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