1/*
   2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
   3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
   4 * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
   5 * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
   6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
   7 * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
   8 *
   9 * This software is available to you under a choice of one of two
  10 * licenses.  You may choose to be licensed under the terms of the GNU
  11 * General Public License (GPL) Version 2, available from the file
  12 * COPYING in the main directory of this source tree, or the
  13 * OpenIB.org BSD license below:
  14 *
  15 *     Redistribution and use in source and binary forms, with or
  16 *     without modification, are permitted provided that the following
  17 *     conditions are met:
  18 *
  19 *      - Redistributions of source code must retain the above
  20 *        copyright notice, this list of conditions and the following
  21 *        disclaimer.
  22 *
  23 *      - Redistributions in binary form must reproduce the above
  24 *        copyright notice, this list of conditions and the following
  25 *        disclaimer in the documentation and/or other materials
  26 *        provided with the distribution.
  27 *
  28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  35 * SOFTWARE.
  36 */
  37
  38#include <linux/sched/signal.h>
  39#include <linux/module.h>
  40#include <crypto/aead.h>
  41
  42#include <net/strparser.h>
  43#include <net/tls.h>
  44
  45static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  46                     unsigned int recursion_level)
  47{
  48        int start = skb_headlen(skb);
  49        int i, chunk = start - offset;
  50        struct sk_buff *frag_iter;
  51        int elt = 0;
  52
  53        if (unlikely(recursion_level >= 24))
  54                return -EMSGSIZE;
  55
  56        if (chunk > 0) {
  57                if (chunk > len)
  58                        chunk = len;
  59                elt++;
  60                len -= chunk;
  61                if (len == 0)
  62                        return elt;
  63                offset += chunk;
  64        }
  65
  66        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
  67                int end;
  68
  69                WARN_ON(start > offset + len);
  70
  71                end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
  72                chunk = end - offset;
  73                if (chunk > 0) {
  74                        if (chunk > len)
  75                                chunk = len;
  76                        elt++;
  77                        len -= chunk;
  78                        if (len == 0)
  79                                return elt;
  80                        offset += chunk;
  81                }
  82                start = end;
  83        }
  84
  85        if (unlikely(skb_has_frag_list(skb))) {
  86                skb_walk_frags(skb, frag_iter) {
  87                        int end, ret;
  88
  89                        WARN_ON(start > offset + len);
  90
  91                        end = start + frag_iter->len;
  92                        chunk = end - offset;
  93                        if (chunk > 0) {
  94                                if (chunk > len)
  95                                        chunk = len;
  96                                ret = __skb_nsg(frag_iter, offset - start, chunk,
  97                                                recursion_level + 1);
  98                                if (unlikely(ret < 0))
  99                                        return ret;
 100                                elt += ret;
 101                                len -= chunk;
 102                                if (len == 0)
 103                                        return elt;
 104                                offset += chunk;
 105                        }
 106                        start = end;
 107                }
 108        }
 109        BUG_ON(len);
 110        return elt;
 111}
 112
 113/* Return the number of scatterlist elements required to completely map the
 114 * skb, or -EMSGSIZE if the recursion depth is exceeded.
 115 */
 116static int skb_nsg(struct sk_buff *skb, int offset, int len)
 117{
 118        return __skb_nsg(skb, offset, len, 0);
 119}
 120
 121static int padding_length(struct tls_sw_context_rx *ctx,
 122			  struct tls_prot_info *prot, struct sk_buff *skb)
 123{
 124	struct strp_msg *rxm = strp_msg(skb);
 125	int sub = 0;
 126
 127	/* Determine zero-padding length */
 128	if (prot->version == TLS_1_3_VERSION) {
 129		char content_type = 0;
 130		int err;
 131		int back = 17;
 132
 133		while (content_type == 0) {
 134			if (back > rxm->full_len - prot->prepend_size)
 135				return -EBADMSG;
 136			err = skb_copy_bits(skb,
 137					    rxm->offset + rxm->full_len - back,
 138					    &content_type, 1);
 139			if (err)
 140				return err;
 141			if (content_type)
 142				break;
 143			sub++;
 144			back++;
 145		}
 146		ctx->control = content_type;
 147	}
 148	return sub;
 149}
 150
 151static void tls_decrypt_done(struct crypto_async_request *req, int err)
 152{
 153	struct aead_request *aead_req = (struct aead_request *)req;
 154	struct scatterlist *sgout = aead_req->dst;
 155	struct scatterlist *sgin = aead_req->src;
 156	struct tls_sw_context_rx *ctx;
 157	struct tls_context *tls_ctx;
 158	struct tls_prot_info *prot;
 159	struct scatterlist *sg;
 160	struct sk_buff *skb;
 161	unsigned int pages;
 162	int pending;
 163
 164	skb = (struct sk_buff *)req->data;
 165	tls_ctx = tls_get_ctx(skb->sk);
 166	ctx = tls_sw_ctx_rx(tls_ctx);
 167	prot = &tls_ctx->prot_info;
 168
 169	/* Propagate if there was an err */
 170	if (err) {
 171		if (err == -EBADMSG)
 172			TLS_INC_STATS(sock_net(skb->sk),
 173				      LINUX_MIB_TLSDECRYPTERROR);
 174		ctx->async_wait.err = err;
 175		tls_err_abort(skb->sk, err);
 176	} else {
 177		struct strp_msg *rxm = strp_msg(skb);
 178		int pad;
 179
 180		pad = padding_length(ctx, prot, skb);
 181		if (pad < 0) {
 182			ctx->async_wait.err = pad;
 183			tls_err_abort(skb->sk, pad);
 184		} else {
 185			rxm->full_len -= pad;
 186			rxm->offset += prot->prepend_size;
 187			rxm->full_len -= prot->overhead_size;
 188		}
 189	}
 190
 191	/* After using skb->sk to propagate sk through crypto async callback
 192	 * we need to NULL it again.
 193	 */
 194	skb->sk = NULL;
 195
 196
 197	/* Free the destination pages if skb was not decrypted inplace */
 198	if (sgout != sgin) {
 199		/* Skip the first S/G entry as it points to AAD */
 200		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 201			if (!sg)
 202				break;
 203			put_page(sg_page(sg));
 204		}
 205	}
 206
 207	kfree(aead_req);
 208
 209	spin_lock_bh(&ctx->decrypt_compl_lock);
 210	pending = atomic_dec_return(&ctx->decrypt_pending);
 211
 212	if (!pending && ctx->async_notify)
 213		complete(&ctx->async_wait.completion);
 214	spin_unlock_bh(&ctx->decrypt_compl_lock);
 215}
 216
 217static int tls_do_decryption(struct sock *sk,
 218			     struct sk_buff *skb,
 219			     struct scatterlist *sgin,
 220			     struct scatterlist *sgout,
 221			     char *iv_recv,
 222			     size_t data_len,
 223			     struct aead_request *aead_req,
 224			     bool async)
 225{
 226	struct tls_context *tls_ctx = tls_get_ctx(sk);
 227	struct tls_prot_info *prot = &tls_ctx->prot_info;
 228	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 229	int ret;
 230
 231	aead_request_set_tfm(aead_req, ctx->aead_recv);
 232	aead_request_set_ad(aead_req, prot->aad_size);
 233	aead_request_set_crypt(aead_req, sgin, sgout,
 234			       data_len + prot->tag_size,
 235			       (u8 *)iv_recv);
 236
 237	if (async) {
 238		/* Using skb->sk to push sk through to crypto async callback
 239		 * handler. This allows propagating errors up to the socket
 240		 * if needed. It _must_ be cleared in the async handler
 241		 * before consume_skb is called. We _know_ skb->sk is NULL
 242		 * because it is a clone from strparser.
 243		 */
 244		skb->sk = sk;
 245		aead_request_set_callback(aead_req,
 246					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 247					  tls_decrypt_done, skb);
 248		atomic_inc(&ctx->decrypt_pending);
 249	} else {
 250		aead_request_set_callback(aead_req,
 251					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 252					  crypto_req_done, &ctx->async_wait);
 253	}
 254
 255	ret = crypto_aead_decrypt(aead_req);
 256	if (ret == -EINPROGRESS) {
 257		if (async)
 258			return ret;
 259
 260		ret = crypto_wait_req(ret, &ctx->async_wait);
 261	}
 262
 263	if (async)
 264		atomic_dec(&ctx->decrypt_pending);
 265
 266	return ret;
 267}
 268
 269static void tls_trim_both_msgs(struct sock *sk, int target_size)
 270{
 271	struct tls_context *tls_ctx = tls_get_ctx(sk);
 272	struct tls_prot_info *prot = &tls_ctx->prot_info;
 273	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 274	struct tls_rec *rec = ctx->open_rec;
 275
 276	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 277	if (target_size > 0)
 278		target_size += prot->overhead_size;
 279	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 280}
 281
 282static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 283{
 284	struct tls_context *tls_ctx = tls_get_ctx(sk);
 285	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 286	struct tls_rec *rec = ctx->open_rec;
 287	struct sk_msg *msg_en = &rec->msg_encrypted;
 288
 289	return sk_msg_alloc(sk, msg_en, len, 0);
 290}
 291
 292static int tls_clone_plaintext_msg(struct sock *sk, int required)
 293{
 294	struct tls_context *tls_ctx = tls_get_ctx(sk);
 295	struct tls_prot_info *prot = &tls_ctx->prot_info;
 296	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 297	struct tls_rec *rec = ctx->open_rec;
 298	struct sk_msg *msg_pl = &rec->msg_plaintext;
 299	struct sk_msg *msg_en = &rec->msg_encrypted;
 300	int skip, len;
 301
 302	/* We add page references worth len bytes from encrypted sg
 303	 * at the end of plaintext sg. It is guaranteed that msg_en
 304	 * has enough required room (ensured by caller).
 305	 */
 306	len = required - msg_pl->sg.size;
 307
 308	/* Skip initial bytes in msg_en's data to be able to use
 309	 * same offset of both plain and encrypted data.
 310	 */
 311	skip = prot->prepend_size + msg_pl->sg.size;
 312
 313	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 314}
 315
 316static struct tls_rec *tls_get_rec(struct sock *sk)
 317{
 318	struct tls_context *tls_ctx = tls_get_ctx(sk);
 319	struct tls_prot_info *prot = &tls_ctx->prot_info;
 320	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 321	struct sk_msg *msg_pl, *msg_en;
 322	struct tls_rec *rec;
 323	int mem_size;
 324
 325	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 326
 327	rec = kzalloc(mem_size, sk->sk_allocation);
 328	if (!rec)
 329		return NULL;
 330
 331	msg_pl = &rec->msg_plaintext;
 332	msg_en = &rec->msg_encrypted;
 333
 334	sk_msg_init(msg_pl);
 335	sk_msg_init(msg_en);
 336
 337	sg_init_table(rec->sg_aead_in, 2);
 338	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 339	sg_unmark_end(&rec->sg_aead_in[1]);
 340
 341	sg_init_table(rec->sg_aead_out, 2);
 342	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 343	sg_unmark_end(&rec->sg_aead_out[1]);
 344
 345	return rec;
 346}
 347
 348static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 349{
 350	sk_msg_free(sk, &rec->msg_encrypted);
 351	sk_msg_free(sk, &rec->msg_plaintext);
 352	kfree(rec);
 353}
 354
 355static void tls_free_open_rec(struct sock *sk)
 356{
 357	struct tls_context *tls_ctx = tls_get_ctx(sk);
 358	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 359	struct tls_rec *rec = ctx->open_rec;
 360
 361	if (rec) {
 362		tls_free_rec(sk, rec);
 363		ctx->open_rec = NULL;
 364	}
 365}
 366
 367int tls_tx_records(struct sock *sk, int flags)
 368{
 369	struct tls_context *tls_ctx = tls_get_ctx(sk);
 370	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 371	struct tls_rec *rec, *tmp;
 372	struct sk_msg *msg_en;
 373	int tx_flags, rc = 0;
 374
 375	if (tls_is_partially_sent_record(tls_ctx)) {
 376		rec = list_first_entry(&ctx->tx_list,
 377				       struct tls_rec, list);
 378
 379		if (flags == -1)
 380			tx_flags = rec->tx_flags;
 381		else
 382			tx_flags = flags;
 383
 384		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 385		if (rc)
 386			goto tx_err;
 387
 388		/* Full record has been transmitted.
 389		 * Remove the head of tx_list
 390		 */
 391		list_del(&rec->list);
 392		sk_msg_free(sk, &rec->msg_plaintext);
 393		kfree(rec);
 394	}
 395
 396	/* Tx all ready records */
 397	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 398		if (READ_ONCE(rec->tx_ready)) {
 399			if (flags == -1)
 400				tx_flags = rec->tx_flags;
 401			else
 402				tx_flags = flags;
 403
 404			msg_en = &rec->msg_encrypted;
 405			rc = tls_push_sg(sk, tls_ctx,
 406					 &msg_en->sg.data[msg_en->sg.curr],
 407					 0, tx_flags);
 408			if (rc)
 409				goto tx_err;
 410
 411			list_del(&rec->list);
 412			sk_msg_free(sk, &rec->msg_plaintext);
 413			kfree(rec);
 414		} else {
 415			break;
 416		}
 417	}
 418
 419tx_err:
 420	if (rc < 0 && rc != -EAGAIN)
 421		tls_err_abort(sk, EBADMSG);
 422
 423	return rc;
 424}
 425
 426static void tls_encrypt_done(struct crypto_async_request *req, int err)
 427{
 428	struct aead_request *aead_req = (struct aead_request *)req;
 429	struct sock *sk = req->data;
 430	struct tls_context *tls_ctx = tls_get_ctx(sk);
 431	struct tls_prot_info *prot = &tls_ctx->prot_info;
 432	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 433	struct scatterlist *sge;
 434	struct sk_msg *msg_en;
 435	struct tls_rec *rec;
 436	bool ready = false;
 437	int pending;
 438
 439	rec = container_of(aead_req, struct tls_rec, aead_req);
 440	msg_en = &rec->msg_encrypted;
 441
 442	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 443	sge->offset -= prot->prepend_size;
 444	sge->length += prot->prepend_size;
 445
 446	/* Check if error is previously set on socket */
 447	if (err || sk->sk_err) {
 448		rec = NULL;
 449
 450		/* If err is already set on socket, return the same code */
 451		if (sk->sk_err) {
 452			ctx->async_wait.err = sk->sk_err;
 453		} else {
 454			ctx->async_wait.err = err;
 455			tls_err_abort(sk, err);
 456		}
 457	}
 458
 459	if (rec) {
 460		struct tls_rec *first_rec;
 461
 462		/* Mark the record as ready for transmission */
 463		smp_store_mb(rec->tx_ready, true);
 464
 465		/* If received record is at head of tx_list, schedule tx */
 466		first_rec = list_first_entry(&ctx->tx_list,
 467					     struct tls_rec, list);
 468		if (rec == first_rec)
 469			ready = true;
 470	}
 471
 472	spin_lock_bh(&ctx->encrypt_compl_lock);
 473	pending = atomic_dec_return(&ctx->encrypt_pending);
 474
 475	if (!pending && ctx->async_notify)
 476		complete(&ctx->async_wait.completion);
 477	spin_unlock_bh(&ctx->encrypt_compl_lock);
 478
 479	if (!ready)
 480		return;
 481
 482	/* Schedule the transmission */
 483	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 484		schedule_delayed_work(&ctx->tx_work.work, 1);
 485}
 486
 487static int tls_do_encryption(struct sock *sk,
 488			     struct tls_context *tls_ctx,
 489			     struct tls_sw_context_tx *ctx,
 490			     struct aead_request *aead_req,
 491			     size_t data_len, u32 start)
 492{
 493	struct tls_prot_info *prot = &tls_ctx->prot_info;
 494	struct tls_rec *rec = ctx->open_rec;
 495	struct sk_msg *msg_en = &rec->msg_encrypted;
 496	struct scatterlist *sge = sk_msg_elem(msg_en, start);
 497	int rc, iv_offset = 0;
 498
 499	/* For CCM based ciphers, first byte of IV is a constant */
 500	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
 501		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 502		iv_offset = 1;
 503	}
 504
 505	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 506	       prot->iv_size + prot->salt_size);
 507
 508	xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq);
 509
 510	sge->offset += prot->prepend_size;
 511	sge->length -= prot->prepend_size;
 512
 513	msg_en->sg.curr = start;
 514
 515	aead_request_set_tfm(aead_req, ctx->aead_send);
 516	aead_request_set_ad(aead_req, prot->aad_size);
 517	aead_request_set_crypt(aead_req, rec->sg_aead_in,
 518			       rec->sg_aead_out,
 519			       data_len, rec->iv_data);
 520
 521	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 522				  tls_encrypt_done, sk);
 523
 524	/* Add the record in tx_list */
 525	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 526	atomic_inc(&ctx->encrypt_pending);
 527
 528	rc = crypto_aead_encrypt(aead_req);
 529	if (!rc || rc != -EINPROGRESS) {
 530		atomic_dec(&ctx->encrypt_pending);
 531		sge->offset -= prot->prepend_size;
 532		sge->length += prot->prepend_size;
 533	}
 534
 535	if (!rc) {
 536		WRITE_ONCE(rec->tx_ready, true);
 537	} else if (rc != -EINPROGRESS) {
 538		list_del(&rec->list);
 539		return rc;
 540	}
 541
 542	/* Unhook the record from context if encryption is not failure */
 543	ctx->open_rec = NULL;
 544	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 545	return rc;
 546}
 547
 548static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 549				 struct tls_rec **to, struct sk_msg *msg_opl,
 550				 struct sk_msg *msg_oen, u32 split_point,
 551				 u32 tx_overhead_size, u32 *orig_end)
 552{
 553	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 554	struct scatterlist *sge, *osge, *nsge;
 555	u32 orig_size = msg_opl->sg.size;
 556	struct scatterlist tmp = { };
 557	struct sk_msg *msg_npl;
 558	struct tls_rec *new;
 559	int ret;
 560
 561	new = tls_get_rec(sk);
 562	if (!new)
 563		return -ENOMEM;
 564	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 565			   tx_overhead_size, 0);
 566	if (ret < 0) {
 567		tls_free_rec(sk, new);
 568		return ret;
 569	}
 570
 571	*orig_end = msg_opl->sg.end;
 572	i = msg_opl->sg.start;
 573	sge = sk_msg_elem(msg_opl, i);
 574	while (apply && sge->length) {
 575		if (sge->length > apply) {
 576			u32 len = sge->length - apply;
 577
 578			get_page(sg_page(sge));
 579			sg_set_page(&tmp, sg_page(sge), len,
 580				    sge->offset + apply);
 581			sge->length = apply;
 582			bytes += apply;
 583			apply = 0;
 584		} else {
 585			apply -= sge->length;
 586			bytes += sge->length;
 587		}
 588
 589		sk_msg_iter_var_next(i);
 590		if (i == msg_opl->sg.end)
 591			break;
 592		sge = sk_msg_elem(msg_opl, i);
 593	}
 594
 595	msg_opl->sg.end = i;
 596	msg_opl->sg.curr = i;
 597	msg_opl->sg.copybreak = 0;
 598	msg_opl->apply_bytes = 0;
 599	msg_opl->sg.size = bytes;
 600
 601	msg_npl = &new->msg_plaintext;
 602	msg_npl->apply_bytes = apply;
 603	msg_npl->sg.size = orig_size - bytes;
 604
 605	j = msg_npl->sg.start;
 606	nsge = sk_msg_elem(msg_npl, j);
 607	if (tmp.length) {
 608		memcpy(nsge, &tmp, sizeof(*nsge));
 609		sk_msg_iter_var_next(j);
 610		nsge = sk_msg_elem(msg_npl, j);
 611	}
 612
 613	osge = sk_msg_elem(msg_opl, i);
 614	while (osge->length) {
 615		memcpy(nsge, osge, sizeof(*nsge));
 616		sg_unmark_end(nsge);
 617		sk_msg_iter_var_next(i);
 618		sk_msg_iter_var_next(j);
 619		if (i == *orig_end)
 620			break;
 621		osge = sk_msg_elem(msg_opl, i);
 622		nsge = sk_msg_elem(msg_npl, j);
 623	}
 624
 625	msg_npl->sg.end = j;
 626	msg_npl->sg.curr = j;
 627	msg_npl->sg.copybreak = 0;
 628
 629	*to = new;
 630	return 0;
 631}
 632
 633static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 634				  struct tls_rec *from, u32 orig_end)
 635{
 636	struct sk_msg *msg_npl = &from->msg_plaintext;
 637	struct sk_msg *msg_opl = &to->msg_plaintext;
 638	struct scatterlist *osge, *nsge;
 639	u32 i, j;
 640
 641	i = msg_opl->sg.end;
 642	sk_msg_iter_var_prev(i);
 643	j = msg_npl->sg.start;
 644
 645	osge = sk_msg_elem(msg_opl, i);
 646	nsge = sk_msg_elem(msg_npl, j);
 647
 648	if (sg_page(osge) == sg_page(nsge) &&
 649	    osge->offset + osge->length == nsge->offset) {
 650		osge->length += nsge->length;
 651		put_page(sg_page(nsge));
 652	}
 653
 654	msg_opl->sg.end = orig_end;
 655	msg_opl->sg.curr = orig_end;
 656	msg_opl->sg.copybreak = 0;
 657	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 658	msg_opl->sg.size += msg_npl->sg.size;
 659
 660	sk_msg_free(sk, &to->msg_encrypted);
 661	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 662
 663	kfree(from);
 664}
 665
 666static int tls_push_record(struct sock *sk, int flags,
 667			   unsigned char record_type)
 668{
 669	struct tls_context *tls_ctx = tls_get_ctx(sk);
 670	struct tls_prot_info *prot = &tls_ctx->prot_info;
 671	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 672	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 673	u32 i, split_point, orig_end;
 674	struct sk_msg *msg_pl, *msg_en;
 675	struct aead_request *req;
 676	bool split;
 677	int rc;
 678
 679	if (!rec)
 680		return 0;
 681
 682	msg_pl = &rec->msg_plaintext;
 683	msg_en = &rec->msg_encrypted;
 684
 685	split_point = msg_pl->apply_bytes;
 686	split = split_point && split_point < msg_pl->sg.size;
 687	if (unlikely((!split &&
 688		      msg_pl->sg.size +
 689		      prot->overhead_size > msg_en->sg.size) ||
 690		     (split &&
 691		      split_point +
 692		      prot->overhead_size > msg_en->sg.size))) {
 693		split = true;
 694		split_point = msg_en->sg.size;
 695	}
 696	if (split) {
 697		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 698					   split_point, prot->overhead_size,
 699					   &orig_end);
 700		if (rc < 0)
 701			return rc;
 702		/* This can happen if above tls_split_open_record allocates
 703		 * a single large encryption buffer instead of two smaller
 704		 * ones. In this case adjust pointers and continue without
 705		 * split.
 706		 */
 707		if (!msg_pl->sg.size) {
 708			tls_merge_open_record(sk, rec, tmp, orig_end);
 709			msg_pl = &rec->msg_plaintext;
 710			msg_en = &rec->msg_encrypted;
 711			split = false;
 712		}
 713		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 714			    prot->overhead_size);
 715	}
 716
 717	rec->tx_flags = flags;
 718	req = &rec->aead_req;
 719
 720	i = msg_pl->sg.end;
 721	sk_msg_iter_var_prev(i);
 722
 723	rec->content_type = record_type;
 724	if (prot->version == TLS_1_3_VERSION) {
 725		/* Add content type to end of message.  No padding added */
 726		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 727		sg_mark_end(&rec->sg_content_type);
 728		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 729			 &rec->sg_content_type);
 730	} else {
 731		sg_mark_end(sk_msg_elem(msg_pl, i));
 732	}
 733
 734	if (msg_pl->sg.end < msg_pl->sg.start) {
 735		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 736			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 737			 msg_pl->sg.data);
 738	}
 739
 740	i = msg_pl->sg.start;
 741	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 
 742
 743	i = msg_en->sg.end;
 744	sk_msg_iter_var_prev(i);
 745	sg_mark_end(sk_msg_elem(msg_en, i));
 746
 747	i = msg_en->sg.start;
 748	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 749
 750	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 751		     tls_ctx->tx.rec_seq, prot->rec_seq_size,
 752		     record_type, prot->version);
 753
 754	tls_fill_prepend(tls_ctx,
 755			 page_address(sg_page(&msg_en->sg.data[i])) +
 756			 msg_en->sg.data[i].offset,
 757			 msg_pl->sg.size + prot->tail_size,
 758			 record_type, prot->version);
 759
 760	tls_ctx->pending_open_record_frags = false;
 761
 762	rc = tls_do_encryption(sk, tls_ctx, ctx, req,
 763			       msg_pl->sg.size + prot->tail_size, i);
 764	if (rc < 0) {
 765		if (rc != -EINPROGRESS) {
 766			tls_err_abort(sk, EBADMSG);
 767			if (split) {
 768				tls_ctx->pending_open_record_frags = true;
 769				tls_merge_open_record(sk, rec, tmp, orig_end);
 770			}
 771		}
 772		ctx->async_capable = 1;
 773		return rc;
 774	} else if (split) {
 775		msg_pl = &tmp->msg_plaintext;
 776		msg_en = &tmp->msg_encrypted;
 777		sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
 778		tls_ctx->pending_open_record_frags = true;
 779		ctx->open_rec = tmp;
 780	}
 781
 782	return tls_tx_records(sk, flags);
 783}
 784
 785static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
 786			       bool full_record, u8 record_type,
 787			       ssize_t *copied, int flags)
 788{
 789	struct tls_context *tls_ctx = tls_get_ctx(sk);
 790	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 791	struct sk_msg msg_redir = { };
 792	struct sk_psock *psock;
 793	struct sock *sk_redir;
 794	struct tls_rec *rec;
 795	bool enospc, policy;
 796	int err = 0, send;
 797	u32 delta = 0;
 798
 799	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
 800	psock = sk_psock_get(sk);
 801	if (!psock || !policy) {
 802		err = tls_push_record(sk, flags, record_type);
 803		if (err && sk->sk_err == EBADMSG) {
 804			*copied -= sk_msg_free(sk, msg);
 805			tls_free_open_rec(sk);
 806			err = -sk->sk_err;
 807		}
 808		if (psock)
 809			sk_psock_put(sk, psock);
 810		return err;
 811	}
 812more_data:
 813	enospc = sk_msg_full(msg);
 814	if (psock->eval == __SK_NONE) {
 815		delta = msg->sg.size;
 816		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
 817		delta -= msg->sg.size;
 
 
 
 818	}
 819	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
 820	    !enospc && !full_record) {
 821		err = -ENOSPC;
 822		goto out_err;
 823	}
 824	msg->cork_bytes = 0;
 825	send = msg->sg.size;
 826	if (msg->apply_bytes && msg->apply_bytes < send)
 827		send = msg->apply_bytes;
 828
 829	switch (psock->eval) {
 830	case __SK_PASS:
 831		err = tls_push_record(sk, flags, record_type);
 832		if (err && sk->sk_err == EBADMSG) {
 833			*copied -= sk_msg_free(sk, msg);
 834			tls_free_open_rec(sk);
 835			err = -sk->sk_err;
 836			goto out_err;
 837		}
 838		break;
 839	case __SK_REDIRECT:
 840		sk_redir = psock->sk_redir;
 841		memcpy(&msg_redir, msg, sizeof(*msg));
 842		if (msg->apply_bytes < send)
 843			msg->apply_bytes = 0;
 844		else
 845			msg->apply_bytes -= send;
 846		sk_msg_return_zero(sk, msg, send);
 847		msg->sg.size -= send;
 848		release_sock(sk);
 849		err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
 850		lock_sock(sk);
 851		if (err < 0) {
 852			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
 853			msg->sg.size = 0;
 854		}
 855		if (msg->sg.size == 0)
 856			tls_free_open_rec(sk);
 857		break;
 858	case __SK_DROP:
 859	default:
 860		sk_msg_free_partial(sk, msg, send);
 861		if (msg->apply_bytes < send)
 862			msg->apply_bytes = 0;
 863		else
 864			msg->apply_bytes -= send;
 865		if (msg->sg.size == 0)
 866			tls_free_open_rec(sk);
 867		*copied -= (send + delta);
 868		err = -EACCES;
 869	}
 870
 871	if (likely(!err)) {
 872		bool reset_eval = !ctx->open_rec;
 873
 874		rec = ctx->open_rec;
 875		if (rec) {
 876			msg = &rec->msg_plaintext;
 877			if (!msg->apply_bytes)
 878				reset_eval = true;
 879		}
 880		if (reset_eval) {
 881			psock->eval = __SK_NONE;
 882			if (psock->sk_redir) {
 883				sock_put(psock->sk_redir);
 884				psock->sk_redir = NULL;
 885			}
 886		}
 887		if (rec)
 888			goto more_data;
 889	}
 890 out_err:
 891	sk_psock_put(sk, psock);
 892	return err;
 893}
 894
 895static int tls_sw_push_pending_record(struct sock *sk, int flags)
 896{
 897	struct tls_context *tls_ctx = tls_get_ctx(sk);
 898	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 899	struct tls_rec *rec = ctx->open_rec;
 900	struct sk_msg *msg_pl;
 901	size_t copied;
 902
 903	if (!rec)
 904		return 0;
 905
 906	msg_pl = &rec->msg_plaintext;
 907	copied = msg_pl->sg.size;
 908	if (!copied)
 909		return 0;
 910
 911	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
 912				   &copied, flags);
 913}
 914
 915int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 916{
 917	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
 918	struct tls_context *tls_ctx = tls_get_ctx(sk);
 919	struct tls_prot_info *prot = &tls_ctx->prot_info;
 920	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 921	bool async_capable = ctx->async_capable;
 922	unsigned char record_type = TLS_RECORD_TYPE_DATA;
 923	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
 924	bool eor = !(msg->msg_flags & MSG_MORE);
 925	size_t try_to_copy;
 926	ssize_t copied = 0;
 927	struct sk_msg *msg_pl, *msg_en;
 928	struct tls_rec *rec;
 929	int required_size;
 930	int num_async = 0;
 931	bool full_record;
 932	int record_room;
 933	int num_zc = 0;
 934	int orig_size;
 935	int ret = 0;
 936	int pending;
 937
 938	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
 939			       MSG_CMSG_COMPAT))
 940		return -EOPNOTSUPP;
 941
 942	mutex_lock(&tls_ctx->tx_lock);
 943	lock_sock(sk);
 944
 945	if (unlikely(msg->msg_controllen)) {
 946		ret = tls_proccess_cmsg(sk, msg, &record_type);
 947		if (ret) {
 948			if (ret == -EINPROGRESS)
 949				num_async++;
 950			else if (ret != -EAGAIN)
 951				goto send_end;
 952		}
 953	}
 954
 955	while (msg_data_left(msg)) {
 956		if (sk->sk_err) {
 957			ret = -sk->sk_err;
 958			goto send_end;
 959		}
 960
 961		if (ctx->open_rec)
 962			rec = ctx->open_rec;
 963		else
 964			rec = ctx->open_rec = tls_get_rec(sk);
 965		if (!rec) {
 966			ret = -ENOMEM;
 967			goto send_end;
 968		}
 969
 970		msg_pl = &rec->msg_plaintext;
 971		msg_en = &rec->msg_encrypted;
 972
 973		orig_size = msg_pl->sg.size;
 974		full_record = false;
 975		try_to_copy = msg_data_left(msg);
 976		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
 977		if (try_to_copy >= record_room) {
 978			try_to_copy = record_room;
 979			full_record = true;
 980		}
 981
 982		required_size = msg_pl->sg.size + try_to_copy +
 983				prot->overhead_size;
 984
 985		if (!sk_stream_memory_free(sk))
 986			goto wait_for_sndbuf;
 987
 988alloc_encrypted:
 989		ret = tls_alloc_encrypted_msg(sk, required_size);
 990		if (ret) {
 991			if (ret != -ENOSPC)
 992				goto wait_for_memory;
 993
 994			/* Adjust try_to_copy according to the amount that was
 995			 * actually allocated. The difference is due
 996			 * to max sg elements limit
 997			 */
 998			try_to_copy -= required_size - msg_en->sg.size;
 999			full_record = true;
1000		}
1001
1002		if (!is_kvec && (full_record || eor) && !async_capable) {
1003			u32 first = msg_pl->sg.end;
1004
1005			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1006							msg_pl, try_to_copy);
1007			if (ret)
1008				goto fallback_to_reg_send;
1009
 
 
1010			num_zc++;
1011			copied += try_to_copy;
1012
1013			sk_msg_sg_copy_set(msg_pl, first);
1014			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1015						  record_type, &copied,
1016						  msg->msg_flags);
1017			if (ret) {
1018				if (ret == -EINPROGRESS)
1019					num_async++;
1020				else if (ret == -ENOMEM)
1021					goto wait_for_memory;
1022				else if (ctx->open_rec && ret == -ENOSPC)
1023					goto rollback_iter;
1024				else if (ret != -EAGAIN)
1025					goto send_end;
1026			}
1027			continue;
1028rollback_iter:
1029			copied -= try_to_copy;
1030			sk_msg_sg_copy_clear(msg_pl, first);
1031			iov_iter_revert(&msg->msg_iter,
1032					msg_pl->sg.size - orig_size);
1033fallback_to_reg_send:
1034			sk_msg_trim(sk, msg_pl, orig_size);
1035		}
1036
1037		required_size = msg_pl->sg.size + try_to_copy;
1038
1039		ret = tls_clone_plaintext_msg(sk, required_size);
1040		if (ret) {
1041			if (ret != -ENOSPC)
1042				goto send_end;
1043
1044			/* Adjust try_to_copy according to the amount that was
1045			 * actually allocated. The difference is due
1046			 * to max sg elements limit
1047			 */
1048			try_to_copy -= required_size - msg_pl->sg.size;
1049			full_record = true;
1050			sk_msg_trim(sk, msg_en,
1051				    msg_pl->sg.size + prot->overhead_size);
1052		}
1053
1054		if (try_to_copy) {
1055			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1056						       msg_pl, try_to_copy);
1057			if (ret < 0)
1058				goto trim_sgl;
1059		}
1060
1061		/* Open records defined only if successfully copied, otherwise
1062		 * we would trim the sg but not reset the open record frags.
1063		 */
1064		tls_ctx->pending_open_record_frags = true;
1065		copied += try_to_copy;
1066		if (full_record || eor) {
1067			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1068						  record_type, &copied,
1069						  msg->msg_flags);
1070			if (ret) {
1071				if (ret == -EINPROGRESS)
1072					num_async++;
1073				else if (ret == -ENOMEM)
1074					goto wait_for_memory;
1075				else if (ret != -EAGAIN) {
1076					if (ret == -ENOSPC)
1077						ret = 0;
1078					goto send_end;
1079				}
1080			}
1081		}
1082
1083		continue;
1084
1085wait_for_sndbuf:
1086		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1087wait_for_memory:
1088		ret = sk_stream_wait_memory(sk, &timeo);
1089		if (ret) {
1090trim_sgl:
1091			if (ctx->open_rec)
1092				tls_trim_both_msgs(sk, orig_size);
1093			goto send_end;
1094		}
1095
1096		if (ctx->open_rec && msg_en->sg.size < required_size)
1097			goto alloc_encrypted;
1098	}
1099
1100	if (!num_async) {
1101		goto send_end;
1102	} else if (num_zc) {
1103		/* Wait for pending encryptions to get completed */
1104		spin_lock_bh(&ctx->encrypt_compl_lock);
1105		ctx->async_notify = true;
1106
1107		pending = atomic_read(&ctx->encrypt_pending);
1108		spin_unlock_bh(&ctx->encrypt_compl_lock);
1109		if (pending)
1110			crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1111		else
1112			reinit_completion(&ctx->async_wait.completion);
1113
1114		/* There can be no concurrent accesses, since we have no
1115		 * pending encrypt operations
1116		 */
1117		WRITE_ONCE(ctx->async_notify, false);
1118
1119		if (ctx->async_wait.err) {
1120			ret = ctx->async_wait.err;
1121			copied = 0;
1122		}
1123	}
1124
1125	/* Transmit if any encryptions have completed */
1126	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1127		cancel_delayed_work(&ctx->tx_work.work);
1128		tls_tx_records(sk, msg->msg_flags);
1129	}
1130
1131send_end:
1132	ret = sk_stream_error(sk, msg->msg_flags, ret);
1133
1134	release_sock(sk);
1135	mutex_unlock(&tls_ctx->tx_lock);
1136	return copied > 0 ? copied : ret;
1137}
1138
1139static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1140			      int offset, size_t size, int flags)
1141{
1142	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1143	struct tls_context *tls_ctx = tls_get_ctx(sk);
1144	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1145	struct tls_prot_info *prot = &tls_ctx->prot_info;
1146	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1147	struct sk_msg *msg_pl;
1148	struct tls_rec *rec;
1149	int num_async = 0;
1150	ssize_t copied = 0;
1151	bool full_record;
1152	int record_room;
1153	int ret = 0;
1154	bool eor;
1155
1156	eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1157	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1158
1159	/* Call the sk_stream functions to manage the sndbuf mem. */
1160	while (size > 0) {
1161		size_t copy, required_size;
1162
1163		if (sk->sk_err) {
1164			ret = -sk->sk_err;
1165			goto sendpage_end;
1166		}
1167
1168		if (ctx->open_rec)
1169			rec = ctx->open_rec;
1170		else
1171			rec = ctx->open_rec = tls_get_rec(sk);
1172		if (!rec) {
1173			ret = -ENOMEM;
1174			goto sendpage_end;
1175		}
1176
1177		msg_pl = &rec->msg_plaintext;
1178
1179		full_record = false;
1180		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1181		copy = size;
1182		if (copy >= record_room) {
1183			copy = record_room;
1184			full_record = true;
1185		}
1186
1187		required_size = msg_pl->sg.size + copy + prot->overhead_size;
1188
1189		if (!sk_stream_memory_free(sk))
1190			goto wait_for_sndbuf;
1191alloc_payload:
1192		ret = tls_alloc_encrypted_msg(sk, required_size);
1193		if (ret) {
1194			if (ret != -ENOSPC)
1195				goto wait_for_memory;
1196
1197			/* Adjust copy according to the amount that was
1198			 * actually allocated. The difference is due
1199			 * to max sg elements limit
1200			 */
1201			copy -= required_size - msg_pl->sg.size;
1202			full_record = true;
1203		}
1204
1205		sk_msg_page_add(msg_pl, page, copy, offset);
1206		sk_mem_charge(sk, copy);
1207
1208		offset += copy;
1209		size -= copy;
1210		copied += copy;
1211
1212		tls_ctx->pending_open_record_frags = true;
1213		if (full_record || eor || sk_msg_full(msg_pl)) {
 
1214			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1215						  record_type, &copied, flags);
1216			if (ret) {
1217				if (ret == -EINPROGRESS)
1218					num_async++;
1219				else if (ret == -ENOMEM)
1220					goto wait_for_memory;
1221				else if (ret != -EAGAIN) {
1222					if (ret == -ENOSPC)
1223						ret = 0;
1224					goto sendpage_end;
1225				}
1226			}
1227		}
1228		continue;
1229wait_for_sndbuf:
1230		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1231wait_for_memory:
1232		ret = sk_stream_wait_memory(sk, &timeo);
1233		if (ret) {
1234			if (ctx->open_rec)
1235				tls_trim_both_msgs(sk, msg_pl->sg.size);
1236			goto sendpage_end;
1237		}
1238
1239		if (ctx->open_rec)
1240			goto alloc_payload;
1241	}
1242
1243	if (num_async) {
1244		/* Transmit if any encryptions have completed */
1245		if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1246			cancel_delayed_work(&ctx->tx_work.work);
1247			tls_tx_records(sk, flags);
1248		}
1249	}
1250sendpage_end:
1251	ret = sk_stream_error(sk, flags, ret);
1252	return copied > 0 ? copied : ret;
1253}
1254
1255int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1256			   int offset, size_t size, int flags)
1257{
1258	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1259		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1260		      MSG_NO_SHARED_FRAGS))
1261		return -EOPNOTSUPP;
1262
1263	return tls_sw_do_sendpage(sk, page, offset, size, flags);
1264}
1265
1266int tls_sw_sendpage(struct sock *sk, struct page *page,
1267		    int offset, size_t size, int flags)
1268{
1269	struct tls_context *tls_ctx = tls_get_ctx(sk);
1270	int ret;
1271
1272	if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273		      MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1274		return -EOPNOTSUPP;
1275
1276	mutex_lock(&tls_ctx->tx_lock);
1277	lock_sock(sk);
1278	ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1279	release_sock(sk);
1280	mutex_unlock(&tls_ctx->tx_lock);
1281	return ret;
1282}
1283
1284static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1285				     int flags, long timeo, int *err)
1286{
1287	struct tls_context *tls_ctx = tls_get_ctx(sk);
1288	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1289	struct sk_buff *skb;
1290	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1291
1292	while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1293		if (sk->sk_err) {
1294			*err = sock_error(sk);
1295			return NULL;
1296		}
1297
1298		if (sk->sk_shutdown & RCV_SHUTDOWN)
1299			return NULL;
1300
1301		if (sock_flag(sk, SOCK_DONE))
1302			return NULL;
1303
1304		if ((flags & MSG_DONTWAIT) || !timeo) {
1305			*err = -EAGAIN;
1306			return NULL;
1307		}
1308
1309		add_wait_queue(sk_sleep(sk), &wait);
1310		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1311		sk_wait_event(sk, &timeo,
1312			      ctx->recv_pkt != skb ||
1313			      !sk_psock_queue_empty(psock),
1314			      &wait);
1315		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1316		remove_wait_queue(sk_sleep(sk), &wait);
1317
1318		/* Handle signals */
1319		if (signal_pending(current)) {
1320			*err = sock_intr_errno(timeo);
1321			return NULL;
1322		}
1323	}
1324
1325	return skb;
1326}
1327
1328static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1329			       int length, int *pages_used,
1330			       unsigned int *size_used,
1331			       struct scatterlist *to,
1332			       int to_max_pages)
1333{
1334	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1335	struct page *pages[MAX_SKB_FRAGS];
1336	unsigned int size = *size_used;
1337	ssize_t copied, use;
1338	size_t offset;
1339
1340	while (length > 0) {
1341		i = 0;
1342		maxpages = to_max_pages - num_elem;
1343		if (maxpages == 0) {
1344			rc = -EFAULT;
1345			goto out;
1346		}
1347		copied = iov_iter_get_pages(from, pages,
1348					    length,
1349					    maxpages, &offset);
1350		if (copied <= 0) {
1351			rc = -EFAULT;
1352			goto out;
1353		}
1354
1355		iov_iter_advance(from, copied);
1356
1357		length -= copied;
1358		size += copied;
1359		while (copied) {
1360			use = min_t(int, copied, PAGE_SIZE - offset);
1361
1362			sg_set_page(&to[num_elem],
1363				    pages[i], use, offset);
1364			sg_unmark_end(&to[num_elem]);
1365			/* We do not uncharge memory from this API */
1366
1367			offset = 0;
1368			copied -= use;
1369
1370			i++;
1371			num_elem++;
1372		}
1373	}
1374	/* Mark the end in the last sg entry if newly added */
1375	if (num_elem > *pages_used)
1376		sg_mark_end(&to[num_elem - 1]);
1377out:
1378	if (rc)
1379		iov_iter_revert(from, size - *size_used);
1380	*size_used = size;
1381	*pages_used = num_elem;
1382
1383	return rc;
1384}
1385
1386/* This function decrypts the input skb into either out_iov or in out_sg
1387 * or in skb buffers itself. The input parameter 'zc' indicates if
1388 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1389 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1390 * NULL, then the decryption happens inside skb buffers itself, i.e.
1391 * zero-copy gets disabled and 'zc' is updated.
1392 */
1393
1394static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1395			    struct iov_iter *out_iov,
1396			    struct scatterlist *out_sg,
1397			    int *chunk, bool *zc, bool async)
1398{
1399	struct tls_context *tls_ctx = tls_get_ctx(sk);
1400	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1401	struct tls_prot_info *prot = &tls_ctx->prot_info;
1402	struct strp_msg *rxm = strp_msg(skb);
1403	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1404	struct aead_request *aead_req;
1405	struct sk_buff *unused;
1406	u8 *aad, *iv, *mem = NULL;
1407	struct scatterlist *sgin = NULL;
1408	struct scatterlist *sgout = NULL;
1409	const int data_len = rxm->full_len - prot->overhead_size +
1410			     prot->tail_size;
1411	int iv_offset = 0;
1412
1413	if (*zc && (out_iov || out_sg)) {
1414		if (out_iov)
1415			n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1416		else
1417			n_sgout = sg_nents(out_sg);
1418		n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1419				 rxm->full_len - prot->prepend_size);
1420	} else {
1421		n_sgout = 0;
1422		*zc = false;
1423		n_sgin = skb_cow_data(skb, 0, &unused);
1424	}
1425
1426	if (n_sgin < 1)
1427		return -EBADMSG;
1428
1429	/* Increment to accommodate AAD */
1430	n_sgin = n_sgin + 1;
1431
1432	nsg = n_sgin + n_sgout;
1433
1434	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1435	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1436	mem_size = mem_size + prot->aad_size;
1437	mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1438
1439	/* Allocate a single block of memory which contains
1440	 * aead_req || sgin[] || sgout[] || aad || iv.
1441	 * This order achieves correct alignment for aead_req, sgin, sgout.
1442	 */
1443	mem = kmalloc(mem_size, sk->sk_allocation);
1444	if (!mem)
1445		return -ENOMEM;
1446
1447	/* Segment the allocated memory */
1448	aead_req = (struct aead_request *)mem;
1449	sgin = (struct scatterlist *)(mem + aead_size);
1450	sgout = sgin + n_sgin;
1451	aad = (u8 *)(sgout + n_sgout);
1452	iv = aad + prot->aad_size;
1453
1454	/* For CCM based ciphers, first byte of nonce+iv is always '2' */
1455	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1456		iv[0] = 2;
1457		iv_offset = 1;
1458	}
1459
1460	/* Prepare IV */
1461	err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1462			    iv + iv_offset + prot->salt_size,
1463			    prot->iv_size);
1464	if (err < 0) {
1465		kfree(mem);
1466		return err;
1467	}
1468	if (prot->version == TLS_1_3_VERSION)
1469		memcpy(iv + iv_offset, tls_ctx->rx.iv,
1470		       crypto_aead_ivsize(ctx->aead_recv));
1471	else
1472		memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1473
1474	xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq);
1475
1476	/* Prepare AAD */
1477	tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1478		     prot->tail_size,
1479		     tls_ctx->rx.rec_seq, prot->rec_seq_size,
1480		     ctx->control, prot->version);
1481
1482	/* Prepare sgin */
1483	sg_init_table(sgin, n_sgin);
1484	sg_set_buf(&sgin[0], aad, prot->aad_size);
1485	err = skb_to_sgvec(skb, &sgin[1],
1486			   rxm->offset + prot->prepend_size,
1487			   rxm->full_len - prot->prepend_size);
1488	if (err < 0) {
1489		kfree(mem);
1490		return err;
1491	}
1492
1493	if (n_sgout) {
1494		if (out_iov) {
1495			sg_init_table(sgout, n_sgout);
1496			sg_set_buf(&sgout[0], aad, prot->aad_size);
1497
1498			*chunk = 0;
1499			err = tls_setup_from_iter(sk, out_iov, data_len,
1500						  &pages, chunk, &sgout[1],
1501						  (n_sgout - 1));
1502			if (err < 0)
1503				goto fallback_to_reg_recv;
1504		} else if (out_sg) {
1505			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1506		} else {
1507			goto fallback_to_reg_recv;
1508		}
1509	} else {
1510fallback_to_reg_recv:
1511		sgout = sgin;
1512		pages = 0;
1513		*chunk = data_len;
1514		*zc = false;
1515	}
1516
1517	/* Prepare and submit AEAD request */
1518	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1519				data_len, aead_req, async);
1520	if (err == -EINPROGRESS)
1521		return err;
1522
1523	/* Release the pages in case iov was mapped to pages */
1524	for (; pages > 0; pages--)
1525		put_page(sg_page(&sgout[pages]));
1526
1527	kfree(mem);
1528	return err;
1529}
1530
1531static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1532			      struct iov_iter *dest, int *chunk, bool *zc,
1533			      bool async)
1534{
1535	struct tls_context *tls_ctx = tls_get_ctx(sk);
1536	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1537	struct tls_prot_info *prot = &tls_ctx->prot_info;
1538	struct strp_msg *rxm = strp_msg(skb);
1539	int pad, err = 0;
1540
1541	if (!ctx->decrypted) {
1542		if (tls_ctx->rx_conf == TLS_HW) {
1543			err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1544			if (err < 0)
1545				return err;
1546		}
1547
1548		/* Still not decrypted after tls_device */
1549		if (!ctx->decrypted) {
1550			err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1551					       async);
1552			if (err < 0) {
1553				if (err == -EINPROGRESS)
1554					tls_advance_record_sn(sk, prot,
1555							      &tls_ctx->rx);
1556				else if (err == -EBADMSG)
1557					TLS_INC_STATS(sock_net(sk),
1558						      LINUX_MIB_TLSDECRYPTERROR);
1559				return err;
1560			}
1561		} else {
1562			*zc = false;
1563		}
1564
1565		pad = padding_length(ctx, prot, skb);
1566		if (pad < 0)
1567			return pad;
1568
1569		rxm->full_len -= pad;
1570		rxm->offset += prot->prepend_size;
1571		rxm->full_len -= prot->overhead_size;
1572		tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1573		ctx->decrypted = 1;
1574		ctx->saved_data_ready(sk);
1575	} else {
1576		*zc = false;
1577	}
1578
1579	return err;
1580}
1581
1582int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1583		struct scatterlist *sgout)
1584{
1585	bool zc = true;
1586	int chunk;
1587
1588	return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1589}
1590
1591static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1592			       unsigned int len)
1593{
1594	struct tls_context *tls_ctx = tls_get_ctx(sk);
1595	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1596
1597	if (skb) {
1598		struct strp_msg *rxm = strp_msg(skb);
1599
1600		if (len < rxm->full_len) {
1601			rxm->offset += len;
1602			rxm->full_len -= len;
1603			return false;
1604		}
1605		consume_skb(skb);
1606	}
1607
1608	/* Finished with message */
1609	ctx->recv_pkt = NULL;
1610	__strp_unpause(&ctx->strp);
1611
1612	return true;
1613}
1614
1615/* This function traverses the rx_list in tls receive context to copies the
1616 * decrypted records into the buffer provided by caller zero copy is not
1617 * true. Further, the records are removed from the rx_list if it is not a peek
1618 * case and the record has been consumed completely.
1619 */
1620static int process_rx_list(struct tls_sw_context_rx *ctx,
1621			   struct msghdr *msg,
1622			   u8 *control,
1623			   bool *cmsg,
1624			   size_t skip,
1625			   size_t len,
1626			   bool zc,
1627			   bool is_peek)
1628{
1629	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1630	u8 ctrl = *control;
1631	u8 msgc = *cmsg;
1632	struct tls_msg *tlm;
1633	ssize_t copied = 0;
1634
1635	/* Set the record type in 'control' if caller didn't pass it */
1636	if (!ctrl && skb) {
1637		tlm = tls_msg(skb);
1638		ctrl = tlm->control;
1639	}
1640
1641	while (skip && skb) {
1642		struct strp_msg *rxm = strp_msg(skb);
1643		tlm = tls_msg(skb);
1644
1645		/* Cannot process a record of different type */
1646		if (ctrl != tlm->control)
1647			return 0;
1648
1649		if (skip < rxm->full_len)
1650			break;
1651
1652		skip = skip - rxm->full_len;
1653		skb = skb_peek_next(skb, &ctx->rx_list);
1654	}
1655
1656	while (len && skb) {
1657		struct sk_buff *next_skb;
1658		struct strp_msg *rxm = strp_msg(skb);
1659		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1660
1661		tlm = tls_msg(skb);
1662
1663		/* Cannot process a record of different type */
1664		if (ctrl != tlm->control)
1665			return 0;
1666
1667		/* Set record type if not already done. For a non-data record,
1668		 * do not proceed if record type could not be copied.
1669		 */
1670		if (!msgc) {
1671			int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1672					    sizeof(ctrl), &ctrl);
1673			msgc = true;
1674			if (ctrl != TLS_RECORD_TYPE_DATA) {
1675				if (cerr || msg->msg_flags & MSG_CTRUNC)
1676					return -EIO;
1677
1678				*cmsg = msgc;
1679			}
1680		}
1681
1682		if (!zc || (rxm->full_len - skip) > len) {
1683			int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1684						    msg, chunk);
1685			if (err < 0)
1686				return err;
1687		}
1688
1689		len = len - chunk;
1690		copied = copied + chunk;
1691
1692		/* Consume the data from record if it is non-peek case*/
1693		if (!is_peek) {
1694			rxm->offset = rxm->offset + chunk;
1695			rxm->full_len = rxm->full_len - chunk;
1696
1697			/* Return if there is unconsumed data in the record */
1698			if (rxm->full_len - skip)
1699				break;
1700		}
1701
1702		/* The remaining skip-bytes must lie in 1st record in rx_list.
1703		 * So from the 2nd record, 'skip' should be 0.
1704		 */
1705		skip = 0;
1706
1707		if (msg)
1708			msg->msg_flags |= MSG_EOR;
1709
1710		next_skb = skb_peek_next(skb, &ctx->rx_list);
1711
1712		if (!is_peek) {
1713			skb_unlink(skb, &ctx->rx_list);
1714			consume_skb(skb);
1715		}
1716
1717		skb = next_skb;
1718	}
1719
1720	*control = ctrl;
1721	return copied;
1722}
1723
1724int tls_sw_recvmsg(struct sock *sk,
1725		   struct msghdr *msg,
1726		   size_t len,
1727		   int nonblock,
1728		   int flags,
1729		   int *addr_len)
1730{
1731	struct tls_context *tls_ctx = tls_get_ctx(sk);
1732	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1733	struct tls_prot_info *prot = &tls_ctx->prot_info;
1734	struct sk_psock *psock;
1735	unsigned char control = 0;
1736	ssize_t decrypted = 0;
1737	struct strp_msg *rxm;
1738	struct tls_msg *tlm;
1739	struct sk_buff *skb;
1740	ssize_t copied = 0;
1741	bool cmsg = false;
1742	int target, err = 0;
1743	long timeo;
1744	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1745	bool is_peek = flags & MSG_PEEK;
1746	bool bpf_strp_enabled;
1747	int num_async = 0;
1748	int pending;
1749
1750	flags |= nonblock;
1751
1752	if (unlikely(flags & MSG_ERRQUEUE))
1753		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1754
1755	psock = sk_psock_get(sk);
1756	lock_sock(sk);
1757	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1758
1759	/* Process pending decrypted records. It must be non-zero-copy */
1760	err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1761			      is_peek);
1762	if (err < 0) {
1763		tls_err_abort(sk, err);
1764		goto end;
1765	} else {
1766		copied = err;
1767	}
1768
1769	if (len <= copied)
1770		goto recv_end;
1771
1772	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1773	len = len - copied;
1774	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1775
1776	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1777		bool retain_skb = false;
1778		bool zc = false;
1779		int to_decrypt;
1780		int chunk = 0;
1781		bool async_capable;
1782		bool async = false;
1783
1784		skb = tls_wait_data(sk, psock, flags, timeo, &err);
1785		if (!skb) {
1786			if (psock) {
1787				int ret = __tcp_bpf_recvmsg(sk, psock,
1788							    msg, len, flags);
1789
1790				if (ret > 0) {
1791					decrypted += ret;
1792					len -= ret;
1793					continue;
1794				}
1795			}
1796			goto recv_end;
1797		} else {
1798			tlm = tls_msg(skb);
1799			if (prot->version == TLS_1_3_VERSION)
1800				tlm->control = 0;
1801			else
1802				tlm->control = ctx->control;
1803		}
1804
1805		rxm = strp_msg(skb);
1806
1807		to_decrypt = rxm->full_len - prot->overhead_size;
1808
1809		if (to_decrypt <= len && !is_kvec && !is_peek &&
1810		    ctx->control == TLS_RECORD_TYPE_DATA &&
1811		    prot->version != TLS_1_3_VERSION &&
1812		    !bpf_strp_enabled)
1813			zc = true;
1814
1815		/* Do not use async mode if record is non-data */
1816		if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1817			async_capable = ctx->async_capable;
1818		else
1819			async_capable = false;
1820
1821		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1822					 &chunk, &zc, async_capable);
1823		if (err < 0 && err != -EINPROGRESS) {
1824			tls_err_abort(sk, EBADMSG);
1825			goto recv_end;
1826		}
1827
1828		if (err == -EINPROGRESS) {
1829			async = true;
1830			num_async++;
1831		} else if (prot->version == TLS_1_3_VERSION) {
1832			tlm->control = ctx->control;
1833		}
1834
1835		/* If the type of records being processed is not known yet,
1836		 * set it to record type just dequeued. If it is already known,
1837		 * but does not match the record type just dequeued, go to end.
1838		 * We always get record type here since for tls1.2, record type
1839		 * is known just after record is dequeued from stream parser.
1840		 * For tls1.3, we disable async.
1841		 */
1842
1843		if (!control)
1844			control = tlm->control;
1845		else if (control != tlm->control)
1846			goto recv_end;
1847
1848		if (!cmsg) {
1849			int cerr;
1850
1851			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1852					sizeof(control), &control);
1853			cmsg = true;
1854			if (control != TLS_RECORD_TYPE_DATA) {
1855				if (cerr || msg->msg_flags & MSG_CTRUNC) {
1856					err = -EIO;
1857					goto recv_end;
1858				}
1859			}
1860		}
1861
1862		if (async)
1863			goto pick_next_record;
1864
1865		if (!zc) {
1866			if (bpf_strp_enabled) {
1867				err = sk_psock_tls_strp_read(psock, skb);
1868				if (err != __SK_PASS) {
1869					rxm->offset = rxm->offset + rxm->full_len;
1870					rxm->full_len = 0;
1871					if (err == __SK_DROP)
1872						consume_skb(skb);
1873					ctx->recv_pkt = NULL;
1874					__strp_unpause(&ctx->strp);
1875					continue;
1876				}
1877			}
1878
1879			if (rxm->full_len > len) {
1880				retain_skb = true;
1881				chunk = len;
1882			} else {
1883				chunk = rxm->full_len;
1884			}
1885
1886			err = skb_copy_datagram_msg(skb, rxm->offset,
1887						    msg, chunk);
1888			if (err < 0)
1889				goto recv_end;
1890
1891			if (!is_peek) {
1892				rxm->offset = rxm->offset + chunk;
1893				rxm->full_len = rxm->full_len - chunk;
1894			}
1895		}
1896
1897pick_next_record:
1898		if (chunk > len)
1899			chunk = len;
1900
1901		decrypted += chunk;
1902		len -= chunk;
1903
1904		/* For async or peek case, queue the current skb */
1905		if (async || is_peek || retain_skb) {
1906			skb_queue_tail(&ctx->rx_list, skb);
1907			skb = NULL;
1908		}
1909
1910		if (tls_sw_advance_skb(sk, skb, chunk)) {
1911			/* Return full control message to
1912			 * userspace before trying to parse
1913			 * another message type
1914			 */
1915			msg->msg_flags |= MSG_EOR;
1916			if (ctx->control != TLS_RECORD_TYPE_DATA)
1917				goto recv_end;
1918		} else {
1919			break;
1920		}
1921	}
1922
1923recv_end:
1924	if (num_async) {
1925		/* Wait for all previously submitted records to be decrypted */
1926		spin_lock_bh(&ctx->decrypt_compl_lock);
1927		ctx->async_notify = true;
1928		pending = atomic_read(&ctx->decrypt_pending);
1929		spin_unlock_bh(&ctx->decrypt_compl_lock);
1930		if (pending) {
1931			err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1932			if (err) {
1933				/* one of async decrypt failed */
1934				tls_err_abort(sk, err);
1935				copied = 0;
1936				decrypted = 0;
1937				goto end;
1938			}
1939		} else {
1940			reinit_completion(&ctx->async_wait.completion);
1941		}
1942
1943		/* There can be no concurrent accesses, since we have no
1944		 * pending decrypt operations
1945		 */
1946		WRITE_ONCE(ctx->async_notify, false);
1947
1948		/* Drain records from the rx_list & copy if required */
1949		if (is_peek || is_kvec)
1950			err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1951					      decrypted, false, is_peek);
1952		else
1953			err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1954					      decrypted, true, is_peek);
1955		if (err < 0) {
1956			tls_err_abort(sk, err);
1957			copied = 0;
1958			goto end;
1959		}
1960	}
1961
1962	copied += decrypted;
1963
1964end:
1965	release_sock(sk);
1966	if (psock)
1967		sk_psock_put(sk, psock);
1968	return copied ? : err;
1969}
1970
1971ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
1972			   struct pipe_inode_info *pipe,
1973			   size_t len, unsigned int flags)
1974{
1975	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1976	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1977	struct strp_msg *rxm = NULL;
1978	struct sock *sk = sock->sk;
1979	struct sk_buff *skb;
1980	ssize_t copied = 0;
1981	int err = 0;
1982	long timeo;
1983	int chunk;
1984	bool zc = false;
1985
1986	lock_sock(sk);
1987
1988	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1989
1990	skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1991	if (!skb)
1992		goto splice_read_end;
1993
1994	if (!ctx->decrypted) {
1995		err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
1996
1997		/* splice does not support reading control messages */
1998		if (ctx->control != TLS_RECORD_TYPE_DATA) {
1999			err = -EINVAL;
2000			goto splice_read_end;
2001		}
2002
2003		if (err < 0) {
2004			tls_err_abort(sk, EBADMSG);
2005			goto splice_read_end;
2006		}
2007		ctx->decrypted = 1;
2008	}
2009	rxm = strp_msg(skb);
2010
2011	chunk = min_t(unsigned int, rxm->full_len, len);
2012	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2013	if (copied < 0)
2014		goto splice_read_end;
2015
2016	if (likely(!(flags & MSG_PEEK)))
2017		tls_sw_advance_skb(sk, skb, copied);
2018
2019splice_read_end:
2020	release_sock(sk);
2021	return copied ? : err;
2022}
2023
2024bool tls_sw_stream_read(const struct sock *sk)
2025{
2026	struct tls_context *tls_ctx = tls_get_ctx(sk);
2027	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2028	bool ingress_empty = true;
2029	struct sk_psock *psock;
2030
2031	rcu_read_lock();
2032	psock = sk_psock(sk);
2033	if (psock)
2034		ingress_empty = list_empty(&psock->ingress_msg);
2035	rcu_read_unlock();
2036
2037	return !ingress_empty || ctx->recv_pkt ||
2038		!skb_queue_empty(&ctx->rx_list);
2039}
2040
2041static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2042{
2043	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2044	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2045	struct tls_prot_info *prot = &tls_ctx->prot_info;
2046	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2047	struct strp_msg *rxm = strp_msg(skb);
2048	size_t cipher_overhead;
2049	size_t data_len = 0;
2050	int ret;
2051
2052	/* Verify that we have a full TLS header, or wait for more data */
2053	if (rxm->offset + prot->prepend_size > skb->len)
2054		return 0;
2055
2056	/* Sanity-check size of on-stack buffer. */
2057	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2058		ret = -EINVAL;
2059		goto read_failure;
2060	}
2061
2062	/* Linearize header to local buffer */
2063	ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2064
2065	if (ret < 0)
2066		goto read_failure;
2067
2068	ctx->control = header[0];
2069
2070	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2071
2072	cipher_overhead = prot->tag_size;
2073	if (prot->version != TLS_1_3_VERSION)
2074		cipher_overhead += prot->iv_size;
2075
2076	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2077	    prot->tail_size) {
2078		ret = -EMSGSIZE;
2079		goto read_failure;
2080	}
2081	if (data_len < cipher_overhead) {
2082		ret = -EBADMSG;
2083		goto read_failure;
2084	}
2085
2086	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2087	if (header[1] != TLS_1_2_VERSION_MINOR ||
2088	    header[2] != TLS_1_2_VERSION_MAJOR) {
2089		ret = -EINVAL;
2090		goto read_failure;
2091	}
2092
2093	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2094				     TCP_SKB_CB(skb)->seq + rxm->offset);
2095	return data_len + TLS_HEADER_SIZE;
2096
2097read_failure:
2098	tls_err_abort(strp->sk, ret);
2099
2100	return ret;
2101}
2102
2103static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2104{
2105	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2106	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2107
2108	ctx->decrypted = 0;
2109
2110	ctx->recv_pkt = skb;
2111	strp_pause(strp);
2112
2113	ctx->saved_data_ready(strp->sk);
2114}
2115
2116static void tls_data_ready(struct sock *sk)
2117{
2118	struct tls_context *tls_ctx = tls_get_ctx(sk);
2119	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2120	struct sk_psock *psock;
2121
2122	strp_data_ready(&ctx->strp);
2123
2124	psock = sk_psock_get(sk);
2125	if (psock) {
2126		if (!list_empty(&psock->ingress_msg))
2127			ctx->saved_data_ready(sk);
2128		sk_psock_put(sk, psock);
2129	}
2130}
2131
2132void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2133{
2134	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2135
2136	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2137	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2138	cancel_delayed_work_sync(&ctx->tx_work.work);
2139}
2140
2141void tls_sw_release_resources_tx(struct sock *sk)
2142{
2143	struct tls_context *tls_ctx = tls_get_ctx(sk);
2144	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2145	struct tls_rec *rec, *tmp;
2146	int pending;
2147
2148	/* Wait for any pending async encryptions to complete */
2149	spin_lock_bh(&ctx->encrypt_compl_lock);
2150	ctx->async_notify = true;
2151	pending = atomic_read(&ctx->encrypt_pending);
2152	spin_unlock_bh(&ctx->encrypt_compl_lock);
2153
2154	if (pending)
2155		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2156
2157	tls_tx_records(sk, -1);
2158
2159	/* Free up un-sent records in tx_list. First, free
2160	 * the partially sent record if any at head of tx_list.
2161	 */
2162	if (tls_ctx->partially_sent_record) {
2163		tls_free_partial_record(sk, tls_ctx);
2164		rec = list_first_entry(&ctx->tx_list,
2165				       struct tls_rec, list);
2166		list_del(&rec->list);
2167		sk_msg_free(sk, &rec->msg_plaintext);
2168		kfree(rec);
2169	}
2170
2171	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2172		list_del(&rec->list);
2173		sk_msg_free(sk, &rec->msg_encrypted);
2174		sk_msg_free(sk, &rec->msg_plaintext);
2175		kfree(rec);
2176	}
2177
2178	crypto_free_aead(ctx->aead_send);
2179	tls_free_open_rec(sk);
2180}
2181
2182void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2183{
2184	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2185
2186	kfree(ctx);
2187}
2188
2189void tls_sw_release_resources_rx(struct sock *sk)
2190{
2191	struct tls_context *tls_ctx = tls_get_ctx(sk);
2192	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2193
2194	kfree(tls_ctx->rx.rec_seq);
2195	kfree(tls_ctx->rx.iv);
2196
2197	if (ctx->aead_recv) {
2198		kfree_skb(ctx->recv_pkt);
2199		ctx->recv_pkt = NULL;
2200		skb_queue_purge(&ctx->rx_list);
2201		crypto_free_aead(ctx->aead_recv);
2202		strp_stop(&ctx->strp);
2203		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2204		 * we still want to strp_stop(), but sk->sk_data_ready was
2205		 * never swapped.
2206		 */
2207		if (ctx->saved_data_ready) {
2208			write_lock_bh(&sk->sk_callback_lock);
2209			sk->sk_data_ready = ctx->saved_data_ready;
2210			write_unlock_bh(&sk->sk_callback_lock);
2211		}
2212	}
2213}
2214
2215void tls_sw_strparser_done(struct tls_context *tls_ctx)
2216{
2217	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2218
2219	strp_done(&ctx->strp);
2220}
2221
2222void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2223{
2224	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2225
2226	kfree(ctx);
2227}
2228
2229void tls_sw_free_resources_rx(struct sock *sk)
2230{
2231	struct tls_context *tls_ctx = tls_get_ctx(sk);
2232
2233	tls_sw_release_resources_rx(sk);
2234	tls_sw_free_ctx_rx(tls_ctx);
2235}
2236
2237/* The work handler to transmitt the encrypted records in tx_list */
2238static void tx_work_handler(struct work_struct *work)
2239{
2240	struct delayed_work *delayed_work = to_delayed_work(work);
2241	struct tx_work *tx_work = container_of(delayed_work,
2242					       struct tx_work, work);
2243	struct sock *sk = tx_work->sk;
2244	struct tls_context *tls_ctx = tls_get_ctx(sk);
2245	struct tls_sw_context_tx *ctx;
2246
2247	if (unlikely(!tls_ctx))
2248		return;
2249
2250	ctx = tls_sw_ctx_tx(tls_ctx);
2251	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2252		return;
2253
2254	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2255		return;
2256	mutex_lock(&tls_ctx->tx_lock);
2257	lock_sock(sk);
2258	tls_tx_records(sk, -1);
2259	release_sock(sk);
2260	mutex_unlock(&tls_ctx->tx_lock);
2261}
2262
2263void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2264{
2265	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2266
2267	/* Schedule the transmission if tx list is ready */
2268	if (is_tx_ready(tx_ctx) &&
2269	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2270		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2271}
2272
2273void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2274{
2275	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2276
2277	write_lock_bh(&sk->sk_callback_lock);
2278	rx_ctx->saved_data_ready = sk->sk_data_ready;
2279	sk->sk_data_ready = tls_data_ready;
2280	write_unlock_bh(&sk->sk_callback_lock);
2281
2282	strp_check_rcv(&rx_ctx->strp);
2283}
2284
2285int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2286{
2287	struct tls_context *tls_ctx = tls_get_ctx(sk);
2288	struct tls_prot_info *prot = &tls_ctx->prot_info;
2289	struct tls_crypto_info *crypto_info;
2290	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2291	struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2292	struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2293	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2294	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2295	struct cipher_context *cctx;
2296	struct crypto_aead **aead;
2297	struct strp_callbacks cb;
2298	u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2299	struct crypto_tfm *tfm;
2300	char *iv, *rec_seq, *key, *salt, *cipher_name;
2301	size_t keysize;
2302	int rc = 0;
2303
2304	if (!ctx) {
2305		rc = -EINVAL;
2306		goto out;
2307	}
2308
2309	if (tx) {
2310		if (!ctx->priv_ctx_tx) {
2311			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2312			if (!sw_ctx_tx) {
2313				rc = -ENOMEM;
2314				goto out;
2315			}
2316			ctx->priv_ctx_tx = sw_ctx_tx;
2317		} else {
2318			sw_ctx_tx =
2319				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2320		}
2321	} else {
2322		if (!ctx->priv_ctx_rx) {
2323			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2324			if (!sw_ctx_rx) {
2325				rc = -ENOMEM;
2326				goto out;
2327			}
2328			ctx->priv_ctx_rx = sw_ctx_rx;
2329		} else {
2330			sw_ctx_rx =
2331				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2332		}
2333	}
2334
2335	if (tx) {
2336		crypto_init_wait(&sw_ctx_tx->async_wait);
2337		spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2338		crypto_info = &ctx->crypto_send.info;
2339		cctx = &ctx->tx;
2340		aead = &sw_ctx_tx->aead_send;
2341		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2342		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2343		sw_ctx_tx->tx_work.sk = sk;
2344	} else {
2345		crypto_init_wait(&sw_ctx_rx->async_wait);
2346		spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2347		crypto_info = &ctx->crypto_recv.info;
2348		cctx = &ctx->rx;
2349		skb_queue_head_init(&sw_ctx_rx->rx_list);
2350		aead = &sw_ctx_rx->aead_recv;
2351	}
2352
2353	switch (crypto_info->cipher_type) {
2354	case TLS_CIPHER_AES_GCM_128: {
2355		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2356		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2357		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2358		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2359		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2360		rec_seq =
2361		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2362		gcm_128_info =
2363			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2364		keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2365		key = gcm_128_info->key;
2366		salt = gcm_128_info->salt;
2367		salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2368		cipher_name = "gcm(aes)";
2369		break;
2370	}
2371	case TLS_CIPHER_AES_GCM_256: {
2372		nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2373		tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2374		iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2375		iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2376		rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2377		rec_seq =
2378		 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2379		gcm_256_info =
2380			(struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2381		keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2382		key = gcm_256_info->key;
2383		salt = gcm_256_info->salt;
2384		salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2385		cipher_name = "gcm(aes)";
2386		break;
2387	}
2388	case TLS_CIPHER_AES_CCM_128: {
2389		nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2390		tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2391		iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2392		iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2393		rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2394		rec_seq =
2395		((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2396		ccm_128_info =
2397		(struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2398		keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2399		key = ccm_128_info->key;
2400		salt = ccm_128_info->salt;
2401		salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2402		cipher_name = "ccm(aes)";
2403		break;
2404	}
2405	default:
2406		rc = -EINVAL;
2407		goto free_priv;
2408	}
2409
2410	/* Sanity-check the sizes for stack allocations. */
2411	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2412	    rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2413		rc = -EINVAL;
2414		goto free_priv;
2415	}
2416
2417	if (crypto_info->version == TLS_1_3_VERSION) {
2418		nonce_size = 0;
2419		prot->aad_size = TLS_HEADER_SIZE;
2420		prot->tail_size = 1;
2421	} else {
2422		prot->aad_size = TLS_AAD_SPACE_SIZE;
2423		prot->tail_size = 0;
2424	}
2425
2426	prot->version = crypto_info->version;
2427	prot->cipher_type = crypto_info->cipher_type;
2428	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2429	prot->tag_size = tag_size;
2430	prot->overhead_size = prot->prepend_size +
2431			      prot->tag_size + prot->tail_size;
2432	prot->iv_size = iv_size;
2433	prot->salt_size = salt_size;
2434	cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2435	if (!cctx->iv) {
2436		rc = -ENOMEM;
2437		goto free_priv;
2438	}
2439	/* Note: 128 & 256 bit salt are the same size */
2440	prot->rec_seq_size = rec_seq_size;
2441	memcpy(cctx->iv, salt, salt_size);
2442	memcpy(cctx->iv + salt_size, iv, iv_size);
2443	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2444	if (!cctx->rec_seq) {
2445		rc = -ENOMEM;
2446		goto free_iv;
2447	}
2448
2449	if (!*aead) {
2450		*aead = crypto_alloc_aead(cipher_name, 0, 0);
2451		if (IS_ERR(*aead)) {
2452			rc = PTR_ERR(*aead);
2453			*aead = NULL;
2454			goto free_rec_seq;
2455		}
2456	}
2457
2458	ctx->push_pending_record = tls_sw_push_pending_record;
2459
2460	rc = crypto_aead_setkey(*aead, key, keysize);
2461
2462	if (rc)
2463		goto free_aead;
2464
2465	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2466	if (rc)
2467		goto free_aead;
2468
2469	if (sw_ctx_rx) {
2470		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2471
2472		if (crypto_info->version == TLS_1_3_VERSION)
2473			sw_ctx_rx->async_capable = 0;
2474		else
2475			sw_ctx_rx->async_capable =
2476				!!(tfm->__crt_alg->cra_flags &
2477				   CRYPTO_ALG_ASYNC);
2478
2479		/* Set up strparser */
2480		memset(&cb, 0, sizeof(cb));
2481		cb.rcv_msg = tls_queue;
2482		cb.parse_msg = tls_read_size;
2483
2484		strp_init(&sw_ctx_rx->strp, sk, &cb);
2485	}
2486
2487	goto out;
2488
2489free_aead:
2490	crypto_free_aead(*aead);
2491	*aead = NULL;
2492free_rec_seq:
2493	kfree(cctx->rec_seq);
2494	cctx->rec_seq = NULL;
2495free_iv:
2496	kfree(cctx->iv);
2497	cctx->iv = NULL;
2498free_priv:
2499	if (tx) {
2500		kfree(ctx->priv_ctx_tx);
2501		ctx->priv_ctx_tx = NULL;
2502	} else {
2503		kfree(ctx->priv_ctx_rx);
2504		ctx->priv_ctx_rx = NULL;
2505	}
2506out:
2507	return rc;
2508}