]> git.ipfire.org Git - thirdparty/kernel/linux.git/blob - net/ceph/messenger_v2.c
libceph: stop checking crypto_shash_alignmask
[thirdparty/kernel/linux.git] / net / ceph / messenger_v2.c
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/hash.h>
12 #include <crypto/sha2.h>
13 #include <crypto/utils.h>
14 #include <linux/bvec.h>
15 #include <linux/crc32c.h>
16 #include <linux/net.h>
17 #include <linux/scatterlist.h>
18 #include <linux/socket.h>
19 #include <linux/sched/mm.h>
20 #include <net/sock.h>
21 #include <net/tcp.h>
22
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/libceph.h>
26 #include <linux/ceph/messenger.h>
27
28 #include "crypto.h" /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */
29
30 #define FRAME_TAG_HELLO 1
31 #define FRAME_TAG_AUTH_REQUEST 2
32 #define FRAME_TAG_AUTH_BAD_METHOD 3
33 #define FRAME_TAG_AUTH_REPLY_MORE 4
34 #define FRAME_TAG_AUTH_REQUEST_MORE 5
35 #define FRAME_TAG_AUTH_DONE 6
36 #define FRAME_TAG_AUTH_SIGNATURE 7
37 #define FRAME_TAG_CLIENT_IDENT 8
38 #define FRAME_TAG_SERVER_IDENT 9
39 #define FRAME_TAG_IDENT_MISSING_FEATURES 10
40 #define FRAME_TAG_SESSION_RECONNECT 11
41 #define FRAME_TAG_SESSION_RESET 12
42 #define FRAME_TAG_SESSION_RETRY 13
43 #define FRAME_TAG_SESSION_RETRY_GLOBAL 14
44 #define FRAME_TAG_SESSION_RECONNECT_OK 15
45 #define FRAME_TAG_WAIT 16
46 #define FRAME_TAG_MESSAGE 17
47 #define FRAME_TAG_KEEPALIVE2 18
48 #define FRAME_TAG_KEEPALIVE2_ACK 19
49 #define FRAME_TAG_ACK 20
50
51 #define FRAME_LATE_STATUS_ABORTED 0x1
52 #define FRAME_LATE_STATUS_COMPLETE 0xe
53 #define FRAME_LATE_STATUS_ABORTED_MASK 0xf
54
55 #define IN_S_HANDLE_PREAMBLE 1
56 #define IN_S_HANDLE_CONTROL 2
57 #define IN_S_HANDLE_CONTROL_REMAINDER 3
58 #define IN_S_PREPARE_READ_DATA 4
59 #define IN_S_PREPARE_READ_DATA_CONT 5
60 #define IN_S_PREPARE_READ_ENC_PAGE 6
61 #define IN_S_PREPARE_SPARSE_DATA 7
62 #define IN_S_PREPARE_SPARSE_DATA_CONT 8
63 #define IN_S_HANDLE_EPILOGUE 9
64 #define IN_S_FINISH_SKIP 10
65
66 #define OUT_S_QUEUE_DATA 1
67 #define OUT_S_QUEUE_DATA_CONT 2
68 #define OUT_S_QUEUE_ENC_PAGE 3
69 #define OUT_S_QUEUE_ZEROS 4
70 #define OUT_S_FINISH_MESSAGE 5
71 #define OUT_S_GET_NEXT 6
72
73 #define CTRL_BODY(p) ((void *)(p) + CEPH_PREAMBLE_LEN)
74 #define FRONT_PAD(p) ((void *)(p) + CEPH_EPILOGUE_SECURE_LEN)
75 #define MIDDLE_PAD(p) (FRONT_PAD(p) + CEPH_GCM_BLOCK_LEN)
76 #define DATA_PAD(p) (MIDDLE_PAD(p) + CEPH_GCM_BLOCK_LEN)
77
78 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
79
80 static int do_recvmsg(struct socket *sock, struct iov_iter *it)
81 {
82 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
83 int ret;
84
85 msg.msg_iter = *it;
86 while (iov_iter_count(it)) {
87 ret = sock_recvmsg(sock, &msg, msg.msg_flags);
88 if (ret <= 0) {
89 if (ret == -EAGAIN)
90 ret = 0;
91 return ret;
92 }
93
94 iov_iter_advance(it, ret);
95 }
96
97 WARN_ON(msg_data_left(&msg));
98 return 1;
99 }
100
101 /*
102 * Read as much as possible.
103 *
104 * Return:
105 * 1 - done, nothing (else) to read
106 * 0 - socket is empty, need to wait
107 * <0 - error
108 */
109 static int ceph_tcp_recv(struct ceph_connection *con)
110 {
111 int ret;
112
113 dout("%s con %p %s %zu\n", __func__, con,
114 iov_iter_is_discard(&con->v2.in_iter) ? "discard" : "need",
115 iov_iter_count(&con->v2.in_iter));
116 ret = do_recvmsg(con->sock, &con->v2.in_iter);
117 dout("%s con %p ret %d left %zu\n", __func__, con, ret,
118 iov_iter_count(&con->v2.in_iter));
119 return ret;
120 }
121
122 static int do_sendmsg(struct socket *sock, struct iov_iter *it)
123 {
124 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
125 int ret;
126
127 msg.msg_iter = *it;
128 while (iov_iter_count(it)) {
129 ret = sock_sendmsg(sock, &msg);
130 if (ret <= 0) {
131 if (ret == -EAGAIN)
132 ret = 0;
133 return ret;
134 }
135
136 iov_iter_advance(it, ret);
137 }
138
139 WARN_ON(msg_data_left(&msg));
140 return 1;
141 }
142
143 static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
144 {
145 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
146 struct bio_vec bv;
147 int ret;
148
149 if (WARN_ON(!iov_iter_is_bvec(it)))
150 return -EINVAL;
151
152 while (iov_iter_count(it)) {
153 /* iov_iter_iovec() for ITER_BVEC */
154 bvec_set_page(&bv, it->bvec->bv_page,
155 min(iov_iter_count(it),
156 it->bvec->bv_len - it->iov_offset),
157 it->bvec->bv_offset + it->iov_offset);
158
159 /*
160 * MSG_SPLICE_PAGES cannot properly handle pages with
161 * page_count == 0, we need to fall back to sendmsg if
162 * that's the case.
163 *
164 * Same goes for slab pages: skb_can_coalesce() allows
165 * coalescing neighboring slab objects into a single frag
166 * which triggers one of hardened usercopy checks.
167 */
168 if (sendpage_ok(bv.bv_page))
169 msg.msg_flags |= MSG_SPLICE_PAGES;
170 else
171 msg.msg_flags &= ~MSG_SPLICE_PAGES;
172
173 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
174 ret = sock_sendmsg(sock, &msg);
175 if (ret <= 0) {
176 if (ret == -EAGAIN)
177 ret = 0;
178 return ret;
179 }
180
181 iov_iter_advance(it, ret);
182 }
183
184 return 1;
185 }
186
187 /*
188 * Write as much as possible. The socket is expected to be corked,
189 * so we don't bother with MSG_MORE here.
190 *
191 * Return:
192 * 1 - done, nothing (else) to write
193 * 0 - socket is full, need to wait
194 * <0 - error
195 */
196 static int ceph_tcp_send(struct ceph_connection *con)
197 {
198 int ret;
199
200 dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
201 iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
202 if (con->v2.out_iter_sendpage)
203 ret = do_try_sendpage(con->sock, &con->v2.out_iter);
204 else
205 ret = do_sendmsg(con->sock, &con->v2.out_iter);
206 dout("%s con %p ret %d left %zu\n", __func__, con, ret,
207 iov_iter_count(&con->v2.out_iter));
208 return ret;
209 }
210
211 static void add_in_kvec(struct ceph_connection *con, void *buf, int len)
212 {
213 BUG_ON(con->v2.in_kvec_cnt >= ARRAY_SIZE(con->v2.in_kvecs));
214 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
215
216 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_base = buf;
217 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_len = len;
218 con->v2.in_kvec_cnt++;
219
220 con->v2.in_iter.nr_segs++;
221 con->v2.in_iter.count += len;
222 }
223
224 static void reset_in_kvecs(struct ceph_connection *con)
225 {
226 WARN_ON(iov_iter_count(&con->v2.in_iter));
227
228 con->v2.in_kvec_cnt = 0;
229 iov_iter_kvec(&con->v2.in_iter, ITER_DEST, con->v2.in_kvecs, 0, 0);
230 }
231
232 static void set_in_bvec(struct ceph_connection *con, const struct bio_vec *bv)
233 {
234 WARN_ON(iov_iter_count(&con->v2.in_iter));
235
236 con->v2.in_bvec = *bv;
237 iov_iter_bvec(&con->v2.in_iter, ITER_DEST, &con->v2.in_bvec, 1, bv->bv_len);
238 }
239
240 static void set_in_skip(struct ceph_connection *con, int len)
241 {
242 WARN_ON(iov_iter_count(&con->v2.in_iter));
243
244 dout("%s con %p len %d\n", __func__, con, len);
245 iov_iter_discard(&con->v2.in_iter, ITER_DEST, len);
246 }
247
248 static void add_out_kvec(struct ceph_connection *con, void *buf, int len)
249 {
250 BUG_ON(con->v2.out_kvec_cnt >= ARRAY_SIZE(con->v2.out_kvecs));
251 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
252 WARN_ON(con->v2.out_zero);
253
254 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_base = buf;
255 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_len = len;
256 con->v2.out_kvec_cnt++;
257
258 con->v2.out_iter.nr_segs++;
259 con->v2.out_iter.count += len;
260 }
261
262 static void reset_out_kvecs(struct ceph_connection *con)
263 {
264 WARN_ON(iov_iter_count(&con->v2.out_iter));
265 WARN_ON(con->v2.out_zero);
266
267 con->v2.out_kvec_cnt = 0;
268
269 iov_iter_kvec(&con->v2.out_iter, ITER_SOURCE, con->v2.out_kvecs, 0, 0);
270 con->v2.out_iter_sendpage = false;
271 }
272
273 static void set_out_bvec(struct ceph_connection *con, const struct bio_vec *bv,
274 bool zerocopy)
275 {
276 WARN_ON(iov_iter_count(&con->v2.out_iter));
277 WARN_ON(con->v2.out_zero);
278
279 con->v2.out_bvec = *bv;
280 con->v2.out_iter_sendpage = zerocopy;
281 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
282 con->v2.out_bvec.bv_len);
283 }
284
285 static void set_out_bvec_zero(struct ceph_connection *con)
286 {
287 WARN_ON(iov_iter_count(&con->v2.out_iter));
288 WARN_ON(!con->v2.out_zero);
289
290 bvec_set_page(&con->v2.out_bvec, ceph_zero_page,
291 min(con->v2.out_zero, (int)PAGE_SIZE), 0);
292 con->v2.out_iter_sendpage = true;
293 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
294 con->v2.out_bvec.bv_len);
295 }
296
297 static void out_zero_add(struct ceph_connection *con, int len)
298 {
299 dout("%s con %p len %d\n", __func__, con, len);
300 con->v2.out_zero += len;
301 }
302
303 static void *alloc_conn_buf(struct ceph_connection *con, int len)
304 {
305 void *buf;
306
307 dout("%s con %p len %d\n", __func__, con, len);
308
309 if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
310 return NULL;
311
312 buf = kvmalloc(len, GFP_NOIO);
313 if (!buf)
314 return NULL;
315
316 con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf;
317 return buf;
318 }
319
320 static void free_conn_bufs(struct ceph_connection *con)
321 {
322 while (con->v2.conn_buf_cnt)
323 kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]);
324 }
325
326 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len)
327 {
328 BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs));
329
330 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf;
331 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len;
332 con->v2.in_sign_kvec_cnt++;
333 }
334
335 static void clear_in_sign_kvecs(struct ceph_connection *con)
336 {
337 con->v2.in_sign_kvec_cnt = 0;
338 }
339
340 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len)
341 {
342 BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs));
343
344 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf;
345 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len;
346 con->v2.out_sign_kvec_cnt++;
347 }
348
349 static void clear_out_sign_kvecs(struct ceph_connection *con)
350 {
351 con->v2.out_sign_kvec_cnt = 0;
352 }
353
354 static bool con_secure(struct ceph_connection *con)
355 {
356 return con->v2.con_mode == CEPH_CON_MODE_SECURE;
357 }
358
359 static int front_len(const struct ceph_msg *msg)
360 {
361 return le32_to_cpu(msg->hdr.front_len);
362 }
363
364 static int middle_len(const struct ceph_msg *msg)
365 {
366 return le32_to_cpu(msg->hdr.middle_len);
367 }
368
369 static int data_len(const struct ceph_msg *msg)
370 {
371 return le32_to_cpu(msg->hdr.data_len);
372 }
373
374 static bool need_padding(int len)
375 {
376 return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN);
377 }
378
379 static int padded_len(int len)
380 {
381 return ALIGN(len, CEPH_GCM_BLOCK_LEN);
382 }
383
384 static int padding_len(int len)
385 {
386 return padded_len(len) - len;
387 }
388
389 /* preamble + control segment */
390 static int head_onwire_len(int ctrl_len, bool secure)
391 {
392 int head_len;
393 int rem_len;
394
395 BUG_ON(ctrl_len < 0 || ctrl_len > CEPH_MSG_MAX_CONTROL_LEN);
396
397 if (secure) {
398 head_len = CEPH_PREAMBLE_SECURE_LEN;
399 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
400 rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
401 head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN;
402 }
403 } else {
404 head_len = CEPH_PREAMBLE_PLAIN_LEN;
405 if (ctrl_len)
406 head_len += ctrl_len + CEPH_CRC_LEN;
407 }
408 return head_len;
409 }
410
411 /* front, middle and data segments + epilogue */
412 static int __tail_onwire_len(int front_len, int middle_len, int data_len,
413 bool secure)
414 {
415 BUG_ON(front_len < 0 || front_len > CEPH_MSG_MAX_FRONT_LEN ||
416 middle_len < 0 || middle_len > CEPH_MSG_MAX_MIDDLE_LEN ||
417 data_len < 0 || data_len > CEPH_MSG_MAX_DATA_LEN);
418
419 if (!front_len && !middle_len && !data_len)
420 return 0;
421
422 if (!secure)
423 return front_len + middle_len + data_len +
424 CEPH_EPILOGUE_PLAIN_LEN;
425
426 return padded_len(front_len) + padded_len(middle_len) +
427 padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN;
428 }
429
430 static int tail_onwire_len(const struct ceph_msg *msg, bool secure)
431 {
432 return __tail_onwire_len(front_len(msg), middle_len(msg),
433 data_len(msg), secure);
434 }
435
436 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */
437 #define MESSAGE_HEAD_PLAIN_LEN (CEPH_PREAMBLE_PLAIN_LEN + \
438 sizeof(struct ceph_msg_header2) + \
439 CEPH_CRC_LEN)
440
441 static const int frame_aligns[] = {
442 sizeof(void *),
443 sizeof(void *),
444 sizeof(void *),
445 PAGE_SIZE
446 };
447
448 /*
449 * Discards trailing empty segments, unless there is just one segment.
450 * A frame always has at least one (possibly empty) segment.
451 */
452 static int calc_segment_count(const int *lens, int len_cnt)
453 {
454 int i;
455
456 for (i = len_cnt - 1; i >= 0; i--) {
457 if (lens[i])
458 return i + 1;
459 }
460
461 return 1;
462 }
463
464 static void init_frame_desc(struct ceph_frame_desc *desc, int tag,
465 const int *lens, int len_cnt)
466 {
467 int i;
468
469 memset(desc, 0, sizeof(*desc));
470
471 desc->fd_tag = tag;
472 desc->fd_seg_cnt = calc_segment_count(lens, len_cnt);
473 BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT);
474 for (i = 0; i < desc->fd_seg_cnt; i++) {
475 desc->fd_lens[i] = lens[i];
476 desc->fd_aligns[i] = frame_aligns[i];
477 }
478 }
479
480 /*
481 * Preamble crc covers everything up to itself (28 bytes) and
482 * is calculated and verified irrespective of the connection mode
483 * (i.e. even if the frame is encrypted).
484 */
485 static void encode_preamble(const struct ceph_frame_desc *desc, void *p)
486 {
487 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
488 void *start = p;
489 int i;
490
491 memset(p, 0, CEPH_PREAMBLE_LEN);
492
493 ceph_encode_8(&p, desc->fd_tag);
494 ceph_encode_8(&p, desc->fd_seg_cnt);
495 for (i = 0; i < desc->fd_seg_cnt; i++) {
496 ceph_encode_32(&p, desc->fd_lens[i]);
497 ceph_encode_16(&p, desc->fd_aligns[i]);
498 }
499
500 put_unaligned_le32(crc32c(0, start, crcp - start), crcp);
501 }
502
503 static int decode_preamble(void *p, struct ceph_frame_desc *desc)
504 {
505 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
506 u32 crc, expected_crc;
507 int i;
508
509 crc = crc32c(0, p, crcp - p);
510 expected_crc = get_unaligned_le32(crcp);
511 if (crc != expected_crc) {
512 pr_err("bad preamble crc, calculated %u, expected %u\n",
513 crc, expected_crc);
514 return -EBADMSG;
515 }
516
517 memset(desc, 0, sizeof(*desc));
518
519 desc->fd_tag = ceph_decode_8(&p);
520 desc->fd_seg_cnt = ceph_decode_8(&p);
521 if (desc->fd_seg_cnt < 1 ||
522 desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) {
523 pr_err("bad segment count %d\n", desc->fd_seg_cnt);
524 return -EINVAL;
525 }
526 for (i = 0; i < desc->fd_seg_cnt; i++) {
527 desc->fd_lens[i] = ceph_decode_32(&p);
528 desc->fd_aligns[i] = ceph_decode_16(&p);
529 }
530
531 if (desc->fd_lens[0] < 0 ||
532 desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) {
533 pr_err("bad control segment length %d\n", desc->fd_lens[0]);
534 return -EINVAL;
535 }
536 if (desc->fd_lens[1] < 0 ||
537 desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) {
538 pr_err("bad front segment length %d\n", desc->fd_lens[1]);
539 return -EINVAL;
540 }
541 if (desc->fd_lens[2] < 0 ||
542 desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) {
543 pr_err("bad middle segment length %d\n", desc->fd_lens[2]);
544 return -EINVAL;
545 }
546 if (desc->fd_lens[3] < 0 ||
547 desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) {
548 pr_err("bad data segment length %d\n", desc->fd_lens[3]);
549 return -EINVAL;
550 }
551
552 /*
553 * This would fire for FRAME_TAG_WAIT (it has one empty
554 * segment), but we should never get it as client.
555 */
556 if (!desc->fd_lens[desc->fd_seg_cnt - 1]) {
557 pr_err("last segment empty, segment count %d\n",
558 desc->fd_seg_cnt);
559 return -EINVAL;
560 }
561
562 return 0;
563 }
564
565 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted)
566 {
567 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
568 FRAME_LATE_STATUS_COMPLETE;
569 cpu_to_le32s(&con->v2.out_epil.front_crc);
570 cpu_to_le32s(&con->v2.out_epil.middle_crc);
571 cpu_to_le32s(&con->v2.out_epil.data_crc);
572 }
573
574 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted)
575 {
576 memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil));
577 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
578 FRAME_LATE_STATUS_COMPLETE;
579 }
580
581 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc,
582 u32 *data_crc)
583 {
584 u8 late_status;
585
586 late_status = ceph_decode_8(&p);
587 if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) !=
588 FRAME_LATE_STATUS_COMPLETE) {
589 /* we should never get an aborted message as client */
590 pr_err("bad late_status 0x%x\n", late_status);
591 return -EINVAL;
592 }
593
594 if (front_crc && middle_crc && data_crc) {
595 *front_crc = ceph_decode_32(&p);
596 *middle_crc = ceph_decode_32(&p);
597 *data_crc = ceph_decode_32(&p);
598 }
599
600 return 0;
601 }
602
603 static void fill_header(struct ceph_msg_header *hdr,
604 const struct ceph_msg_header2 *hdr2,
605 int front_len, int middle_len, int data_len,
606 const struct ceph_entity_name *peer_name)
607 {
608 hdr->seq = hdr2->seq;
609 hdr->tid = hdr2->tid;
610 hdr->type = hdr2->type;
611 hdr->priority = hdr2->priority;
612 hdr->version = hdr2->version;
613 hdr->front_len = cpu_to_le32(front_len);
614 hdr->middle_len = cpu_to_le32(middle_len);
615 hdr->data_len = cpu_to_le32(data_len);
616 hdr->data_off = hdr2->data_off;
617 hdr->src = *peer_name;
618 hdr->compat_version = hdr2->compat_version;
619 hdr->reserved = 0;
620 hdr->crc = 0;
621 }
622
623 static void fill_header2(struct ceph_msg_header2 *hdr2,
624 const struct ceph_msg_header *hdr, u64 ack_seq)
625 {
626 hdr2->seq = hdr->seq;
627 hdr2->tid = hdr->tid;
628 hdr2->type = hdr->type;
629 hdr2->priority = hdr->priority;
630 hdr2->version = hdr->version;
631 hdr2->data_pre_padding_len = 0;
632 hdr2->data_off = hdr->data_off;
633 hdr2->ack_seq = cpu_to_le64(ack_seq);
634 hdr2->flags = 0;
635 hdr2->compat_version = hdr->compat_version;
636 hdr2->reserved = 0;
637 }
638
639 static int verify_control_crc(struct ceph_connection *con)
640 {
641 int ctrl_len = con->v2.in_desc.fd_lens[0];
642 u32 crc, expected_crc;
643
644 WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len);
645 WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN);
646
647 crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len);
648 expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base);
649 if (crc != expected_crc) {
650 pr_err("bad control crc, calculated %u, expected %u\n",
651 crc, expected_crc);
652 return -EBADMSG;
653 }
654
655 return 0;
656 }
657
658 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc,
659 u32 middle_crc, u32 data_crc)
660 {
661 if (front_len(con->in_msg)) {
662 con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base,
663 front_len(con->in_msg));
664 } else {
665 WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg));
666 con->in_front_crc = -1;
667 }
668
669 if (middle_len(con->in_msg))
670 con->in_middle_crc = crc32c(-1,
671 con->in_msg->middle->vec.iov_base,
672 middle_len(con->in_msg));
673 else if (data_len(con->in_msg))
674 con->in_middle_crc = -1;
675 else
676 con->in_middle_crc = 0;
677
678 if (!data_len(con->in_msg))
679 con->in_data_crc = 0;
680
681 dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg,
682 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
683
684 if (con->in_front_crc != front_crc) {
685 pr_err("bad front crc, calculated %u, expected %u\n",
686 con->in_front_crc, front_crc);
687 return -EBADMSG;
688 }
689 if (con->in_middle_crc != middle_crc) {
690 pr_err("bad middle crc, calculated %u, expected %u\n",
691 con->in_middle_crc, middle_crc);
692 return -EBADMSG;
693 }
694 if (con->in_data_crc != data_crc) {
695 pr_err("bad data crc, calculated %u, expected %u\n",
696 con->in_data_crc, data_crc);
697 return -EBADMSG;
698 }
699
700 return 0;
701 }
702
703 static int setup_crypto(struct ceph_connection *con,
704 const u8 *session_key, int session_key_len,
705 const u8 *con_secret, int con_secret_len)
706 {
707 unsigned int noio_flag;
708 int ret;
709
710 dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n",
711 __func__, con, con->v2.con_mode, session_key_len, con_secret_len);
712 WARN_ON(con->v2.hmac_tfm || con->v2.gcm_tfm || con->v2.gcm_req);
713
714 if (con->v2.con_mode != CEPH_CON_MODE_CRC &&
715 con->v2.con_mode != CEPH_CON_MODE_SECURE) {
716 pr_err("bad con_mode %d\n", con->v2.con_mode);
717 return -EINVAL;
718 }
719
720 if (!session_key_len) {
721 WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC);
722 WARN_ON(con_secret_len);
723 return 0; /* auth_none */
724 }
725
726 noio_flag = memalloc_noio_save();
727 con->v2.hmac_tfm = crypto_alloc_shash("hmac(sha256)", 0, 0);
728 memalloc_noio_restore(noio_flag);
729 if (IS_ERR(con->v2.hmac_tfm)) {
730 ret = PTR_ERR(con->v2.hmac_tfm);
731 con->v2.hmac_tfm = NULL;
732 pr_err("failed to allocate hmac tfm context: %d\n", ret);
733 return ret;
734 }
735
736 ret = crypto_shash_setkey(con->v2.hmac_tfm, session_key,
737 session_key_len);
738 if (ret) {
739 pr_err("failed to set hmac key: %d\n", ret);
740 return ret;
741 }
742
743 if (con->v2.con_mode == CEPH_CON_MODE_CRC) {
744 WARN_ON(con_secret_len);
745 return 0; /* auth_x, plain mode */
746 }
747
748 if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) {
749 pr_err("con_secret too small %d\n", con_secret_len);
750 return -EINVAL;
751 }
752
753 noio_flag = memalloc_noio_save();
754 con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0);
755 memalloc_noio_restore(noio_flag);
756 if (IS_ERR(con->v2.gcm_tfm)) {
757 ret = PTR_ERR(con->v2.gcm_tfm);
758 con->v2.gcm_tfm = NULL;
759 pr_err("failed to allocate gcm tfm context: %d\n", ret);
760 return ret;
761 }
762
763 WARN_ON((unsigned long)con_secret &
764 crypto_aead_alignmask(con->v2.gcm_tfm));
765 ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN);
766 if (ret) {
767 pr_err("failed to set gcm key: %d\n", ret);
768 return ret;
769 }
770
771 WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN);
772 ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN);
773 if (ret) {
774 pr_err("failed to set gcm tag size: %d\n", ret);
775 return ret;
776 }
777
778 con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO);
779 if (!con->v2.gcm_req) {
780 pr_err("failed to allocate gcm request\n");
781 return -ENOMEM;
782 }
783
784 crypto_init_wait(&con->v2.gcm_wait);
785 aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
786 crypto_req_done, &con->v2.gcm_wait);
787
788 memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN,
789 CEPH_GCM_IV_LEN);
790 memcpy(&con->v2.out_gcm_nonce,
791 con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN,
792 CEPH_GCM_IV_LEN);
793 return 0; /* auth_x, secure mode */
794 }
795
796 static int hmac_sha256(struct ceph_connection *con, const struct kvec *kvecs,
797 int kvec_cnt, u8 *hmac)
798 {
799 SHASH_DESC_ON_STACK(desc, con->v2.hmac_tfm); /* tfm arg is ignored */
800 int ret;
801 int i;
802
803 dout("%s con %p hmac_tfm %p kvec_cnt %d\n", __func__, con,
804 con->v2.hmac_tfm, kvec_cnt);
805
806 if (!con->v2.hmac_tfm) {
807 memset(hmac, 0, SHA256_DIGEST_SIZE);
808 return 0; /* auth_none */
809 }
810
811 desc->tfm = con->v2.hmac_tfm;
812 ret = crypto_shash_init(desc);
813 if (ret)
814 goto out;
815
816 for (i = 0; i < kvec_cnt; i++) {
817 ret = crypto_shash_update(desc, kvecs[i].iov_base,
818 kvecs[i].iov_len);
819 if (ret)
820 goto out;
821 }
822
823 ret = crypto_shash_final(desc, hmac);
824
825 out:
826 shash_desc_zero(desc);
827 return ret; /* auth_x, both plain and secure modes */
828 }
829
830 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce)
831 {
832 u64 counter;
833
834 counter = le64_to_cpu(nonce->counter);
835 nonce->counter = cpu_to_le64(counter + 1);
836 }
837
838 static int gcm_crypt(struct ceph_connection *con, bool encrypt,
839 struct scatterlist *src, struct scatterlist *dst,
840 int src_len)
841 {
842 struct ceph_gcm_nonce *nonce;
843 int ret;
844
845 nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce;
846
847 aead_request_set_ad(con->v2.gcm_req, 0); /* no AAD */
848 aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce);
849 ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) :
850 crypto_aead_decrypt(con->v2.gcm_req),
851 &con->v2.gcm_wait);
852 if (ret)
853 return ret;
854
855 gcm_inc_nonce(nonce);
856 return 0;
857 }
858
859 static void get_bvec_at(struct ceph_msg_data_cursor *cursor,
860 struct bio_vec *bv)
861 {
862 struct page *page;
863 size_t off, len;
864
865 WARN_ON(!cursor->total_resid);
866
867 /* skip zero-length data items */
868 while (!cursor->resid)
869 ceph_msg_data_advance(cursor, 0);
870
871 /* get a piece of data, cursor isn't advanced */
872 page = ceph_msg_data_next(cursor, &off, &len);
873 bvec_set_page(bv, page, len, off);
874 }
875
876 static int calc_sg_cnt(void *buf, int buf_len)
877 {
878 int sg_cnt;
879
880 if (!buf_len)
881 return 0;
882
883 sg_cnt = need_padding(buf_len) ? 1 : 0;
884 if (is_vmalloc_addr(buf)) {
885 WARN_ON(offset_in_page(buf));
886 sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT;
887 } else {
888 sg_cnt++;
889 }
890
891 return sg_cnt;
892 }
893
894 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor)
895 {
896 int data_len = cursor->total_resid;
897 struct bio_vec bv;
898 int sg_cnt;
899
900 if (!data_len)
901 return 0;
902
903 sg_cnt = need_padding(data_len) ? 1 : 0;
904 do {
905 get_bvec_at(cursor, &bv);
906 sg_cnt++;
907
908 ceph_msg_data_advance(cursor, bv.bv_len);
909 } while (cursor->total_resid);
910
911 return sg_cnt;
912 }
913
914 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad)
915 {
916 void *end = buf + buf_len;
917 struct page *page;
918 int len;
919 void *p;
920
921 if (!buf_len)
922 return;
923
924 if (is_vmalloc_addr(buf)) {
925 p = buf;
926 do {
927 page = vmalloc_to_page(p);
928 len = min_t(int, end - p, PAGE_SIZE);
929 WARN_ON(!page || !len || offset_in_page(p));
930 sg_set_page(*sg, page, len, 0);
931 *sg = sg_next(*sg);
932 p += len;
933 } while (p != end);
934 } else {
935 sg_set_buf(*sg, buf, buf_len);
936 *sg = sg_next(*sg);
937 }
938
939 if (need_padding(buf_len)) {
940 sg_set_buf(*sg, pad, padding_len(buf_len));
941 *sg = sg_next(*sg);
942 }
943 }
944
945 static void init_sgs_cursor(struct scatterlist **sg,
946 struct ceph_msg_data_cursor *cursor, u8 *pad)
947 {
948 int data_len = cursor->total_resid;
949 struct bio_vec bv;
950
951 if (!data_len)
952 return;
953
954 do {
955 get_bvec_at(cursor, &bv);
956 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
957 *sg = sg_next(*sg);
958
959 ceph_msg_data_advance(cursor, bv.bv_len);
960 } while (cursor->total_resid);
961
962 if (need_padding(data_len)) {
963 sg_set_buf(*sg, pad, padding_len(data_len));
964 *sg = sg_next(*sg);
965 }
966 }
967
968 /**
969 * init_sgs_pages: set up scatterlist on an array of page pointers
970 * @sg: scatterlist to populate
971 * @pages: pointer to page array
972 * @dpos: position in the array to start (bytes)
973 * @dlen: len to add to sg (bytes)
974 * @pad: pointer to pad destination (if any)
975 *
976 * Populate the scatterlist from the page array, starting at an arbitrary
977 * byte in the array and running for a specified length.
978 */
979 static void init_sgs_pages(struct scatterlist **sg, struct page **pages,
980 int dpos, int dlen, u8 *pad)
981 {
982 int idx = dpos >> PAGE_SHIFT;
983 int off = offset_in_page(dpos);
984 int resid = dlen;
985
986 do {
987 int len = min(resid, (int)PAGE_SIZE - off);
988
989 sg_set_page(*sg, pages[idx], len, off);
990 *sg = sg_next(*sg);
991 off = 0;
992 ++idx;
993 resid -= len;
994 } while (resid);
995
996 if (need_padding(dlen)) {
997 sg_set_buf(*sg, pad, padding_len(dlen));
998 *sg = sg_next(*sg);
999 }
1000 }
1001
1002 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
1003 u8 *front_pad, u8 *middle_pad, u8 *data_pad,
1004 void *epilogue, struct page **pages, int dpos,
1005 bool add_tag)
1006 {
1007 struct ceph_msg_data_cursor cursor;
1008 struct scatterlist *cur_sg;
1009 int dlen = data_len(msg);
1010 int sg_cnt;
1011 int ret;
1012
1013 if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
1014 return 0;
1015
1016 sg_cnt = 1; /* epilogue + [auth tag] */
1017 if (front_len(msg))
1018 sg_cnt += calc_sg_cnt(msg->front.iov_base,
1019 front_len(msg));
1020 if (middle_len(msg))
1021 sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
1022 middle_len(msg));
1023 if (dlen) {
1024 if (pages) {
1025 sg_cnt += calc_pages_for(dpos, dlen);
1026 if (need_padding(dlen))
1027 sg_cnt++;
1028 } else {
1029 ceph_msg_data_cursor_init(&cursor, msg, dlen);
1030 sg_cnt += calc_sg_cnt_cursor(&cursor);
1031 }
1032 }
1033
1034 ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
1035 if (ret)
1036 return ret;
1037
1038 cur_sg = sgt->sgl;
1039 if (front_len(msg))
1040 init_sgs(&cur_sg, msg->front.iov_base, front_len(msg),
1041 front_pad);
1042 if (middle_len(msg))
1043 init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
1044 middle_pad);
1045 if (dlen) {
1046 if (pages) {
1047 init_sgs_pages(&cur_sg, pages, dpos, dlen, data_pad);
1048 } else {
1049 ceph_msg_data_cursor_init(&cursor, msg, dlen);
1050 init_sgs_cursor(&cur_sg, &cursor, data_pad);
1051 }
1052 }
1053
1054 WARN_ON(!sg_is_last(cur_sg));
1055 sg_set_buf(cur_sg, epilogue,
1056 CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0));
1057 return 0;
1058 }
1059
1060 static int decrypt_preamble(struct ceph_connection *con)
1061 {
1062 struct scatterlist sg;
1063
1064 sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN);
1065 return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN);
1066 }
1067
1068 static int decrypt_control_remainder(struct ceph_connection *con)
1069 {
1070 int ctrl_len = con->v2.in_desc.fd_lens[0];
1071 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1072 int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN;
1073 struct scatterlist sgs[2];
1074
1075 WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len);
1076 WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len);
1077
1078 sg_init_table(sgs, 2);
1079 sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len);
1080 sg_set_buf(&sgs[1], con->v2.in_buf, pt_len);
1081
1082 return gcm_crypt(con, false, sgs, sgs,
1083 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
1084 }
1085
1086 /* Process sparse read data that lives in a buffer */
1087 static int process_v2_sparse_read(struct ceph_connection *con,
1088 struct page **pages, int spos)
1089 {
1090 struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
1091 int ret;
1092
1093 for (;;) {
1094 char *buf = NULL;
1095
1096 ret = con->ops->sparse_read(con, cursor, &buf);
1097 if (ret <= 0)
1098 return ret;
1099
1100 dout("%s: sparse_read return %x buf %p\n", __func__, ret, buf);
1101
1102 do {
1103 int idx = spos >> PAGE_SHIFT;
1104 int soff = offset_in_page(spos);
1105 struct page *spage = con->v2.in_enc_pages[idx];
1106 int len = min_t(int, ret, PAGE_SIZE - soff);
1107
1108 if (buf) {
1109 memcpy_from_page(buf, spage, soff, len);
1110 buf += len;
1111 } else {
1112 struct bio_vec bv;
1113
1114 get_bvec_at(cursor, &bv);
1115 len = min_t(int, len, bv.bv_len);
1116 memcpy_page(bv.bv_page, bv.bv_offset,
1117 spage, soff, len);
1118 ceph_msg_data_advance(cursor, len);
1119 }
1120 spos += len;
1121 ret -= len;
1122 } while (ret);
1123 }
1124 }
1125
1126 static int decrypt_tail(struct ceph_connection *con)
1127 {
1128 struct sg_table enc_sgt = {};
1129 struct sg_table sgt = {};
1130 struct page **pages = NULL;
1131 bool sparse = con->in_msg->sparse_read;
1132 int dpos = 0;
1133 int tail_len;
1134 int ret;
1135
1136 tail_len = tail_onwire_len(con->in_msg, true);
1137 ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages,
1138 con->v2.in_enc_page_cnt, 0, tail_len,
1139 GFP_NOIO);
1140 if (ret)
1141 goto out;
1142
1143 if (sparse) {
1144 dpos = padded_len(front_len(con->in_msg) + padded_len(middle_len(con->in_msg)));
1145 pages = con->v2.in_enc_pages;
1146 }
1147
1148 ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
1149 MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
1150 con->v2.in_buf, pages, dpos, true);
1151 if (ret)
1152 goto out;
1153
1154 dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con,
1155 con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents);
1156 ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len);
1157 if (ret)
1158 goto out;
1159
1160 if (sparse && data_len(con->in_msg)) {
1161 ret = process_v2_sparse_read(con, con->v2.in_enc_pages, dpos);
1162 if (ret)
1163 goto out;
1164 }
1165
1166 WARN_ON(!con->v2.in_enc_page_cnt);
1167 ceph_release_page_vector(con->v2.in_enc_pages,
1168 con->v2.in_enc_page_cnt);
1169 con->v2.in_enc_pages = NULL;
1170 con->v2.in_enc_page_cnt = 0;
1171
1172 out:
1173 sg_free_table(&sgt);
1174 sg_free_table(&enc_sgt);
1175 return ret;
1176 }
1177
1178 static int prepare_banner(struct ceph_connection *con)
1179 {
1180 int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8;
1181 void *buf, *p;
1182
1183 buf = alloc_conn_buf(con, buf_len);
1184 if (!buf)
1185 return -ENOMEM;
1186
1187 p = buf;
1188 ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN);
1189 ceph_encode_16(&p, sizeof(u64) + sizeof(u64));
1190 ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES);
1191 ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES);
1192 WARN_ON(p != buf + buf_len);
1193
1194 add_out_kvec(con, buf, buf_len);
1195 add_out_sign_kvec(con, buf, buf_len);
1196 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1197 return 0;
1198 }
1199
1200 /*
1201 * base:
1202 * preamble
1203 * control body (ctrl_len bytes)
1204 * space for control crc
1205 *
1206 * extdata (optional):
1207 * control body (extdata_len bytes)
1208 *
1209 * Compute control crc and gather base and extdata into:
1210 *
1211 * preamble
1212 * control body (ctrl_len + extdata_len bytes)
1213 * control crc
1214 *
1215 * Preamble should already be encoded at the start of base.
1216 */
1217 static void prepare_head_plain(struct ceph_connection *con, void *base,
1218 int ctrl_len, void *extdata, int extdata_len,
1219 bool to_be_signed)
1220 {
1221 int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN;
1222 void *crcp = base + base_len - CEPH_CRC_LEN;
1223 u32 crc;
1224
1225 crc = crc32c(-1, CTRL_BODY(base), ctrl_len);
1226 if (extdata_len)
1227 crc = crc32c(crc, extdata, extdata_len);
1228 put_unaligned_le32(crc, crcp);
1229
1230 if (!extdata_len) {
1231 add_out_kvec(con, base, base_len);
1232 if (to_be_signed)
1233 add_out_sign_kvec(con, base, base_len);
1234 return;
1235 }
1236
1237 add_out_kvec(con, base, crcp - base);
1238 add_out_kvec(con, extdata, extdata_len);
1239 add_out_kvec(con, crcp, CEPH_CRC_LEN);
1240 if (to_be_signed) {
1241 add_out_sign_kvec(con, base, crcp - base);
1242 add_out_sign_kvec(con, extdata, extdata_len);
1243 add_out_sign_kvec(con, crcp, CEPH_CRC_LEN);
1244 }
1245 }
1246
1247 static int prepare_head_secure_small(struct ceph_connection *con,
1248 void *base, int ctrl_len)
1249 {
1250 struct scatterlist sg;
1251 int ret;
1252
1253 /* inline buffer padding? */
1254 if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN)
1255 memset(CTRL_BODY(base) + ctrl_len, 0,
1256 CEPH_PREAMBLE_INLINE_LEN - ctrl_len);
1257
1258 sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN);
1259 ret = gcm_crypt(con, true, &sg, &sg,
1260 CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN);
1261 if (ret)
1262 return ret;
1263
1264 add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN);
1265 return 0;
1266 }
1267
1268 /*
1269 * base:
1270 * preamble
1271 * control body (ctrl_len bytes)
1272 * space for padding, if needed
1273 * space for control remainder auth tag
1274 * space for preamble auth tag
1275 *
1276 * Encrypt preamble and the inline portion, then encrypt the remainder
1277 * and gather into:
1278 *
1279 * preamble
1280 * control body (48 bytes)
1281 * preamble auth tag
1282 * control body (ctrl_len - 48 bytes)
1283 * zero padding, if needed
1284 * control remainder auth tag
1285 *
1286 * Preamble should already be encoded at the start of base.
1287 */
1288 static int prepare_head_secure_big(struct ceph_connection *con,
1289 void *base, int ctrl_len)
1290 {
1291 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1292 void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN;
1293 void *rem_tag = rem + padded_len(rem_len);
1294 void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN;
1295 struct scatterlist sgs[2];
1296 int ret;
1297
1298 sg_init_table(sgs, 2);
1299 sg_set_buf(&sgs[0], base, rem - base);
1300 sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN);
1301 ret = gcm_crypt(con, true, sgs, sgs, rem - base);
1302 if (ret)
1303 return ret;
1304
1305 /* control remainder padding? */
1306 if (need_padding(rem_len))
1307 memset(rem + rem_len, 0, padding_len(rem_len));
1308
1309 sg_init_one(&sgs[0], rem, pmbl_tag - rem);
1310 ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem);
1311 if (ret)
1312 return ret;
1313
1314 add_out_kvec(con, base, rem - base);
1315 add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN);
1316 add_out_kvec(con, rem, pmbl_tag - rem);
1317 return 0;
1318 }
1319
1320 static int __prepare_control(struct ceph_connection *con, int tag,
1321 void *base, int ctrl_len, void *extdata,
1322 int extdata_len, bool to_be_signed)
1323 {
1324 int total_len = ctrl_len + extdata_len;
1325 struct ceph_frame_desc desc;
1326 int ret;
1327
1328 dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag,
1329 total_len, ctrl_len, extdata_len);
1330
1331 /* extdata may be vmalloc'ed but not base */
1332 if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len))
1333 return -EINVAL;
1334
1335 init_frame_desc(&desc, tag, &total_len, 1);
1336 encode_preamble(&desc, base);
1337
1338 if (con_secure(con)) {
1339 if (WARN_ON(extdata_len || to_be_signed))
1340 return -EINVAL;
1341
1342 if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN)
1343 /* fully inlined, inline buffer may need padding */
1344 ret = prepare_head_secure_small(con, base, ctrl_len);
1345 else
1346 /* partially inlined, inline buffer is full */
1347 ret = prepare_head_secure_big(con, base, ctrl_len);
1348 if (ret)
1349 return ret;
1350 } else {
1351 prepare_head_plain(con, base, ctrl_len, extdata, extdata_len,
1352 to_be_signed);
1353 }
1354
1355 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1356 return 0;
1357 }
1358
1359 static int prepare_control(struct ceph_connection *con, int tag,
1360 void *base, int ctrl_len)
1361 {
1362 return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false);
1363 }
1364
1365 static int prepare_hello(struct ceph_connection *con)
1366 {
1367 void *buf, *p;
1368 int ctrl_len;
1369
1370 ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr);
1371 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1372 if (!buf)
1373 return -ENOMEM;
1374
1375 p = CTRL_BODY(buf);
1376 ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT);
1377 ceph_encode_entity_addr(&p, &con->peer_addr);
1378 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1379
1380 return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len,
1381 NULL, 0, true);
1382 }
1383
1384 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */
1385 #define AUTH_BUF_LEN (512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN)
1386
1387 static int prepare_auth_request(struct ceph_connection *con)
1388 {
1389 void *authorizer, *authorizer_copy;
1390 int ctrl_len, authorizer_len;
1391 void *buf;
1392 int ret;
1393
1394 ctrl_len = AUTH_BUF_LEN;
1395 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1396 if (!buf)
1397 return -ENOMEM;
1398
1399 mutex_unlock(&con->mutex);
1400 ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len,
1401 &authorizer, &authorizer_len);
1402 mutex_lock(&con->mutex);
1403 if (con->state != CEPH_CON_S_V2_HELLO) {
1404 dout("%s con %p state changed to %d\n", __func__, con,
1405 con->state);
1406 return -EAGAIN;
1407 }
1408
1409 dout("%s con %p get_auth_request ret %d\n", __func__, con, ret);
1410 if (ret)
1411 return ret;
1412
1413 authorizer_copy = alloc_conn_buf(con, authorizer_len);
1414 if (!authorizer_copy)
1415 return -ENOMEM;
1416
1417 memcpy(authorizer_copy, authorizer, authorizer_len);
1418
1419 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len,
1420 authorizer_copy, authorizer_len, true);
1421 }
1422
1423 static int prepare_auth_request_more(struct ceph_connection *con,
1424 void *reply, int reply_len)
1425 {
1426 int ctrl_len, authorizer_len;
1427 void *authorizer;
1428 void *buf;
1429 int ret;
1430
1431 ctrl_len = AUTH_BUF_LEN;
1432 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1433 if (!buf)
1434 return -ENOMEM;
1435
1436 mutex_unlock(&con->mutex);
1437 ret = con->ops->handle_auth_reply_more(con, reply, reply_len,
1438 CTRL_BODY(buf), &ctrl_len,
1439 &authorizer, &authorizer_len);
1440 mutex_lock(&con->mutex);
1441 if (con->state != CEPH_CON_S_V2_AUTH) {
1442 dout("%s con %p state changed to %d\n", __func__, con,
1443 con->state);
1444 return -EAGAIN;
1445 }
1446
1447 dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret);
1448 if (ret)
1449 return ret;
1450
1451 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf,
1452 ctrl_len, authorizer, authorizer_len, true);
1453 }
1454
1455 static int prepare_auth_signature(struct ceph_connection *con)
1456 {
1457 void *buf;
1458 int ret;
1459
1460 buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE,
1461 con_secure(con)));
1462 if (!buf)
1463 return -ENOMEM;
1464
1465 ret = hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt,
1466 CTRL_BODY(buf));
1467 if (ret)
1468 return ret;
1469
1470 return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf,
1471 SHA256_DIGEST_SIZE);
1472 }
1473
1474 static int prepare_client_ident(struct ceph_connection *con)
1475 {
1476 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1477 struct ceph_client *client = from_msgr(con->msgr);
1478 u64 global_id = ceph_client_gid(client);
1479 void *buf, *p;
1480 int ctrl_len;
1481
1482 WARN_ON(con->v2.server_cookie);
1483 WARN_ON(con->v2.connect_seq);
1484 WARN_ON(con->v2.peer_global_seq);
1485
1486 if (!con->v2.client_cookie) {
1487 do {
1488 get_random_bytes(&con->v2.client_cookie,
1489 sizeof(con->v2.client_cookie));
1490 } while (!con->v2.client_cookie);
1491 dout("%s con %p generated cookie 0x%llx\n", __func__, con,
1492 con->v2.client_cookie);
1493 } else {
1494 dout("%s con %p cookie already set 0x%llx\n", __func__, con,
1495 con->v2.client_cookie);
1496 }
1497
1498 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",
1499 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1500 ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce),
1501 global_id, con->v2.global_seq, client->supported_features,
1502 client->required_features, con->v2.client_cookie);
1503
1504 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) +
1505 ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8;
1506 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1507 if (!buf)
1508 return -ENOMEM;
1509
1510 p = CTRL_BODY(buf);
1511 ceph_encode_8(&p, 2); /* addrvec marker */
1512 ceph_encode_32(&p, 1); /* addr_cnt */
1513 ceph_encode_entity_addr(&p, my_addr);
1514 ceph_encode_entity_addr(&p, &con->peer_addr);
1515 ceph_encode_64(&p, global_id);
1516 ceph_encode_64(&p, con->v2.global_seq);
1517 ceph_encode_64(&p, client->supported_features);
1518 ceph_encode_64(&p, client->required_features);
1519 ceph_encode_64(&p, 0); /* flags */
1520 ceph_encode_64(&p, con->v2.client_cookie);
1521 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1522
1523 return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len);
1524 }
1525
1526 static int prepare_session_reconnect(struct ceph_connection *con)
1527 {
1528 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1529 void *buf, *p;
1530 int ctrl_len;
1531
1532 WARN_ON(!con->v2.client_cookie);
1533 WARN_ON(!con->v2.server_cookie);
1534 WARN_ON(!con->v2.connect_seq);
1535 WARN_ON(!con->v2.peer_global_seq);
1536
1537 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",
1538 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1539 con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq,
1540 con->v2.connect_seq, con->in_seq);
1541
1542 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8;
1543 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1544 if (!buf)
1545 return -ENOMEM;
1546
1547 p = CTRL_BODY(buf);
1548 ceph_encode_8(&p, 2); /* entity_addrvec_t marker */
1549 ceph_encode_32(&p, 1); /* my_addrs len */
1550 ceph_encode_entity_addr(&p, my_addr);
1551 ceph_encode_64(&p, con->v2.client_cookie);
1552 ceph_encode_64(&p, con->v2.server_cookie);
1553 ceph_encode_64(&p, con->v2.global_seq);
1554 ceph_encode_64(&p, con->v2.connect_seq);
1555 ceph_encode_64(&p, con->in_seq);
1556 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1557
1558 return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len);
1559 }
1560
1561 static int prepare_keepalive2(struct ceph_connection *con)
1562 {
1563 struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
1564 struct timespec64 now;
1565
1566 ktime_get_real_ts64(&now);
1567 dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec,
1568 now.tv_nsec);
1569
1570 ceph_encode_timespec64(ts, &now);
1571
1572 reset_out_kvecs(con);
1573 return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf,
1574 sizeof(struct ceph_timespec));
1575 }
1576
1577 static int prepare_ack(struct ceph_connection *con)
1578 {
1579 void *p;
1580
1581 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1582 con->in_seq_acked, con->in_seq);
1583 con->in_seq_acked = con->in_seq;
1584
1585 p = CTRL_BODY(con->v2.out_buf);
1586 ceph_encode_64(&p, con->in_seq_acked);
1587
1588 reset_out_kvecs(con);
1589 return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8);
1590 }
1591
1592 static void prepare_epilogue_plain(struct ceph_connection *con, bool aborted)
1593 {
1594 dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con,
1595 con->out_msg, aborted, con->v2.out_epil.front_crc,
1596 con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc);
1597
1598 encode_epilogue_plain(con, aborted);
1599 add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN);
1600 }
1601
1602 /*
1603 * For "used" empty segments, crc is -1. For unused (trailing)
1604 * segments, crc is 0.
1605 */
1606 static void prepare_message_plain(struct ceph_connection *con)
1607 {
1608 struct ceph_msg *msg = con->out_msg;
1609
1610 prepare_head_plain(con, con->v2.out_buf,
1611 sizeof(struct ceph_msg_header2), NULL, 0, false);
1612
1613 if (!front_len(msg) && !middle_len(msg)) {
1614 if (!data_len(msg)) {
1615 /*
1616 * Empty message: once the head is written,
1617 * we are done -- there is no epilogue.
1618 */
1619 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1620 return;
1621 }
1622
1623 con->v2.out_epil.front_crc = -1;
1624 con->v2.out_epil.middle_crc = -1;
1625 con->v2.out_state = OUT_S_QUEUE_DATA;
1626 return;
1627 }
1628
1629 if (front_len(msg)) {
1630 con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base,
1631 front_len(msg));
1632 add_out_kvec(con, msg->front.iov_base, front_len(msg));
1633 } else {
1634 /* middle (at least) is there, checked above */
1635 con->v2.out_epil.front_crc = -1;
1636 }
1637
1638 if (middle_len(msg)) {
1639 con->v2.out_epil.middle_crc =
1640 crc32c(-1, msg->middle->vec.iov_base, middle_len(msg));
1641 add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1642 } else {
1643 con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0;
1644 }
1645
1646 if (data_len(msg)) {
1647 con->v2.out_state = OUT_S_QUEUE_DATA;
1648 } else {
1649 con->v2.out_epil.data_crc = 0;
1650 prepare_epilogue_plain(con, false);
1651 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1652 }
1653 }
1654
1655 /*
1656 * Unfortunately the kernel crypto API doesn't support streaming
1657 * (piecewise) operation for AEAD algorithms, so we can't get away
1658 * with a fixed size buffer and a couple sgs. Instead, we have to
1659 * allocate pages for the entire tail of the message (currently up
1660 * to ~32M) and two sgs arrays (up to ~256K each)...
1661 */
1662 static int prepare_message_secure(struct ceph_connection *con)
1663 {
1664 void *zerop = page_address(ceph_zero_page);
1665 struct sg_table enc_sgt = {};
1666 struct sg_table sgt = {};
1667 struct page **enc_pages;
1668 int enc_page_cnt;
1669 int tail_len;
1670 int ret;
1671
1672 ret = prepare_head_secure_small(con, con->v2.out_buf,
1673 sizeof(struct ceph_msg_header2));
1674 if (ret)
1675 return ret;
1676
1677 tail_len = tail_onwire_len(con->out_msg, true);
1678 if (!tail_len) {
1679 /*
1680 * Empty message: once the head is written,
1681 * we are done -- there is no epilogue.
1682 */
1683 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1684 return 0;
1685 }
1686
1687 encode_epilogue_secure(con, false);
1688 ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop,
1689 &con->v2.out_epil, NULL, 0, false);
1690 if (ret)
1691 goto out;
1692
1693 enc_page_cnt = calc_pages_for(0, tail_len);
1694 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1695 if (IS_ERR(enc_pages)) {
1696 ret = PTR_ERR(enc_pages);
1697 goto out;
1698 }
1699
1700 WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt);
1701 con->v2.out_enc_pages = enc_pages;
1702 con->v2.out_enc_page_cnt = enc_page_cnt;
1703 con->v2.out_enc_resid = tail_len;
1704 con->v2.out_enc_i = 0;
1705
1706 ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt,
1707 0, tail_len, GFP_NOIO);
1708 if (ret)
1709 goto out;
1710
1711 ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl,
1712 tail_len - CEPH_GCM_TAG_LEN);
1713 if (ret)
1714 goto out;
1715
1716 dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con,
1717 con->out_msg, sgt.orig_nents, enc_page_cnt);
1718 con->v2.out_state = OUT_S_QUEUE_ENC_PAGE;
1719
1720 out:
1721 sg_free_table(&sgt);
1722 sg_free_table(&enc_sgt);
1723 return ret;
1724 }
1725
1726 static int prepare_message(struct ceph_connection *con)
1727 {
1728 int lens[] = {
1729 sizeof(struct ceph_msg_header2),
1730 front_len(con->out_msg),
1731 middle_len(con->out_msg),
1732 data_len(con->out_msg)
1733 };
1734 struct ceph_frame_desc desc;
1735 int ret;
1736
1737 dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con,
1738 con->out_msg, lens[0], lens[1], lens[2], lens[3]);
1739
1740 if (con->in_seq > con->in_seq_acked) {
1741 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1742 con->in_seq_acked, con->in_seq);
1743 con->in_seq_acked = con->in_seq;
1744 }
1745
1746 reset_out_kvecs(con);
1747 init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4);
1748 encode_preamble(&desc, con->v2.out_buf);
1749 fill_header2(CTRL_BODY(con->v2.out_buf), &con->out_msg->hdr,
1750 con->in_seq_acked);
1751
1752 if (con_secure(con)) {
1753 ret = prepare_message_secure(con);
1754 if (ret)
1755 return ret;
1756 } else {
1757 prepare_message_plain(con);
1758 }
1759
1760 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1761 return 0;
1762 }
1763
1764 static int prepare_read_banner_prefix(struct ceph_connection *con)
1765 {
1766 void *buf;
1767
1768 buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN);
1769 if (!buf)
1770 return -ENOMEM;
1771
1772 reset_in_kvecs(con);
1773 add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1774 add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1775 con->state = CEPH_CON_S_V2_BANNER_PREFIX;
1776 return 0;
1777 }
1778
1779 static int prepare_read_banner_payload(struct ceph_connection *con,
1780 int payload_len)
1781 {
1782 void *buf;
1783
1784 buf = alloc_conn_buf(con, payload_len);
1785 if (!buf)
1786 return -ENOMEM;
1787
1788 reset_in_kvecs(con);
1789 add_in_kvec(con, buf, payload_len);
1790 add_in_sign_kvec(con, buf, payload_len);
1791 con->state = CEPH_CON_S_V2_BANNER_PAYLOAD;
1792 return 0;
1793 }
1794
1795 static void prepare_read_preamble(struct ceph_connection *con)
1796 {
1797 reset_in_kvecs(con);
1798 add_in_kvec(con, con->v2.in_buf,
1799 con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN :
1800 CEPH_PREAMBLE_PLAIN_LEN);
1801 con->v2.in_state = IN_S_HANDLE_PREAMBLE;
1802 }
1803
1804 static int prepare_read_control(struct ceph_connection *con)
1805 {
1806 int ctrl_len = con->v2.in_desc.fd_lens[0];
1807 int head_len;
1808 void *buf;
1809
1810 reset_in_kvecs(con);
1811 if (con->state == CEPH_CON_S_V2_HELLO ||
1812 con->state == CEPH_CON_S_V2_AUTH) {
1813 head_len = head_onwire_len(ctrl_len, false);
1814 buf = alloc_conn_buf(con, head_len);
1815 if (!buf)
1816 return -ENOMEM;
1817
1818 /* preserve preamble */
1819 memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN);
1820
1821 add_in_kvec(con, CTRL_BODY(buf), ctrl_len);
1822 add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN);
1823 add_in_sign_kvec(con, buf, head_len);
1824 } else {
1825 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
1826 buf = alloc_conn_buf(con, ctrl_len);
1827 if (!buf)
1828 return -ENOMEM;
1829
1830 add_in_kvec(con, buf, ctrl_len);
1831 } else {
1832 add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len);
1833 }
1834 add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN);
1835 }
1836 con->v2.in_state = IN_S_HANDLE_CONTROL;
1837 return 0;
1838 }
1839
1840 static int prepare_read_control_remainder(struct ceph_connection *con)
1841 {
1842 int ctrl_len = con->v2.in_desc.fd_lens[0];
1843 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1844 void *buf;
1845
1846 buf = alloc_conn_buf(con, ctrl_len);
1847 if (!buf)
1848 return -ENOMEM;
1849
1850 memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN);
1851
1852 reset_in_kvecs(con);
1853 add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len);
1854 add_in_kvec(con, con->v2.in_buf,
1855 padding_len(rem_len) + CEPH_GCM_TAG_LEN);
1856 con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER;
1857 return 0;
1858 }
1859
1860 static int prepare_read_data(struct ceph_connection *con)
1861 {
1862 struct bio_vec bv;
1863
1864 con->in_data_crc = -1;
1865 ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg,
1866 data_len(con->in_msg));
1867
1868 get_bvec_at(&con->v2.in_cursor, &bv);
1869 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1870 if (unlikely(!con->bounce_page)) {
1871 con->bounce_page = alloc_page(GFP_NOIO);
1872 if (!con->bounce_page) {
1873 pr_err("failed to allocate bounce page\n");
1874 return -ENOMEM;
1875 }
1876 }
1877
1878 bv.bv_page = con->bounce_page;
1879 bv.bv_offset = 0;
1880 }
1881 set_in_bvec(con, &bv);
1882 con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT;
1883 return 0;
1884 }
1885
1886 static void prepare_read_data_cont(struct ceph_connection *con)
1887 {
1888 struct bio_vec bv;
1889
1890 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1891 con->in_data_crc = crc32c(con->in_data_crc,
1892 page_address(con->bounce_page),
1893 con->v2.in_bvec.bv_len);
1894
1895 get_bvec_at(&con->v2.in_cursor, &bv);
1896 memcpy_to_page(bv.bv_page, bv.bv_offset,
1897 page_address(con->bounce_page),
1898 con->v2.in_bvec.bv_len);
1899 } else {
1900 con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1901 con->v2.in_bvec.bv_page,
1902 con->v2.in_bvec.bv_offset,
1903 con->v2.in_bvec.bv_len);
1904 }
1905
1906 ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len);
1907 if (con->v2.in_cursor.total_resid) {
1908 get_bvec_at(&con->v2.in_cursor, &bv);
1909 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1910 bv.bv_page = con->bounce_page;
1911 bv.bv_offset = 0;
1912 }
1913 set_in_bvec(con, &bv);
1914 WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT);
1915 return;
1916 }
1917
1918 /*
1919 * We've read all data. Prepare to read epilogue.
1920 */
1921 reset_in_kvecs(con);
1922 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1923 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1924 }
1925
1926 static int prepare_sparse_read_cont(struct ceph_connection *con)
1927 {
1928 int ret;
1929 struct bio_vec bv;
1930 char *buf = NULL;
1931 struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
1932
1933 WARN_ON(con->v2.in_state != IN_S_PREPARE_SPARSE_DATA_CONT);
1934
1935 if (iov_iter_is_bvec(&con->v2.in_iter)) {
1936 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1937 con->in_data_crc = crc32c(con->in_data_crc,
1938 page_address(con->bounce_page),
1939 con->v2.in_bvec.bv_len);
1940 get_bvec_at(cursor, &bv);
1941 memcpy_to_page(bv.bv_page, bv.bv_offset,
1942 page_address(con->bounce_page),
1943 con->v2.in_bvec.bv_len);
1944 } else {
1945 con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1946 con->v2.in_bvec.bv_page,
1947 con->v2.in_bvec.bv_offset,
1948 con->v2.in_bvec.bv_len);
1949 }
1950
1951 ceph_msg_data_advance(cursor, con->v2.in_bvec.bv_len);
1952 cursor->sr_resid -= con->v2.in_bvec.bv_len;
1953 dout("%s: advance by 0x%x sr_resid 0x%x\n", __func__,
1954 con->v2.in_bvec.bv_len, cursor->sr_resid);
1955 WARN_ON_ONCE(cursor->sr_resid > cursor->total_resid);
1956 if (cursor->sr_resid) {
1957 get_bvec_at(cursor, &bv);
1958 if (bv.bv_len > cursor->sr_resid)
1959 bv.bv_len = cursor->sr_resid;
1960 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1961 bv.bv_page = con->bounce_page;
1962 bv.bv_offset = 0;
1963 }
1964 set_in_bvec(con, &bv);
1965 con->v2.data_len_remain -= bv.bv_len;
1966 return 0;
1967 }
1968 } else if (iov_iter_is_kvec(&con->v2.in_iter)) {
1969 /* On first call, we have no kvec so don't compute crc */
1970 if (con->v2.in_kvec_cnt) {
1971 WARN_ON_ONCE(con->v2.in_kvec_cnt > 1);
1972 con->in_data_crc = crc32c(con->in_data_crc,
1973 con->v2.in_kvecs[0].iov_base,
1974 con->v2.in_kvecs[0].iov_len);
1975 }
1976 } else {
1977 return -EIO;
1978 }
1979
1980 /* get next extent */
1981 ret = con->ops->sparse_read(con, cursor, &buf);
1982 if (ret <= 0) {
1983 if (ret < 0)
1984 return ret;
1985
1986 reset_in_kvecs(con);
1987 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1988 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1989 return 0;
1990 }
1991
1992 if (buf) {
1993 /* receive into buffer */
1994 reset_in_kvecs(con);
1995 add_in_kvec(con, buf, ret);
1996 con->v2.data_len_remain -= ret;
1997 return 0;
1998 }
1999
2000 if (ret > cursor->total_resid) {
2001 pr_warn("%s: ret 0x%x total_resid 0x%zx resid 0x%zx\n",
2002 __func__, ret, cursor->total_resid, cursor->resid);
2003 return -EIO;
2004 }
2005 get_bvec_at(cursor, &bv);
2006 if (bv.bv_len > cursor->sr_resid)
2007 bv.bv_len = cursor->sr_resid;
2008 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
2009 if (unlikely(!con->bounce_page)) {
2010 con->bounce_page = alloc_page(GFP_NOIO);
2011 if (!con->bounce_page) {
2012 pr_err("failed to allocate bounce page\n");
2013 return -ENOMEM;
2014 }
2015 }
2016
2017 bv.bv_page = con->bounce_page;
2018 bv.bv_offset = 0;
2019 }
2020 set_in_bvec(con, &bv);
2021 con->v2.data_len_remain -= ret;
2022 return ret;
2023 }
2024
2025 static int prepare_sparse_read_data(struct ceph_connection *con)
2026 {
2027 struct ceph_msg *msg = con->in_msg;
2028
2029 dout("%s: starting sparse read\n", __func__);
2030
2031 if (WARN_ON_ONCE(!con->ops->sparse_read))
2032 return -EOPNOTSUPP;
2033
2034 if (!con_secure(con))
2035 con->in_data_crc = -1;
2036
2037 reset_in_kvecs(con);
2038 con->v2.in_state = IN_S_PREPARE_SPARSE_DATA_CONT;
2039 con->v2.data_len_remain = data_len(msg);
2040 return prepare_sparse_read_cont(con);
2041 }
2042
2043 static int prepare_read_tail_plain(struct ceph_connection *con)
2044 {
2045 struct ceph_msg *msg = con->in_msg;
2046
2047 if (!front_len(msg) && !middle_len(msg)) {
2048 WARN_ON(!data_len(msg));
2049 return prepare_read_data(con);
2050 }
2051
2052 reset_in_kvecs(con);
2053 if (front_len(msg)) {
2054 add_in_kvec(con, msg->front.iov_base, front_len(msg));
2055 WARN_ON(msg->front.iov_len != front_len(msg));
2056 }
2057 if (middle_len(msg)) {
2058 add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
2059 WARN_ON(msg->middle->vec.iov_len != middle_len(msg));
2060 }
2061
2062 if (data_len(msg)) {
2063 if (msg->sparse_read)
2064 con->v2.in_state = IN_S_PREPARE_SPARSE_DATA;
2065 else
2066 con->v2.in_state = IN_S_PREPARE_READ_DATA;
2067 } else {
2068 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
2069 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2070 }
2071 return 0;
2072 }
2073
2074 static void prepare_read_enc_page(struct ceph_connection *con)
2075 {
2076 struct bio_vec bv;
2077
2078 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i,
2079 con->v2.in_enc_resid);
2080 WARN_ON(!con->v2.in_enc_resid);
2081
2082 bvec_set_page(&bv, con->v2.in_enc_pages[con->v2.in_enc_i],
2083 min(con->v2.in_enc_resid, (int)PAGE_SIZE), 0);
2084
2085 set_in_bvec(con, &bv);
2086 con->v2.in_enc_i++;
2087 con->v2.in_enc_resid -= bv.bv_len;
2088
2089 if (con->v2.in_enc_resid) {
2090 con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE;
2091 return;
2092 }
2093
2094 /*
2095 * We are set to read the last piece of ciphertext (ending
2096 * with epilogue) + auth tag.
2097 */
2098 WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt);
2099 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2100 }
2101
2102 static int prepare_read_tail_secure(struct ceph_connection *con)
2103 {
2104 struct page **enc_pages;
2105 int enc_page_cnt;
2106 int tail_len;
2107
2108 tail_len = tail_onwire_len(con->in_msg, true);
2109 WARN_ON(!tail_len);
2110
2111 enc_page_cnt = calc_pages_for(0, tail_len);
2112 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
2113 if (IS_ERR(enc_pages))
2114 return PTR_ERR(enc_pages);
2115
2116 WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt);
2117 con->v2.in_enc_pages = enc_pages;
2118 con->v2.in_enc_page_cnt = enc_page_cnt;
2119 con->v2.in_enc_resid = tail_len;
2120 con->v2.in_enc_i = 0;
2121
2122 prepare_read_enc_page(con);
2123 return 0;
2124 }
2125
2126 static void __finish_skip(struct ceph_connection *con)
2127 {
2128 con->in_seq++;
2129 prepare_read_preamble(con);
2130 }
2131
2132 static void prepare_skip_message(struct ceph_connection *con)
2133 {
2134 struct ceph_frame_desc *desc = &con->v2.in_desc;
2135 int tail_len;
2136
2137 dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1],
2138 desc->fd_lens[2], desc->fd_lens[3]);
2139
2140 tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2],
2141 desc->fd_lens[3], con_secure(con));
2142 if (!tail_len) {
2143 __finish_skip(con);
2144 } else {
2145 set_in_skip(con, tail_len);
2146 con->v2.in_state = IN_S_FINISH_SKIP;
2147 }
2148 }
2149
2150 static int process_banner_prefix(struct ceph_connection *con)
2151 {
2152 int payload_len;
2153 void *p;
2154
2155 WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN);
2156
2157 p = con->v2.in_kvecs[0].iov_base;
2158 if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) {
2159 if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN))
2160 con->error_msg = "server is speaking msgr1 protocol";
2161 else
2162 con->error_msg = "protocol error, bad banner";
2163 return -EINVAL;
2164 }
2165
2166 p += CEPH_BANNER_V2_LEN;
2167 payload_len = ceph_decode_16(&p);
2168 dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2169
2170 return prepare_read_banner_payload(con, payload_len);
2171 }
2172
2173 static int process_banner_payload(struct ceph_connection *con)
2174 {
2175 void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len;
2176 u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES;
2177 u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES;
2178 u64 server_feat, server_req_feat;
2179 void *p;
2180 int ret;
2181
2182 p = con->v2.in_kvecs[0].iov_base;
2183 ceph_decode_64_safe(&p, end, server_feat, bad);
2184 ceph_decode_64_safe(&p, end, server_req_feat, bad);
2185
2186 dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n",
2187 __func__, con, server_feat, server_req_feat);
2188
2189 if (req_feat & ~server_feat) {
2190 pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2191 server_feat, req_feat & ~server_feat);
2192 con->error_msg = "missing required protocol features";
2193 return -EINVAL;
2194 }
2195 if (server_req_feat & ~feat) {
2196 pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2197 feat, server_req_feat & ~feat);
2198 con->error_msg = "missing required protocol features";
2199 return -EINVAL;
2200 }
2201
2202 /* no reset_out_kvecs() as our banner may still be pending */
2203 ret = prepare_hello(con);
2204 if (ret) {
2205 pr_err("prepare_hello failed: %d\n", ret);
2206 return ret;
2207 }
2208
2209 con->state = CEPH_CON_S_V2_HELLO;
2210 prepare_read_preamble(con);
2211 return 0;
2212
2213 bad:
2214 pr_err("failed to decode banner payload\n");
2215 return -EINVAL;
2216 }
2217
2218 static int process_hello(struct ceph_connection *con, void *p, void *end)
2219 {
2220 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
2221 struct ceph_entity_addr addr_for_me;
2222 u8 entity_type;
2223 int ret;
2224
2225 if (con->state != CEPH_CON_S_V2_HELLO) {
2226 con->error_msg = "protocol error, unexpected hello";
2227 return -EINVAL;
2228 }
2229
2230 ceph_decode_8_safe(&p, end, entity_type, bad);
2231 ret = ceph_decode_entity_addr(&p, end, &addr_for_me);
2232 if (ret) {
2233 pr_err("failed to decode addr_for_me: %d\n", ret);
2234 return ret;
2235 }
2236
2237 dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con,
2238 entity_type, ceph_pr_addr(&addr_for_me));
2239
2240 if (entity_type != con->peer_name.type) {
2241 pr_err("bad peer type, want %d, got %d\n",
2242 con->peer_name.type, entity_type);
2243 con->error_msg = "wrong peer at address";
2244 return -EINVAL;
2245 }
2246
2247 /*
2248 * Set our address to the address our first peer (i.e. monitor)
2249 * sees that we are connecting from. If we are behind some sort
2250 * of NAT and want to be identified by some private (not NATed)
2251 * address, ip option should be used.
2252 */
2253 if (ceph_addr_is_blank(my_addr)) {
2254 memcpy(&my_addr->in_addr, &addr_for_me.in_addr,
2255 sizeof(my_addr->in_addr));
2256 ceph_addr_set_port(my_addr, 0);
2257 dout("%s con %p set my addr %s, as seen by peer %s\n",
2258 __func__, con, ceph_pr_addr(my_addr),
2259 ceph_pr_addr(&con->peer_addr));
2260 } else {
2261 dout("%s con %p my addr already set %s\n",
2262 __func__, con, ceph_pr_addr(my_addr));
2263 }
2264
2265 WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr));
2266 WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY);
2267 WARN_ON(!my_addr->nonce);
2268
2269 /* no reset_out_kvecs() as our hello may still be pending */
2270 ret = prepare_auth_request(con);
2271 if (ret) {
2272 if (ret != -EAGAIN)
2273 pr_err("prepare_auth_request failed: %d\n", ret);
2274 return ret;
2275 }
2276
2277 con->state = CEPH_CON_S_V2_AUTH;
2278 return 0;
2279
2280 bad:
2281 pr_err("failed to decode hello\n");
2282 return -EINVAL;
2283 }
2284
2285 static int process_auth_bad_method(struct ceph_connection *con,
2286 void *p, void *end)
2287 {
2288 int allowed_protos[8], allowed_modes[8];
2289 int allowed_proto_cnt, allowed_mode_cnt;
2290 int used_proto, result;
2291 int ret;
2292 int i;
2293
2294 if (con->state != CEPH_CON_S_V2_AUTH) {
2295 con->error_msg = "protocol error, unexpected auth_bad_method";
2296 return -EINVAL;
2297 }
2298
2299 ceph_decode_32_safe(&p, end, used_proto, bad);
2300 ceph_decode_32_safe(&p, end, result, bad);
2301 dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto,
2302 result);
2303
2304 ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad);
2305 if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) {
2306 pr_err("allowed_protos too big %d\n", allowed_proto_cnt);
2307 return -EINVAL;
2308 }
2309 for (i = 0; i < allowed_proto_cnt; i++) {
2310 ceph_decode_32_safe(&p, end, allowed_protos[i], bad);
2311 dout("%s con %p allowed_protos[%d] %d\n", __func__, con,
2312 i, allowed_protos[i]);
2313 }
2314
2315 ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad);
2316 if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) {
2317 pr_err("allowed_modes too big %d\n", allowed_mode_cnt);
2318 return -EINVAL;
2319 }
2320 for (i = 0; i < allowed_mode_cnt; i++) {
2321 ceph_decode_32_safe(&p, end, allowed_modes[i], bad);
2322 dout("%s con %p allowed_modes[%d] %d\n", __func__, con,
2323 i, allowed_modes[i]);
2324 }
2325
2326 mutex_unlock(&con->mutex);
2327 ret = con->ops->handle_auth_bad_method(con, used_proto, result,
2328 allowed_protos,
2329 allowed_proto_cnt,
2330 allowed_modes,
2331 allowed_mode_cnt);
2332 mutex_lock(&con->mutex);
2333 if (con->state != CEPH_CON_S_V2_AUTH) {
2334 dout("%s con %p state changed to %d\n", __func__, con,
2335 con->state);
2336 return -EAGAIN;
2337 }
2338
2339 dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret);
2340 return ret;
2341
2342 bad:
2343 pr_err("failed to decode auth_bad_method\n");
2344 return -EINVAL;
2345 }
2346
2347 static int process_auth_reply_more(struct ceph_connection *con,
2348 void *p, void *end)
2349 {
2350 int payload_len;
2351 int ret;
2352
2353 if (con->state != CEPH_CON_S_V2_AUTH) {
2354 con->error_msg = "protocol error, unexpected auth_reply_more";
2355 return -EINVAL;
2356 }
2357
2358 ceph_decode_32_safe(&p, end, payload_len, bad);
2359 ceph_decode_need(&p, end, payload_len, bad);
2360
2361 dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2362
2363 reset_out_kvecs(con);
2364 ret = prepare_auth_request_more(con, p, payload_len);
2365 if (ret) {
2366 if (ret != -EAGAIN)
2367 pr_err("prepare_auth_request_more failed: %d\n", ret);
2368 return ret;
2369 }
2370
2371 return 0;
2372
2373 bad:
2374 pr_err("failed to decode auth_reply_more\n");
2375 return -EINVAL;
2376 }
2377
2378 /*
2379 * Align session_key and con_secret to avoid GFP_ATOMIC allocation
2380 * inside crypto_shash_setkey() and crypto_aead_setkey() called from
2381 * setup_crypto(). __aligned(16) isn't guaranteed to work for stack
2382 * objects, so do it by hand.
2383 */
2384 static int process_auth_done(struct ceph_connection *con, void *p, void *end)
2385 {
2386 u8 session_key_buf[CEPH_KEY_LEN + 16];
2387 u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16];
2388 u8 *session_key = PTR_ALIGN(&session_key_buf[0], 16);
2389 u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16);
2390 int session_key_len, con_secret_len;
2391 int payload_len;
2392 u64 global_id;
2393 int ret;
2394
2395 if (con->state != CEPH_CON_S_V2_AUTH) {
2396 con->error_msg = "protocol error, unexpected auth_done";
2397 return -EINVAL;
2398 }
2399
2400 ceph_decode_64_safe(&p, end, global_id, bad);
2401 ceph_decode_32_safe(&p, end, con->v2.con_mode, bad);
2402 ceph_decode_32_safe(&p, end, payload_len, bad);
2403
2404 dout("%s con %p global_id %llu con_mode %d payload_len %d\n",
2405 __func__, con, global_id, con->v2.con_mode, payload_len);
2406
2407 mutex_unlock(&con->mutex);
2408 session_key_len = 0;
2409 con_secret_len = 0;
2410 ret = con->ops->handle_auth_done(con, global_id, p, payload_len,
2411 session_key, &session_key_len,
2412 con_secret, &con_secret_len);
2413 mutex_lock(&con->mutex);
2414 if (con->state != CEPH_CON_S_V2_AUTH) {
2415 dout("%s con %p state changed to %d\n", __func__, con,
2416 con->state);
2417 ret = -EAGAIN;
2418 goto out;
2419 }
2420
2421 dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret);
2422 if (ret)
2423 goto out;
2424
2425 ret = setup_crypto(con, session_key, session_key_len, con_secret,
2426 con_secret_len);
2427 if (ret)
2428 goto out;
2429
2430 reset_out_kvecs(con);
2431 ret = prepare_auth_signature(con);
2432 if (ret) {
2433 pr_err("prepare_auth_signature failed: %d\n", ret);
2434 goto out;
2435 }
2436
2437 con->state = CEPH_CON_S_V2_AUTH_SIGNATURE;
2438
2439 out:
2440 memzero_explicit(session_key_buf, sizeof(session_key_buf));
2441 memzero_explicit(con_secret_buf, sizeof(con_secret_buf));
2442 return ret;
2443
2444 bad:
2445 pr_err("failed to decode auth_done\n");
2446 return -EINVAL;
2447 }
2448
2449 static int process_auth_signature(struct ceph_connection *con,
2450 void *p, void *end)
2451 {
2452 u8 hmac[SHA256_DIGEST_SIZE];
2453 int ret;
2454
2455 if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) {
2456 con->error_msg = "protocol error, unexpected auth_signature";
2457 return -EINVAL;
2458 }
2459
2460 ret = hmac_sha256(con, con->v2.out_sign_kvecs,
2461 con->v2.out_sign_kvec_cnt, hmac);
2462 if (ret)
2463 return ret;
2464
2465 ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad);
2466 if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) {
2467 con->error_msg = "integrity error, bad auth signature";
2468 return -EBADMSG;
2469 }
2470
2471 dout("%s con %p auth signature ok\n", __func__, con);
2472
2473 /* no reset_out_kvecs() as our auth_signature may still be pending */
2474 if (!con->v2.server_cookie) {
2475 ret = prepare_client_ident(con);
2476 if (ret) {
2477 pr_err("prepare_client_ident failed: %d\n", ret);
2478 return ret;
2479 }
2480
2481 con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2482 } else {
2483 ret = prepare_session_reconnect(con);
2484 if (ret) {
2485 pr_err("prepare_session_reconnect failed: %d\n", ret);
2486 return ret;
2487 }
2488
2489 con->state = CEPH_CON_S_V2_SESSION_RECONNECT;
2490 }
2491
2492 return 0;
2493
2494 bad:
2495 pr_err("failed to decode auth_signature\n");
2496 return -EINVAL;
2497 }
2498
2499 static int process_server_ident(struct ceph_connection *con,
2500 void *p, void *end)
2501 {
2502 struct ceph_client *client = from_msgr(con->msgr);
2503 u64 features, required_features;
2504 struct ceph_entity_addr addr;
2505 u64 global_seq;
2506 u64 global_id;
2507 u64 cookie;
2508 u64 flags;
2509 int ret;
2510
2511 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2512 con->error_msg = "protocol error, unexpected server_ident";
2513 return -EINVAL;
2514 }
2515
2516 ret = ceph_decode_entity_addrvec(&p, end, true, &addr);
2517 if (ret) {
2518 pr_err("failed to decode server addrs: %d\n", ret);
2519 return ret;
2520 }
2521
2522 ceph_decode_64_safe(&p, end, global_id, bad);
2523 ceph_decode_64_safe(&p, end, global_seq, bad);
2524 ceph_decode_64_safe(&p, end, features, bad);
2525 ceph_decode_64_safe(&p, end, required_features, bad);
2526 ceph_decode_64_safe(&p, end, flags, bad);
2527 ceph_decode_64_safe(&p, end, cookie, bad);
2528
2529 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",
2530 __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce),
2531 global_id, global_seq, features, required_features, flags, cookie);
2532
2533 /* is this who we intended to talk to? */
2534 if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
2535 pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
2536 ceph_pr_addr(&con->peer_addr),
2537 le32_to_cpu(con->peer_addr.nonce),
2538 ceph_pr_addr(&addr), le32_to_cpu(addr.nonce));
2539 con->error_msg = "wrong peer at address";
2540 return -EINVAL;
2541 }
2542
2543 if (client->required_features & ~features) {
2544 pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2545 features, client->required_features & ~features);
2546 con->error_msg = "missing required protocol features";
2547 return -EINVAL;
2548 }
2549
2550 /*
2551 * Both name->type and name->num are set in ceph_con_open() but
2552 * name->num may be bogus in the initial monmap. name->type is
2553 * verified in handle_hello().
2554 */
2555 WARN_ON(!con->peer_name.type);
2556 con->peer_name.num = cpu_to_le64(global_id);
2557 con->v2.peer_global_seq = global_seq;
2558 con->peer_features = features;
2559 WARN_ON(required_features & ~client->supported_features);
2560 con->v2.server_cookie = cookie;
2561
2562 if (flags & CEPH_MSG_CONNECT_LOSSY) {
2563 ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
2564 WARN_ON(con->v2.server_cookie);
2565 } else {
2566 WARN_ON(!con->v2.server_cookie);
2567 }
2568
2569 clear_in_sign_kvecs(con);
2570 clear_out_sign_kvecs(con);
2571 free_conn_bufs(con);
2572 con->delay = 0; /* reset backoff memory */
2573
2574 con->state = CEPH_CON_S_OPEN;
2575 con->v2.out_state = OUT_S_GET_NEXT;
2576 return 0;
2577
2578 bad:
2579 pr_err("failed to decode server_ident\n");
2580 return -EINVAL;
2581 }
2582
2583 static int process_ident_missing_features(struct ceph_connection *con,
2584 void *p, void *end)
2585 {
2586 struct ceph_client *client = from_msgr(con->msgr);
2587 u64 missing_features;
2588
2589 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2590 con->error_msg = "protocol error, unexpected ident_missing_features";
2591 return -EINVAL;
2592 }
2593
2594 ceph_decode_64_safe(&p, end, missing_features, bad);
2595 pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2596 client->supported_features, missing_features);
2597 con->error_msg = "missing required protocol features";
2598 return -EINVAL;
2599
2600 bad:
2601 pr_err("failed to decode ident_missing_features\n");
2602 return -EINVAL;
2603 }
2604
2605 static int process_session_reconnect_ok(struct ceph_connection *con,
2606 void *p, void *end)
2607 {
2608 u64 seq;
2609
2610 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2611 con->error_msg = "protocol error, unexpected session_reconnect_ok";
2612 return -EINVAL;
2613 }
2614
2615 ceph_decode_64_safe(&p, end, seq, bad);
2616
2617 dout("%s con %p seq %llu\n", __func__, con, seq);
2618 ceph_con_discard_requeued(con, seq);
2619
2620 clear_in_sign_kvecs(con);
2621 clear_out_sign_kvecs(con);
2622 free_conn_bufs(con);
2623 con->delay = 0; /* reset backoff memory */
2624
2625 con->state = CEPH_CON_S_OPEN;
2626 con->v2.out_state = OUT_S_GET_NEXT;
2627 return 0;
2628
2629 bad:
2630 pr_err("failed to decode session_reconnect_ok\n");
2631 return -EINVAL;
2632 }
2633
2634 static int process_session_retry(struct ceph_connection *con,
2635 void *p, void *end)
2636 {
2637 u64 connect_seq;
2638 int ret;
2639
2640 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2641 con->error_msg = "protocol error, unexpected session_retry";
2642 return -EINVAL;
2643 }
2644
2645 ceph_decode_64_safe(&p, end, connect_seq, bad);
2646
2647 dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq);
2648 WARN_ON(connect_seq <= con->v2.connect_seq);
2649 con->v2.connect_seq = connect_seq + 1;
2650
2651 free_conn_bufs(con);
2652
2653 reset_out_kvecs(con);
2654 ret = prepare_session_reconnect(con);
2655 if (ret) {
2656 pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret);
2657 return ret;
2658 }
2659
2660 return 0;
2661
2662 bad:
2663 pr_err("failed to decode session_retry\n");
2664 return -EINVAL;
2665 }
2666
2667 static int process_session_retry_global(struct ceph_connection *con,
2668 void *p, void *end)
2669 {
2670 u64 global_seq;
2671 int ret;
2672
2673 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2674 con->error_msg = "protocol error, unexpected session_retry_global";
2675 return -EINVAL;
2676 }
2677
2678 ceph_decode_64_safe(&p, end, global_seq, bad);
2679
2680 dout("%s con %p global_seq %llu\n", __func__, con, global_seq);
2681 WARN_ON(global_seq <= con->v2.global_seq);
2682 con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq);
2683
2684 free_conn_bufs(con);
2685
2686 reset_out_kvecs(con);
2687 ret = prepare_session_reconnect(con);
2688 if (ret) {
2689 pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret);
2690 return ret;
2691 }
2692
2693 return 0;
2694
2695 bad:
2696 pr_err("failed to decode session_retry_global\n");
2697 return -EINVAL;
2698 }
2699
2700 static int process_session_reset(struct ceph_connection *con,
2701 void *p, void *end)
2702 {
2703 bool full;
2704 int ret;
2705
2706 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2707 con->error_msg = "protocol error, unexpected session_reset";
2708 return -EINVAL;
2709 }
2710
2711 ceph_decode_8_safe(&p, end, full, bad);
2712 if (!full) {
2713 con->error_msg = "protocol error, bad session_reset";
2714 return -EINVAL;
2715 }
2716
2717 pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name),
2718 ceph_pr_addr(&con->peer_addr));
2719 ceph_con_reset_session(con);
2720
2721 mutex_unlock(&con->mutex);
2722 if (con->ops->peer_reset)
2723 con->ops->peer_reset(con);
2724 mutex_lock(&con->mutex);
2725 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2726 dout("%s con %p state changed to %d\n", __func__, con,
2727 con->state);
2728 return -EAGAIN;
2729 }
2730
2731 free_conn_bufs(con);
2732
2733 reset_out_kvecs(con);
2734 ret = prepare_client_ident(con);
2735 if (ret) {
2736 pr_err("prepare_client_ident (rst) failed: %d\n", ret);
2737 return ret;
2738 }
2739
2740 con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2741 return 0;
2742
2743 bad:
2744 pr_err("failed to decode session_reset\n");
2745 return -EINVAL;
2746 }
2747
2748 static int process_keepalive2_ack(struct ceph_connection *con,
2749 void *p, void *end)
2750 {
2751 if (con->state != CEPH_CON_S_OPEN) {
2752 con->error_msg = "protocol error, unexpected keepalive2_ack";
2753 return -EINVAL;
2754 }
2755
2756 ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad);
2757 ceph_decode_timespec64(&con->last_keepalive_ack, p);
2758
2759 dout("%s con %p timestamp %lld.%09ld\n", __func__, con,
2760 con->last_keepalive_ack.tv_sec, con->last_keepalive_ack.tv_nsec);
2761
2762 return 0;
2763
2764 bad:
2765 pr_err("failed to decode keepalive2_ack\n");
2766 return -EINVAL;
2767 }
2768
2769 static int process_ack(struct ceph_connection *con, void *p, void *end)
2770 {
2771 u64 seq;
2772
2773 if (con->state != CEPH_CON_S_OPEN) {
2774 con->error_msg = "protocol error, unexpected ack";
2775 return -EINVAL;
2776 }
2777
2778 ceph_decode_64_safe(&p, end, seq, bad);
2779
2780 dout("%s con %p seq %llu\n", __func__, con, seq);
2781 ceph_con_discard_sent(con, seq);
2782 return 0;
2783
2784 bad:
2785 pr_err("failed to decode ack\n");
2786 return -EINVAL;
2787 }
2788
2789 static int process_control(struct ceph_connection *con, void *p, void *end)
2790 {
2791 int tag = con->v2.in_desc.fd_tag;
2792 int ret;
2793
2794 dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p));
2795
2796 switch (tag) {
2797 case FRAME_TAG_HELLO:
2798 ret = process_hello(con, p, end);
2799 break;
2800 case FRAME_TAG_AUTH_BAD_METHOD:
2801 ret = process_auth_bad_method(con, p, end);
2802 break;
2803 case FRAME_TAG_AUTH_REPLY_MORE:
2804 ret = process_auth_reply_more(con, p, end);
2805 break;
2806 case FRAME_TAG_AUTH_DONE:
2807 ret = process_auth_done(con, p, end);
2808 break;
2809 case FRAME_TAG_AUTH_SIGNATURE:
2810 ret = process_auth_signature(con, p, end);
2811 break;
2812 case FRAME_TAG_SERVER_IDENT:
2813 ret = process_server_ident(con, p, end);
2814 break;
2815 case FRAME_TAG_IDENT_MISSING_FEATURES:
2816 ret = process_ident_missing_features(con, p, end);
2817 break;
2818 case FRAME_TAG_SESSION_RECONNECT_OK:
2819 ret = process_session_reconnect_ok(con, p, end);
2820 break;
2821 case FRAME_TAG_SESSION_RETRY:
2822 ret = process_session_retry(con, p, end);
2823 break;
2824 case FRAME_TAG_SESSION_RETRY_GLOBAL:
2825 ret = process_session_retry_global(con, p, end);
2826 break;
2827 case FRAME_TAG_SESSION_RESET:
2828 ret = process_session_reset(con, p, end);
2829 break;
2830 case FRAME_TAG_KEEPALIVE2_ACK:
2831 ret = process_keepalive2_ack(con, p, end);
2832 break;
2833 case FRAME_TAG_ACK:
2834 ret = process_ack(con, p, end);
2835 break;
2836 default:
2837 pr_err("bad tag %d\n", tag);
2838 con->error_msg = "protocol error, bad tag";
2839 return -EINVAL;
2840 }
2841 if (ret) {
2842 dout("%s con %p error %d\n", __func__, con, ret);
2843 return ret;
2844 }
2845
2846 prepare_read_preamble(con);
2847 return 0;
2848 }
2849
2850 /*
2851 * Return:
2852 * 1 - con->in_msg set, read message
2853 * 0 - skip message
2854 * <0 - error
2855 */
2856 static int process_message_header(struct ceph_connection *con,
2857 void *p, void *end)
2858 {
2859 struct ceph_frame_desc *desc = &con->v2.in_desc;
2860 struct ceph_msg_header2 *hdr2 = p;
2861 struct ceph_msg_header hdr;
2862 int skip;
2863 int ret;
2864 u64 seq;
2865
2866 /* verify seq# */
2867 seq = le64_to_cpu(hdr2->seq);
2868 if ((s64)seq - (s64)con->in_seq < 1) {
2869 pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n",
2870 ENTITY_NAME(con->peer_name),
2871 ceph_pr_addr(&con->peer_addr),
2872 seq, con->in_seq + 1);
2873 return 0;
2874 }
2875 if ((s64)seq - (s64)con->in_seq > 1) {
2876 pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1);
2877 con->error_msg = "bad message sequence # for incoming message";
2878 return -EBADE;
2879 }
2880
2881 ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq));
2882
2883 fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2],
2884 desc->fd_lens[3], &con->peer_name);
2885 ret = ceph_con_in_msg_alloc(con, &hdr, &skip);
2886 if (ret)
2887 return ret;
2888
2889 WARN_ON(!con->in_msg ^ skip);
2890 if (skip)
2891 return 0;
2892
2893 WARN_ON(!con->in_msg);
2894 WARN_ON(con->in_msg->con != con);
2895 return 1;
2896 }
2897
2898 static int process_message(struct ceph_connection *con)
2899 {
2900 ceph_con_process_message(con);
2901
2902 /*
2903 * We could have been closed by ceph_con_close() because
2904 * ceph_con_process_message() temporarily drops con->mutex.
2905 */
2906 if (con->state != CEPH_CON_S_OPEN) {
2907 dout("%s con %p state changed to %d\n", __func__, con,
2908 con->state);
2909 return -EAGAIN;
2910 }
2911
2912 prepare_read_preamble(con);
2913 return 0;
2914 }
2915
2916 static int __handle_control(struct ceph_connection *con, void *p)
2917 {
2918 void *end = p + con->v2.in_desc.fd_lens[0];
2919 struct ceph_msg *msg;
2920 int ret;
2921
2922 if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE)
2923 return process_control(con, p, end);
2924
2925 ret = process_message_header(con, p, end);
2926 if (ret < 0)
2927 return ret;
2928 if (ret == 0) {
2929 prepare_skip_message(con);
2930 return 0;
2931 }
2932
2933 msg = con->in_msg; /* set in process_message_header() */
2934 if (front_len(msg)) {
2935 WARN_ON(front_len(msg) > msg->front_alloc_len);
2936 msg->front.iov_len = front_len(msg);
2937 } else {
2938 msg->front.iov_len = 0;
2939 }
2940 if (middle_len(msg)) {
2941 WARN_ON(middle_len(msg) > msg->middle->alloc_len);
2942 msg->middle->vec.iov_len = middle_len(msg);
2943 } else if (msg->middle) {
2944 msg->middle->vec.iov_len = 0;
2945 }
2946
2947 if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
2948 return process_message(con);
2949
2950 if (con_secure(con))
2951 return prepare_read_tail_secure(con);
2952
2953 return prepare_read_tail_plain(con);
2954 }
2955
2956 static int handle_preamble(struct ceph_connection *con)
2957 {
2958 struct ceph_frame_desc *desc = &con->v2.in_desc;
2959 int ret;
2960
2961 if (con_secure(con)) {
2962 ret = decrypt_preamble(con);
2963 if (ret) {
2964 if (ret == -EBADMSG)
2965 con->error_msg = "integrity error, bad preamble auth tag";
2966 return ret;
2967 }
2968 }
2969
2970 ret = decode_preamble(con->v2.in_buf, desc);
2971 if (ret) {
2972 if (ret == -EBADMSG)
2973 con->error_msg = "integrity error, bad crc";
2974 else
2975 con->error_msg = "protocol error, bad preamble";
2976 return ret;
2977 }
2978
2979 dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__,
2980 con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0],
2981 desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]);
2982
2983 if (!con_secure(con))
2984 return prepare_read_control(con);
2985
2986 if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN)
2987 return prepare_read_control_remainder(con);
2988
2989 return __handle_control(con, CTRL_BODY(con->v2.in_buf));
2990 }
2991
2992 static int handle_control(struct ceph_connection *con)
2993 {
2994 int ctrl_len = con->v2.in_desc.fd_lens[0];
2995 void *buf;
2996 int ret;
2997
2998 WARN_ON(con_secure(con));
2999
3000 ret = verify_control_crc(con);
3001 if (ret) {
3002 con->error_msg = "integrity error, bad crc";
3003 return ret;
3004 }
3005
3006 if (con->state == CEPH_CON_S_V2_AUTH) {
3007 buf = alloc_conn_buf(con, ctrl_len);
3008 if (!buf)
3009 return -ENOMEM;
3010
3011 memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len);
3012 return __handle_control(con, buf);
3013 }
3014
3015 return __handle_control(con, con->v2.in_kvecs[0].iov_base);
3016 }
3017
3018 static int handle_control_remainder(struct ceph_connection *con)
3019 {
3020 int ret;
3021
3022 WARN_ON(!con_secure(con));
3023
3024 ret = decrypt_control_remainder(con);
3025 if (ret) {
3026 if (ret == -EBADMSG)
3027 con->error_msg = "integrity error, bad control remainder auth tag";
3028 return ret;
3029 }
3030
3031 return __handle_control(con, con->v2.in_kvecs[0].iov_base -
3032 CEPH_PREAMBLE_INLINE_LEN);
3033 }
3034
3035 static int handle_epilogue(struct ceph_connection *con)
3036 {
3037 u32 front_crc, middle_crc, data_crc;
3038 int ret;
3039
3040 if (con_secure(con)) {
3041 ret = decrypt_tail(con);
3042 if (ret) {
3043 if (ret == -EBADMSG)
3044 con->error_msg = "integrity error, bad epilogue auth tag";
3045 return ret;
3046 }
3047
3048 /* just late_status */
3049 ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL);
3050 if (ret) {
3051 con->error_msg = "protocol error, bad epilogue";
3052 return ret;
3053 }
3054 } else {
3055 ret = decode_epilogue(con->v2.in_buf, &front_crc,
3056 &middle_crc, &data_crc);
3057 if (ret) {
3058 con->error_msg = "protocol error, bad epilogue";
3059 return ret;
3060 }
3061
3062 ret = verify_epilogue_crcs(con, front_crc, middle_crc,
3063 data_crc);
3064 if (ret) {
3065 con->error_msg = "integrity error, bad crc";
3066 return ret;
3067 }
3068 }
3069
3070 return process_message(con);
3071 }
3072
3073 static void finish_skip(struct ceph_connection *con)
3074 {
3075 dout("%s con %p\n", __func__, con);
3076
3077 if (con_secure(con))
3078 gcm_inc_nonce(&con->v2.in_gcm_nonce);
3079
3080 __finish_skip(con);
3081 }
3082
3083 static int populate_in_iter(struct ceph_connection *con)
3084 {
3085 int ret;
3086
3087 dout("%s con %p state %d in_state %d\n", __func__, con, con->state,
3088 con->v2.in_state);
3089 WARN_ON(iov_iter_count(&con->v2.in_iter));
3090
3091 if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) {
3092 ret = process_banner_prefix(con);
3093 } else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) {
3094 ret = process_banner_payload(con);
3095 } else if ((con->state >= CEPH_CON_S_V2_HELLO &&
3096 con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) ||
3097 con->state == CEPH_CON_S_OPEN) {
3098 switch (con->v2.in_state) {
3099 case IN_S_HANDLE_PREAMBLE:
3100 ret = handle_preamble(con);
3101 break;
3102 case IN_S_HANDLE_CONTROL:
3103 ret = handle_control(con);
3104 break;
3105 case IN_S_HANDLE_CONTROL_REMAINDER:
3106 ret = handle_control_remainder(con);
3107 break;
3108 case IN_S_PREPARE_READ_DATA:
3109 ret = prepare_read_data(con);
3110 break;
3111 case IN_S_PREPARE_READ_DATA_CONT:
3112 prepare_read_data_cont(con);
3113 ret = 0;
3114 break;
3115 case IN_S_PREPARE_READ_ENC_PAGE:
3116 prepare_read_enc_page(con);
3117 ret = 0;
3118 break;
3119 case IN_S_PREPARE_SPARSE_DATA:
3120 ret = prepare_sparse_read_data(con);
3121 break;
3122 case IN_S_PREPARE_SPARSE_DATA_CONT:
3123 ret = prepare_sparse_read_cont(con);
3124 break;
3125 case IN_S_HANDLE_EPILOGUE:
3126 ret = handle_epilogue(con);
3127 break;
3128 case IN_S_FINISH_SKIP:
3129 finish_skip(con);
3130 ret = 0;
3131 break;
3132 default:
3133 WARN(1, "bad in_state %d", con->v2.in_state);
3134 return -EINVAL;
3135 }
3136 } else {
3137 WARN(1, "bad state %d", con->state);
3138 return -EINVAL;
3139 }
3140 if (ret) {
3141 dout("%s con %p error %d\n", __func__, con, ret);
3142 return ret;
3143 }
3144
3145 if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3146 return -ENODATA;
3147 dout("%s con %p populated %zu\n", __func__, con,
3148 iov_iter_count(&con->v2.in_iter));
3149 return 1;
3150 }
3151
3152 int ceph_con_v2_try_read(struct ceph_connection *con)
3153 {
3154 int ret;
3155
3156 dout("%s con %p state %d need %zu\n", __func__, con, con->state,
3157 iov_iter_count(&con->v2.in_iter));
3158
3159 if (con->state == CEPH_CON_S_PREOPEN)
3160 return 0;
3161
3162 /*
3163 * We should always have something pending here. If not,
3164 * avoid calling populate_in_iter() as if we read something
3165 * (ceph_tcp_recv() would immediately return 1).
3166 */
3167 if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3168 return -ENODATA;
3169
3170 for (;;) {
3171 ret = ceph_tcp_recv(con);
3172 if (ret <= 0)
3173 return ret;
3174
3175 ret = populate_in_iter(con);
3176 if (ret <= 0) {
3177 if (ret && ret != -EAGAIN && !con->error_msg)
3178 con->error_msg = "read processing error";
3179 return ret;
3180 }
3181 }
3182 }
3183
3184 static void queue_data(struct ceph_connection *con)
3185 {
3186 struct bio_vec bv;
3187
3188 con->v2.out_epil.data_crc = -1;
3189 ceph_msg_data_cursor_init(&con->v2.out_cursor, con->out_msg,
3190 data_len(con->out_msg));
3191
3192 get_bvec_at(&con->v2.out_cursor, &bv);
3193 set_out_bvec(con, &bv, true);
3194 con->v2.out_state = OUT_S_QUEUE_DATA_CONT;
3195 }
3196
3197 static void queue_data_cont(struct ceph_connection *con)
3198 {
3199 struct bio_vec bv;
3200
3201 con->v2.out_epil.data_crc = ceph_crc32c_page(
3202 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3203 con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len);
3204
3205 ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len);
3206 if (con->v2.out_cursor.total_resid) {
3207 get_bvec_at(&con->v2.out_cursor, &bv);
3208 set_out_bvec(con, &bv, true);
3209 WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT);
3210 return;
3211 }
3212
3213 /*
3214 * We've written all data. Queue epilogue. Once it's written,
3215 * we are done.
3216 */
3217 reset_out_kvecs(con);
3218 prepare_epilogue_plain(con, false);
3219 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3220 }
3221
3222 static void queue_enc_page(struct ceph_connection *con)
3223 {
3224 struct bio_vec bv;
3225
3226 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i,
3227 con->v2.out_enc_resid);
3228 WARN_ON(!con->v2.out_enc_resid);
3229
3230 bvec_set_page(&bv, con->v2.out_enc_pages[con->v2.out_enc_i],
3231 min(con->v2.out_enc_resid, (int)PAGE_SIZE), 0);
3232
3233 set_out_bvec(con, &bv, false);
3234 con->v2.out_enc_i++;
3235 con->v2.out_enc_resid -= bv.bv_len;
3236
3237 if (con->v2.out_enc_resid) {
3238 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE);
3239 return;
3240 }
3241
3242 /*
3243 * We've queued the last piece of ciphertext (ending with
3244 * epilogue) + auth tag. Once it's written, we are done.
3245 */
3246 WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt);
3247 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3248 }
3249
3250 static void queue_zeros(struct ceph_connection *con)
3251 {
3252 dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero);
3253
3254 if (con->v2.out_zero) {
3255 set_out_bvec_zero(con);
3256 con->v2.out_zero -= con->v2.out_bvec.bv_len;
3257 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3258 return;
3259 }
3260
3261 /*
3262 * We've zero-filled everything up to epilogue. Queue epilogue
3263 * with late_status set to ABORTED and crcs adjusted for zeros.
3264 * Once it's written, we are done patching up for the revoke.
3265 */
3266 reset_out_kvecs(con);
3267 prepare_epilogue_plain(con, true);
3268 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3269 }
3270
3271 static void finish_message(struct ceph_connection *con)
3272 {
3273 dout("%s con %p msg %p\n", __func__, con, con->out_msg);
3274
3275 /* we end up here both plain and secure modes */
3276 if (con->v2.out_enc_pages) {
3277 WARN_ON(!con->v2.out_enc_page_cnt);
3278 ceph_release_page_vector(con->v2.out_enc_pages,
3279 con->v2.out_enc_page_cnt);
3280 con->v2.out_enc_pages = NULL;
3281 con->v2.out_enc_page_cnt = 0;
3282 }
3283 /* message may have been revoked */
3284 if (con->out_msg) {
3285 ceph_msg_put(con->out_msg);
3286 con->out_msg = NULL;
3287 }
3288
3289 con->v2.out_state = OUT_S_GET_NEXT;
3290 }
3291
3292 static int populate_out_iter(struct ceph_connection *con)
3293 {
3294 int ret;
3295
3296 dout("%s con %p state %d out_state %d\n", __func__, con, con->state,
3297 con->v2.out_state);
3298 WARN_ON(iov_iter_count(&con->v2.out_iter));
3299
3300 if (con->state != CEPH_CON_S_OPEN) {
3301 WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX ||
3302 con->state > CEPH_CON_S_V2_SESSION_RECONNECT);
3303 goto nothing_pending;
3304 }
3305
3306 switch (con->v2.out_state) {
3307 case OUT_S_QUEUE_DATA:
3308 WARN_ON(!con->out_msg);
3309 queue_data(con);
3310 goto populated;
3311 case OUT_S_QUEUE_DATA_CONT:
3312 WARN_ON(!con->out_msg);
3313 queue_data_cont(con);
3314 goto populated;
3315 case OUT_S_QUEUE_ENC_PAGE:
3316 queue_enc_page(con);
3317 goto populated;
3318 case OUT_S_QUEUE_ZEROS:
3319 WARN_ON(con->out_msg); /* revoked */
3320 queue_zeros(con);
3321 goto populated;
3322 case OUT_S_FINISH_MESSAGE:
3323 finish_message(con);
3324 break;
3325 case OUT_S_GET_NEXT:
3326 break;
3327 default:
3328 WARN(1, "bad out_state %d", con->v2.out_state);
3329 return -EINVAL;
3330 }
3331
3332 WARN_ON(con->v2.out_state != OUT_S_GET_NEXT);
3333 if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
3334 ret = prepare_keepalive2(con);
3335 if (ret) {
3336 pr_err("prepare_keepalive2 failed: %d\n", ret);
3337 return ret;
3338 }
3339 } else if (!list_empty(&con->out_queue)) {
3340 ceph_con_get_out_msg(con);
3341 ret = prepare_message(con);
3342 if (ret) {
3343 pr_err("prepare_message failed: %d\n", ret);
3344 return ret;
3345 }
3346 } else if (con->in_seq > con->in_seq_acked) {
3347 ret = prepare_ack(con);
3348 if (ret) {
3349 pr_err("prepare_ack failed: %d\n", ret);
3350 return ret;
3351 }
3352 } else {
3353 goto nothing_pending;
3354 }
3355
3356 populated:
3357 if (WARN_ON(!iov_iter_count(&con->v2.out_iter)))
3358 return -ENODATA;
3359 dout("%s con %p populated %zu\n", __func__, con,
3360 iov_iter_count(&con->v2.out_iter));
3361 return 1;
3362
3363 nothing_pending:
3364 WARN_ON(iov_iter_count(&con->v2.out_iter));
3365 dout("%s con %p nothing pending\n", __func__, con);
3366 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
3367 return 0;
3368 }
3369
3370 int ceph_con_v2_try_write(struct ceph_connection *con)
3371 {
3372 int ret;
3373
3374 dout("%s con %p state %d have %zu\n", __func__, con, con->state,
3375 iov_iter_count(&con->v2.out_iter));
3376
3377 /* open the socket first? */
3378 if (con->state == CEPH_CON_S_PREOPEN) {
3379 WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2);
3380
3381 /*
3382 * Always bump global_seq. Bump connect_seq only if
3383 * there is a session (i.e. we are reconnecting and will
3384 * send session_reconnect instead of client_ident).
3385 */
3386 con->v2.global_seq = ceph_get_global_seq(con->msgr, 0);
3387 if (con->v2.server_cookie)
3388 con->v2.connect_seq++;
3389
3390 ret = prepare_read_banner_prefix(con);
3391 if (ret) {
3392 pr_err("prepare_read_banner_prefix failed: %d\n", ret);
3393 con->error_msg = "connect error";
3394 return ret;
3395 }
3396
3397 reset_out_kvecs(con);
3398 ret = prepare_banner(con);
3399 if (ret) {
3400 pr_err("prepare_banner failed: %d\n", ret);
3401 con->error_msg = "connect error";
3402 return ret;
3403 }
3404
3405 ret = ceph_tcp_connect(con);
3406 if (ret) {
3407 pr_err("ceph_tcp_connect failed: %d\n", ret);
3408 con->error_msg = "connect error";
3409 return ret;
3410 }
3411 }
3412
3413 if (!iov_iter_count(&con->v2.out_iter)) {
3414 ret = populate_out_iter(con);
3415 if (ret <= 0) {
3416 if (ret && ret != -EAGAIN && !con->error_msg)
3417 con->error_msg = "write processing error";
3418 return ret;
3419 }
3420 }
3421
3422 tcp_sock_set_cork(con->sock->sk, true);
3423 for (;;) {
3424 ret = ceph_tcp_send(con);
3425 if (ret <= 0)
3426 break;
3427
3428 ret = populate_out_iter(con);
3429 if (ret <= 0) {
3430 if (ret && ret != -EAGAIN && !con->error_msg)
3431 con->error_msg = "write processing error";
3432 break;
3433 }
3434 }
3435
3436 tcp_sock_set_cork(con->sock->sk, false);
3437 return ret;
3438 }
3439
3440 static u32 crc32c_zeros(u32 crc, int zero_len)
3441 {
3442 int len;
3443
3444 while (zero_len) {
3445 len = min(zero_len, (int)PAGE_SIZE);
3446 crc = crc32c(crc, page_address(ceph_zero_page), len);
3447 zero_len -= len;
3448 }
3449
3450 return crc;
3451 }
3452
3453 static void prepare_zero_front(struct ceph_connection *con, int resid)
3454 {
3455 int sent;
3456
3457 WARN_ON(!resid || resid > front_len(con->out_msg));
3458 sent = front_len(con->out_msg) - resid;
3459 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3460
3461 if (sent) {
3462 con->v2.out_epil.front_crc =
3463 crc32c(-1, con->out_msg->front.iov_base, sent);
3464 con->v2.out_epil.front_crc =
3465 crc32c_zeros(con->v2.out_epil.front_crc, resid);
3466 } else {
3467 con->v2.out_epil.front_crc = crc32c_zeros(-1, resid);
3468 }
3469
3470 con->v2.out_iter.count -= resid;
3471 out_zero_add(con, resid);
3472 }
3473
3474 static void prepare_zero_middle(struct ceph_connection *con, int resid)
3475 {
3476 int sent;
3477
3478 WARN_ON(!resid || resid > middle_len(con->out_msg));
3479 sent = middle_len(con->out_msg) - resid;
3480 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3481
3482 if (sent) {
3483 con->v2.out_epil.middle_crc =
3484 crc32c(-1, con->out_msg->middle->vec.iov_base, sent);
3485 con->v2.out_epil.middle_crc =
3486 crc32c_zeros(con->v2.out_epil.middle_crc, resid);
3487 } else {
3488 con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid);
3489 }
3490
3491 con->v2.out_iter.count -= resid;
3492 out_zero_add(con, resid);
3493 }
3494
3495 static void prepare_zero_data(struct ceph_connection *con)
3496 {
3497 dout("%s con %p\n", __func__, con);
3498 con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(con->out_msg));
3499 out_zero_add(con, data_len(con->out_msg));
3500 }
3501
3502 static void revoke_at_queue_data(struct ceph_connection *con)
3503 {
3504 int boundary;
3505 int resid;
3506
3507 WARN_ON(!data_len(con->out_msg));
3508 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3509 resid = iov_iter_count(&con->v2.out_iter);
3510
3511 boundary = front_len(con->out_msg) + middle_len(con->out_msg);
3512 if (resid > boundary) {
3513 resid -= boundary;
3514 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3515 dout("%s con %p was sending head\n", __func__, con);
3516 if (front_len(con->out_msg))
3517 prepare_zero_front(con, front_len(con->out_msg));
3518 if (middle_len(con->out_msg))
3519 prepare_zero_middle(con, middle_len(con->out_msg));
3520 prepare_zero_data(con);
3521 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3522 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3523 return;
3524 }
3525
3526 boundary = middle_len(con->out_msg);
3527 if (resid > boundary) {
3528 resid -= boundary;
3529 dout("%s con %p was sending front\n", __func__, con);
3530 prepare_zero_front(con, resid);
3531 if (middle_len(con->out_msg))
3532 prepare_zero_middle(con, middle_len(con->out_msg));
3533 prepare_zero_data(con);
3534 queue_zeros(con);
3535 return;
3536 }
3537
3538 WARN_ON(!resid);
3539 dout("%s con %p was sending middle\n", __func__, con);
3540 prepare_zero_middle(con, resid);
3541 prepare_zero_data(con);
3542 queue_zeros(con);
3543 }
3544
3545 static void revoke_at_queue_data_cont(struct ceph_connection *con)
3546 {
3547 int sent, resid; /* current piece of data */
3548
3549 WARN_ON(!data_len(con->out_msg));
3550 WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter));
3551 resid = iov_iter_count(&con->v2.out_iter);
3552 WARN_ON(!resid || resid > con->v2.out_bvec.bv_len);
3553 sent = con->v2.out_bvec.bv_len - resid;
3554 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3555
3556 if (sent) {
3557 con->v2.out_epil.data_crc = ceph_crc32c_page(
3558 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3559 con->v2.out_bvec.bv_offset, sent);
3560 ceph_msg_data_advance(&con->v2.out_cursor, sent);
3561 }
3562 WARN_ON(resid > con->v2.out_cursor.total_resid);
3563 con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc,
3564 con->v2.out_cursor.total_resid);
3565
3566 con->v2.out_iter.count -= resid;
3567 out_zero_add(con, con->v2.out_cursor.total_resid);
3568 queue_zeros(con);
3569 }
3570
3571 static void revoke_at_finish_message(struct ceph_connection *con)
3572 {
3573 int boundary;
3574 int resid;
3575
3576 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3577 resid = iov_iter_count(&con->v2.out_iter);
3578
3579 if (!front_len(con->out_msg) && !middle_len(con->out_msg) &&
3580 !data_len(con->out_msg)) {
3581 WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN);
3582 dout("%s con %p was sending head (empty message) - noop\n",
3583 __func__, con);
3584 return;
3585 }
3586
3587 boundary = front_len(con->out_msg) + middle_len(con->out_msg) +
3588 CEPH_EPILOGUE_PLAIN_LEN;
3589 if (resid > boundary) {
3590 resid -= boundary;
3591 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3592 dout("%s con %p was sending head\n", __func__, con);
3593 if (front_len(con->out_msg))
3594 prepare_zero_front(con, front_len(con->out_msg));
3595 if (middle_len(con->out_msg))
3596 prepare_zero_middle(con, middle_len(con->out_msg));
3597 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3598 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3599 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3600 return;
3601 }
3602
3603 boundary = middle_len(con->out_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3604 if (resid > boundary) {
3605 resid -= boundary;
3606 dout("%s con %p was sending front\n", __func__, con);
3607 prepare_zero_front(con, resid);
3608 if (middle_len(con->out_msg))
3609 prepare_zero_middle(con, middle_len(con->out_msg));
3610 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3611 queue_zeros(con);
3612 return;
3613 }
3614
3615 boundary = CEPH_EPILOGUE_PLAIN_LEN;
3616 if (resid > boundary) {
3617 resid -= boundary;
3618 dout("%s con %p was sending middle\n", __func__, con);
3619 prepare_zero_middle(con, resid);
3620 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3621 queue_zeros(con);
3622 return;
3623 }
3624
3625 WARN_ON(!resid);
3626 dout("%s con %p was sending epilogue - noop\n", __func__, con);
3627 }
3628
3629 void ceph_con_v2_revoke(struct ceph_connection *con)
3630 {
3631 WARN_ON(con->v2.out_zero);
3632
3633 if (con_secure(con)) {
3634 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE &&
3635 con->v2.out_state != OUT_S_FINISH_MESSAGE);
3636 dout("%s con %p secure - noop\n", __func__, con);
3637 return;
3638 }
3639
3640 switch (con->v2.out_state) {
3641 case OUT_S_QUEUE_DATA:
3642 revoke_at_queue_data(con);
3643 break;
3644 case OUT_S_QUEUE_DATA_CONT:
3645 revoke_at_queue_data_cont(con);
3646 break;
3647 case OUT_S_FINISH_MESSAGE:
3648 revoke_at_finish_message(con);
3649 break;
3650 default:
3651 WARN(1, "bad out_state %d", con->v2.out_state);
3652 break;
3653 }
3654 }
3655
3656 static void revoke_at_prepare_read_data(struct ceph_connection *con)
3657 {
3658 int remaining;
3659 int resid;
3660
3661 WARN_ON(con_secure(con));
3662 WARN_ON(!data_len(con->in_msg));
3663 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
3664 resid = iov_iter_count(&con->v2.in_iter);
3665 WARN_ON(!resid);
3666
3667 remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3668 dout("%s con %p resid %d remaining %d\n", __func__, con, resid,
3669 remaining);
3670 con->v2.in_iter.count -= resid;
3671 set_in_skip(con, resid + remaining);
3672 con->v2.in_state = IN_S_FINISH_SKIP;
3673 }
3674
3675 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con)
3676 {
3677 int recved, resid; /* current piece of data */
3678 int remaining;
3679
3680 WARN_ON(con_secure(con));
3681 WARN_ON(!data_len(con->in_msg));
3682 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3683 resid = iov_iter_count(&con->v2.in_iter);
3684 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3685 recved = con->v2.in_bvec.bv_len - resid;
3686 dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid);
3687
3688 if (recved)
3689 ceph_msg_data_advance(&con->v2.in_cursor, recved);
3690 WARN_ON(resid > con->v2.in_cursor.total_resid);
3691
3692 remaining = CEPH_EPILOGUE_PLAIN_LEN;
3693 dout("%s con %p total_resid %zu remaining %d\n", __func__, con,
3694 con->v2.in_cursor.total_resid, remaining);
3695 con->v2.in_iter.count -= resid;
3696 set_in_skip(con, con->v2.in_cursor.total_resid + remaining);
3697 con->v2.in_state = IN_S_FINISH_SKIP;
3698 }
3699
3700 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con)
3701 {
3702 int resid; /* current enc page (not necessarily data) */
3703
3704 WARN_ON(!con_secure(con));
3705 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3706 resid = iov_iter_count(&con->v2.in_iter);
3707 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3708
3709 dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid,
3710 con->v2.in_enc_resid);
3711 con->v2.in_iter.count -= resid;
3712 set_in_skip(con, resid + con->v2.in_enc_resid);
3713 con->v2.in_state = IN_S_FINISH_SKIP;
3714 }
3715
3716 static void revoke_at_prepare_sparse_data(struct ceph_connection *con)
3717 {
3718 int resid; /* current piece of data */
3719 int remaining;
3720
3721 WARN_ON(con_secure(con));
3722 WARN_ON(!data_len(con->in_msg));
3723 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3724 resid = iov_iter_count(&con->v2.in_iter);
3725 dout("%s con %p resid %d\n", __func__, con, resid);
3726
3727 remaining = CEPH_EPILOGUE_PLAIN_LEN + con->v2.data_len_remain;
3728 con->v2.in_iter.count -= resid;
3729 set_in_skip(con, resid + remaining);
3730 con->v2.in_state = IN_S_FINISH_SKIP;
3731 }
3732
3733 static void revoke_at_handle_epilogue(struct ceph_connection *con)
3734 {
3735 int resid;
3736
3737 resid = iov_iter_count(&con->v2.in_iter);
3738 WARN_ON(!resid);
3739
3740 dout("%s con %p resid %d\n", __func__, con, resid);
3741 con->v2.in_iter.count -= resid;
3742 set_in_skip(con, resid);
3743 con->v2.in_state = IN_S_FINISH_SKIP;
3744 }
3745
3746 void ceph_con_v2_revoke_incoming(struct ceph_connection *con)
3747 {
3748 switch (con->v2.in_state) {
3749 case IN_S_PREPARE_SPARSE_DATA:
3750 case IN_S_PREPARE_READ_DATA:
3751 revoke_at_prepare_read_data(con);
3752 break;
3753 case IN_S_PREPARE_READ_DATA_CONT:
3754 revoke_at_prepare_read_data_cont(con);
3755 break;
3756 case IN_S_PREPARE_READ_ENC_PAGE:
3757 revoke_at_prepare_read_enc_page(con);
3758 break;
3759 case IN_S_PREPARE_SPARSE_DATA_CONT:
3760 revoke_at_prepare_sparse_data(con);
3761 break;
3762 case IN_S_HANDLE_EPILOGUE:
3763 revoke_at_handle_epilogue(con);
3764 break;
3765 default:
3766 WARN(1, "bad in_state %d", con->v2.in_state);
3767 break;
3768 }
3769 }
3770
3771 bool ceph_con_v2_opened(struct ceph_connection *con)
3772 {
3773 return con->v2.peer_global_seq;
3774 }
3775
3776 void ceph_con_v2_reset_session(struct ceph_connection *con)
3777 {
3778 con->v2.client_cookie = 0;
3779 con->v2.server_cookie = 0;
3780 con->v2.global_seq = 0;
3781 con->v2.connect_seq = 0;
3782 con->v2.peer_global_seq = 0;
3783 }
3784
3785 void ceph_con_v2_reset_protocol(struct ceph_connection *con)
3786 {
3787 iov_iter_truncate(&con->v2.in_iter, 0);
3788 iov_iter_truncate(&con->v2.out_iter, 0);
3789 con->v2.out_zero = 0;
3790
3791 clear_in_sign_kvecs(con);
3792 clear_out_sign_kvecs(con);
3793 free_conn_bufs(con);
3794
3795 if (con->v2.in_enc_pages) {
3796 WARN_ON(!con->v2.in_enc_page_cnt);
3797 ceph_release_page_vector(con->v2.in_enc_pages,
3798 con->v2.in_enc_page_cnt);
3799 con->v2.in_enc_pages = NULL;
3800 con->v2.in_enc_page_cnt = 0;
3801 }
3802 if (con->v2.out_enc_pages) {
3803 WARN_ON(!con->v2.out_enc_page_cnt);
3804 ceph_release_page_vector(con->v2.out_enc_pages,
3805 con->v2.out_enc_page_cnt);
3806 con->v2.out_enc_pages = NULL;
3807 con->v2.out_enc_page_cnt = 0;
3808 }
3809
3810 con->v2.con_mode = CEPH_CON_MODE_UNKNOWN;
3811 memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN);
3812 memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN);
3813
3814 if (con->v2.hmac_tfm) {
3815 crypto_free_shash(con->v2.hmac_tfm);
3816 con->v2.hmac_tfm = NULL;
3817 }
3818 if (con->v2.gcm_req) {
3819 aead_request_free(con->v2.gcm_req);
3820 con->v2.gcm_req = NULL;
3821 }
3822 if (con->v2.gcm_tfm) {
3823 crypto_free_aead(con->v2.gcm_tfm);
3824 con->v2.gcm_tfm = NULL;
3825 }
3826 }