvecs(con);
2544 	free_conn_bufs(con);
2545 	con->delay = 0;  /* reset backoff memory */
2546 
2547 	con->state = CEPH_CON_S_OPEN;
2548 	con->v2.out_state = OUT_S_GET_NEXT;
2549 	return 0;
2550 
2551 bad:
2552 	pr_err("failed to decode server_ident\n");
2553 	return -EINVAL;
2554 }
2555 
process_ident_missing_features(struct ceph_connection * con,void * p,void * end)2556 static int process_ident_missing_features(struct ceph_connection *con,
2557 					  void *p, void *end)
2558 {
2559 	struct ceph_client *client = from_msgr(con->msgr);
2560 	u64 missing_features;
2561 
2562 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2563 		con->error_msg = "protocol error, unexpected ident_missing_features";
2564 		return -EINVAL;
2565 	}
2566 
2567 	ceph_decode_64_safe(&p, end, missing_features, bad);
2568 	pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2569 	       client->supported_features, missing_features);
2570 	con->error_msg = "missing required protocol features";
2571 	return -EINVAL;
2572 
2573 bad:
2574 	pr_err("failed to decode ident_missing_features\n");
2575 	return -EINVAL;
2576 }
2577 
process_session_reconnect_ok(struct ceph_connection * con,void * p,void * end)2578 static int process_session_reconnect_ok(struct ceph_connection *con,
2579 					void *p, void *end)
2580 {
2581 	u64 seq;
2582 
2583 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2584 		con->error_msg = "protocol error, unexpected session_reconnect_ok";
2585 		return -EINVAL;
2586 	}
2587 
2588 	ceph_decode_64_safe(&p, end, seq, bad);
2589 
2590 	dout("%s con %p seq %llu\n", __func__, con, seq);
2591 	ceph_con_discard_requeued(con, seq);
2592 
2593 	clear_in_sign_kvecs(con);
2594 	clear_out_sign_kvecs(con);
2595 	free_conn_bufs(con);
2596 	con->delay = 0;  /* reset backoff memory */
2597 
2598 	con->state = CEPH_CON_S_OPEN;
2599 	con->v2.out_state = OUT_S_GET_NEXT;
2600 	return 0;
2601 
2602 bad:
2603 	pr_err("failed to decode session_reconnect_ok\n");
2604 	return -EINVAL;
2605 }
2606 
process_session_retry(struct ceph_connection * con,void * p,void * end)2607 static int process_session_retry(struct ceph_connection *con,
2608 				 void *p, void *end)
2609 {
2610 	u64 connect_seq;
2611 	int ret;
2612 
2613 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2614 		con->error_msg = "protocol error, unexpected session_retry";
2615 		return -EINVAL;
2616 	}
2617 
2618 	ceph_decode_64_safe(&p, end, connect_seq, bad);
2619 
2620 	dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq);
2621 	WARN_ON(connect_seq <= con->v2.connect_seq);
2622 	con->v2.connect_seq = connect_seq + 1;
2623 
2624 	free_conn_bufs(con);
2625 
2626 	reset_out_kvecs(con);
2627 	ret = prepare_session_reconnect(con);
2628 	if (ret) {
2629 		pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret);
2630 		return ret;
2631 	}
2632 
2633 	return 0;
2634 
2635 bad:
2636 	pr_err("failed to decode session_retry\n");
2637 	return -EINVAL;
2638 }
2639 
process_session_retry_global(struct ceph_connection * con,void * p,void * end)2640 static int process_session_retry_global(struct ceph_connection *con,
2641 					void *p, void *end)
2642 {
2643 	u64 global_seq;
2644 	int ret;
2645 
2646 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2647 		con->error_msg = "protocol error, unexpected session_retry_global";
2648 		return -EINVAL;
2649 	}
2650 
2651 	ceph_decode_64_safe(&p, end, global_seq, bad);
2652 
2653 	dout("%s con %p global_seq %llu\n", __func__, con, global_seq);
2654 	WARN_ON(global_seq <= con->v2.global_seq);
2655 	con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq);
2656 
2657 	free_conn_bufs(con);
2658 
2659 	reset_out_kvecs(con);
2660 	ret = prepare_session_reconnect(con);
2661 	if (ret) {
2662 		pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret);
2663 		return ret;
2664 	}
2665 
2666 	return 0;
2667 
2668 bad:
2669 	pr_err("failed to decode session_retry_global\n");
2670 	return -EINVAL;
2671 }
2672 
process_session_reset(struct ceph_connection * con,void * p,void * end)2673 static int process_session_reset(struct ceph_connection *con,
2674 				 void *p, void *end)
2675 {
2676 	bool full;
2677 	int ret;
2678 
2679 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2680 		con->error_msg = "protocol error, unexpected session_reset";
2681 		return -EINVAL;
2682 	}
2683 
2684 	ceph_decode_8_safe(&p, end, full, bad);
2685 	if (!full) {
2686 		con->error_msg = "protocol error, bad session_reset";
2687 		return -EINVAL;
2688 	}
2689 
2690 	pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name),
2691 		ceph_pr_addr(&con->peer_addr));
2692 	ceph_con_reset_session(con);
2693 
2694 	mutex_unlock(&con->mutex);
2695 	if (con->ops->peer_reset)
2696 		con->ops->peer_reset(con);
2697 	mutex_lock(&con->mutex);
2698 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2699 		dout("%s con %p state changed to %d\n", __func__, con,
2700 		     con->state);
2701 		return -EAGAIN;
2702 	}
2703 
2704 	free_conn_bufs(con);
2705 
2706 	reset_out_kvecs(con);
2707 	ret = prepare_client_ident(con);
2708 	if (ret) {
2709 		pr_err("prepare_client_ident (rst) failed: %d\n", ret);
2710 		return ret;
2711 	}
2712 
2713 	con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2714 	return 0;
2715 
2716 bad:
2717 	pr_err("failed to decode session_reset\n");
2718 	return -EINVAL;
2719 }
2720 
process_keepalive2_ack(struct ceph_connection * con,void * p,void * end)2721 static int process_keepalive2_ack(struct ceph_connection *con,
2722 				  void *p, void *end)
2723 {
2724 	if (con->state != CEPH_CON_S_OPEN) {
2725 		con->error_msg = "protocol error, unexpected keepalive2_ack";
2726 		return -EINVAL;
2727 	}
2728 
2729 	ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad);
2730 	ceph_decode_timespec64(&con->last_keepalive_ack, p);
2731 
2732 	dout("%s con %p timestamp %ptSp\n", __func__, con, &con->last_keepalive_ack);
2733 
2734 	return 0;
2735 
2736 bad:
2737 	pr_err("failed to decode keepalive2_ack\n");
2738 	return -EINVAL;
2739 }
2740 
process_ack(struct ceph_connection * con,void * p,void * end)2741 static int process_ack(struct ceph_connection *con, void *p, void *end)
2742 {
2743 	u64 seq;
2744 
2745 	if (con->state != CEPH_CON_S_OPEN) {
2746 		con->error_msg = "protocol error, unexpected ack";
2747 		return -EINVAL;
2748 	}
2749 
2750 	ceph_decode_64_safe(&p, end, seq, bad);
2751 
2752 	dout("%s con %p seq %llu\n", __func__, con, seq);
2753 	ceph_con_discard_sent(con, seq);
2754 	return 0;
2755 
2756 bad:
2757 	pr_err("failed to decode ack\n");
2758 	return -EINVAL;
2759 }
2760 
process_control(struct ceph_connection * con,void * p,void * end)2761 static int process_control(struct ceph_connection *con, void *p, void *end)
2762 {
2763 	int tag = con->v2.in_desc.fd_tag;
2764 	int ret;
2765 
2766 	dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p));
2767 
2768 	switch (tag) {
2769 	case FRAME_TAG_HELLO:
2770 		ret = process_hello(con, p, end);
2771 		break;
2772 	case FRAME_TAG_AUTH_BAD_METHOD:
2773 		ret = process_auth_bad_method(con, p, end);
2774 		break;
2775 	case FRAME_TAG_AUTH_REPLY_MORE:
2776 		ret = process_auth_reply_more(con, p, end);
2777 		break;
2778 	case FRAME_TAG_AUTH_DONE:
2779 		ret = process_auth_done(con, p, end);
2780 		break;
2781 	case FRAME_TAG_AUTH_SIGNATURE:
2782 		ret = process_auth_signature(con, p, end);
2783 		break;
2784 	case FRAME_TAG_SERVER_IDENT:
2785 		ret = process_server_ident(con, p, end);
2786 		break;
2787 	case FRAME_TAG_IDENT_MISSING_FEATURES:
2788 		ret = process_ident_missing_features(con, p, end);
2789 		break;
2790 	case FRAME_TAG_SESSION_RECONNECT_OK:
2791 		ret = process_session_reconnect_ok(con, p, end);
2792 		break;
2793 	case FRAME_TAG_SESSION_RETRY:
2794 		ret = process_session_retry(con, p, end);
2795 		break;
2796 	case FRAME_TAG_SESSION_RETRY_GLOBAL:
2797 		ret = process_session_retry_global(con, p, end);
2798 		break;
2799 	case FRAME_TAG_SESSION_RESET:
2800 		ret = process_session_reset(con, p, end);
2801 		break;
2802 	case FRAME_TAG_KEEPALIVE2_ACK:
2803 		ret = process_keepalive2_ack(con, p, end);
2804 		break;
2805 	case FRAME_TAG_ACK:
2806 		ret = process_ack(con, p, end);
2807 		break;
2808 	default:
2809 		pr_err("bad tag %d\n", tag);
2810 		con->error_msg = "protocol error, bad tag";
2811 		return -EINVAL;
2812 	}
2813 	if (ret) {
2814 		dout("%s con %p error %d\n", __func__, con, ret);
2815 		return ret;
2816 	}
2817 
2818 	prepare_read_preamble(con);
2819 	return 0;
2820 }
2821 
2822 /*
2823  * Return:
2824  *   1 - con->in_msg set, read message
2825  *   0 - skip message
2826  *  <0 - error
2827  */
process_message_header(struct ceph_connection * con,void * p,void * end)2828 static int process_message_header(struct ceph_connection *con,
2829 				  void *p, void *end)
2830 {
2831 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2832 	struct ceph_msg_header2 *hdr2;
2833 	struct ceph_msg_header hdr;
2834 	int skip;
2835 	int ret;
2836 	u64 seq;
2837 
2838 	ceph_decode_need(&p, end, sizeof(*hdr2), bad);
2839 	hdr2 = p;
2840 
2841 	/* verify seq# */
2842 	seq = le64_to_cpu(hdr2->seq);
2843 	if ((s64)seq - (s64)con->in_seq < 1) {
2844 		pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n",
2845 			ENTITY_NAME(con->peer_name),
2846 			ceph_pr_addr(&con->peer_addr),
2847 			seq, con->in_seq + 1);
2848 		return 0;
2849 	}
2850 	if ((s64)seq - (s64)con->in_seq > 1) {
2851 		pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1);
2852 		con->error_msg = "bad message sequence # for incoming message";
2853 		return -EBADE;
2854 	}
2855 
2856 	ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq));
2857 
2858 	fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2],
2859 		    desc->fd_lens[3], &con->peer_name);
2860 	ret = ceph_con_in_msg_alloc(con, &hdr, &skip);
2861 	if (ret)
2862 		return ret;
2863 
2864 	WARN_ON(!con->in_msg ^ skip);
2865 	if (skip)
2866 		return 0;
2867 
2868 	WARN_ON(!con->in_msg);
2869 	WARN_ON(con->in_msg->con != con);
2870 	return 1;
2871 
2872 bad:
2873 	pr_err("failed to decode message header\n");
2874 	return -EINVAL;
2875 }
2876 
process_message(struct ceph_connection * con)2877 static int process_message(struct ceph_connection *con)
2878 {
2879 	ceph_con_process_message(con);
2880 
2881 	/*
2882 	 * We could have been closed by ceph_con_close() because
2883 	 * ceph_con_process_message() temporarily drops con->mutex.
2884 	 */
2885 	if (con->state != CEPH_CON_S_OPEN) {
2886 		dout("%s con %p state changed to %d\n", __func__, con,
2887 		     con->state);
2888 		return -EAGAIN;
2889 	}
2890 
2891 	prepare_read_preamble(con);
2892 	return 0;
2893 }
2894 
__handle_control(struct ceph_connection * con,void * p)2895 static int __handle_control(struct ceph_connection *con, void *p)
2896 {
2897 	void *end = p + con->v2.in_desc.fd_lens[0];
2898 	struct ceph_msg *msg;
2899 	int ret;
2900 
2901 	if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE)
2902 		return process_control(con, p, end);
2903 
2904 	if (con->state != CEPH_CON_S_OPEN) {
2905 		con->error_msg = "protocol error, unexpected message";
2906 		return -EINVAL;
2907 	}
2908 
2909 	ret = process_message_header(con, p, end);
2910 	if (ret < 0)
2911 		return ret;
2912 	if (ret == 0) {
2913 		prepare_skip_message(con);
2914 		return 0;
2915 	}
2916 
2917 	msg = con->in_msg;  /* set in process_message_header() */
2918 	if (front_len(msg)) {
2919 		WARN_ON(front_len(msg) > msg->front_alloc_len);
2920 		msg->front.iov_len = front_len(msg);
2921 	} else {
2922 		msg->front.iov_len = 0;
2923 	}
2924 	if (middle_len(msg)) {
2925 		WARN_ON(middle_len(msg) > msg->middle->alloc_len);
2926 		msg->middle->vec.iov_len = middle_len(msg);
2927 	} else if (msg->middle) {
2928 		msg->middle->vec.iov_len = 0;
2929 	}
2930 
2931 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
2932 		return process_message(con);
2933 
2934 	if (con_secure(con))
2935 		return prepare_read_tail_secure(con);
2936 
2937 	return prepare_read_tail_plain(con);
2938 }
2939 
handle_preamble(struct ceph_connection * con)2940 static int handle_preamble(struct ceph_connection *con)
2941 {
2942 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2943 	int ret;
2944 
2945 	if (con_secure(con)) {
2946 		ret = decrypt_preamble(con);
2947 		if (ret) {
2948 			if (ret == -EBADMSG)
2949 				con->error_msg = "integrity error, bad preamble auth tag";
2950 			return ret;
2951 		}
2952 	}
2953 
2954 	ret = decode_preamble(con->v2.in_buf, desc);
2955 	if (ret) {
2956 		if (ret == -EBADMSG)
2957 			con->error_msg = "integrity error, bad crc";
2958 		else
2959 			con->error_msg = "protocol error, bad preamble";
2960 		return ret;
2961 	}
2962 
2963 	dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__,
2964 	     con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0],
2965 	     desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]);
2966 
2967 	if (!con_secure(con))
2968 		return prepare_read_control(con);
2969 
2970 	if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN)
2971 		return prepare_read_control_remainder(con);
2972 
2973 	return __handle_control(con, CTRL_BODY(con->v2.in_buf));
2974 }
2975 
handle_control(struct ceph_connection * con)2976 static int handle_control(struct ceph_connection *con)
2977 {
2978 	int ctrl_len = con->v2.in_desc.fd_lens[0];
2979 	void *buf;
2980 	int ret;
2981 
2982 	WARN_ON(con_secure(con));
2983 
2984 	ret = verify_control_crc(con);
2985 	if (ret) {
2986 		con->error_msg = "integrity error, bad crc";
2987 		return ret;
2988 	}
2989 
2990 	if (con->state == CEPH_CON_S_V2_AUTH) {
2991 		buf = alloc_conn_buf(con, ctrl_len);
2992 		if (!buf)
2993 			return -ENOMEM;
2994 
2995 		memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len);
2996 		return __handle_control(con, buf);
2997 	}
2998 
2999 	return __handle_control(con, con->v2.in_kvecs[0].iov_base);
3000 }
3001 
handle_control_remainder(struct ceph_connection * con)3002 static int handle_control_remainder(struct ceph_connection *con)
3003 {
3004 	int ret;
3005 
3006 	WARN_ON(!con_secure(con));
3007 
3008 	ret = decrypt_control_remainder(con);
3009 	if (ret) {
3010 		if (ret == -EBADMSG)
3011 			con->error_msg = "integrity error, bad control remainder auth tag";
3012 		return ret;
3013 	}
3014 
3015 	return __handle_control(con, con->v2.in_kvecs[0].iov_base -
3016 				     CEPH_PREAMBLE_INLINE_LEN);
3017 }
3018 
handle_epilogue(struct ceph_connection * con)3019 static int handle_epilogue(struct ceph_connection *con)
3020 {
3021 	u32 front_crc, middle_crc, data_crc;
3022 	int ret;
3023 
3024 	if (con_secure(con)) {
3025 		ret = decrypt_tail(con);
3026 		if (ret) {
3027 			if (ret == -EBADMSG)
3028 				con->error_msg = "integrity error, bad epilogue auth tag";
3029 			return ret;
3030 		}
3031 
3032 		/* just late_status */
3033 		ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL);
3034 		if (ret) {
3035 			con->error_msg = "protocol error, bad epilogue";
3036 			return ret;
3037 		}
3038 	} else {
3039 		ret = decode_epilogue(con->v2.in_buf, &front_crc,
3040 				      &middle_crc, &data_crc);
3041 		if (ret) {
3042 			con->error_msg = "protocol error, bad epilogue";
3043 			return ret;
3044 		}
3045 
3046 		ret = verify_epilogue_crcs(con, front_crc, middle_crc,
3047 					   data_crc);
3048 		if (ret) {
3049 			con->error_msg = "integrity error, bad crc";
3050 			return ret;
3051 		}
3052 	}
3053 
3054 	return process_message(con);
3055 }
3056 
finish_skip(struct ceph_connection * con)3057 static void finish_skip(struct ceph_connection *con)
3058 {
3059 	dout("%s con %p\n", __func__, con);
3060 
3061 	if (con_secure(con))
3062 		gcm_inc_nonce(&con->v2.in_gcm_nonce);
3063 
3064 	__finish_skip(con);
3065 }
3066 
populate_in_iter(struct ceph_connection * con)3067 static int populate_in_iter(struct ceph_connection *con)
3068 {
3069 	int ret;
3070 
3071 	dout("%s con %p state %d in_state %d\n", __func__, con, con->state,
3072 	     con->v2.in_state);
3073 	WARN_ON(iov_iter_count(&con->v2.in_iter));
3074 
3075 	if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) {
3076 		ret = process_banner_prefix(con);
3077 	} else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) {
3078 		ret = process_banner_payload(con);
3079 	} else if ((con->state >= CEPH_CON_S_V2_HELLO &&
3080 		    con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) ||
3081 		   con->state == CEPH_CON_S_OPEN) {
3082 		switch (con->v2.in_state) {
3083 		case IN_S_HANDLE_PREAMBLE:
3084 			ret = handle_preamble(con);
3085 			break;
3086 		case IN_S_HANDLE_CONTROL:
3087 			ret = handle_control(con);
3088 			break;
3089 		case IN_S_HANDLE_CONTROL_REMAINDER:
3090 			ret = handle_control_remainder(con);
3091 			break;
3092 		case IN_S_PREPARE_READ_DATA:
3093 			ret = prepare_read_data(con);
3094 			break;
3095 		case IN_S_PREPARE_READ_DATA_CONT:
3096 			prepare_read_data_cont(con);
3097 			ret = 0;
3098 			break;
3099 		case IN_S_PREPARE_READ_ENC_PAGE:
3100 			prepare_read_enc_page(con);
3101 			ret = 0;
3102 			break;
3103 		case IN_S_PREPARE_SPARSE_DATA:
3104 			ret = prepare_sparse_read_data(con);
3105 			break;
3106 		case IN_S_PREPARE_SPARSE_DATA_CONT:
3107 			ret = prepare_sparse_read_cont(con);
3108 			break;
3109 		case IN_S_HANDLE_EPILOGUE:
3110 			ret = handle_epilogue(con);
3111 			break;
3112 		case IN_S_FINISH_SKIP:
3113 			finish_skip(con);
3114 			ret = 0;
3115 			break;
3116 		default:
3117 			WARN(1, "bad in_state %d", con->v2.in_state);
3118 			return -EINVAL;
3119 		}
3120 	} else {
3121 		WARN(1, "bad state %d", con->state);
3122 		return -EINVAL;
3123 	}
3124 	if (ret) {
3125 		dout("%s con %p error %d\n", __func__, con, ret);
3126 		return ret;
3127 	}
3128 
3129 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3130 		return -ENODATA;
3131 	dout("%s con %p populated %zu\n", __func__, con,
3132 	     iov_iter_count(&con->v2.in_iter));
3133 	return 1;
3134 }
3135 
ceph_con_v2_try_read(struct ceph_connection * con)3136 int ceph_con_v2_try_read(struct ceph_connection *con)
3137 {
3138 	int ret;
3139 
3140 	dout("%s con %p state %d need %zu\n", __func__, con, con->state,
3141 	     iov_iter_count(&con->v2.in_iter));
3142 
3143 	if (con->state == CEPH_CON_S_PREOPEN)
3144 		return 0;
3145 
3146 	/*
3147 	 * We should always have something pending here.  If not,
3148 	 * avoid calling populate_in_iter() as if we read something
3149 	 * (ceph_tcp_recv() would immediately return 1).
3150 	 */
3151 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3152 		return -ENODATA;
3153 
3154 	for (;;) {
3155 		ret = ceph_tcp_recv(con);
3156 		if (ret <= 0)
3157 			return ret;
3158 
3159 		ret = populate_in_iter(con);
3160 		if (ret <= 0) {
3161 			if (ret && ret != -EAGAIN && !con->error_msg)
3162 				con->error_msg = "read processing error";
3163 			return ret;
3164 		}
3165 	}
3166 }
3167 
queue_data(struct ceph_connection * con,struct ceph_msg * msg)3168 static void queue_data(struct ceph_connection *con, struct ceph_msg *msg)
3169 {
3170 	struct bio_vec bv;
3171 
3172 	con->v2.out_epil.data_crc = -1;
3173 	ceph_msg_data_cursor_init(&con->v2.out_cursor, msg,
3174 				  data_len(msg));
3175 
3176 	get_bvec_at(&con->v2.out_cursor, &bv);
3177 	set_out_bvec(con, &bv, true);
3178 	con->v2.out_state = OUT_S_QUEUE_DATA_CONT;
3179 }
3180 
queue_data_cont(struct ceph_connection * con,struct ceph_msg * msg)3181 static void queue_data_cont(struct ceph_connection *con, struct ceph_msg *msg)
3182 {
3183 	struct bio_vec bv;
3184 
3185 	con->v2.out_epil.data_crc = ceph_crc32c_page(
3186 		con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3187 		con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len);
3188 
3189 	ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len);
3190 	if (con->v2.out_cursor.total_resid) {
3191 		get_bvec_at(&con->v2.out_cursor, &bv);
3192 		set_out_bvec(con, &bv, true);
3193 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT);
3194 		return;
3195 	}
3196 
3197 	/*
3198 	 * We've written all data.  Queue epilogue.  Once it's written,
3199 	 * we are done.
3200 	 */
3201 	reset_out_kvecs(con);
3202 	prepare_epilogue_plain(con, msg, false);
3203 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3204 }
3205 
queue_enc_page(struct ceph_connection * con)3206 static void queue_enc_page(struct ceph_connection *con)
3207 {
3208 	struct bio_vec bv;
3209 
3210 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i,
3211 	     con->v2.out_enc_resid);
3212 	WARN_ON(!con->v2.out_enc_resid);
3213 
3214 	bvec_set_page(&bv, con->v2.out_enc_pages[con->v2.out_enc_i],
3215 		      min(con->v2.out_enc_resid, (int)PAGE_SIZE), 0);
3216 
3217 	set_out_bvec(con, &bv, false);
3218 	con->v2.out_enc_i++;
3219 	con->v2.out_enc_resid -= bv.bv_len;
3220 
3221 	if (con->v2.out_enc_resid) {
3222 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE);
3223 		return;
3224 	}
3225 
3226 	/*
3227 	 * We've queued the last piece of ciphertext (ending with
3228 	 * epilogue) + auth tag.  Once it's written, we are done.
3229 	 */
3230 	WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt);
3231 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3232 }
3233 
queue_zeros(struct ceph_connection * con,struct ceph_msg * msg)3234 static void queue_zeros(struct ceph_connection *con, struct ceph_msg *msg)
3235 {
3236 	dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero);
3237 
3238 	if (con->v2.out_zero) {
3239 		set_out_bvec_zero(con);
3240 		con->v2.out_zero -= con->v2.out_bvec.bv_len;
3241 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3242 		return;
3243 	}
3244 
3245 	/*
3246 	 * We've zero-filled everything up to epilogue.  Queue epilogue
3247 	 * with late_status set to ABORTED and crcs adjusted for zeros.
3248 	 * Once it's written, we are done patching up for the revoke.
3249 	 */
3250 	reset_out_kvecs(con);
3251 	prepare_epilogue_plain(con, msg, true);
3252 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3253 }
3254 
finish_message(struct ceph_connection * con)3255 static void finish_message(struct ceph_connection *con)
3256 {
3257 	dout("%s con %p msg %p\n", __func__, con, con->out_msg);
3258 
3259 	/* we end up here both plain and secure modes */
3260 	if (con->v2.out_enc_pages) {
3261 		WARN_ON(!con->v2.out_enc_page_cnt);
3262 		ceph_release_page_vector(con->v2.out_enc_pages,
3263 					 con->v2.out_enc_page_cnt);
3264 		con->v2.out_enc_pages = NULL;
3265 		con->v2.out_enc_page_cnt = 0;
3266 	}
3267 	/* message may have been revoked */
3268 	if (con->out_msg) {
3269 		ceph_msg_put(con->out_msg);
3270 		con->out_msg = NULL;
3271 	}
3272 
3273 	con->v2.out_state = OUT_S_GET_NEXT;
3274 }
3275 
populate_out_iter(struct ceph_connection * con)3276 static int populate_out_iter(struct ceph_connection *con)
3277 {
3278 	struct ceph_msg *msg;
3279 	int ret;
3280 
3281 	dout("%s con %p state %d out_state %d\n", __func__, con, con->state,
3282 	     con->v2.out_state);
3283 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3284 
3285 	if (con->state != CEPH_CON_S_OPEN) {
3286 		WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX ||
3287 			con->state > CEPH_CON_S_V2_SESSION_RECONNECT);
3288 		goto nothing_pending;
3289 	}
3290 
3291 	switch (con->v2.out_state) {
3292 	case OUT_S_QUEUE_DATA:
3293 		WARN_ON(!con->out_msg);
3294 		queue_data(con, con->out_msg);
3295 		goto populated;
3296 	case OUT_S_QUEUE_DATA_CONT:
3297 		WARN_ON(!con->out_msg);
3298 		queue_data_cont(con, con->out_msg);
3299 		goto populated;
3300 	case OUT_S_QUEUE_ENC_PAGE:
3301 		queue_enc_page(con);
3302 		goto populated;
3303 	case OUT_S_QUEUE_ZEROS:
3304 		WARN_ON(con->out_msg);  /* revoked */
3305 		queue_zeros(con, con->out_msg);
3306 		goto populated;
3307 	case OUT_S_FINISH_MESSAGE:
3308 		finish_message(con);
3309 		break;
3310 	case OUT_S_GET_NEXT:
3311 		break;
3312 	default:
3313 		WARN(1, "bad out_state %d", con->v2.out_state);
3314 		return -EINVAL;
3315 	}
3316 
3317 	WARN_ON(con->v2.out_state != OUT_S_GET_NEXT);
3318 	if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
3319 		ret = prepare_keepalive2(con);
3320 		if (ret) {
3321 			pr_err("prepare_keepalive2 failed: %d\n", ret);
3322 			return ret;
3323 		}
3324 	} else if ((msg = ceph_con_get_out_msg(con)) != NULL) {
3325 		ret = prepare_message(con, msg);
3326 		if (ret) {
3327 			pr_err("prepare_message failed: %d\n", ret);
3328 			return ret;
3329 		}
3330 	} else if (con->in_seq > con->in_seq_acked) {
3331 		ret = prepare_ack(con);
3332 		if (ret) {
3333 			pr_err("prepare_ack failed: %d\n", ret);
3334 			return ret;
3335 		}
3336 	} else {
3337 		goto nothing_pending;
3338 	}
3339 
3340 populated:
3341 	if (WARN_ON(!iov_iter_count(&con->v2.out_iter)))
3342 		return -ENODATA;
3343 	dout("%s con %p populated %zu\n", __func__, con,
3344 	     iov_iter_count(&con->v2.out_iter));
3345 	return 1;
3346 
3347 nothing_pending:
3348 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3349 	dout("%s con %p nothing pending\n", __func__, con);
3350 	ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
3351 	return 0;
3352 }
3353 
ceph_con_v2_try_write(struct ceph_connection * con)3354 int ceph_con_v2_try_write(struct ceph_connection *con)
3355 {
3356 	int ret;
3357 
3358 	dout("%s con %p state %d have %zu\n", __func__, con, con->state,
3359 	     iov_iter_count(&con->v2.out_iter));
3360 
3361 	/* open the socket first? */
3362 	if (con->state == CEPH_CON_S_PREOPEN) {
3363 		WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2);
3364 
3365 		/*
3366 		 * Always bump global_seq.  Bump connect_seq only if
3367 		 * there is a session (i.e. we are reconnecting and will
3368 		 * send session_reconnect instead of client_ident).
3369 		 */
3370 		con->v2.global_seq = ceph_get_global_seq(con->msgr, 0);
3371 		if (con->v2.server_cookie)
3372 			con->v2.connect_seq++;
3373 
3374 		ret = prepare_read_banner_prefix(con);
3375 		if (ret) {
3376 			pr_err("prepare_read_banner_prefix failed: %d\n", ret);
3377 			con->error_msg = "connect error";
3378 			return ret;
3379 		}
3380 
3381 		reset_out_kvecs(con);
3382 		ret = prepare_banner(con);
3383 		if (ret) {
3384 			pr_err("prepare_banner failed: %d\n", ret);
3385 			con->error_msg = "connect error";
3386 			return ret;
3387 		}
3388 
3389 		ret = ceph_tcp_connect(con);
3390 		if (ret) {
3391 			pr_err("ceph_tcp_connect failed: %d\n", ret);
3392 			con->error_msg = "connect error";
3393 			return ret;
3394 		}
3395 	}
3396 
3397 	if (!iov_iter_count(&con->v2.out_iter)) {
3398 		ret = populate_out_iter(con);
3399 		if (ret <= 0) {
3400 			if (ret && ret != -EAGAIN && !con->error_msg)
3401 				con->error_msg = "write processing error";
3402 			return ret;
3403 		}
3404 	}
3405 
3406 	tcp_sock_set_cork(con->sock->sk, true);
3407 	for (;;) {
3408 		ret = ceph_tcp_send(con);
3409 		if (ret <= 0)
3410 			break;
3411 
3412 		ret = populate_out_iter(con);
3413 		if (ret <= 0) {
3414 			if (ret && ret != -EAGAIN && !con->error_msg)
3415 				con->error_msg = "write processing error";
3416 			break;
3417 		}
3418 	}
3419 
3420 	tcp_sock_set_cork(con->sock->sk, false);
3421 	return ret;
3422 }
3423 
crc32c_zeros(u32 crc,int zero_len)3424 static u32 crc32c_zeros(u32 crc, int zero_len)
3425 {
3426 	int len;
3427 
3428 	while (zero_len) {
3429 		len = min(zero_len, (int)PAGE_SIZE);
3430 		crc = crc32c(crc, page_address(ceph_zero_page), len);
3431 		zero_len -= len;
3432 	}
3433 
3434 	return crc;
3435 }
3436 
prepare_zero_front(struct ceph_connection * con,struct ceph_msg * msg,int resid)3437 static void prepare_zero_front(struct ceph_connection *con,
3438 			       struct ceph_msg *msg, int resid)
3439 {
3440 	int sent;
3441 
3442 	WARN_ON(!resid || resid > front_len(msg));
3443 	sent = front_len(msg) - resid;
3444 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3445 
3446 	if (sent) {
3447 		con->v2.out_epil.front_crc =
3448 			crc32c(-1, msg->front.iov_base, sent);
3449 		con->v2.out_epil.front_crc =
3450 			crc32c_zeros(con->v2.out_epil.front_crc, resid);
3451 	} else {
3452 		con->v2.out_epil.front_crc = crc32c_zeros(-1, resid);
3453 	}
3454 
3455 	con->v2.out_iter.count -= resid;
3456 	out_zero_add(con, resid);
3457 }
3458 
prepare_zero_middle(struct ceph_connection * con,struct ceph_msg * msg,int resid)3459 static void prepare_zero_middle(struct ceph_connection *con,
3460 				struct ceph_msg *msg, int resid)
3461 {
3462 	int sent;
3463 
3464 	WARN_ON(!resid || resid > middle_len(msg));
3465 	sent = middle_len(msg) - resid;
3466 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3467 
3468 	if (sent) {
3469 		con->v2.out_epil.middle_crc =
3470 			crc32c(-1, msg->middle->vec.iov_base, sent);
3471 		con->v2.out_epil.middle_crc =
3472 			crc32c_zeros(con->v2.out_epil.middle_crc, resid);
3473 	} else {
3474 		con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid);
3475 	}
3476 
3477 	con->v2.out_iter.count -= resid;
3478 	out_zero_add(con, resid);
3479 }
3480 
prepare_zero_data(struct ceph_connection * con,struct ceph_msg * msg)3481 static void prepare_zero_data(struct ceph_connection *con,
3482 			      struct ceph_msg *msg)
3483 {
3484 	dout("%s con %p\n", __func__, con);
3485 	con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(msg));
3486 	out_zero_add(con, data_len(msg));
3487 }
3488 
revoke_at_queue_data(struct ceph_connection * con,struct ceph_msg * msg)3489 static void revoke_at_queue_data(struct ceph_connection *con,
3490 				 struct ceph_msg *msg)
3491 {
3492 	int boundary;
3493 	int resid;
3494 
3495 	WARN_ON(!data_len(msg));
3496 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3497 	resid = iov_iter_count(&con->v2.out_iter);
3498 
3499 	boundary = front_len(msg) + middle_len(msg);
3500 	if (resid > boundary) {
3501 		resid -= boundary;
3502 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3503 		dout("%s con %p was sending head\n", __func__, con);
3504 		if (front_len(msg))
3505 			prepare_zero_front(con, msg, front_len(msg));
3506 		if (middle_len(msg))
3507 			prepare_zero_middle(con, msg, middle_len(msg));
3508 		prepare_zero_data(con, msg);
3509 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3510 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3511 		return;
3512 	}
3513 
3514 	boundary = middle_len(msg);
3515 	if (resid > boundary) {
3516 		resid -= boundary;
3517 		dout("%s con %p was sending front\n", __func__, con);
3518 		prepare_zero_front(con, msg, resid);
3519 		if (middle_len(msg))
3520 			prepare_zero_middle(con, msg, middle_len(msg));
3521 		prepare_zero_data(con, msg);
3522 		queue_zeros(con, msg);
3523 		return;
3524 	}
3525 
3526 	WARN_ON(!resid);
3527 	dout("%s con %p was sending middle\n", __func__, con);
3528 	prepare_zero_middle(con, msg, resid);
3529 	prepare_zero_data(con, msg);
3530 	queue_zeros(con, msg);
3531 }
3532 
revoke_at_queue_data_cont(struct ceph_connection * con,struct ceph_msg * msg)3533 static void revoke_at_queue_data_cont(struct ceph_connection *con,
3534 				      struct ceph_msg *msg)
3535 {
3536 	int sent, resid;  /* current piece of data */
3537 
3538 	WARN_ON(!data_len(msg));
3539 	WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter));
3540 	resid = iov_iter_count(&con->v2.out_iter);
3541 	WARN_ON(!resid || resid > con->v2.out_bvec.bv_len);
3542 	sent = con->v2.out_bvec.bv_len - resid;
3543 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3544 
3545 	if (sent) {
3546 		con->v2.out_epil.data_crc = ceph_crc32c_page(
3547 			con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3548 			con->v2.out_bvec.bv_offset, sent);
3549 		ceph_msg_data_advance(&con->v2.out_cursor, sent);
3550 	}
3551 	WARN_ON(resid > con->v2.out_cursor.total_resid);
3552 	con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc,
3553 						con->v2.out_cursor.total_resid);
3554 
3555 	con->v2.out_iter.count -= resid;
3556 	out_zero_add(con, con->v2.out_cursor.total_resid);
3557 	queue_zeros(con, msg);
3558 }
3559 
revoke_at_finish_message(struct ceph_connection * con,struct ceph_msg * msg)3560 static void revoke_at_finish_message(struct ceph_connection *con,
3561 				     struct ceph_msg *msg)
3562 {
3563 	int boundary;
3564 	int resid;
3565 
3566 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3567 	resid = iov_iter_count(&con->v2.out_iter);
3568 
3569 	if (!front_len(msg) && !middle_len(msg) &&
3570 	    !data_len(msg)) {
3571 		WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN);
3572 		dout("%s con %p was sending head (empty message) - noop\n",
3573 		     __func__, con);
3574 		return;
3575 	}
3576 
3577 	boundary = front_len(msg) + middle_len(msg) +
3578 		   CEPH_EPILOGUE_PLAIN_LEN;
3579 	if (resid > boundary) {
3580 		resid -= boundary;
3581 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3582 		dout("%s con %p was sending head\n", __func__, con);
3583 		if (front_len(msg))
3584 			prepare_zero_front(con, msg, front_len(msg));
3585 		if (middle_len(msg))
3586 			prepare_zero_middle(con, msg, middle_len(msg));
3587 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3588 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3589 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3590 		return;
3591 	}
3592 
3593 	boundary = middle_len(msg) + CEPH_EPILOGUE_PLAIN_LEN;
3594 	if (resid > boundary) {
3595 		resid -= boundary;
3596 		dout("%s con %p was sending front\n", __func__, con);
3597 		prepare_zero_front(con, msg, resid);
3598 		if (middle_len(msg))
3599 			prepare_zero_middle(con, msg, middle_len(msg));
3600 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3601 		queue_zeros(con, msg);
3602 		return;
3603 	}
3604 
3605 	boundary = CEPH_EPILOGUE_PLAIN_LEN;
3606 	if (resid > boundary) {
3607 		resid -= boundary;
3608 		dout("%s con %p was sending middle\n", __func__, con);
3609 		prepare_zero_middle(con, msg, resid);
3610 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3611 		queue_zeros(con, msg);
3612 		return;
3613 	}
3614 
3615 	WARN_ON(!resid);
3616 	dout("%s con %p was sending epilogue - noop\n", __func__, con);
3617 }
3618 
ceph_con_v2_revoke(struct ceph_connection * con,struct ceph_msg * msg)3619 void ceph_con_v2_revoke(struct ceph_connection *con, struct ceph_msg *msg)
3620 {
3621 	WARN_ON(con->v2.out_zero);
3622 
3623 	if (con_secure(con)) {
3624 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE &&
3625 			con->v2.out_state != OUT_S_FINISH_MESSAGE);
3626 		dout("%s con %p secure - noop\n", __func__, con);
3627 		return;
3628 	}
3629 
3630 	switch (con->v2.out_state) {
3631 	case OUT_S_QUEUE_DATA:
3632 		revoke_at_queue_data(con, msg);
3633 		break;
3634 	case OUT_S_QUEUE_DATA_CONT:
3635 		revoke_at_queue_data_cont(con, msg);
3636 		break;
3637 	case OUT_S_FINISH_MESSAGE:
3638 		revoke_at_finish_message(con, msg);
3639 		break;
3640 	default:
3641 		WARN(1, "bad out_state %d", con->v2.out_state);
3642 		break;
3643 	}
3644 }
3645 
revoke_at_prepare_read_data(struct ceph_connection * con)3646 static void revoke_at_prepare_read_data(struct ceph_connection *con)
3647 {
3648 	int remaining;
3649 	int resid;
3650 
3651 	WARN_ON(con_secure(con));
3652 	WARN_ON(!data_len(con->in_msg));
3653 	WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
3654 	resid = iov_iter_count(&con->v2.in_iter);
3655 	WARN_ON(!resid);
3656 
3657 	remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3658 	dout("%s con %p resid %d remaining %d\n", __func__, con, resid,
3659 	     remaining);
3660 	con->v2.in_iter.count -= resid;
3661 	set_in_skip(con, resid + remaining);
3662 	con->v2.in_state = IN_S_FINISH_SKIP;
3663 }
3664 
revoke_at_prepare_read_data_cont(struct ceph_connection * con)3665 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con)
3666 {
3667 	int recved, resid;  /* current piece of data */
3668 	int remaining;
3669 
3670 	WARN_ON(con_secure(con));
3671 	WARN_ON(!data_len(con->in_msg));
3672 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3673 	resid = iov_iter_count(&con->v2.in_iter);
3674 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3675 	recved = con->v2.in_bvec.bv_len - resid;
3676 	dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid);
3677 
3678 	if (recved)
3679 		ceph_msg_data_advance(&con->v2.in_cursor, recved);
3680 	WARN_ON(resid > con->v2.in_cursor.total_resid);
3681 
3682 	remaining = CEPH_EPILOGUE_PLAIN_LEN;
3683 	dout("%s con %p total_resid %zu remaining %d\n", __func__, con,
3684 	     con->v2.in_cursor.total_resid, remaining);
3685 	con->v2.in_iter.count -= resid;
3686 	set_in_skip(con, con->v2.in_cursor.total_resid + remaining);
3687 	con->v2.in_state = IN_S_FINISH_SKIP;
3688 }
3689 
revoke_at_prepare_read_enc_page(struct ceph_connection * con)3690 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con)
3691 {
3692 	int resid;  /* current enc page (not necessarily data) */
3693 
3694 	WARN_ON(!con_secure(con));
3695 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3696 	resid = iov_iter_count(&con->v2.in_iter);
3697 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3698 
3699 	dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid,
3700 	     con->v2.in_enc_resid);
3701 	con->v2.in_iter.count -= resid;
3702 	set_in_skip(con, resid + con->v2.in_enc_resid);
3703 	con->v2.in_state = IN_S_FINISH_SKIP;
3704 }
3705 
revoke_at_prepare_sparse_data(struct ceph_connection * con)3706 static void revoke_at_prepare_sparse_data(struct ceph_connection *con)
3707 {
3708 	int resid;  /* current piece of data */
3709 	int remaining;
3710 
3711 	WARN_ON(con_secure(con));
3712 	WARN_ON(!data_len(con->in_msg));
3713 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3714 	resid = iov_iter_count(&con->v2.in_iter);
3715 	dout("%s con %p resid %d\n", __func__, con, resid);
3716 
3717 	remaining = CEPH_EPILOGUE_PLAIN_LEN + con->v2.data_len_remain;
3718 	con->v2.in_iter.count -= resid;
3719 	set_in_skip(con, resid + remaining);
3720 	con->v2.in_state = IN_S_FINISH_SKIP;
3721 }
3722 
revoke_at_handle_epilogue(struct ceph_connection * con)3723 static void revoke_at_handle_epilogue(struct ceph_connection *con)
3724 {
3725 	int resid;
3726 
3727 	resid = iov_iter_count(&con->v2.in_iter);
3728 	WARN_ON(!resid);
3729 
3730 	dout("%s con %p resid %d\n", __func__, con, resid);
3731 	con->v2.in_iter.count -= resid;
3732 	set_in_skip(con, resid);
3733 	con->v2.in_state = IN_S_FINISH_SKIP;
3734 }
3735 
ceph_con_v2_revoke_incoming(struct ceph_connection * con)3736 void ceph_con_v2_revoke_incoming(struct ceph_connection *con)
3737 {
3738 	switch (con->v2.in_state) {
3739 	case IN_S_PREPARE_SPARSE_DATA:
3740 	case IN_S_PREPARE_READ_DATA:
3741 		revoke_at_prepare_read_data(con);
3742 		break;
3743 	case IN_S_PREPARE_READ_DATA_CONT:
3744 		revoke_at_prepare_read_data_cont(con);
3745 		break;
3746 	case IN_S_PREPARE_READ_ENC_PAGE:
3747 		revoke_at_prepare_read_enc_page(con);
3748 		break;
3749 	case IN_S_PREPARE_SPARSE_DATA_CONT:
3750 		revoke_at_prepare_sparse_data(con);
3751 		break;
3752 	case IN_S_HANDLE_EPILOGUE:
3753 		revoke_at_handle_epilogue(con);
3754 		break;
3755 	default:
3756 		WARN(1, "bad in_state %d", con->v2.in_state);
3757 		break;
3758 	}
3759 }
3760 
ceph_con_v2_opened(struct ceph_connection * con)3761 bool ceph_con_v2_opened(struct ceph_connection *con)
3762 {
3763 	return con->v2.peer_global_seq;
3764 }
3765 
ceph_con_v2_reset_session(struct ceph_connection * con)3766 void ceph_con_v2_reset_session(struct ceph_connection *con)
3767 {
3768 	con->v2.client_cookie = 0;
3769 	con->v2.server_cookie = 0;
3770 	con->v2.global_seq = 0;
3771 	con->v2.connect_seq = 0;
3772 	con->v2.peer_global_seq = 0;
3773 }
3774 
ceph_con_v2_reset_protocol(struct ceph_connection * con)3775 void ceph_con_v2_reset_protocol(struct ceph_connection *con)
3776 {
3777 	iov_iter_truncate(&con->v2.in_iter, 0);
3778 	iov_iter_truncate(&con->v2.out_iter, 0);
3779 	con->v2.out_zero = 0;
3780 
3781 	clear_in_sign_kvecs(con);
3782 	clear_out_sign_kvecs(con);
3783 	free_conn_bufs(con);
3784 
3785 	if (con->v2.in_enc_pages) {
3786 		WARN_ON(!con->v2.in_enc_page_cnt);
3787 		ceph_release_page_vector(con->v2.in_enc_pages,
3788 					 con->v2.in_enc_page_cnt);
3789 		con->v2.in_enc_pages = NULL;
3790 		con->v2.in_enc_page_cnt = 0;
3791 	}
3792 	if (con->v2.out_enc_pages) {
3793 		WARN_ON(!con->v2.out_enc_page_cnt);
3794 		ceph_release_page_vector(con->v2.out_enc_pages,
3795 					 con->v2.out_enc_page_cnt);
3796 		con->v2.out_enc_pages = NULL;
3797 		con->v2.out_enc_page_cnt = 0;
3798 	}
3799 
3800 	con->v2.con_mode = CEPH_CON_MODE_UNKNOWN;
3801 	memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN);
3802 	memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN);
3803 
3804 	memzero_explicit(&con->v2.hmac_key, sizeof(con->v2.hmac_key));
3805 	con->v2.hmac_key_set = false;
3806 	if (con->v2.gcm_req) {
3807 		aead_request_free(con->v2.gcm_req);
3808 		con->v2.gcm_req = NULL;
3809 	}
3810 	if (con->v2.gcm_tfm) {
3811 		crypto_free_aead(con->v2.gcm_tfm);
3812 		con->v2.gcm_tfm = NULL;
3813 	}
3814 }
3815