Linux Audio

Check our new training course

Loading...
v5.14.15
   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 <linux/splice.h>
  41#include <crypto/aead.h>
  42
  43#include <net/strparser.h>
  44#include <net/tls.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  45
  46static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  47                     unsigned int recursion_level)
  48{
  49        int start = skb_headlen(skb);
  50        int i, chunk = start - offset;
  51        struct sk_buff *frag_iter;
  52        int elt = 0;
  53
  54        if (unlikely(recursion_level >= 24))
  55                return -EMSGSIZE;
  56
  57        if (chunk > 0) {
  58                if (chunk > len)
  59                        chunk = len;
  60                elt++;
  61                len -= chunk;
  62                if (len == 0)
  63                        return elt;
  64                offset += chunk;
  65        }
  66
  67        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
  68                int end;
  69
  70                WARN_ON(start > offset + len);
  71
  72                end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
  73                chunk = end - offset;
  74                if (chunk > 0) {
  75                        if (chunk > len)
  76                                chunk = len;
  77                        elt++;
  78                        len -= chunk;
  79                        if (len == 0)
  80                                return elt;
  81                        offset += chunk;
  82                }
  83                start = end;
  84        }
  85
  86        if (unlikely(skb_has_frag_list(skb))) {
  87                skb_walk_frags(skb, frag_iter) {
  88                        int end, ret;
  89
  90                        WARN_ON(start > offset + len);
  91
  92                        end = start + frag_iter->len;
  93                        chunk = end - offset;
  94                        if (chunk > 0) {
  95                                if (chunk > len)
  96                                        chunk = len;
  97                                ret = __skb_nsg(frag_iter, offset - start, chunk,
  98                                                recursion_level + 1);
  99                                if (unlikely(ret < 0))
 100                                        return ret;
 101                                elt += ret;
 102                                len -= chunk;
 103                                if (len == 0)
 104                                        return elt;
 105                                offset += chunk;
 106                        }
 107                        start = end;
 108                }
 109        }
 110        BUG_ON(len);
 111        return elt;
 112}
 113
 114/* Return the number of scatterlist elements required to completely map the
 115 * skb, or -EMSGSIZE if the recursion depth is exceeded.
 116 */
 117static int skb_nsg(struct sk_buff *skb, int offset, int len)
 118{
 119        return __skb_nsg(skb, offset, len, 0);
 120}
 121
 122static int padding_length(struct tls_sw_context_rx *ctx,
 123			  struct tls_prot_info *prot, struct sk_buff *skb)
 124{
 125	struct strp_msg *rxm = strp_msg(skb);
 
 126	int sub = 0;
 127
 128	/* Determine zero-padding length */
 129	if (prot->version == TLS_1_3_VERSION) {
 130		char content_type = 0;
 
 131		int err;
 132		int back = 17;
 133
 134		while (content_type == 0) {
 135			if (back > rxm->full_len - prot->prepend_size)
 136				return -EBADMSG;
 137			err = skb_copy_bits(skb,
 138					    rxm->offset + rxm->full_len - back,
 139					    &content_type, 1);
 140			if (err)
 141				return err;
 142			if (content_type)
 143				break;
 144			sub++;
 145			back++;
 146		}
 147		ctx->control = content_type;
 148	}
 149	return sub;
 150}
 151
 152static void tls_decrypt_done(struct crypto_async_request *req, int err)
 153{
 154	struct aead_request *aead_req = (struct aead_request *)req;
 
 155	struct scatterlist *sgout = aead_req->dst;
 156	struct scatterlist *sgin = aead_req->src;
 157	struct tls_sw_context_rx *ctx;
 
 158	struct tls_context *tls_ctx;
 159	struct tls_prot_info *prot;
 160	struct scatterlist *sg;
 161	struct sk_buff *skb;
 162	unsigned int pages;
 163	int pending;
 
 164
 165	skb = (struct sk_buff *)req->data;
 166	tls_ctx = tls_get_ctx(skb->sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 167	ctx = tls_sw_ctx_rx(tls_ctx);
 168	prot = &tls_ctx->prot_info;
 169
 170	/* Propagate if there was an err */
 171	if (err) {
 172		if (err == -EBADMSG)
 173			TLS_INC_STATS(sock_net(skb->sk),
 174				      LINUX_MIB_TLSDECRYPTERROR);
 175		ctx->async_wait.err = err;
 176		tls_err_abort(skb->sk, err);
 177	} else {
 178		struct strp_msg *rxm = strp_msg(skb);
 179		int pad;
 180
 181		pad = padding_length(ctx, prot, skb);
 182		if (pad < 0) {
 183			ctx->async_wait.err = pad;
 184			tls_err_abort(skb->sk, pad);
 185		} else {
 186			rxm->full_len -= pad;
 187			rxm->offset += prot->prepend_size;
 188			rxm->full_len -= prot->overhead_size;
 189		}
 190	}
 191
 192	/* After using skb->sk to propagate sk through crypto async callback
 193	 * we need to NULL it again.
 194	 */
 195	skb->sk = NULL;
 196
 197
 198	/* Free the destination pages if skb was not decrypted inplace */
 199	if (sgout != sgin) {
 200		/* Skip the first S/G entry as it points to AAD */
 201		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 202			if (!sg)
 203				break;
 204			put_page(sg_page(sg));
 205		}
 206	}
 207
 208	kfree(aead_req);
 209
 210	spin_lock_bh(&ctx->decrypt_compl_lock);
 211	pending = atomic_dec_return(&ctx->decrypt_pending);
 212
 213	if (!pending && ctx->async_notify)
 214		complete(&ctx->async_wait.completion);
 215	spin_unlock_bh(&ctx->decrypt_compl_lock);
 
 
 
 
 
 
 
 
 216}
 217
 218static int tls_do_decryption(struct sock *sk,
 219			     struct sk_buff *skb,
 220			     struct scatterlist *sgin,
 221			     struct scatterlist *sgout,
 222			     char *iv_recv,
 223			     size_t data_len,
 224			     struct aead_request *aead_req,
 225			     bool async)
 226{
 227	struct tls_context *tls_ctx = tls_get_ctx(sk);
 228	struct tls_prot_info *prot = &tls_ctx->prot_info;
 229	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 230	int ret;
 231
 232	aead_request_set_tfm(aead_req, ctx->aead_recv);
 233	aead_request_set_ad(aead_req, prot->aad_size);
 234	aead_request_set_crypt(aead_req, sgin, sgout,
 235			       data_len + prot->tag_size,
 236			       (u8 *)iv_recv);
 237
 238	if (async) {
 239		/* Using skb->sk to push sk through to crypto async callback
 240		 * handler. This allows propagating errors up to the socket
 241		 * if needed. It _must_ be cleared in the async handler
 242		 * before consume_skb is called. We _know_ skb->sk is NULL
 243		 * because it is a clone from strparser.
 244		 */
 245		skb->sk = sk;
 246		aead_request_set_callback(aead_req,
 247					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 248					  tls_decrypt_done, skb);
 
 249		atomic_inc(&ctx->decrypt_pending);
 250	} else {
 
 
 251		aead_request_set_callback(aead_req,
 252					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 253					  crypto_req_done, &ctx->async_wait);
 
 
 
 
 254	}
 255
 256	ret = crypto_aead_decrypt(aead_req);
 257	if (ret == -EINPROGRESS) {
 258		if (async)
 259			return ret;
 260
 261		ret = crypto_wait_req(ret, &ctx->async_wait);
 
 
 
 
 
 262	}
 263
 264	if (async)
 265		atomic_dec(&ctx->decrypt_pending);
 266
 267	return ret;
 268}
 269
 270static void tls_trim_both_msgs(struct sock *sk, int target_size)
 271{
 272	struct tls_context *tls_ctx = tls_get_ctx(sk);
 273	struct tls_prot_info *prot = &tls_ctx->prot_info;
 274	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 275	struct tls_rec *rec = ctx->open_rec;
 276
 277	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 278	if (target_size > 0)
 279		target_size += prot->overhead_size;
 280	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 281}
 282
 283static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 284{
 285	struct tls_context *tls_ctx = tls_get_ctx(sk);
 286	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 287	struct tls_rec *rec = ctx->open_rec;
 288	struct sk_msg *msg_en = &rec->msg_encrypted;
 289
 290	return sk_msg_alloc(sk, msg_en, len, 0);
 291}
 292
 293static int tls_clone_plaintext_msg(struct sock *sk, int required)
 294{
 295	struct tls_context *tls_ctx = tls_get_ctx(sk);
 296	struct tls_prot_info *prot = &tls_ctx->prot_info;
 297	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 298	struct tls_rec *rec = ctx->open_rec;
 299	struct sk_msg *msg_pl = &rec->msg_plaintext;
 300	struct sk_msg *msg_en = &rec->msg_encrypted;
 301	int skip, len;
 302
 303	/* We add page references worth len bytes from encrypted sg
 304	 * at the end of plaintext sg. It is guaranteed that msg_en
 305	 * has enough required room (ensured by caller).
 306	 */
 307	len = required - msg_pl->sg.size;
 308
 309	/* Skip initial bytes in msg_en's data to be able to use
 310	 * same offset of both plain and encrypted data.
 311	 */
 312	skip = prot->prepend_size + msg_pl->sg.size;
 313
 314	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 315}
 316
 317static struct tls_rec *tls_get_rec(struct sock *sk)
 318{
 319	struct tls_context *tls_ctx = tls_get_ctx(sk);
 320	struct tls_prot_info *prot = &tls_ctx->prot_info;
 321	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 322	struct sk_msg *msg_pl, *msg_en;
 323	struct tls_rec *rec;
 324	int mem_size;
 325
 326	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 327
 328	rec = kzalloc(mem_size, sk->sk_allocation);
 329	if (!rec)
 330		return NULL;
 331
 332	msg_pl = &rec->msg_plaintext;
 333	msg_en = &rec->msg_encrypted;
 334
 335	sk_msg_init(msg_pl);
 336	sk_msg_init(msg_en);
 337
 338	sg_init_table(rec->sg_aead_in, 2);
 339	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 340	sg_unmark_end(&rec->sg_aead_in[1]);
 341
 342	sg_init_table(rec->sg_aead_out, 2);
 343	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 344	sg_unmark_end(&rec->sg_aead_out[1]);
 345
 
 
 346	return rec;
 347}
 348
 349static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 350{
 351	sk_msg_free(sk, &rec->msg_encrypted);
 352	sk_msg_free(sk, &rec->msg_plaintext);
 353	kfree(rec);
 354}
 355
 356static void tls_free_open_rec(struct sock *sk)
 357{
 358	struct tls_context *tls_ctx = tls_get_ctx(sk);
 359	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 360	struct tls_rec *rec = ctx->open_rec;
 361
 362	if (rec) {
 363		tls_free_rec(sk, rec);
 364		ctx->open_rec = NULL;
 365	}
 366}
 367
 368int tls_tx_records(struct sock *sk, int flags)
 369{
 370	struct tls_context *tls_ctx = tls_get_ctx(sk);
 371	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 372	struct tls_rec *rec, *tmp;
 373	struct sk_msg *msg_en;
 374	int tx_flags, rc = 0;
 375
 376	if (tls_is_partially_sent_record(tls_ctx)) {
 377		rec = list_first_entry(&ctx->tx_list,
 378				       struct tls_rec, list);
 379
 380		if (flags == -1)
 381			tx_flags = rec->tx_flags;
 382		else
 383			tx_flags = flags;
 384
 385		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 386		if (rc)
 387			goto tx_err;
 388
 389		/* Full record has been transmitted.
 390		 * Remove the head of tx_list
 391		 */
 392		list_del(&rec->list);
 393		sk_msg_free(sk, &rec->msg_plaintext);
 394		kfree(rec);
 395	}
 396
 397	/* Tx all ready records */
 398	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 399		if (READ_ONCE(rec->tx_ready)) {
 400			if (flags == -1)
 401				tx_flags = rec->tx_flags;
 402			else
 403				tx_flags = flags;
 404
 405			msg_en = &rec->msg_encrypted;
 406			rc = tls_push_sg(sk, tls_ctx,
 407					 &msg_en->sg.data[msg_en->sg.curr],
 408					 0, tx_flags);
 409			if (rc)
 410				goto tx_err;
 411
 412			list_del(&rec->list);
 413			sk_msg_free(sk, &rec->msg_plaintext);
 414			kfree(rec);
 415		} else {
 416			break;
 417		}
 418	}
 419
 420tx_err:
 421	if (rc < 0 && rc != -EAGAIN)
 422		tls_err_abort(sk, EBADMSG);
 423
 424	return rc;
 425}
 426
 427static void tls_encrypt_done(struct crypto_async_request *req, int err)
 428{
 429	struct aead_request *aead_req = (struct aead_request *)req;
 430	struct sock *sk = req->data;
 431	struct tls_context *tls_ctx = tls_get_ctx(sk);
 432	struct tls_prot_info *prot = &tls_ctx->prot_info;
 433	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 434	struct scatterlist *sge;
 435	struct sk_msg *msg_en;
 436	struct tls_rec *rec;
 437	bool ready = false;
 438	int pending;
 
 439
 440	rec = container_of(aead_req, struct tls_rec, aead_req);
 441	msg_en = &rec->msg_encrypted;
 442
 
 
 
 
 
 443	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 444	sge->offset -= prot->prepend_size;
 445	sge->length += prot->prepend_size;
 446
 447	/* Check if error is previously set on socket */
 448	if (err || sk->sk_err) {
 449		rec = NULL;
 450
 451		/* If err is already set on socket, return the same code */
 452		if (sk->sk_err) {
 453			ctx->async_wait.err = sk->sk_err;
 454		} else {
 455			ctx->async_wait.err = err;
 456			tls_err_abort(sk, err);
 457		}
 458	}
 459
 460	if (rec) {
 461		struct tls_rec *first_rec;
 462
 463		/* Mark the record as ready for transmission */
 464		smp_store_mb(rec->tx_ready, true);
 465
 466		/* If received record is at head of tx_list, schedule tx */
 467		first_rec = list_first_entry(&ctx->tx_list,
 468					     struct tls_rec, list);
 469		if (rec == first_rec)
 470			ready = true;
 
 
 
 
 471	}
 472
 473	spin_lock_bh(&ctx->encrypt_compl_lock);
 474	pending = atomic_dec_return(&ctx->encrypt_pending);
 475
 476	if (!pending && ctx->async_notify)
 477		complete(&ctx->async_wait.completion);
 478	spin_unlock_bh(&ctx->encrypt_compl_lock);
 479
 480	if (!ready)
 481		return;
 
 
 
 482
 483	/* Schedule the transmission */
 484	if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
 485		schedule_delayed_work(&ctx->tx_work.work, 1);
 486}
 487
 488static int tls_do_encryption(struct sock *sk,
 489			     struct tls_context *tls_ctx,
 490			     struct tls_sw_context_tx *ctx,
 491			     struct aead_request *aead_req,
 492			     size_t data_len, u32 start)
 493{
 494	struct tls_prot_info *prot = &tls_ctx->prot_info;
 495	struct tls_rec *rec = ctx->open_rec;
 496	struct sk_msg *msg_en = &rec->msg_encrypted;
 497	struct scatterlist *sge = sk_msg_elem(msg_en, start);
 498	int rc, iv_offset = 0;
 499
 500	/* For CCM based ciphers, first byte of IV is a constant */
 501	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
 
 502		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 503		iv_offset = 1;
 
 
 
 
 
 504	}
 505
 506	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 507	       prot->iv_size + prot->salt_size);
 508
 509	xor_iv_with_seq(prot, rec->iv_data, tls_ctx->tx.rec_seq);
 
 510
 511	sge->offset += prot->prepend_size;
 512	sge->length -= prot->prepend_size;
 513
 514	msg_en->sg.curr = start;
 515
 516	aead_request_set_tfm(aead_req, ctx->aead_send);
 517	aead_request_set_ad(aead_req, prot->aad_size);
 518	aead_request_set_crypt(aead_req, rec->sg_aead_in,
 519			       rec->sg_aead_out,
 520			       data_len, rec->iv_data);
 521
 522	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 523				  tls_encrypt_done, sk);
 524
 525	/* Add the record in tx_list */
 526	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 
 527	atomic_inc(&ctx->encrypt_pending);
 528
 529	rc = crypto_aead_encrypt(aead_req);
 
 
 
 
 530	if (!rc || rc != -EINPROGRESS) {
 531		atomic_dec(&ctx->encrypt_pending);
 532		sge->offset -= prot->prepend_size;
 533		sge->length += prot->prepend_size;
 534	}
 535
 536	if (!rc) {
 537		WRITE_ONCE(rec->tx_ready, true);
 538	} else if (rc != -EINPROGRESS) {
 539		list_del(&rec->list);
 540		return rc;
 541	}
 542
 543	/* Unhook the record from context if encryption is not failure */
 544	ctx->open_rec = NULL;
 545	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 546	return rc;
 547}
 548
 549static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 550				 struct tls_rec **to, struct sk_msg *msg_opl,
 551				 struct sk_msg *msg_oen, u32 split_point,
 552				 u32 tx_overhead_size, u32 *orig_end)
 553{
 554	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 555	struct scatterlist *sge, *osge, *nsge;
 556	u32 orig_size = msg_opl->sg.size;
 557	struct scatterlist tmp = { };
 558	struct sk_msg *msg_npl;
 559	struct tls_rec *new;
 560	int ret;
 561
 562	new = tls_get_rec(sk);
 563	if (!new)
 564		return -ENOMEM;
 565	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 566			   tx_overhead_size, 0);
 567	if (ret < 0) {
 568		tls_free_rec(sk, new);
 569		return ret;
 570	}
 571
 572	*orig_end = msg_opl->sg.end;
 573	i = msg_opl->sg.start;
 574	sge = sk_msg_elem(msg_opl, i);
 575	while (apply && sge->length) {
 576		if (sge->length > apply) {
 577			u32 len = sge->length - apply;
 578
 579			get_page(sg_page(sge));
 580			sg_set_page(&tmp, sg_page(sge), len,
 581				    sge->offset + apply);
 582			sge->length = apply;
 583			bytes += apply;
 584			apply = 0;
 585		} else {
 586			apply -= sge->length;
 587			bytes += sge->length;
 588		}
 589
 590		sk_msg_iter_var_next(i);
 591		if (i == msg_opl->sg.end)
 592			break;
 593		sge = sk_msg_elem(msg_opl, i);
 594	}
 595
 596	msg_opl->sg.end = i;
 597	msg_opl->sg.curr = i;
 598	msg_opl->sg.copybreak = 0;
 599	msg_opl->apply_bytes = 0;
 600	msg_opl->sg.size = bytes;
 601
 602	msg_npl = &new->msg_plaintext;
 603	msg_npl->apply_bytes = apply;
 604	msg_npl->sg.size = orig_size - bytes;
 605
 606	j = msg_npl->sg.start;
 607	nsge = sk_msg_elem(msg_npl, j);
 608	if (tmp.length) {
 609		memcpy(nsge, &tmp, sizeof(*nsge));
 610		sk_msg_iter_var_next(j);
 611		nsge = sk_msg_elem(msg_npl, j);
 612	}
 613
 614	osge = sk_msg_elem(msg_opl, i);
 615	while (osge->length) {
 616		memcpy(nsge, osge, sizeof(*nsge));
 617		sg_unmark_end(nsge);
 618		sk_msg_iter_var_next(i);
 619		sk_msg_iter_var_next(j);
 620		if (i == *orig_end)
 621			break;
 622		osge = sk_msg_elem(msg_opl, i);
 623		nsge = sk_msg_elem(msg_npl, j);
 624	}
 625
 626	msg_npl->sg.end = j;
 627	msg_npl->sg.curr = j;
 628	msg_npl->sg.copybreak = 0;
 629
 630	*to = new;
 631	return 0;
 632}
 633
 634static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 635				  struct tls_rec *from, u32 orig_end)
 636{
 637	struct sk_msg *msg_npl = &from->msg_plaintext;
 638	struct sk_msg *msg_opl = &to->msg_plaintext;
 639	struct scatterlist *osge, *nsge;
 640	u32 i, j;
 641
 642	i = msg_opl->sg.end;
 643	sk_msg_iter_var_prev(i);
 644	j = msg_npl->sg.start;
 645
 646	osge = sk_msg_elem(msg_opl, i);
 647	nsge = sk_msg_elem(msg_npl, j);
 648
 649	if (sg_page(osge) == sg_page(nsge) &&
 650	    osge->offset + osge->length == nsge->offset) {
 651		osge->length += nsge->length;
 652		put_page(sg_page(nsge));
 653	}
 654
 655	msg_opl->sg.end = orig_end;
 656	msg_opl->sg.curr = orig_end;
 657	msg_opl->sg.copybreak = 0;
 658	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 659	msg_opl->sg.size += msg_npl->sg.size;
 660
 661	sk_msg_free(sk, &to->msg_encrypted);
 662	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 663
 664	kfree(from);
 665}
 666
 667static int tls_push_record(struct sock *sk, int flags,
 668			   unsigned char record_type)
 669{
 670	struct tls_context *tls_ctx = tls_get_ctx(sk);
 671	struct tls_prot_info *prot = &tls_ctx->prot_info;
 672	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 673	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 674	u32 i, split_point, orig_end;
 675	struct sk_msg *msg_pl, *msg_en;
 676	struct aead_request *req;
 677	bool split;
 678	int rc;
 679
 680	if (!rec)
 681		return 0;
 682
 683	msg_pl = &rec->msg_plaintext;
 684	msg_en = &rec->msg_encrypted;
 685
 686	split_point = msg_pl->apply_bytes;
 687	split = split_point && split_point < msg_pl->sg.size;
 688	if (unlikely((!split &&
 689		      msg_pl->sg.size +
 690		      prot->overhead_size > msg_en->sg.size) ||
 691		     (split &&
 692		      split_point +
 693		      prot->overhead_size > msg_en->sg.size))) {
 694		split = true;
 695		split_point = msg_en->sg.size;
 696	}
 697	if (split) {
 698		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 699					   split_point, prot->overhead_size,
 700					   &orig_end);
 701		if (rc < 0)
 702			return rc;
 703		/* This can happen if above tls_split_open_record allocates
 704		 * a single large encryption buffer instead of two smaller
 705		 * ones. In this case adjust pointers and continue without
 706		 * split.
 707		 */
 708		if (!msg_pl->sg.size) {
 709			tls_merge_open_record(sk, rec, tmp, orig_end);
 710			msg_pl = &rec->msg_plaintext;
 711			msg_en = &rec->msg_encrypted;
 712			split = false;
 713		}
 714		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 715			    prot->overhead_size);
 716	}
 717
 718	rec->tx_flags = flags;
 719	req = &rec->aead_req;
 720
 721	i = msg_pl->sg.end;
 722	sk_msg_iter_var_prev(i);
 723
 724	rec->content_type = record_type;
 725	if (prot->version == TLS_1_3_VERSION) {
 726		/* Add content type to end of message.  No padding added */
 727		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 728		sg_mark_end(&rec->sg_content_type);
 729		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 730			 &rec->sg_content_type);
 731	} else {
 732		sg_mark_end(sk_msg_elem(msg_pl, i));
 733	}
 734
 735	if (msg_pl->sg.end < msg_pl->sg.start) {
 736		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 737			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 738			 msg_pl->sg.data);
 739	}
 740
 741	i = msg_pl->sg.start;
 742	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 743
 744	i = msg_en->sg.end;
 745	sk_msg_iter_var_prev(i);
 746	sg_mark_end(sk_msg_elem(msg_en, i));
 747
 748	i = msg_en->sg.start;
 749	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 750
 751	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 752		     tls_ctx->tx.rec_seq, record_type, prot);
 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);
 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_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				     bool nonblock, 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 (!skb_queue_empty(&sk->sk_receive_queue)) {
1299			__strp_unpause(&ctx->strp);
1300			if (ctx->recv_pkt)
1301				return ctx->recv_pkt;
1302		}
1303
1304		if (sk->sk_shutdown & RCV_SHUTDOWN)
1305			return NULL;
1306
1307		if (sock_flag(sk, SOCK_DONE))
1308			return NULL;
1309
1310		if (nonblock || !timeo) {
1311			*err = -EAGAIN;
1312			return NULL;
1313		}
1314
 
1315		add_wait_queue(sk_sleep(sk), &wait);
1316		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1317		sk_wait_event(sk, &timeo,
1318			      ctx->recv_pkt != skb ||
1319			      !sk_psock_queue_empty(psock),
1320			      &wait);
1321		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1322		remove_wait_queue(sk_sleep(sk), &wait);
1323
1324		/* Handle signals */
1325		if (signal_pending(current)) {
1326			*err = sock_intr_errno(timeo);
1327			return NULL;
1328		}
1329	}
1330
1331	return skb;
 
 
1332}
1333
1334static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1335			       int length, int *pages_used,
1336			       unsigned int *size_used,
1337			       struct scatterlist *to,
1338			       int to_max_pages)
1339{
1340	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1341	struct page *pages[MAX_SKB_FRAGS];
1342	unsigned int size = *size_used;
1343	ssize_t copied, use;
1344	size_t offset;
1345
1346	while (length > 0) {
1347		i = 0;
1348		maxpages = to_max_pages - num_elem;
1349		if (maxpages == 0) {
1350			rc = -EFAULT;
1351			goto out;
1352		}
1353		copied = iov_iter_get_pages(from, pages,
1354					    length,
1355					    maxpages, &offset);
1356		if (copied <= 0) {
1357			rc = -EFAULT;
1358			goto out;
1359		}
1360
1361		iov_iter_advance(from, copied);
1362
1363		length -= copied;
1364		size += copied;
1365		while (copied) {
1366			use = min_t(int, copied, PAGE_SIZE - offset);
1367
1368			sg_set_page(&to[num_elem],
1369				    pages[i], use, offset);
1370			sg_unmark_end(&to[num_elem]);
1371			/* We do not uncharge memory from this API */
1372
1373			offset = 0;
1374			copied -= use;
1375
1376			i++;
1377			num_elem++;
1378		}
1379	}
1380	/* Mark the end in the last sg entry if newly added */
1381	if (num_elem > *pages_used)
1382		sg_mark_end(&to[num_elem - 1]);
1383out:
1384	if (rc)
1385		iov_iter_revert(from, size - *size_used);
1386	*size_used = size;
1387	*pages_used = num_elem;
1388
1389	return rc;
1390}
1391
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1392/* This function decrypts the input skb into either out_iov or in out_sg
1393 * or in skb buffers itself. The input parameter 'zc' indicates if
1394 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1395 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1396 * NULL, then the decryption happens inside skb buffers itself, i.e.
1397 * zero-copy gets disabled and 'zc' is updated.
1398 */
1399
1400static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1401			    struct iov_iter *out_iov,
1402			    struct scatterlist *out_sg,
1403			    int *chunk, bool *zc, bool async)
1404{
1405	struct tls_context *tls_ctx = tls_get_ctx(sk);
1406	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1407	struct tls_prot_info *prot = &tls_ctx->prot_info;
1408	struct strp_msg *rxm = strp_msg(skb);
1409	int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
 
 
1410	struct aead_request *aead_req;
1411	struct sk_buff *unused;
1412	u8 *aad, *iv, *mem = NULL;
1413	struct scatterlist *sgin = NULL;
1414	struct scatterlist *sgout = NULL;
1415	const int data_len = rxm->full_len - prot->overhead_size +
1416			     prot->tail_size;
 
 
1417	int iv_offset = 0;
 
 
 
 
 
 
 
 
 
1418
1419	if (*zc && (out_iov || out_sg)) {
1420		if (out_iov)
1421			n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
 
1422		else
1423			n_sgout = sg_nents(out_sg);
1424		n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1425				 rxm->full_len - prot->prepend_size);
1426	} else {
1427		n_sgout = 0;
1428		*zc = false;
1429		n_sgin = skb_cow_data(skb, 0, &unused);
1430	}
1431
1432	if (n_sgin < 1)
1433		return -EBADMSG;
 
 
 
 
1434
1435	/* Increment to accommodate AAD */
1436	n_sgin = n_sgin + 1;
1437
1438	nsg = n_sgin + n_sgout;
1439
1440	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1441	mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1442	mem_size = mem_size + prot->aad_size;
1443	mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1444
1445	/* Allocate a single block of memory which contains
1446	 * aead_req || sgin[] || sgout[] || aad || iv.
1447	 * This order achieves correct alignment for aead_req, sgin, sgout.
1448	 */
1449	mem = kmalloc(mem_size, sk->sk_allocation);
1450	if (!mem)
1451		return -ENOMEM;
 
 
 
 
 
1452
1453	/* Segment the allocated memory */
1454	aead_req = (struct aead_request *)mem;
1455	sgin = (struct scatterlist *)(mem + aead_size);
1456	sgout = sgin + n_sgin;
1457	aad = (u8 *)(sgout + n_sgout);
1458	iv = aad + prot->aad_size;
1459
1460	/* For CCM based ciphers, first byte of nonce+iv is always '2' */
1461	if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1462		iv[0] = 2;
 
1463		iv_offset = 1;
 
 
 
 
 
1464	}
1465
1466	/* Prepare IV */
1467	err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1468			    iv + iv_offset + prot->salt_size,
1469			    prot->iv_size);
1470	if (err < 0) {
1471		kfree(mem);
1472		return err;
1473	}
1474	if (prot->version == TLS_1_3_VERSION ||
1475	    prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1476		memcpy(iv + iv_offset, tls_ctx->rx.iv,
1477		       crypto_aead_ivsize(ctx->aead_recv));
1478	else
1479		memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1480
1481	xor_iv_with_seq(prot, iv, tls_ctx->rx.rec_seq);
 
 
 
 
 
1482
1483	/* Prepare AAD */
1484	tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1485		     prot->tail_size,
1486		     tls_ctx->rx.rec_seq, ctx->control, prot);
1487
1488	/* Prepare sgin */
1489	sg_init_table(sgin, n_sgin);
1490	sg_set_buf(&sgin[0], aad, prot->aad_size);
1491	err = skb_to_sgvec(skb, &sgin[1],
1492			   rxm->offset + prot->prepend_size,
1493			   rxm->full_len - prot->prepend_size);
1494	if (err < 0) {
1495		kfree(mem);
1496		return err;
1497	}
1498
1499	if (n_sgout) {
1500		if (out_iov) {
1501			sg_init_table(sgout, n_sgout);
1502			sg_set_buf(&sgout[0], aad, prot->aad_size);
1503
1504			*chunk = 0;
1505			err = tls_setup_from_iter(sk, out_iov, data_len,
1506						  &pages, chunk, &sgout[1],
1507						  (n_sgout - 1));
1508			if (err < 0)
1509				goto fallback_to_reg_recv;
1510		} else if (out_sg) {
1511			memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1512		} else {
1513			goto fallback_to_reg_recv;
 
 
 
 
 
 
 
1514		}
1515	} else {
1516fallback_to_reg_recv:
1517		sgout = sgin;
1518		pages = 0;
1519		*chunk = data_len;
1520		*zc = false;
1521	}
 
1522
1523	/* Prepare and submit AEAD request */
1524	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1525				data_len, aead_req, async);
1526	if (err == -EINPROGRESS)
 
 
 
 
 
 
 
 
 
 
 
 
1527		return err;
 
 
 
 
1528
 
 
 
 
1529	/* Release the pages in case iov was mapped to pages */
1530	for (; pages > 0; pages--)
1531		put_page(sg_page(&sgout[pages]));
1532
1533	kfree(mem);
 
 
1534	return err;
1535}
1536
1537static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1538			      struct iov_iter *dest, int *chunk, bool *zc,
1539			      bool async)
1540{
1541	struct tls_context *tls_ctx = tls_get_ctx(sk);
1542	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1543	struct tls_prot_info *prot = &tls_ctx->prot_info;
1544	struct strp_msg *rxm = strp_msg(skb);
1545	int pad, err = 0;
1546
1547	if (!ctx->decrypted) {
1548		if (tls_ctx->rx_conf == TLS_HW) {
1549			err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1550			if (err < 0)
1551				return err;
1552		}
 
1553
1554		/* Still not decrypted after tls_device */
1555		if (!ctx->decrypted) {
1556			err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1557					       async);
1558			if (err < 0) {
1559				if (err == -EINPROGRESS)
1560					tls_advance_record_sn(sk, prot,
1561							      &tls_ctx->rx);
1562				else if (err == -EBADMSG)
1563					TLS_INC_STATS(sock_net(sk),
1564						      LINUX_MIB_TLSDECRYPTERROR);
1565				return err;
1566			}
1567		} else {
1568			*zc = false;
1569		}
 
 
 
1570
1571		pad = padding_length(ctx, prot, skb);
1572		if (pad < 0)
1573			return pad;
1574
1575		rxm->full_len -= pad;
1576		rxm->offset += prot->prepend_size;
1577		rxm->full_len -= prot->overhead_size;
1578		tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1579		ctx->decrypted = 1;
1580		ctx->saved_data_ready(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1581	} else {
1582		*zc = false;
 
 
 
 
 
 
 
 
 
 
1583	}
 
 
1584
1585	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1586}
1587
1588int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1589		struct scatterlist *sgout)
1590{
1591	bool zc = true;
1592	int chunk;
1593
1594	return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1595}
1596
1597static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1598			       unsigned int len)
1599{
1600	struct tls_context *tls_ctx = tls_get_ctx(sk);
1601	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1602
1603	if (skb) {
1604		struct strp_msg *rxm = strp_msg(skb);
 
 
1605
1606		if (len < rxm->full_len) {
1607			rxm->offset += len;
1608			rxm->full_len -= len;
1609			return false;
 
1610		}
1611		consume_skb(skb);
 
1612	}
1613
1614	/* Finished with message */
1615	ctx->recv_pkt = NULL;
1616	__strp_unpause(&ctx->strp);
1617
1618	return true;
 
 
1619}
1620
1621/* This function traverses the rx_list in tls receive context to copies the
1622 * decrypted records into the buffer provided by caller zero copy is not
1623 * true. Further, the records are removed from the rx_list if it is not a peek
1624 * case and the record has been consumed completely.
1625 */
1626static int process_rx_list(struct tls_sw_context_rx *ctx,
1627			   struct msghdr *msg,
1628			   u8 *control,
1629			   bool *cmsg,
1630			   size_t skip,
1631			   size_t len,
1632			   bool zc,
1633			   bool is_peek)
1634{
1635	struct sk_buff *skb = skb_peek(&ctx->rx_list);
1636	u8 ctrl = *control;
1637	u8 msgc = *cmsg;
1638	struct tls_msg *tlm;
1639	ssize_t copied = 0;
1640
1641	/* Set the record type in 'control' if caller didn't pass it */
1642	if (!ctrl && skb) {
1643		tlm = tls_msg(skb);
1644		ctrl = tlm->control;
1645	}
1646
1647	while (skip && skb) {
1648		struct strp_msg *rxm = strp_msg(skb);
1649		tlm = tls_msg(skb);
1650
1651		/* Cannot process a record of different type */
1652		if (ctrl != tlm->control)
1653			return 0;
1654
1655		if (skip < rxm->full_len)
1656			break;
1657
1658		skip = skip - rxm->full_len;
1659		skb = skb_peek_next(skb, &ctx->rx_list);
1660	}
1661
1662	while (len && skb) {
1663		struct sk_buff *next_skb;
1664		struct strp_msg *rxm = strp_msg(skb);
1665		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1666
1667		tlm = tls_msg(skb);
1668
1669		/* Cannot process a record of different type */
1670		if (ctrl != tlm->control)
1671			return 0;
1672
1673		/* Set record type if not already done. For a non-data record,
1674		 * do not proceed if record type could not be copied.
1675		 */
1676		if (!msgc) {
1677			int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1678					    sizeof(ctrl), &ctrl);
1679			msgc = true;
1680			if (ctrl != TLS_RECORD_TYPE_DATA) {
1681				if (cerr || msg->msg_flags & MSG_CTRUNC)
1682					return -EIO;
1683
1684				*cmsg = msgc;
1685			}
1686		}
1687
1688		if (!zc || (rxm->full_len - skip) > len) {
1689			int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1690						    msg, chunk);
1691			if (err < 0)
1692				return err;
1693		}
1694
1695		len = len - chunk;
1696		copied = copied + chunk;
1697
1698		/* Consume the data from record if it is non-peek case*/
1699		if (!is_peek) {
1700			rxm->offset = rxm->offset + chunk;
1701			rxm->full_len = rxm->full_len - chunk;
1702
1703			/* Return if there is unconsumed data in the record */
1704			if (rxm->full_len - skip)
1705				break;
1706		}
1707
1708		/* The remaining skip-bytes must lie in 1st record in rx_list.
1709		 * So from the 2nd record, 'skip' should be 0.
1710		 */
1711		skip = 0;
1712
1713		if (msg)
1714			msg->msg_flags |= MSG_EOR;
1715
1716		next_skb = skb_peek_next(skb, &ctx->rx_list);
1717
1718		if (!is_peek) {
1719			skb_unlink(skb, &ctx->rx_list);
1720			consume_skb(skb);
1721		}
1722
1723		skb = next_skb;
1724	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1725
1726	*control = ctrl;
1727	return copied;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1728}
1729
1730int tls_sw_recvmsg(struct sock *sk,
1731		   struct msghdr *msg,
1732		   size_t len,
1733		   int nonblock,
1734		   int flags,
1735		   int *addr_len)
1736{
1737	struct tls_context *tls_ctx = tls_get_ctx(sk);
1738	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1739	struct tls_prot_info *prot = &tls_ctx->prot_info;
 
1740	struct sk_psock *psock;
1741	unsigned char control = 0;
1742	ssize_t decrypted = 0;
1743	struct strp_msg *rxm;
1744	struct tls_msg *tlm;
1745	struct sk_buff *skb;
1746	ssize_t copied = 0;
1747	bool cmsg = false;
1748	int target, err = 0;
1749	long timeo;
1750	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1751	bool is_peek = flags & MSG_PEEK;
 
 
1752	bool bpf_strp_enabled;
1753	int num_async = 0;
1754	int pending;
1755
1756	flags |= nonblock;
1757
1758	if (unlikely(flags & MSG_ERRQUEUE))
1759		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1760
1761	psock = sk_psock_get(sk);
1762	lock_sock(sk);
 
 
1763	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1764
 
 
 
 
 
1765	/* Process pending decrypted records. It must be non-zero-copy */
1766	err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1767			      is_peek);
1768	if (err < 0) {
1769		tls_err_abort(sk, err);
1770		goto end;
1771	} else {
1772		copied = err;
1773	}
1774
1775	if (len <= copied)
1776		goto recv_end;
 
1777
1778	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1779	len = len - copied;
1780	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1781
1782	while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1783		bool retain_skb = false;
1784		bool zc = false;
1785		int to_decrypt;
1786		int chunk = 0;
1787		bool async_capable;
1788		bool async = false;
1789
1790		skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1791		if (!skb) {
 
 
 
 
 
 
 
 
1792			if (psock) {
1793				int ret = sk_msg_recvmsg(sk, psock, msg, len,
1794							 flags);
1795
1796				if (ret > 0) {
1797					decrypted += ret;
1798					len -= ret;
1799					continue;
1800				}
1801			}
1802			goto recv_end;
1803		} else {
1804			tlm = tls_msg(skb);
1805			if (prot->version == TLS_1_3_VERSION)
1806				tlm->control = 0;
1807			else
1808				tlm->control = ctx->control;
1809		}
1810
1811		rxm = strp_msg(skb);
 
 
 
1812
1813		to_decrypt = rxm->full_len - prot->overhead_size;
1814
1815		if (to_decrypt <= len && !is_kvec && !is_peek &&
1816		    ctx->control == TLS_RECORD_TYPE_DATA &&
1817		    prot->version != TLS_1_3_VERSION &&
1818		    !bpf_strp_enabled)
1819			zc = true;
1820
1821		/* Do not use async mode if record is non-data */
1822		if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1823			async_capable = ctx->async_capable;
1824		else
1825			async_capable = false;
1826
1827		err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1828					 &chunk, &zc, async_capable);
1829		if (err < 0 && err != -EINPROGRESS) {
1830			tls_err_abort(sk, EBADMSG);
1831			goto recv_end;
1832		}
1833
1834		if (err == -EINPROGRESS) {
1835			async = true;
1836			num_async++;
1837		} else if (prot->version == TLS_1_3_VERSION) {
1838			tlm->control = ctx->control;
1839		}
1840
1841		/* If the type of records being processed is not known yet,
1842		 * set it to record type just dequeued. If it is already known,
1843		 * but does not match the record type just dequeued, go to end.
1844		 * We always get record type here since for tls1.2, record type
1845		 * is known just after record is dequeued from stream parser.
1846		 * For tls1.3, we disable async.
1847		 */
1848
1849		if (!control)
1850			control = tlm->control;
1851		else if (control != tlm->control)
 
 
1852			goto recv_end;
1853
1854		if (!cmsg) {
1855			int cerr;
1856
1857			cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1858					sizeof(control), &control);
1859			cmsg = true;
1860			if (control != TLS_RECORD_TYPE_DATA) {
1861				if (cerr || msg->msg_flags & MSG_CTRUNC) {
1862					err = -EIO;
1863					goto recv_end;
1864				}
1865			}
1866		}
1867
1868		if (async)
1869			goto pick_next_record;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1870
1871		if (!zc) {
1872			if (bpf_strp_enabled) {
 
1873				err = sk_psock_tls_strp_read(psock, skb);
1874				if (err != __SK_PASS) {
1875					rxm->offset = rxm->offset + rxm->full_len;
1876					rxm->full_len = 0;
1877					if (err == __SK_DROP)
1878						consume_skb(skb);
1879					ctx->recv_pkt = NULL;
1880					__strp_unpause(&ctx->strp);
1881					continue;
1882				}
1883			}
1884
1885			if (rxm->full_len > len) {
1886				retain_skb = true;
1887				chunk = len;
1888			} else {
1889				chunk = rxm->full_len;
1890			}
1891
1892			err = skb_copy_datagram_msg(skb, rxm->offset,
1893						    msg, chunk);
1894			if (err < 0)
1895				goto recv_end;
1896
1897			if (!is_peek) {
1898				rxm->offset = rxm->offset + chunk;
1899				rxm->full_len = rxm->full_len - chunk;
 
 
 
 
 
 
1900			}
1901		}
1902
1903pick_next_record:
1904		if (chunk > len)
1905			chunk = len;
1906
1907		decrypted += chunk;
1908		len -= chunk;
1909
1910		/* For async or peek case, queue the current skb */
1911		if (async || is_peek || retain_skb) {
1912			skb_queue_tail(&ctx->rx_list, skb);
1913			skb = NULL;
1914		}
1915
1916		if (tls_sw_advance_skb(sk, skb, chunk)) {
1917			/* Return full control message to
1918			 * userspace before trying to parse
1919			 * another message type
1920			 */
1921			msg->msg_flags |= MSG_EOR;
1922			if (control != TLS_RECORD_TYPE_DATA)
1923				goto recv_end;
1924		} else {
1925			break;
1926		}
1927	}
1928
1929recv_end:
1930	if (num_async) {
 
 
1931		/* Wait for all previously submitted records to be decrypted */
1932		spin_lock_bh(&ctx->decrypt_compl_lock);
1933		ctx->async_notify = true;
1934		pending = atomic_read(&ctx->decrypt_pending);
1935		spin_unlock_bh(&ctx->decrypt_compl_lock);
1936		if (pending) {
1937			err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1938			if (err) {
1939				/* one of async decrypt failed */
1940				tls_err_abort(sk, err);
1941				copied = 0;
1942				decrypted = 0;
1943				goto end;
1944			}
1945		} else {
1946			reinit_completion(&ctx->async_wait.completion);
1947		}
1948
1949		/* There can be no concurrent accesses, since we have no
1950		 * pending decrypt operations
1951		 */
1952		WRITE_ONCE(ctx->async_notify, false);
 
 
1953
1954		/* Drain records from the rx_list & copy if required */
1955		if (is_peek || is_kvec)
1956			err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1957					      decrypted, false, is_peek);
1958		else
1959			err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1960					      decrypted, true, is_peek);
1961		if (err < 0) {
1962			tls_err_abort(sk, err);
1963			copied = 0;
1964			goto end;
1965		}
1966	}
1967
1968	copied += decrypted;
1969
1970end:
1971	release_sock(sk);
1972	if (psock)
1973		sk_psock_put(sk, psock);
1974	return copied ? : err;
1975}
1976
1977ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
1978			   struct pipe_inode_info *pipe,
1979			   size_t len, unsigned int flags)
1980{
1981	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1982	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1983	struct strp_msg *rxm = NULL;
1984	struct sock *sk = sock->sk;
 
1985	struct sk_buff *skb;
1986	ssize_t copied = 0;
1987	int err = 0;
1988	long timeo;
1989	int chunk;
1990	bool zc = false;
1991
1992	lock_sock(sk);
1993
1994	timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
1995
1996	skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo, &err);
1997	if (!skb)
1998		goto splice_read_end;
1999
2000	if (!ctx->decrypted) {
2001		err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
 
 
2002
2003		/* splice does not support reading control messages */
2004		if (ctx->control != TLS_RECORD_TYPE_DATA) {
2005			err = -EINVAL;
2006			goto splice_read_end;
2007		}
2008
 
 
 
2009		if (err < 0) {
2010			tls_err_abort(sk, EBADMSG);
2011			goto splice_read_end;
2012		}
2013		ctx->decrypted = 1;
 
 
2014	}
 
2015	rxm = strp_msg(skb);
 
 
 
 
 
 
 
2016
2017	chunk = min_t(unsigned int, rxm->full_len, len);
2018	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2019	if (copied < 0)
2020		goto splice_read_end;
2021
2022	tls_sw_advance_skb(sk, skb, copied);
 
 
 
 
 
 
2023
2024splice_read_end:
2025	release_sock(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2026	return copied ? : err;
 
 
 
 
2027}
2028
2029bool tls_sw_stream_read(const struct sock *sk)
2030{
2031	struct tls_context *tls_ctx = tls_get_ctx(sk);
2032	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2033	bool ingress_empty = true;
2034	struct sk_psock *psock;
2035
2036	rcu_read_lock();
2037	psock = sk_psock(sk);
2038	if (psock)
2039		ingress_empty = list_empty(&psock->ingress_msg);
2040	rcu_read_unlock();
2041
2042	return !ingress_empty || ctx->recv_pkt ||
2043		!skb_queue_empty(&ctx->rx_list);
2044}
2045
2046static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2047{
2048	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2049	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2050	struct tls_prot_info *prot = &tls_ctx->prot_info;
2051	char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2052	struct strp_msg *rxm = strp_msg(skb);
2053	size_t cipher_overhead;
2054	size_t data_len = 0;
2055	int ret;
2056
2057	/* Verify that we have a full TLS header, or wait for more data */
2058	if (rxm->offset + prot->prepend_size > skb->len)
2059		return 0;
2060
2061	/* Sanity-check size of on-stack buffer. */
2062	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2063		ret = -EINVAL;
2064		goto read_failure;
2065	}
2066
2067	/* Linearize header to local buffer */
2068	ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2069
2070	if (ret < 0)
2071		goto read_failure;
2072
2073	ctx->control = header[0];
2074
2075	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2076
2077	cipher_overhead = prot->tag_size;
2078	if (prot->version != TLS_1_3_VERSION &&
2079	    prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2080		cipher_overhead += prot->iv_size;
2081
2082	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2083	    prot->tail_size) {
2084		ret = -EMSGSIZE;
2085		goto read_failure;
2086	}
2087	if (data_len < cipher_overhead) {
2088		ret = -EBADMSG;
2089		goto read_failure;
2090	}
2091
2092	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2093	if (header[1] != TLS_1_2_VERSION_MINOR ||
2094	    header[2] != TLS_1_2_VERSION_MAJOR) {
2095		ret = -EINVAL;
2096		goto read_failure;
2097	}
2098
2099	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2100				     TCP_SKB_CB(skb)->seq + rxm->offset);
2101	return data_len + TLS_HEADER_SIZE;
2102
2103read_failure:
2104	tls_err_abort(strp->sk, ret);
2105
2106	return ret;
2107}
2108
2109static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2110{
2111	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2112	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2113
2114	ctx->decrypted = 0;
2115
2116	ctx->recv_pkt = skb;
2117	strp_pause(strp);
2118
 
2119	ctx->saved_data_ready(strp->sk);
2120}
2121
2122static void tls_data_ready(struct sock *sk)
2123{
2124	struct tls_context *tls_ctx = tls_get_ctx(sk);
2125	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2126	struct sk_psock *psock;
 
 
 
2127
2128	strp_data_ready(&ctx->strp);
 
 
 
2129
2130	psock = sk_psock_get(sk);
2131	if (psock) {
2132		if (!list_empty(&psock->ingress_msg))
2133			ctx->saved_data_ready(sk);
2134		sk_psock_put(sk, psock);
2135	}
2136}
2137
2138void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2139{
2140	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2141
2142	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2143	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2144	cancel_delayed_work_sync(&ctx->tx_work.work);
2145}
2146
2147void tls_sw_release_resources_tx(struct sock *sk)
2148{
2149	struct tls_context *tls_ctx = tls_get_ctx(sk);
2150	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2151	struct tls_rec *rec, *tmp;
2152	int pending;
2153
2154	/* Wait for any pending async encryptions to complete */
2155	spin_lock_bh(&ctx->encrypt_compl_lock);
2156	ctx->async_notify = true;
2157	pending = atomic_read(&ctx->encrypt_pending);
2158	spin_unlock_bh(&ctx->encrypt_compl_lock);
2159
2160	if (pending)
2161		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2162
2163	tls_tx_records(sk, -1);
2164
2165	/* Free up un-sent records in tx_list. First, free
2166	 * the partially sent record if any at head of tx_list.
2167	 */
2168	if (tls_ctx->partially_sent_record) {
2169		tls_free_partial_record(sk, tls_ctx);
2170		rec = list_first_entry(&ctx->tx_list,
2171				       struct tls_rec, list);
2172		list_del(&rec->list);
2173		sk_msg_free(sk, &rec->msg_plaintext);
2174		kfree(rec);
2175	}
2176
2177	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2178		list_del(&rec->list);
2179		sk_msg_free(sk, &rec->msg_encrypted);
2180		sk_msg_free(sk, &rec->msg_plaintext);
2181		kfree(rec);
2182	}
2183
2184	crypto_free_aead(ctx->aead_send);
2185	tls_free_open_rec(sk);
2186}
2187
2188void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2189{
2190	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2191
2192	kfree(ctx);
2193}
2194
2195void tls_sw_release_resources_rx(struct sock *sk)
2196{
2197	struct tls_context *tls_ctx = tls_get_ctx(sk);
2198	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2199
2200	kfree(tls_ctx->rx.rec_seq);
2201	kfree(tls_ctx->rx.iv);
2202
2203	if (ctx->aead_recv) {
2204		kfree_skb(ctx->recv_pkt);
2205		ctx->recv_pkt = NULL;
2206		skb_queue_purge(&ctx->rx_list);
2207		crypto_free_aead(ctx->aead_recv);
2208		strp_stop(&ctx->strp);
2209		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2210		 * we still want to strp_stop(), but sk->sk_data_ready was
2211		 * never swapped.
2212		 */
2213		if (ctx->saved_data_ready) {
2214			write_lock_bh(&sk->sk_callback_lock);
2215			sk->sk_data_ready = ctx->saved_data_ready;
2216			write_unlock_bh(&sk->sk_callback_lock);
2217		}
2218	}
2219}
2220
2221void tls_sw_strparser_done(struct tls_context *tls_ctx)
2222{
2223	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2224
2225	strp_done(&ctx->strp);
2226}
2227
2228void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2229{
2230	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2231
2232	kfree(ctx);
2233}
2234
2235void tls_sw_free_resources_rx(struct sock *sk)
2236{
2237	struct tls_context *tls_ctx = tls_get_ctx(sk);
2238
2239	tls_sw_release_resources_rx(sk);
2240	tls_sw_free_ctx_rx(tls_ctx);
2241}
2242
2243/* The work handler to transmitt the encrypted records in tx_list */
2244static void tx_work_handler(struct work_struct *work)
2245{
2246	struct delayed_work *delayed_work = to_delayed_work(work);
2247	struct tx_work *tx_work = container_of(delayed_work,
2248					       struct tx_work, work);
2249	struct sock *sk = tx_work->sk;
2250	struct tls_context *tls_ctx = tls_get_ctx(sk);
2251	struct tls_sw_context_tx *ctx;
2252
2253	if (unlikely(!tls_ctx))
2254		return;
2255
2256	ctx = tls_sw_ctx_tx(tls_ctx);
2257	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2258		return;
2259
2260	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2261		return;
2262	mutex_lock(&tls_ctx->tx_lock);
2263	lock_sock(sk);
2264	tls_tx_records(sk, -1);
2265	release_sock(sk);
2266	mutex_unlock(&tls_ctx->tx_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2267}
2268
2269void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2270{
2271	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2272
2273	/* Schedule the transmission if tx list is ready */
2274	if (is_tx_ready(tx_ctx) &&
2275	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2276		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2277}
2278
2279void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2280{
2281	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2282
2283	write_lock_bh(&sk->sk_callback_lock);
2284	rx_ctx->saved_data_ready = sk->sk_data_ready;
2285	sk->sk_data_ready = tls_data_ready;
2286	write_unlock_bh(&sk->sk_callback_lock);
 
 
 
 
 
2287
2288	strp_check_rcv(&rx_ctx->strp);
 
2289}
2290
2291int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2292{
2293	struct tls_context *tls_ctx = tls_get_ctx(sk);
2294	struct tls_prot_info *prot = &tls_ctx->prot_info;
2295	struct tls_crypto_info *crypto_info;
2296	struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2297	struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2298	struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2299	struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2300	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2301	struct tls_sw_context_rx *sw_ctx_rx = NULL;
 
 
 
2302	struct cipher_context *cctx;
 
2303	struct crypto_aead **aead;
2304	struct strp_callbacks cb;
2305	u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2306	struct crypto_tfm *tfm;
2307	char *iv, *rec_seq, *key, *salt, *cipher_name;
2308	size_t keysize;
2309	int rc = 0;
2310
2311	if (!ctx) {
2312		rc = -EINVAL;
2313		goto out;
2314	}
2315
2316	if (tx) {
2317		if (!ctx->priv_ctx_tx) {
2318			sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2319			if (!sw_ctx_tx) {
2320				rc = -ENOMEM;
2321				goto out;
2322			}
2323			ctx->priv_ctx_tx = sw_ctx_tx;
2324		} else {
2325			sw_ctx_tx =
2326				(struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2327		}
2328	} else {
2329		if (!ctx->priv_ctx_rx) {
2330			sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2331			if (!sw_ctx_rx) {
2332				rc = -ENOMEM;
2333				goto out;
2334			}
2335			ctx->priv_ctx_rx = sw_ctx_rx;
2336		} else {
2337			sw_ctx_rx =
2338				(struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2339		}
2340	}
2341
2342	if (tx) {
2343		crypto_init_wait(&sw_ctx_tx->async_wait);
2344		spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2345		crypto_info = &ctx->crypto_send.info;
2346		cctx = &ctx->tx;
2347		aead = &sw_ctx_tx->aead_send;
2348		INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2349		INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2350		sw_ctx_tx->tx_work.sk = sk;
2351	} else {
2352		crypto_init_wait(&sw_ctx_rx->async_wait);
2353		spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
 
 
 
2354		crypto_info = &ctx->crypto_recv.info;
2355		cctx = &ctx->rx;
2356		skb_queue_head_init(&sw_ctx_rx->rx_list);
2357		aead = &sw_ctx_rx->aead_recv;
2358	}
2359
2360	switch (crypto_info->cipher_type) {
2361	case TLS_CIPHER_AES_GCM_128: {
2362		nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2363		tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2364		iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2365		iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2366		rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2367		rec_seq =
2368		 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2369		gcm_128_info =
2370			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2371		keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2372		key = gcm_128_info->key;
2373		salt = gcm_128_info->salt;
2374		salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2375		cipher_name = "gcm(aes)";
2376		break;
2377	}
2378	case TLS_CIPHER_AES_GCM_256: {
2379		nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2380		tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2381		iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2382		iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2383		rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2384		rec_seq =
2385		 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2386		gcm_256_info =
2387			(struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2388		keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2389		key = gcm_256_info->key;
2390		salt = gcm_256_info->salt;
2391		salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2392		cipher_name = "gcm(aes)";
2393		break;
2394	}
2395	case TLS_CIPHER_AES_CCM_128: {
2396		nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2397		tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2398		iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2399		iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2400		rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2401		rec_seq =
2402		((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2403		ccm_128_info =
2404		(struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2405		keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2406		key = ccm_128_info->key;
2407		salt = ccm_128_info->salt;
2408		salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2409		cipher_name = "ccm(aes)";
2410		break;
2411	}
2412	case TLS_CIPHER_CHACHA20_POLY1305: {
2413		chacha20_poly1305_info = (void *)crypto_info;
2414		nonce_size = 0;
2415		tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2416		iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2417		iv = chacha20_poly1305_info->iv;
2418		rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2419		rec_seq = chacha20_poly1305_info->rec_seq;
2420		keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2421		key = chacha20_poly1305_info->key;
2422		salt = chacha20_poly1305_info->salt;
2423		salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2424		cipher_name = "rfc7539(chacha20,poly1305)";
2425		break;
2426	}
2427	default:
2428		rc = -EINVAL;
2429		goto free_priv;
2430	}
2431
2432	/* Sanity-check the sizes for stack allocations. */
2433	if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2434	    rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2435		rc = -EINVAL;
2436		goto free_priv;
2437	}
2438
2439	if (crypto_info->version == TLS_1_3_VERSION) {
2440		nonce_size = 0;
2441		prot->aad_size = TLS_HEADER_SIZE;
2442		prot->tail_size = 1;
2443	} else {
2444		prot->aad_size = TLS_AAD_SPACE_SIZE;
2445		prot->tail_size = 0;
2446	}
2447
2448	prot->version = crypto_info->version;
2449	prot->cipher_type = crypto_info->cipher_type;
2450	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2451	prot->tag_size = tag_size;
2452	prot->overhead_size = prot->prepend_size +
2453			      prot->tag_size + prot->tail_size;
2454	prot->iv_size = iv_size;
2455	prot->salt_size = salt_size;
2456	cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2457	if (!cctx->iv) {
2458		rc = -ENOMEM;
2459		goto free_priv;
2460	}
2461	/* Note: 128 & 256 bit salt are the same size */
2462	prot->rec_seq_size = rec_seq_size;
2463	memcpy(cctx->iv, salt, salt_size);
2464	memcpy(cctx->iv + salt_size, iv, iv_size);
2465	cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2466	if (!cctx->rec_seq) {
2467		rc = -ENOMEM;
2468		goto free_iv;
2469	}
2470
2471	if (!*aead) {
2472		*aead = crypto_alloc_aead(cipher_name, 0, 0);
2473		if (IS_ERR(*aead)) {
2474			rc = PTR_ERR(*aead);
2475			*aead = NULL;
2476			goto free_rec_seq;
2477		}
2478	}
2479
2480	ctx->push_pending_record = tls_sw_push_pending_record;
2481
2482	rc = crypto_aead_setkey(*aead, key, keysize);
2483
2484	if (rc)
2485		goto free_aead;
2486
2487	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2488	if (rc)
2489		goto free_aead;
2490
2491	if (sw_ctx_rx) {
2492		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2493
2494		if (crypto_info->version == TLS_1_3_VERSION)
2495			sw_ctx_rx->async_capable = 0;
2496		else
2497			sw_ctx_rx->async_capable =
2498				!!(tfm->__crt_alg->cra_flags &
2499				   CRYPTO_ALG_ASYNC);
2500
2501		/* Set up strparser */
2502		memset(&cb, 0, sizeof(cb));
2503		cb.rcv_msg = tls_queue;
2504		cb.parse_msg = tls_read_size;
2505
2506		strp_init(&sw_ctx_rx->strp, sk, &cb);
 
 
2507	}
2508
2509	goto out;
2510
2511free_aead:
2512	crypto_free_aead(*aead);
2513	*aead = NULL;
2514free_rec_seq:
2515	kfree(cctx->rec_seq);
2516	cctx->rec_seq = NULL;
2517free_iv:
2518	kfree(cctx->iv);
2519	cctx->iv = NULL;
2520free_priv:
2521	if (tx) {
2522		kfree(ctx->priv_ctx_tx);
2523		ctx->priv_ctx_tx = NULL;
2524	} else {
2525		kfree(ctx->priv_ctx_rx);
2526		ctx->priv_ctx_rx = NULL;
2527	}
2528out:
2529	return rc;
2530}
v6.8
   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/bug.h>
  39#include <linux/sched/signal.h>
  40#include <linux/module.h>
  41#include <linux/kernel.h>
  42#include <linux/splice.h>
  43#include <crypto/aead.h>
  44
  45#include <net/strparser.h>
  46#include <net/tls.h>
  47#include <trace/events/sock.h>
  48
  49#include "tls.h"
  50
  51struct tls_decrypt_arg {
  52	struct_group(inargs,
  53	bool zc;
  54	bool async;
  55	bool async_done;
  56	u8 tail;
  57	);
  58
  59	struct sk_buff *skb;
  60};
  61
  62struct tls_decrypt_ctx {
  63	struct sock *sk;
  64	u8 iv[TLS_MAX_IV_SIZE];
  65	u8 aad[TLS_MAX_AAD_SIZE];
  66	u8 tail;
  67	bool free_sgout;
  68	struct scatterlist sg[];
  69};
  70
  71noinline void tls_err_abort(struct sock *sk, int err)
  72{
  73	WARN_ON_ONCE(err >= 0);
  74	/* sk->sk_err should contain a positive error code. */
  75	WRITE_ONCE(sk->sk_err, -err);
  76	/* Paired with smp_rmb() in tcp_poll() */
  77	smp_wmb();
  78	sk_error_report(sk);
  79}
  80
  81static int __skb_nsg(struct sk_buff *skb, int offset, int len,
  82                     unsigned int recursion_level)
  83{
  84        int start = skb_headlen(skb);
  85        int i, chunk = start - offset;
  86        struct sk_buff *frag_iter;
  87        int elt = 0;
  88
  89        if (unlikely(recursion_level >= 24))
  90                return -EMSGSIZE;
  91
  92        if (chunk > 0) {
  93                if (chunk > len)
  94                        chunk = len;
  95                elt++;
  96                len -= chunk;
  97                if (len == 0)
  98                        return elt;
  99                offset += chunk;
 100        }
 101
 102        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 103                int end;
 104
 105                WARN_ON(start > offset + len);
 106
 107                end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
 108                chunk = end - offset;
 109                if (chunk > 0) {
 110                        if (chunk > len)
 111                                chunk = len;
 112                        elt++;
 113                        len -= chunk;
 114                        if (len == 0)
 115                                return elt;
 116                        offset += chunk;
 117                }
 118                start = end;
 119        }
 120
 121        if (unlikely(skb_has_frag_list(skb))) {
 122                skb_walk_frags(skb, frag_iter) {
 123                        int end, ret;
 124
 125                        WARN_ON(start > offset + len);
 126
 127                        end = start + frag_iter->len;
 128                        chunk = end - offset;
 129                        if (chunk > 0) {
 130                                if (chunk > len)
 131                                        chunk = len;
 132                                ret = __skb_nsg(frag_iter, offset - start, chunk,
 133                                                recursion_level + 1);
 134                                if (unlikely(ret < 0))
 135                                        return ret;
 136                                elt += ret;
 137                                len -= chunk;
 138                                if (len == 0)
 139                                        return elt;
 140                                offset += chunk;
 141                        }
 142                        start = end;
 143                }
 144        }
 145        BUG_ON(len);
 146        return elt;
 147}
 148
 149/* Return the number of scatterlist elements required to completely map the
 150 * skb, or -EMSGSIZE if the recursion depth is exceeded.
 151 */
 152static int skb_nsg(struct sk_buff *skb, int offset, int len)
 153{
 154        return __skb_nsg(skb, offset, len, 0);
 155}
 156
 157static int tls_padding_length(struct tls_prot_info *prot, struct sk_buff *skb,
 158			      struct tls_decrypt_arg *darg)
 159{
 160	struct strp_msg *rxm = strp_msg(skb);
 161	struct tls_msg *tlm = tls_msg(skb);
 162	int sub = 0;
 163
 164	/* Determine zero-padding length */
 165	if (prot->version == TLS_1_3_VERSION) {
 166		int offset = rxm->full_len - TLS_TAG_SIZE - 1;
 167		char content_type = darg->zc ? darg->tail : 0;
 168		int err;
 
 169
 170		while (content_type == 0) {
 171			if (offset < prot->prepend_size)
 172				return -EBADMSG;
 173			err = skb_copy_bits(skb, rxm->offset + offset,
 
 174					    &content_type, 1);
 175			if (err)
 176				return err;
 177			if (content_type)
 178				break;
 179			sub++;
 180			offset--;
 181		}
 182		tlm->control = content_type;
 183	}
 184	return sub;
 185}
 186
 187static void tls_decrypt_done(void *data, int err)
 188{
 189	struct aead_request *aead_req = data;
 190	struct crypto_aead *aead = crypto_aead_reqtfm(aead_req);
 191	struct scatterlist *sgout = aead_req->dst;
 
 192	struct tls_sw_context_rx *ctx;
 193	struct tls_decrypt_ctx *dctx;
 194	struct tls_context *tls_ctx;
 
 195	struct scatterlist *sg;
 
 196	unsigned int pages;
 197	struct sock *sk;
 198	int aead_size;
 199
 200	/* If requests get too backlogged crypto API returns -EBUSY and calls
 201	 * ->complete(-EINPROGRESS) immediately followed by ->complete(0)
 202	 * to make waiting for backlog to flush with crypto_wait_req() easier.
 203	 * First wait converts -EBUSY -> -EINPROGRESS, and the second one
 204	 * -EINPROGRESS -> 0.
 205	 * We have a single struct crypto_async_request per direction, this
 206	 * scheme doesn't help us, so just ignore the first ->complete().
 207	 */
 208	if (err == -EINPROGRESS)
 209		return;
 210
 211	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(aead);
 212	aead_size = ALIGN(aead_size, __alignof__(*dctx));
 213	dctx = (void *)((u8 *)aead_req + aead_size);
 214
 215	sk = dctx->sk;
 216	tls_ctx = tls_get_ctx(sk);
 217	ctx = tls_sw_ctx_rx(tls_ctx);
 
 218
 219	/* Propagate if there was an err */
 220	if (err) {
 221		if (err == -EBADMSG)
 222			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
 
 223		ctx->async_wait.err = err;
 224		tls_err_abort(sk, err);
 
 
 
 
 
 
 
 
 
 
 
 
 
 225	}
 226
 
 
 
 
 
 
 227	/* Free the destination pages if skb was not decrypted inplace */
 228	if (dctx->free_sgout) {
 229		/* Skip the first S/G entry as it points to AAD */
 230		for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
 231			if (!sg)
 232				break;
 233			put_page(sg_page(sg));
 234		}
 235	}
 236
 237	kfree(aead_req);
 238
 239	if (atomic_dec_and_test(&ctx->decrypt_pending))
 
 
 
 240		complete(&ctx->async_wait.completion);
 241}
 242
 243static int tls_decrypt_async_wait(struct tls_sw_context_rx *ctx)
 244{
 245	if (!atomic_dec_and_test(&ctx->decrypt_pending))
 246		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
 247	atomic_inc(&ctx->decrypt_pending);
 248
 249	return ctx->async_wait.err;
 250}
 251
 252static int tls_do_decryption(struct sock *sk,
 
 253			     struct scatterlist *sgin,
 254			     struct scatterlist *sgout,
 255			     char *iv_recv,
 256			     size_t data_len,
 257			     struct aead_request *aead_req,
 258			     struct tls_decrypt_arg *darg)
 259{
 260	struct tls_context *tls_ctx = tls_get_ctx(sk);
 261	struct tls_prot_info *prot = &tls_ctx->prot_info;
 262	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 263	int ret;
 264
 265	aead_request_set_tfm(aead_req, ctx->aead_recv);
 266	aead_request_set_ad(aead_req, prot->aad_size);
 267	aead_request_set_crypt(aead_req, sgin, sgout,
 268			       data_len + prot->tag_size,
 269			       (u8 *)iv_recv);
 270
 271	if (darg->async) {
 
 
 
 
 
 
 
 272		aead_request_set_callback(aead_req,
 273					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 274					  tls_decrypt_done, aead_req);
 275		DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->decrypt_pending) < 1);
 276		atomic_inc(&ctx->decrypt_pending);
 277	} else {
 278		DECLARE_CRYPTO_WAIT(wait);
 279
 280		aead_request_set_callback(aead_req,
 281					  CRYPTO_TFM_REQ_MAY_BACKLOG,
 282					  crypto_req_done, &wait);
 283		ret = crypto_aead_decrypt(aead_req);
 284		if (ret == -EINPROGRESS || ret == -EBUSY)
 285			ret = crypto_wait_req(ret, &wait);
 286		return ret;
 287	}
 288
 289	ret = crypto_aead_decrypt(aead_req);
 290	if (ret == -EINPROGRESS)
 291		return 0;
 
 292
 293	if (ret == -EBUSY) {
 294		ret = tls_decrypt_async_wait(ctx);
 295		darg->async_done = true;
 296		/* all completions have run, we're not doing async anymore */
 297		darg->async = false;
 298		return ret;
 299	}
 300
 301	atomic_dec(&ctx->decrypt_pending);
 302	darg->async = false;
 303
 304	return ret;
 305}
 306
 307static void tls_trim_both_msgs(struct sock *sk, int target_size)
 308{
 309	struct tls_context *tls_ctx = tls_get_ctx(sk);
 310	struct tls_prot_info *prot = &tls_ctx->prot_info;
 311	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 312	struct tls_rec *rec = ctx->open_rec;
 313
 314	sk_msg_trim(sk, &rec->msg_plaintext, target_size);
 315	if (target_size > 0)
 316		target_size += prot->overhead_size;
 317	sk_msg_trim(sk, &rec->msg_encrypted, target_size);
 318}
 319
 320static int tls_alloc_encrypted_msg(struct sock *sk, int len)
 321{
 322	struct tls_context *tls_ctx = tls_get_ctx(sk);
 323	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 324	struct tls_rec *rec = ctx->open_rec;
 325	struct sk_msg *msg_en = &rec->msg_encrypted;
 326
 327	return sk_msg_alloc(sk, msg_en, len, 0);
 328}
 329
 330static int tls_clone_plaintext_msg(struct sock *sk, int required)
 331{
 332	struct tls_context *tls_ctx = tls_get_ctx(sk);
 333	struct tls_prot_info *prot = &tls_ctx->prot_info;
 334	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 335	struct tls_rec *rec = ctx->open_rec;
 336	struct sk_msg *msg_pl = &rec->msg_plaintext;
 337	struct sk_msg *msg_en = &rec->msg_encrypted;
 338	int skip, len;
 339
 340	/* We add page references worth len bytes from encrypted sg
 341	 * at the end of plaintext sg. It is guaranteed that msg_en
 342	 * has enough required room (ensured by caller).
 343	 */
 344	len = required - msg_pl->sg.size;
 345
 346	/* Skip initial bytes in msg_en's data to be able to use
 347	 * same offset of both plain and encrypted data.
 348	 */
 349	skip = prot->prepend_size + msg_pl->sg.size;
 350
 351	return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
 352}
 353
 354static struct tls_rec *tls_get_rec(struct sock *sk)
 355{
 356	struct tls_context *tls_ctx = tls_get_ctx(sk);
 357	struct tls_prot_info *prot = &tls_ctx->prot_info;
 358	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 359	struct sk_msg *msg_pl, *msg_en;
 360	struct tls_rec *rec;
 361	int mem_size;
 362
 363	mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
 364
 365	rec = kzalloc(mem_size, sk->sk_allocation);
 366	if (!rec)
 367		return NULL;
 368
 369	msg_pl = &rec->msg_plaintext;
 370	msg_en = &rec->msg_encrypted;
 371
 372	sk_msg_init(msg_pl);
 373	sk_msg_init(msg_en);
 374
 375	sg_init_table(rec->sg_aead_in, 2);
 376	sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
 377	sg_unmark_end(&rec->sg_aead_in[1]);
 378
 379	sg_init_table(rec->sg_aead_out, 2);
 380	sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
 381	sg_unmark_end(&rec->sg_aead_out[1]);
 382
 383	rec->sk = sk;
 384
 385	return rec;
 386}
 387
 388static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
 389{
 390	sk_msg_free(sk, &rec->msg_encrypted);
 391	sk_msg_free(sk, &rec->msg_plaintext);
 392	kfree(rec);
 393}
 394
 395static void tls_free_open_rec(struct sock *sk)
 396{
 397	struct tls_context *tls_ctx = tls_get_ctx(sk);
 398	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 399	struct tls_rec *rec = ctx->open_rec;
 400
 401	if (rec) {
 402		tls_free_rec(sk, rec);
 403		ctx->open_rec = NULL;
 404	}
 405}
 406
 407int tls_tx_records(struct sock *sk, int flags)
 408{
 409	struct tls_context *tls_ctx = tls_get_ctx(sk);
 410	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 411	struct tls_rec *rec, *tmp;
 412	struct sk_msg *msg_en;
 413	int tx_flags, rc = 0;
 414
 415	if (tls_is_partially_sent_record(tls_ctx)) {
 416		rec = list_first_entry(&ctx->tx_list,
 417				       struct tls_rec, list);
 418
 419		if (flags == -1)
 420			tx_flags = rec->tx_flags;
 421		else
 422			tx_flags = flags;
 423
 424		rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
 425		if (rc)
 426			goto tx_err;
 427
 428		/* Full record has been transmitted.
 429		 * Remove the head of tx_list
 430		 */
 431		list_del(&rec->list);
 432		sk_msg_free(sk, &rec->msg_plaintext);
 433		kfree(rec);
 434	}
 435
 436	/* Tx all ready records */
 437	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
 438		if (READ_ONCE(rec->tx_ready)) {
 439			if (flags == -1)
 440				tx_flags = rec->tx_flags;
 441			else
 442				tx_flags = flags;
 443
 444			msg_en = &rec->msg_encrypted;
 445			rc = tls_push_sg(sk, tls_ctx,
 446					 &msg_en->sg.data[msg_en->sg.curr],
 447					 0, tx_flags);
 448			if (rc)
 449				goto tx_err;
 450
 451			list_del(&rec->list);
 452			sk_msg_free(sk, &rec->msg_plaintext);
 453			kfree(rec);
 454		} else {
 455			break;
 456		}
 457	}
 458
 459tx_err:
 460	if (rc < 0 && rc != -EAGAIN)
 461		tls_err_abort(sk, -EBADMSG);
 462
 463	return rc;
 464}
 465
 466static void tls_encrypt_done(void *data, int err)
 467{
 468	struct tls_sw_context_tx *ctx;
 469	struct tls_context *tls_ctx;
 470	struct tls_prot_info *prot;
 471	struct tls_rec *rec = data;
 
 472	struct scatterlist *sge;
 473	struct sk_msg *msg_en;
 474	struct sock *sk;
 475
 476	if (err == -EINPROGRESS) /* see the comment in tls_decrypt_done() */
 477		return;
 478
 
 479	msg_en = &rec->msg_encrypted;
 480
 481	sk = rec->sk;
 482	tls_ctx = tls_get_ctx(sk);
 483	prot = &tls_ctx->prot_info;
 484	ctx = tls_sw_ctx_tx(tls_ctx);
 485
 486	sge = sk_msg_elem(msg_en, msg_en->sg.curr);
 487	sge->offset -= prot->prepend_size;
 488	sge->length += prot->prepend_size;
 489
 490	/* Check if error is previously set on socket */
 491	if (err || sk->sk_err) {
 492		rec = NULL;
 493
 494		/* If err is already set on socket, return the same code */
 495		if (sk->sk_err) {
 496			ctx->async_wait.err = -sk->sk_err;
 497		} else {
 498			ctx->async_wait.err = err;
 499			tls_err_abort(sk, err);
 500		}
 501	}
 502
 503	if (rec) {
 504		struct tls_rec *first_rec;
 505
 506		/* Mark the record as ready for transmission */
 507		smp_store_mb(rec->tx_ready, true);
 508
 509		/* If received record is at head of tx_list, schedule tx */
 510		first_rec = list_first_entry(&ctx->tx_list,
 511					     struct tls_rec, list);
 512		if (rec == first_rec) {
 513			/* Schedule the transmission */
 514			if (!test_and_set_bit(BIT_TX_SCHEDULED,
 515					      &ctx->tx_bitmask))
 516				schedule_delayed_work(&ctx->tx_work.work, 1);
 517		}
 518	}
 519
 520	if (atomic_dec_and_test(&ctx->encrypt_pending))
 
 
 
 521		complete(&ctx->async_wait.completion);
 522}
 523
 524static int tls_encrypt_async_wait(struct tls_sw_context_tx *ctx)
 525{
 526	if (!atomic_dec_and_test(&ctx->encrypt_pending))
 527		crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
 528	atomic_inc(&ctx->encrypt_pending);
 529
 530	return ctx->async_wait.err;
 
 
 531}
 532
 533static int tls_do_encryption(struct sock *sk,
 534			     struct tls_context *tls_ctx,
 535			     struct tls_sw_context_tx *ctx,
 536			     struct aead_request *aead_req,
 537			     size_t data_len, u32 start)
 538{
 539	struct tls_prot_info *prot = &tls_ctx->prot_info;
 540	struct tls_rec *rec = ctx->open_rec;
 541	struct sk_msg *msg_en = &rec->msg_encrypted;
 542	struct scatterlist *sge = sk_msg_elem(msg_en, start);
 543	int rc, iv_offset = 0;
 544
 545	/* For CCM based ciphers, first byte of IV is a constant */
 546	switch (prot->cipher_type) {
 547	case TLS_CIPHER_AES_CCM_128:
 548		rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
 549		iv_offset = 1;
 550		break;
 551	case TLS_CIPHER_SM4_CCM:
 552		rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
 553		iv_offset = 1;
 554		break;
 555	}
 556
 557	memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
 558	       prot->iv_size + prot->salt_size);
 559
 560	tls_xor_iv_with_seq(prot, rec->iv_data + iv_offset,
 561			    tls_ctx->tx.rec_seq);
 562
 563	sge->offset += prot->prepend_size;
 564	sge->length -= prot->prepend_size;
 565
 566	msg_en->sg.curr = start;
 567
 568	aead_request_set_tfm(aead_req, ctx->aead_send);
 569	aead_request_set_ad(aead_req, prot->aad_size);
 570	aead_request_set_crypt(aead_req, rec->sg_aead_in,
 571			       rec->sg_aead_out,
 572			       data_len, rec->iv_data);
 573
 574	aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 575				  tls_encrypt_done, rec);
 576
 577	/* Add the record in tx_list */
 578	list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
 579	DEBUG_NET_WARN_ON_ONCE(atomic_read(&ctx->encrypt_pending) < 1);
 580	atomic_inc(&ctx->encrypt_pending);
 581
 582	rc = crypto_aead_encrypt(aead_req);
 583	if (rc == -EBUSY) {
 584		rc = tls_encrypt_async_wait(ctx);
 585		rc = rc ?: -EINPROGRESS;
 586	}
 587	if (!rc || rc != -EINPROGRESS) {
 588		atomic_dec(&ctx->encrypt_pending);
 589		sge->offset -= prot->prepend_size;
 590		sge->length += prot->prepend_size;
 591	}
 592
 593	if (!rc) {
 594		WRITE_ONCE(rec->tx_ready, true);
 595	} else if (rc != -EINPROGRESS) {
 596		list_del(&rec->list);
 597		return rc;
 598	}
 599
 600	/* Unhook the record from context if encryption is not failure */
 601	ctx->open_rec = NULL;
 602	tls_advance_record_sn(sk, prot, &tls_ctx->tx);
 603	return rc;
 604}
 605
 606static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
 607				 struct tls_rec **to, struct sk_msg *msg_opl,
 608				 struct sk_msg *msg_oen, u32 split_point,
 609				 u32 tx_overhead_size, u32 *orig_end)
 610{
 611	u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
 612	struct scatterlist *sge, *osge, *nsge;
 613	u32 orig_size = msg_opl->sg.size;
 614	struct scatterlist tmp = { };
 615	struct sk_msg *msg_npl;
 616	struct tls_rec *new;
 617	int ret;
 618
 619	new = tls_get_rec(sk);
 620	if (!new)
 621		return -ENOMEM;
 622	ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
 623			   tx_overhead_size, 0);
 624	if (ret < 0) {
 625		tls_free_rec(sk, new);
 626		return ret;
 627	}
 628
 629	*orig_end = msg_opl->sg.end;
 630	i = msg_opl->sg.start;
 631	sge = sk_msg_elem(msg_opl, i);
 632	while (apply && sge->length) {
 633		if (sge->length > apply) {
 634			u32 len = sge->length - apply;
 635
 636			get_page(sg_page(sge));
 637			sg_set_page(&tmp, sg_page(sge), len,
 638				    sge->offset + apply);
 639			sge->length = apply;
 640			bytes += apply;
 641			apply = 0;
 642		} else {
 643			apply -= sge->length;
 644			bytes += sge->length;
 645		}
 646
 647		sk_msg_iter_var_next(i);
 648		if (i == msg_opl->sg.end)
 649			break;
 650		sge = sk_msg_elem(msg_opl, i);
 651	}
 652
 653	msg_opl->sg.end = i;
 654	msg_opl->sg.curr = i;
 655	msg_opl->sg.copybreak = 0;
 656	msg_opl->apply_bytes = 0;
 657	msg_opl->sg.size = bytes;
 658
 659	msg_npl = &new->msg_plaintext;
 660	msg_npl->apply_bytes = apply;
 661	msg_npl->sg.size = orig_size - bytes;
 662
 663	j = msg_npl->sg.start;
 664	nsge = sk_msg_elem(msg_npl, j);
 665	if (tmp.length) {
 666		memcpy(nsge, &tmp, sizeof(*nsge));
 667		sk_msg_iter_var_next(j);
 668		nsge = sk_msg_elem(msg_npl, j);
 669	}
 670
 671	osge = sk_msg_elem(msg_opl, i);
 672	while (osge->length) {
 673		memcpy(nsge, osge, sizeof(*nsge));
 674		sg_unmark_end(nsge);
 675		sk_msg_iter_var_next(i);
 676		sk_msg_iter_var_next(j);
 677		if (i == *orig_end)
 678			break;
 679		osge = sk_msg_elem(msg_opl, i);
 680		nsge = sk_msg_elem(msg_npl, j);
 681	}
 682
 683	msg_npl->sg.end = j;
 684	msg_npl->sg.curr = j;
 685	msg_npl->sg.copybreak = 0;
 686
 687	*to = new;
 688	return 0;
 689}
 690
 691static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
 692				  struct tls_rec *from, u32 orig_end)
 693{
 694	struct sk_msg *msg_npl = &from->msg_plaintext;
 695	struct sk_msg *msg_opl = &to->msg_plaintext;
 696	struct scatterlist *osge, *nsge;
 697	u32 i, j;
 698
 699	i = msg_opl->sg.end;
 700	sk_msg_iter_var_prev(i);
 701	j = msg_npl->sg.start;
 702
 703	osge = sk_msg_elem(msg_opl, i);
 704	nsge = sk_msg_elem(msg_npl, j);
 705
 706	if (sg_page(osge) == sg_page(nsge) &&
 707	    osge->offset + osge->length == nsge->offset) {
 708		osge->length += nsge->length;
 709		put_page(sg_page(nsge));
 710	}
 711
 712	msg_opl->sg.end = orig_end;
 713	msg_opl->sg.curr = orig_end;
 714	msg_opl->sg.copybreak = 0;
 715	msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
 716	msg_opl->sg.size += msg_npl->sg.size;
 717
 718	sk_msg_free(sk, &to->msg_encrypted);
 719	sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
 720
 721	kfree(from);
 722}
 723
 724static int tls_push_record(struct sock *sk, int flags,
 725			   unsigned char record_type)
 726{
 727	struct tls_context *tls_ctx = tls_get_ctx(sk);
 728	struct tls_prot_info *prot = &tls_ctx->prot_info;
 729	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 730	struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
 731	u32 i, split_point, orig_end;
 732	struct sk_msg *msg_pl, *msg_en;
 733	struct aead_request *req;
 734	bool split;
 735	int rc;
 736
 737	if (!rec)
 738		return 0;
 739
 740	msg_pl = &rec->msg_plaintext;
 741	msg_en = &rec->msg_encrypted;
 742
 743	split_point = msg_pl->apply_bytes;
 744	split = split_point && split_point < msg_pl->sg.size;
 745	if (unlikely((!split &&
 746		      msg_pl->sg.size +
 747		      prot->overhead_size > msg_en->sg.size) ||
 748		     (split &&
 749		      split_point +
 750		      prot->overhead_size > msg_en->sg.size))) {
 751		split = true;
 752		split_point = msg_en->sg.size;
 753	}
 754	if (split) {
 755		rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
 756					   split_point, prot->overhead_size,
 757					   &orig_end);
 758		if (rc < 0)
 759			return rc;
 760		/* This can happen if above tls_split_open_record allocates
 761		 * a single large encryption buffer instead of two smaller
 762		 * ones. In this case adjust pointers and continue without
 763		 * split.
 764		 */
 765		if (!msg_pl->sg.size) {
 766			tls_merge_open_record(sk, rec, tmp, orig_end);
 767			msg_pl = &rec->msg_plaintext;
 768			msg_en = &rec->msg_encrypted;
 769			split = false;
 770		}
 771		sk_msg_trim(sk, msg_en, msg_pl->sg.size +
 772			    prot->overhead_size);
 773	}
 774
 775	rec->tx_flags = flags;
 776	req = &rec->aead_req;
 777
 778	i = msg_pl->sg.end;
 779	sk_msg_iter_var_prev(i);
 780
 781	rec->content_type = record_type;
 782	if (prot->version == TLS_1_3_VERSION) {
 783		/* Add content type to end of message.  No padding added */
 784		sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
 785		sg_mark_end(&rec->sg_content_type);
 786		sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
 787			 &rec->sg_content_type);
 788	} else {
 789		sg_mark_end(sk_msg_elem(msg_pl, i));
 790	}
 791
 792	if (msg_pl->sg.end < msg_pl->sg.start) {
 793		sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
 794			 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
 795			 msg_pl->sg.data);
 796	}
 797
 798	i = msg_pl->sg.start;
 799	sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
 800
 801	i = msg_en->sg.end;
 802	sk_msg_iter_var_prev(i);
 803	sg_mark_end(sk_msg_elem(msg_en, i));
 804
 805	i = msg_en->sg.start;
 806	sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
 807
 808	tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
 809		     tls_ctx->tx.rec_seq, record_type, prot);
 810
 811	tls_fill_prepend(tls_ctx,
 812			 page_address(sg_page(&msg_en->sg.data[i])) +
 813			 msg_en->sg.data[i].offset,
 814			 msg_pl->sg.size + prot->tail_size,
 815			 record_type);
 816
 817	tls_ctx->pending_open_record_frags = false;
 818
 819	rc = tls_do_encryption(sk, tls_ctx, ctx, req,
 820			       msg_pl->sg.size + prot->tail_size, i);
 821	if (rc < 0) {
 822		if (rc != -EINPROGRESS) {
 823			tls_err_abort(sk, -EBADMSG);
 824			if (split) {
 825				tls_ctx->pending_open_record_frags = true;
 826				tls_merge_open_record(sk, rec, tmp, orig_end);
 827			}
 828		}
 829		ctx->async_capable = 1;
 830		return rc;
 831	} else if (split) {
 832		msg_pl = &tmp->msg_plaintext;
 833		msg_en = &tmp->msg_encrypted;
 834		sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
 835		tls_ctx->pending_open_record_frags = true;
 836		ctx->open_rec = tmp;
 837	}
 838
 839	return tls_tx_records(sk, flags);
 840}
 841
 842static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
 843			       bool full_record, u8 record_type,
 844			       ssize_t *copied, int flags)
 845{
 846	struct tls_context *tls_ctx = tls_get_ctx(sk);
 847	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 848	struct sk_msg msg_redir = { };
 849	struct sk_psock *psock;
 850	struct sock *sk_redir;
 851	struct tls_rec *rec;
 852	bool enospc, policy, redir_ingress;
 853	int err = 0, send;
 854	u32 delta = 0;
 855
 856	policy = !(flags & MSG_SENDPAGE_NOPOLICY);
 857	psock = sk_psock_get(sk);
 858	if (!psock || !policy) {
 859		err = tls_push_record(sk, flags, record_type);
 860		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
 861			*copied -= sk_msg_free(sk, msg);
 862			tls_free_open_rec(sk);
 863			err = -sk->sk_err;
 864		}
 865		if (psock)
 866			sk_psock_put(sk, psock);
 867		return err;
 868	}
 869more_data:
 870	enospc = sk_msg_full(msg);
 871	if (psock->eval == __SK_NONE) {
 872		delta = msg->sg.size;
 873		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
 874		delta -= msg->sg.size;
 875	}
 876	if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
 877	    !enospc && !full_record) {
 878		err = -ENOSPC;
 879		goto out_err;
 880	}
 881	msg->cork_bytes = 0;
 882	send = msg->sg.size;
 883	if (msg->apply_bytes && msg->apply_bytes < send)
 884		send = msg->apply_bytes;
 885
 886	switch (psock->eval) {
 887	case __SK_PASS:
 888		err = tls_push_record(sk, flags, record_type);
 889		if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
 890			*copied -= sk_msg_free(sk, msg);
 891			tls_free_open_rec(sk);
 892			err = -sk->sk_err;
 893			goto out_err;
 894		}
 895		break;
 896	case __SK_REDIRECT:
 897		redir_ingress = psock->redir_ingress;
 898		sk_redir = psock->sk_redir;
 899		memcpy(&msg_redir, msg, sizeof(*msg));
 900		if (msg->apply_bytes < send)
 901			msg->apply_bytes = 0;
 902		else
 903			msg->apply_bytes -= send;
 904		sk_msg_return_zero(sk, msg, send);
 905		msg->sg.size -= send;
 906		release_sock(sk);
 907		err = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
 908					    &msg_redir, send, flags);
 909		lock_sock(sk);
 910		if (err < 0) {
 911			*copied -= sk_msg_free_nocharge(sk, &msg_redir);
 912			msg->sg.size = 0;
 913		}
 914		if (msg->sg.size == 0)
 915			tls_free_open_rec(sk);
 916		break;
 917	case __SK_DROP:
 918	default:
 919		sk_msg_free_partial(sk, msg, send);
 920		if (msg->apply_bytes < send)
 921			msg->apply_bytes = 0;
 922		else
 923			msg->apply_bytes -= send;
 924		if (msg->sg.size == 0)
 925			tls_free_open_rec(sk);
 926		*copied -= (send + delta);
 927		err = -EACCES;
 928	}
 929
 930	if (likely(!err)) {
 931		bool reset_eval = !ctx->open_rec;
 932
 933		rec = ctx->open_rec;
 934		if (rec) {
 935			msg = &rec->msg_plaintext;
 936			if (!msg->apply_bytes)
 937				reset_eval = true;
 938		}
 939		if (reset_eval) {
 940			psock->eval = __SK_NONE;
 941			if (psock->sk_redir) {
 942				sock_put(psock->sk_redir);
 943				psock->sk_redir = NULL;
 944			}
 945		}
 946		if (rec)
 947			goto more_data;
 948	}
 949 out_err:
 950	sk_psock_put(sk, psock);
 951	return err;
 952}
 953
 954static int tls_sw_push_pending_record(struct sock *sk, int flags)
 955{
 956	struct tls_context *tls_ctx = tls_get_ctx(sk);
 957	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 958	struct tls_rec *rec = ctx->open_rec;
 959	struct sk_msg *msg_pl;
 960	size_t copied;
 961
 962	if (!rec)
 963		return 0;
 964
 965	msg_pl = &rec->msg_plaintext;
 966	copied = msg_pl->sg.size;
 967	if (!copied)
 968		return 0;
 969
 970	return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
 971				   &copied, flags);
 972}
 973
 974static int tls_sw_sendmsg_splice(struct sock *sk, struct msghdr *msg,
 975				 struct sk_msg *msg_pl, size_t try_to_copy,
 976				 ssize_t *copied)
 977{
 978	struct page *page = NULL, **pages = &page;
 979
 980	do {
 981		ssize_t part;
 982		size_t off;
 983
 984		part = iov_iter_extract_pages(&msg->msg_iter, &pages,
 985					      try_to_copy, 1, 0, &off);
 986		if (part <= 0)
 987			return part ?: -EIO;
 988
 989		if (WARN_ON_ONCE(!sendpage_ok(page))) {
 990			iov_iter_revert(&msg->msg_iter, part);
 991			return -EIO;
 992		}
 993
 994		sk_msg_page_add(msg_pl, page, part, off);
 995		msg_pl->sg.copybreak = 0;
 996		msg_pl->sg.curr = msg_pl->sg.end;
 997		sk_mem_charge(sk, part);
 998		*copied += part;
 999		try_to_copy -= part;
1000	} while (try_to_copy && !sk_msg_full(msg_pl));
1001
1002	return 0;
1003}
1004
1005static int tls_sw_sendmsg_locked(struct sock *sk, struct msghdr *msg,
1006				 size_t size)
1007{
1008	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1009	struct tls_context *tls_ctx = tls_get_ctx(sk);
1010	struct tls_prot_info *prot = &tls_ctx->prot_info;
1011	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1012	bool async_capable = ctx->async_capable;
1013	unsigned char record_type = TLS_RECORD_TYPE_DATA;
1014	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1015	bool eor = !(msg->msg_flags & MSG_MORE);
1016	size_t try_to_copy;
1017	ssize_t copied = 0;
1018	struct sk_msg *msg_pl, *msg_en;
1019	struct tls_rec *rec;
1020	int required_size;
1021	int num_async = 0;
1022	bool full_record;
1023	int record_room;
1024	int num_zc = 0;
1025	int orig_size;
1026	int ret = 0;
 
 
 
 
 
1027
1028	if (!eor && (msg->msg_flags & MSG_EOR))
1029		return -EINVAL;
1030
1031	if (unlikely(msg->msg_controllen)) {
1032		ret = tls_process_cmsg(sk, msg, &record_type);
1033		if (ret) {
1034			if (ret == -EINPROGRESS)
1035				num_async++;
1036			else if (ret != -EAGAIN)
1037				goto send_end;
1038		}
1039	}
1040
1041	while (msg_data_left(msg)) {
1042		if (sk->sk_err) {
1043			ret = -sk->sk_err;
1044			goto send_end;
1045		}
1046
1047		if (ctx->open_rec)
1048			rec = ctx->open_rec;
1049		else
1050			rec = ctx->open_rec = tls_get_rec(sk);
1051		if (!rec) {
1052			ret = -ENOMEM;
1053			goto send_end;
1054		}
1055
1056		msg_pl = &rec->msg_plaintext;
1057		msg_en = &rec->msg_encrypted;
1058
1059		orig_size = msg_pl->sg.size;
1060		full_record = false;
1061		try_to_copy = msg_data_left(msg);
1062		record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1063		if (try_to_copy >= record_room) {
1064			try_to_copy = record_room;
1065			full_record = true;
1066		}
1067
1068		required_size = msg_pl->sg.size + try_to_copy +
1069				prot->overhead_size;
1070
1071		if (!sk_stream_memory_free(sk))
1072			goto wait_for_sndbuf;
1073
1074alloc_encrypted:
1075		ret = tls_alloc_encrypted_msg(sk, required_size);
1076		if (ret) {
1077			if (ret != -ENOSPC)
1078				goto wait_for_memory;
1079
1080			/* Adjust try_to_copy according to the amount that was
1081			 * actually allocated. The difference is due
1082			 * to max sg elements limit
1083			 */
1084			try_to_copy -= required_size - msg_en->sg.size;
1085			full_record = true;
1086		}
1087
1088		if (try_to_copy && (msg->msg_flags & MSG_SPLICE_PAGES)) {
1089			ret = tls_sw_sendmsg_splice(sk, msg, msg_pl,
1090						    try_to_copy, &copied);
1091			if (ret < 0)
1092				goto send_end;
1093			tls_ctx->pending_open_record_frags = true;
1094
1095			if (sk_msg_full(msg_pl))
1096				full_record = true;
1097
1098			if (full_record || eor)
1099				goto copied;
1100			continue;
1101		}
1102
1103		if (!is_kvec && (full_record || eor) && !async_capable) {
1104			u32 first = msg_pl->sg.end;
1105
1106			ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1107							msg_pl, try_to_copy);
1108			if (ret)
1109				goto fallback_to_reg_send;
1110
1111			num_zc++;
1112			copied += try_to_copy;
1113
1114			sk_msg_sg_copy_set(msg_pl, first);
1115			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1116						  record_type, &copied,
1117						  msg->msg_flags);
1118			if (ret) {
1119				if (ret == -EINPROGRESS)
1120					num_async++;
1121				else if (ret == -ENOMEM)
1122					goto wait_for_memory;
1123				else if (ctx->open_rec && ret == -ENOSPC)
1124					goto rollback_iter;
1125				else if (ret != -EAGAIN)
1126					goto send_end;
1127			}
1128			continue;
1129rollback_iter:
1130			copied -= try_to_copy;
1131			sk_msg_sg_copy_clear(msg_pl, first);
1132			iov_iter_revert(&msg->msg_iter,
1133					msg_pl->sg.size - orig_size);
1134fallback_to_reg_send:
1135			sk_msg_trim(sk, msg_pl, orig_size);
1136		}
1137
1138		required_size = msg_pl->sg.size + try_to_copy;
1139
1140		ret = tls_clone_plaintext_msg(sk, required_size);
1141		if (ret) {
1142			if (ret != -ENOSPC)
1143				goto send_end;
1144
1145			/* Adjust try_to_copy according to the amount that was
1146			 * actually allocated. The difference is due
1147			 * to max sg elements limit
1148			 */
1149			try_to_copy -= required_size - msg_pl->sg.size;
1150			full_record = true;
1151			sk_msg_trim(sk, msg_en,
1152				    msg_pl->sg.size + prot->overhead_size);
1153		}
1154
1155		if (try_to_copy) {
1156			ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1157						       msg_pl, try_to_copy);
1158			if (ret < 0)
1159				goto trim_sgl;
1160		}
1161
1162		/* Open records defined only if successfully copied, otherwise
1163		 * we would trim the sg but not reset the open record frags.
1164		 */
1165		tls_ctx->pending_open_record_frags = true;
1166		copied += try_to_copy;
1167copied:
1168		if (full_record || eor) {
1169			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1170						  record_type, &copied,
1171						  msg->msg_flags);
1172			if (ret) {
1173				if (ret == -EINPROGRESS)
1174					num_async++;
1175				else if (ret == -ENOMEM)
1176					goto wait_for_memory;
1177				else if (ret != -EAGAIN) {
1178					if (ret == -ENOSPC)
1179						ret = 0;
1180					goto send_end;
1181				}
1182			}
1183		}
1184
1185		continue;
1186
1187wait_for_sndbuf:
1188		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1189wait_for_memory:
1190		ret = sk_stream_wait_memory(sk, &timeo);
1191		if (ret) {
1192trim_sgl:
1193			if (ctx->open_rec)
1194				tls_trim_both_msgs(sk, orig_size);
1195			goto send_end;
1196		}
1197
1198		if (ctx->open_rec && msg_en->sg.size < required_size)
1199			goto alloc_encrypted;
1200	}
1201
1202	if (!num_async) {
1203		goto send_end;
1204	} else if (num_zc) {
1205		int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1206
1207		/* Wait for pending encryptions to get completed */
1208		err = tls_encrypt_async_wait(ctx);
1209		if (err) {
1210			ret = err;
1211			copied = 0;
1212		}
1213	}
1214
1215	/* Transmit if any encryptions have completed */
1216	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1217		cancel_delayed_work(&ctx->tx_work.work);
1218		tls_tx_records(sk, msg->msg_flags);
1219	}
1220
1221send_end:
1222	ret = sk_stream_error(sk, msg->msg_flags, ret);
1223	return copied > 0 ? copied : ret;
1224}
1225
1226int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1227{
1228	struct tls_context *tls_ctx = tls_get_ctx(sk);
1229	int ret;
1230
1231	if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1232			       MSG_CMSG_COMPAT | MSG_SPLICE_PAGES | MSG_EOR |
1233			       MSG_SENDPAGE_NOPOLICY))
1234		return -EOPNOTSUPP;
1235
1236	ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1237	if (ret)
1238		return ret;
1239	lock_sock(sk);
1240	ret = tls_sw_sendmsg_locked(sk, msg, size);
1241	release_sock(sk);
1242	mutex_unlock(&tls_ctx->tx_lock);
1243	return ret;
1244}
1245
1246/*
1247 * Handle unexpected EOF during splice without SPLICE_F_MORE set.
1248 */
1249void tls_sw_splice_eof(struct socket *sock)
1250{
1251	struct sock *sk = sock->sk;
1252	struct tls_context *tls_ctx = tls_get_ctx(sk);
1253	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
 
 
 
1254	struct tls_rec *rec;
1255	struct sk_msg *msg_pl;
1256	ssize_t copied = 0;
1257	bool retrying = false;
 
1258	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1259
1260	if (!ctx->open_rec)
1261		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1262
1263	mutex_lock(&tls_ctx->tx_lock);
1264	lock_sock(sk);
 
1265
1266retry:
1267	/* same checks as in tls_sw_push_pending_record() */
1268	rec = ctx->open_rec;
1269	if (!rec)
1270		goto unlock;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1271
1272	msg_pl = &rec->msg_plaintext;
1273	if (msg_pl->sg.size == 0)
1274		goto unlock;
1275
1276	/* Check the BPF advisor and perform transmission. */
1277	ret = bpf_exec_tx_verdict(msg_pl, sk, false, TLS_RECORD_TYPE_DATA,
1278				  &copied, 0);
1279	switch (ret) {
1280	case 0:
1281	case -EAGAIN:
1282		if (retrying)
1283			goto unlock;
1284		retrying = true;
1285		goto retry;
1286	case -EINPROGRESS:
1287		break;
1288	default:
1289		goto unlock;
1290	}
 
 
 
 
 
 
 
 
 
 
 
 
1291
1292	/* Wait for pending encryptions to get completed */
1293	if (tls_encrypt_async_wait(ctx))
1294		goto unlock;
1295
1296	/* Transmit if any encryptions have completed */
1297	if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1298		cancel_delayed_work(&ctx->tx_work.work);
1299		tls_tx_records(sk, 0);
1300	}
 
 
 
 
1301
1302unlock:
 
 
1303	release_sock(sk);
1304	mutex_unlock(&tls_ctx->tx_lock);
 
1305}
1306
1307static int
1308tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
1309		bool released)
1310{
1311	struct tls_context *tls_ctx = tls_get_ctx(sk);
1312	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
 
1313	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1314	int ret = 0;
1315	long timeo;
1316
1317	timeo = sock_rcvtimeo(sk, nonblock);
1318
1319	while (!tls_strp_msg_ready(ctx)) {
1320		if (!sk_psock_queue_empty(psock))
1321			return 0;
1322
1323		if (sk->sk_err)
1324			return sock_error(sk);
1325
1326		if (ret < 0)
1327			return ret;
1328
1329		if (!skb_queue_empty(&sk->sk_receive_queue)) {
1330			tls_strp_check_rcv(&ctx->strp);
1331			if (tls_strp_msg_ready(ctx))
1332				break;
1333		}
1334
1335		if (sk->sk_shutdown & RCV_SHUTDOWN)
1336			return 0;
1337
1338		if (sock_flag(sk, SOCK_DONE))
1339			return 0;
1340
1341		if (!timeo)
1342			return -EAGAIN;
 
 
1343
1344		released = true;
1345		add_wait_queue(sk_sleep(sk), &wait);
1346		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1347		ret = sk_wait_event(sk, &timeo,
1348				    tls_strp_msg_ready(ctx) ||
1349				    !sk_psock_queue_empty(psock),
1350				    &wait);
1351		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1352		remove_wait_queue(sk_sleep(sk), &wait);
1353
1354		/* Handle signals */
1355		if (signal_pending(current))
1356			return sock_intr_errno(timeo);
 
 
1357	}
1358
1359	tls_strp_msg_load(&ctx->strp, released);
1360
1361	return 1;
1362}
1363
1364static int tls_setup_from_iter(struct iov_iter *from,
1365			       int length, int *pages_used,
 
1366			       struct scatterlist *to,
1367			       int to_max_pages)
1368{
1369	int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1370	struct page *pages[MAX_SKB_FRAGS];
1371	unsigned int size = 0;
1372	ssize_t copied, use;
1373	size_t offset;
1374
1375	while (length > 0) {
1376		i = 0;
1377		maxpages = to_max_pages - num_elem;
1378		if (maxpages == 0) {
1379			rc = -EFAULT;
1380			goto out;
1381		}
1382		copied = iov_iter_get_pages2(from, pages,
1383					    length,
1384					    maxpages, &offset);
1385		if (copied <= 0) {
1386			rc = -EFAULT;
1387			goto out;
1388		}
1389
 
 
1390		length -= copied;
1391		size += copied;
1392		while (copied) {
1393			use = min_t(int, copied, PAGE_SIZE - offset);
1394
1395			sg_set_page(&to[num_elem],
1396				    pages[i], use, offset);
1397			sg_unmark_end(&to[num_elem]);
1398			/* We do not uncharge memory from this API */
1399
1400			offset = 0;
1401			copied -= use;
1402
1403			i++;
1404			num_elem++;
1405		}
1406	}
1407	/* Mark the end in the last sg entry if newly added */
1408	if (num_elem > *pages_used)
1409		sg_mark_end(&to[num_elem - 1]);
1410out:
1411	if (rc)
1412		iov_iter_revert(from, size);
 
1413	*pages_used = num_elem;
1414
1415	return rc;
1416}
1417
1418static struct sk_buff *
1419tls_alloc_clrtxt_skb(struct sock *sk, struct sk_buff *skb,
1420		     unsigned int full_len)
1421{
1422	struct strp_msg *clr_rxm;
1423	struct sk_buff *clr_skb;
1424	int err;
1425
1426	clr_skb = alloc_skb_with_frags(0, full_len, TLS_PAGE_ORDER,
1427				       &err, sk->sk_allocation);
1428	if (!clr_skb)
1429		return NULL;
1430
1431	skb_copy_header(clr_skb, skb);
1432	clr_skb->len = full_len;
1433	clr_skb->data_len = full_len;
1434
1435	clr_rxm = strp_msg(clr_skb);
1436	clr_rxm->offset = 0;
1437
1438	return clr_skb;
1439}
1440
1441/* Decrypt handlers
1442 *
1443 * tls_decrypt_sw() and tls_decrypt_device() are decrypt handlers.
1444 * They must transform the darg in/out argument are as follows:
1445 *       |          Input            |         Output
1446 * -------------------------------------------------------------------
1447 *    zc | Zero-copy decrypt allowed | Zero-copy performed
1448 * async | Async decrypt allowed     | Async crypto used / in progress
1449 *   skb |            *              | Output skb
1450 *
1451 * If ZC decryption was performed darg.skb will point to the input skb.
1452 */
1453
1454/* This function decrypts the input skb into either out_iov or in out_sg
1455 * or in skb buffers itself. The input parameter 'darg->zc' indicates if
1456 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1457 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1458 * NULL, then the decryption happens inside skb buffers itself, i.e.
1459 * zero-copy gets disabled and 'darg->zc' is updated.
1460 */
1461static int tls_decrypt_sg(struct sock *sk, struct iov_iter *out_iov,
1462			  struct scatterlist *out_sg,
1463			  struct tls_decrypt_arg *darg)
 
 
1464{
1465	struct tls_context *tls_ctx = tls_get_ctx(sk);
1466	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1467	struct tls_prot_info *prot = &tls_ctx->prot_info;
1468	int n_sgin, n_sgout, aead_size, err, pages = 0;
1469	struct sk_buff *skb = tls_strp_msg(ctx);
1470	const struct strp_msg *rxm = strp_msg(skb);
1471	const struct tls_msg *tlm = tls_msg(skb);
1472	struct aead_request *aead_req;
 
 
1473	struct scatterlist *sgin = NULL;
1474	struct scatterlist *sgout = NULL;
1475	const int data_len = rxm->full_len - prot->overhead_size;
1476	int tail_pages = !!prot->tail_size;
1477	struct tls_decrypt_ctx *dctx;
1478	struct sk_buff *clear_skb;
1479	int iv_offset = 0;
1480	u8 *mem;
1481
1482	n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1483			 rxm->full_len - prot->prepend_size);
1484	if (n_sgin < 1)
1485		return n_sgin ?: -EBADMSG;
1486
1487	if (darg->zc && (out_iov || out_sg)) {
1488		clear_skb = NULL;
1489
 
1490		if (out_iov)
1491			n_sgout = 1 + tail_pages +
1492				iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1493		else
1494			n_sgout = sg_nents(out_sg);
 
 
1495	} else {
1496		darg->zc = false;
 
 
 
1497
1498		clear_skb = tls_alloc_clrtxt_skb(sk, skb, rxm->full_len);
1499		if (!clear_skb)
1500			return -ENOMEM;
1501
1502		n_sgout = 1 + skb_shinfo(clear_skb)->nr_frags;
1503	}
1504
1505	/* Increment to accommodate AAD */
1506	n_sgin = n_sgin + 1;
1507
 
 
 
 
 
 
 
1508	/* Allocate a single block of memory which contains
1509	 *   aead_req || tls_decrypt_ctx.
1510	 * Both structs are variable length.
1511	 */
1512	aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1513	aead_size = ALIGN(aead_size, __alignof__(*dctx));
1514	mem = kmalloc(aead_size + struct_size(dctx, sg, size_add(n_sgin, n_sgout)),
1515		      sk->sk_allocation);
1516	if (!mem) {
1517		err = -ENOMEM;
1518		goto exit_free_skb;
1519	}
1520
1521	/* Segment the allocated memory */
1522	aead_req = (struct aead_request *)mem;
1523	dctx = (struct tls_decrypt_ctx *)(mem + aead_size);
1524	dctx->sk = sk;
1525	sgin = &dctx->sg[0];
1526	sgout = &dctx->sg[n_sgin];
1527
1528	/* For CCM based ciphers, first byte of nonce+iv is a constant */
1529	switch (prot->cipher_type) {
1530	case TLS_CIPHER_AES_CCM_128:
1531		dctx->iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1532		iv_offset = 1;
1533		break;
1534	case TLS_CIPHER_SM4_CCM:
1535		dctx->iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1536		iv_offset = 1;
1537		break;
1538	}
1539
1540	/* Prepare IV */
 
 
 
 
 
 
 
1541	if (prot->version == TLS_1_3_VERSION ||
1542	    prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1543		memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv,
1544		       prot->iv_size + prot->salt_size);
1545	} else {
1546		err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1547				    &dctx->iv[iv_offset] + prot->salt_size,
1548				    prot->iv_size);
1549		if (err < 0)
1550			goto exit_free;
1551		memcpy(&dctx->iv[iv_offset], tls_ctx->rx.iv, prot->salt_size);
1552	}
1553	tls_xor_iv_with_seq(prot, &dctx->iv[iv_offset], tls_ctx->rx.rec_seq);
1554
1555	/* Prepare AAD */
1556	tls_make_aad(dctx->aad, rxm->full_len - prot->overhead_size +
1557		     prot->tail_size,
1558		     tls_ctx->rx.rec_seq, tlm->control, prot);
1559
1560	/* Prepare sgin */
1561	sg_init_table(sgin, n_sgin);
1562	sg_set_buf(&sgin[0], dctx->aad, prot->aad_size);
1563	err = skb_to_sgvec(skb, &sgin[1],
1564			   rxm->offset + prot->prepend_size,
1565			   rxm->full_len - prot->prepend_size);
1566	if (err < 0)
1567		goto exit_free;
 
 
1568
1569	if (clear_skb) {
1570		sg_init_table(sgout, n_sgout);
1571		sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1572
1573		err = skb_to_sgvec(clear_skb, &sgout[1], prot->prepend_size,
1574				   data_len + prot->tail_size);
1575		if (err < 0)
1576			goto exit_free;
1577	} else if (out_iov) {
1578		sg_init_table(sgout, n_sgout);
1579		sg_set_buf(&sgout[0], dctx->aad, prot->aad_size);
1580
1581		err = tls_setup_from_iter(out_iov, data_len, &pages, &sgout[1],
1582					  (n_sgout - 1 - tail_pages));
1583		if (err < 0)
1584			goto exit_free_pages;
1585
1586		if (prot->tail_size) {
1587			sg_unmark_end(&sgout[pages]);
1588			sg_set_buf(&sgout[pages + 1], &dctx->tail,
1589				   prot->tail_size);
1590			sg_mark_end(&sgout[pages + 1]);
1591		}
1592	} else if (out_sg) {
1593		memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
 
 
 
 
1594	}
1595	dctx->free_sgout = !!pages;
1596
1597	/* Prepare and submit AEAD request */
1598	err = tls_do_decryption(sk, sgin, sgout, dctx->iv,
1599				data_len + prot->tail_size, aead_req, darg);
1600	if (err) {
1601		if (darg->async_done)
1602			goto exit_free_skb;
1603		goto exit_free_pages;
1604	}
1605
1606	darg->skb = clear_skb ?: tls_strp_msg(ctx);
1607	clear_skb = NULL;
1608
1609	if (unlikely(darg->async)) {
1610		err = tls_strp_msg_hold(&ctx->strp, &ctx->async_hold);
1611		if (err)
1612			__skb_queue_tail(&ctx->async_hold, darg->skb);
1613		return err;
1614	}
1615
1616	if (unlikely(darg->async_done))
1617		return 0;
1618
1619	if (prot->tail_size)
1620		darg->tail = dctx->tail;
1621
1622exit_free_pages:
1623	/* Release the pages in case iov was mapped to pages */
1624	for (; pages > 0; pages--)
1625		put_page(sg_page(&sgout[pages]));
1626exit_free:
1627	kfree(mem);
1628exit_free_skb:
1629	consume_skb(clear_skb);
1630	return err;
1631}
1632
1633static int
1634tls_decrypt_sw(struct sock *sk, struct tls_context *tls_ctx,
1635	       struct msghdr *msg, struct tls_decrypt_arg *darg)
1636{
 
1637	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1638	struct tls_prot_info *prot = &tls_ctx->prot_info;
1639	struct strp_msg *rxm;
1640	int pad, err;
1641
1642	err = tls_decrypt_sg(sk, &msg->msg_iter, NULL, darg);
1643	if (err < 0) {
1644		if (err == -EBADMSG)
1645			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
1646		return err;
1647	}
1648	/* keep going even for ->async, the code below is TLS 1.3 */
1649
1650	/* If opportunistic TLS 1.3 ZC failed retry without ZC */
1651	if (unlikely(darg->zc && prot->version == TLS_1_3_VERSION &&
1652		     darg->tail != TLS_RECORD_TYPE_DATA)) {
1653		darg->zc = false;
1654		if (!darg->tail)
1655			TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXNOPADVIOL);
1656		TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTRETRY);
1657		return tls_decrypt_sw(sk, tls_ctx, msg, darg);
1658	}
1659
1660	pad = tls_padding_length(prot, darg->skb, darg);
1661	if (pad < 0) {
1662		if (darg->skb != tls_strp_msg(ctx))
1663			consume_skb(darg->skb);
1664		return pad;
1665	}
1666
1667	rxm = strp_msg(darg->skb);
1668	rxm->full_len -= pad;
1669
1670	return 0;
1671}
1672
1673static int
1674tls_decrypt_device(struct sock *sk, struct msghdr *msg,
1675		   struct tls_context *tls_ctx, struct tls_decrypt_arg *darg)
1676{
1677	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1678	struct tls_prot_info *prot = &tls_ctx->prot_info;
1679	struct strp_msg *rxm;
1680	int pad, err;
1681
1682	if (tls_ctx->rx_conf != TLS_HW)
1683		return 0;
1684
1685	err = tls_device_decrypted(sk, tls_ctx);
1686	if (err <= 0)
1687		return err;
1688
1689	pad = tls_padding_length(prot, tls_strp_msg(ctx), darg);
1690	if (pad < 0)
1691		return pad;
1692
1693	darg->async = false;
1694	darg->skb = tls_strp_msg(ctx);
1695	/* ->zc downgrade check, in case TLS 1.3 gets here */
1696	darg->zc &= !(prot->version == TLS_1_3_VERSION &&
1697		      tls_msg(darg->skb)->control != TLS_RECORD_TYPE_DATA);
1698
1699	rxm = strp_msg(darg->skb);
1700	rxm->full_len -= pad;
1701
1702	if (!darg->zc) {
1703		/* Non-ZC case needs a real skb */
1704		darg->skb = tls_strp_msg_detach(ctx);
1705		if (!darg->skb)
1706			return -ENOMEM;
1707	} else {
1708		unsigned int off, len;
1709
1710		/* In ZC case nobody cares about the output skb.
1711		 * Just copy the data here. Note the skb is not fully trimmed.
1712		 */
1713		off = rxm->offset + prot->prepend_size;
1714		len = rxm->full_len - prot->overhead_size;
1715
1716		err = skb_copy_datagram_msg(darg->skb, off, msg, len);
1717		if (err)
1718			return err;
1719	}
1720	return 1;
1721}
1722
1723static int tls_rx_one_record(struct sock *sk, struct msghdr *msg,
1724			     struct tls_decrypt_arg *darg)
1725{
1726	struct tls_context *tls_ctx = tls_get_ctx(sk);
1727	struct tls_prot_info *prot = &tls_ctx->prot_info;
1728	struct strp_msg *rxm;
1729	int err;
1730
1731	err = tls_decrypt_device(sk, msg, tls_ctx, darg);
1732	if (!err)
1733		err = tls_decrypt_sw(sk, tls_ctx, msg, darg);
1734	if (err < 0)
1735		return err;
1736
1737	rxm = strp_msg(darg->skb);
1738	rxm->offset += prot->prepend_size;
1739	rxm->full_len -= prot->overhead_size;
1740	tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1741
1742	return 0;
1743}
1744
1745int decrypt_skb(struct sock *sk, struct scatterlist *sgout)
 
1746{
1747	struct tls_decrypt_arg darg = { .zc = true, };
 
1748
1749	return tls_decrypt_sg(sk, NULL, sgout, &darg);
1750}
1751
1752static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1753				   u8 *control)
1754{
1755	int err;
 
1756
1757	if (!*control) {
1758		*control = tlm->control;
1759		if (!*control)
1760			return -EBADMSG;
1761
1762		err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1763			       sizeof(*control), control);
1764		if (*control != TLS_RECORD_TYPE_DATA) {
1765			if (err || msg->msg_flags & MSG_CTRUNC)
1766				return -EIO;
1767		}
1768	} else if (*control != tlm->control) {
1769		return 0;
1770	}
1771
1772	return 1;
1773}
 
1774
1775static void tls_rx_rec_done(struct tls_sw_context_rx *ctx)
1776{
1777	tls_strp_msg_done(&ctx->strp);
1778}
1779
1780/* This function traverses the rx_list in tls receive context to copies the
1781 * decrypted records into the buffer provided by caller zero copy is not
1782 * true. Further, the records are removed from the rx_list if it is not a peek
1783 * case and the record has been consumed completely.
1784 */
1785static int process_rx_list(struct tls_sw_context_rx *ctx,
1786			   struct msghdr *msg,
1787			   u8 *control,
 
1788			   size_t skip,
1789			   size_t len,
1790			   bool is_peek,
1791			   bool *more)
1792{
1793	struct sk_buff *skb = skb_peek(&ctx->rx_list);
 
 
1794	struct tls_msg *tlm;
1795	ssize_t copied = 0;
1796	int err;
 
 
 
 
 
1797
1798	while (skip && skb) {
1799		struct strp_msg *rxm = strp_msg(skb);
1800		tlm = tls_msg(skb);
1801
1802		err = tls_record_content_type(msg, tlm, control);
1803		if (err <= 0)
1804			goto more;
1805
1806		if (skip < rxm->full_len)
1807			break;
1808
1809		skip = skip - rxm->full_len;
1810		skb = skb_peek_next(skb, &ctx->rx_list);
1811	}
1812
1813	while (len && skb) {
1814		struct sk_buff *next_skb;
1815		struct strp_msg *rxm = strp_msg(skb);
1816		int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1817
1818		tlm = tls_msg(skb);
1819
1820		err = tls_record_content_type(msg, tlm, control);
1821		if (err <= 0)
1822			goto more;
1823
1824		err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1825					    msg, chunk);
1826		if (err < 0)
1827			goto more;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1828
1829		len = len - chunk;
1830		copied = copied + chunk;
1831
1832		/* Consume the data from record if it is non-peek case*/
1833		if (!is_peek) {
1834			rxm->offset = rxm->offset + chunk;
1835			rxm->full_len = rxm->full_len - chunk;
1836
1837			/* Return if there is unconsumed data in the record */
1838			if (rxm->full_len - skip)
1839				break;
1840		}
1841
1842		/* The remaining skip-bytes must lie in 1st record in rx_list.
1843		 * So from the 2nd record, 'skip' should be 0.
1844		 */
1845		skip = 0;
1846
1847		if (msg)
1848			msg->msg_flags |= MSG_EOR;
1849
1850		next_skb = skb_peek_next(skb, &ctx->rx_list);
1851
1852		if (!is_peek) {
1853			__skb_unlink(skb, &ctx->rx_list);
1854			consume_skb(skb);
1855		}
1856
1857		skb = next_skb;
1858	}
1859	err = 0;
1860
1861out:
1862	return copied ? : err;
1863more:
1864	if (more)
1865		*more = true;
1866	goto out;
1867}
1868
1869static bool
1870tls_read_flush_backlog(struct sock *sk, struct tls_prot_info *prot,
1871		       size_t len_left, size_t decrypted, ssize_t done,
1872		       size_t *flushed_at)
1873{
1874	size_t max_rec;
1875
1876	if (len_left <= decrypted)
1877		return false;
1878
1879	max_rec = prot->overhead_size - prot->tail_size + TLS_MAX_PAYLOAD_SIZE;
1880	if (done - *flushed_at < SZ_128K && tcp_inq(sk) > max_rec)
1881		return false;
1882
1883	*flushed_at = done;
1884	return sk_flush_backlog(sk);
1885}
1886
1887static int tls_rx_reader_acquire(struct sock *sk, struct tls_sw_context_rx *ctx,
1888				 bool nonblock)
1889{
1890	long timeo;
1891	int ret;
1892
1893	timeo = sock_rcvtimeo(sk, nonblock);
1894
1895	while (unlikely(ctx->reader_present)) {
1896		DEFINE_WAIT_FUNC(wait, woken_wake_function);
1897
1898		ctx->reader_contended = 1;
1899
1900		add_wait_queue(&ctx->wq, &wait);
1901		ret = sk_wait_event(sk, &timeo,
1902				    !READ_ONCE(ctx->reader_present), &wait);
1903		remove_wait_queue(&ctx->wq, &wait);
1904
1905		if (timeo <= 0)
1906			return -EAGAIN;
1907		if (signal_pending(current))
1908			return sock_intr_errno(timeo);
1909		if (ret < 0)
1910			return ret;
1911	}
1912
1913	WRITE_ONCE(ctx->reader_present, 1);
1914
1915	return 0;
1916}
1917
1918static int tls_rx_reader_lock(struct sock *sk, struct tls_sw_context_rx *ctx,
1919			      bool nonblock)
1920{
1921	int err;
1922
1923	lock_sock(sk);
1924	err = tls_rx_reader_acquire(sk, ctx, nonblock);
1925	if (err)
1926		release_sock(sk);
1927	return err;
1928}
1929
1930static void tls_rx_reader_release(struct sock *sk, struct tls_sw_context_rx *ctx)
1931{
1932	if (unlikely(ctx->reader_contended)) {
1933		if (wq_has_sleeper(&ctx->wq))
1934			wake_up(&ctx->wq);
1935		else
1936			ctx->reader_contended = 0;
1937
1938		WARN_ON_ONCE(!ctx->reader_present);
1939	}
1940
1941	WRITE_ONCE(ctx->reader_present, 0);
1942}
1943
1944static void tls_rx_reader_unlock(struct sock *sk, struct tls_sw_context_rx *ctx)
1945{
1946	tls_rx_reader_release(sk, ctx);
1947	release_sock(sk);
1948}
1949
1950int tls_sw_recvmsg(struct sock *sk,
1951		   struct msghdr *msg,
1952		   size_t len,
 
1953		   int flags,
1954		   int *addr_len)
1955{
1956	struct tls_context *tls_ctx = tls_get_ctx(sk);
1957	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1958	struct tls_prot_info *prot = &tls_ctx->prot_info;
1959	ssize_t decrypted = 0, async_copy_bytes = 0;
1960	struct sk_psock *psock;
1961	unsigned char control = 0;
1962	size_t flushed_at = 0;
1963	struct strp_msg *rxm;
1964	struct tls_msg *tlm;
 
1965	ssize_t copied = 0;
1966	ssize_t peeked = 0;
1967	bool async = false;
1968	int target, err;
1969	bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1970	bool is_peek = flags & MSG_PEEK;
1971	bool rx_more = false;
1972	bool released = true;
1973	bool bpf_strp_enabled;
1974	bool zc_capable;
 
 
 
1975
1976	if (unlikely(flags & MSG_ERRQUEUE))
1977		return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1978
1979	psock = sk_psock_get(sk);
1980	err = tls_rx_reader_lock(sk, ctx, flags & MSG_DONTWAIT);
1981	if (err < 0)
1982		return err;
1983	bpf_strp_enabled = sk_psock_strp_enabled(psock);
1984
1985	/* If crypto failed the connection is broken */
1986	err = ctx->async_wait.err;
1987	if (err)
1988		goto end;
1989
1990	/* Process pending decrypted records. It must be non-zero-copy */
1991	err = process_rx_list(ctx, msg, &control, 0, len, is_peek, &rx_more);
1992	if (err < 0)
 
 
1993		goto end;
 
 
 
1994
1995	copied = err;
1996	if (len <= copied || (copied && control != TLS_RECORD_TYPE_DATA) || rx_more)
1997		goto end;
1998
1999	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2000	len = len - copied;
 
 
 
 
 
 
 
 
 
2001
2002	zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
2003		ctx->zc_capable;
2004	decrypted = 0;
2005	while (len && (decrypted + copied < target || tls_strp_msg_ready(ctx))) {
2006		struct tls_decrypt_arg darg;
2007		int to_decrypt, chunk;
2008
2009		err = tls_rx_rec_wait(sk, psock, flags & MSG_DONTWAIT,
2010				      released);
2011		if (err <= 0) {
2012			if (psock) {
2013				chunk = sk_msg_recvmsg(sk, psock, msg, len,
2014						       flags);
2015				if (chunk > 0) {
2016					decrypted += chunk;
2017					len -= chunk;
 
2018					continue;
2019				}
2020			}
2021			goto recv_end;
 
 
 
 
 
 
2022		}
2023
2024		memset(&darg.inargs, 0, sizeof(darg.inargs));
2025
2026		rxm = strp_msg(tls_strp_msg(ctx));
2027		tlm = tls_msg(tls_strp_msg(ctx));
2028
2029		to_decrypt = rxm->full_len - prot->overhead_size;
2030
2031		if (zc_capable && to_decrypt <= len &&
2032		    tlm->control == TLS_RECORD_TYPE_DATA)
2033			darg.zc = true;
 
 
2034
2035		/* Do not use async mode if record is non-data */
2036		if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
2037			darg.async = ctx->async_capable;
2038		else
2039			darg.async = false;
2040
2041		err = tls_rx_one_record(sk, msg, &darg);
2042		if (err < 0) {
2043			tls_err_abort(sk, -EBADMSG);
 
2044			goto recv_end;
2045		}
2046
2047		async |= darg.async;
 
 
 
 
 
2048
2049		/* If the type of records being processed is not known yet,
2050		 * set it to record type just dequeued. If it is already known,
2051		 * but does not match the record type just dequeued, go to end.
2052		 * We always get record type here since for tls1.2, record type
2053		 * is known just after record is dequeued from stream parser.
2054		 * For tls1.3, we disable async.
2055		 */
2056		err = tls_record_content_type(msg, tls_msg(darg.skb), &control);
2057		if (err <= 0) {
2058			DEBUG_NET_WARN_ON_ONCE(darg.zc);
2059			tls_rx_rec_done(ctx);
2060put_on_rx_list_err:
2061			__skb_queue_tail(&ctx->rx_list, darg.skb);
2062			goto recv_end;
 
 
 
 
 
 
 
 
 
 
 
 
 
2063		}
2064
2065		/* periodically flush backlog, and feed strparser */
2066		released = tls_read_flush_backlog(sk, prot, len, to_decrypt,
2067						  decrypted + copied,
2068						  &flushed_at);
2069
2070		/* TLS 1.3 may have updated the length by more than overhead */
2071		rxm = strp_msg(darg.skb);
2072		chunk = rxm->full_len;
2073		tls_rx_rec_done(ctx);
2074
2075		if (!darg.zc) {
2076			bool partially_consumed = chunk > len;
2077			struct sk_buff *skb = darg.skb;
2078
2079			DEBUG_NET_WARN_ON_ONCE(darg.skb == ctx->strp.anchor);
2080
2081			if (async) {
2082				/* TLS 1.2-only, to_decrypt must be text len */
2083				chunk = min_t(int, to_decrypt, len);
2084				async_copy_bytes += chunk;
2085put_on_rx_list:
2086				decrypted += chunk;
2087				len -= chunk;
2088				__skb_queue_tail(&ctx->rx_list, skb);
2089				if (unlikely(control != TLS_RECORD_TYPE_DATA))
2090					break;
2091				continue;
2092			}
2093
 
2094			if (bpf_strp_enabled) {
2095				released = true;
2096				err = sk_psock_tls_strp_read(psock, skb);
2097				if (err != __SK_PASS) {
2098					rxm->offset = rxm->offset + rxm->full_len;
2099					rxm->full_len = 0;
2100					if (err == __SK_DROP)
2101						consume_skb(skb);
 
 
2102					continue;
2103				}
2104			}
2105
2106			if (partially_consumed)
 
2107				chunk = len;
 
 
 
2108
2109			err = skb_copy_datagram_msg(skb, rxm->offset,
2110						    msg, chunk);
2111			if (err < 0)
2112				goto put_on_rx_list_err;
2113
2114			if (is_peek) {
2115				peeked += chunk;
2116				goto put_on_rx_list;
2117			}
2118
2119			if (partially_consumed) {
2120				rxm->offset += chunk;
2121				rxm->full_len -= chunk;
2122				goto put_on_rx_list;
2123			}
 
2124
2125			consume_skb(skb);
2126		}
 
2127
2128		decrypted += chunk;
2129		len -= chunk;
2130
2131		/* Return full control message to userspace before trying
2132		 * to parse another message type
2133		 */
2134		msg->msg_flags |= MSG_EOR;
2135		if (control != TLS_RECORD_TYPE_DATA)
 
 
 
 
 
 
 
 
 
 
2136			break;
 
2137	}
2138
2139recv_end:
2140	if (async) {
2141		int ret;
2142
2143		/* Wait for all previously submitted records to be decrypted */
2144		ret = tls_decrypt_async_wait(ctx);
2145		__skb_queue_purge(&ctx->async_hold);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2146
2147		if (ret) {
2148			if (err >= 0 || err == -EINPROGRESS)
2149				err = ret;
2150			decrypted = 0;
2151			goto end;
2152		}
2153
2154		/* Drain records from the rx_list & copy if required */
2155		if (is_peek || is_kvec)
2156			err = process_rx_list(ctx, msg, &control, copied + peeked,
2157					      decrypted - peeked, is_peek, NULL);
2158		else
2159			err = process_rx_list(ctx, msg, &control, 0,
2160					      async_copy_bytes, is_peek, NULL);
 
 
 
 
 
2161	}
2162
2163	copied += decrypted;
2164
2165end:
2166	tls_rx_reader_unlock(sk, ctx);
2167	if (psock)
2168		sk_psock_put(sk, psock);
2169	return copied ? : err;
2170}
2171
2172ssize_t tls_sw_splice_read(struct socket *sock,  loff_t *ppos,
2173			   struct pipe_inode_info *pipe,
2174			   size_t len, unsigned int flags)
2175{
2176	struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
2177	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2178	struct strp_msg *rxm = NULL;
2179	struct sock *sk = sock->sk;
2180	struct tls_msg *tlm;
2181	struct sk_buff *skb;
2182	ssize_t copied = 0;
 
 
2183	int chunk;
2184	int err;
 
 
2185
2186	err = tls_rx_reader_lock(sk, ctx, flags & SPLICE_F_NONBLOCK);
2187	if (err < 0)
2188		return err;
 
 
2189
2190	if (!skb_queue_empty(&ctx->rx_list)) {
2191		skb = __skb_dequeue(&ctx->rx_list);
2192	} else {
2193		struct tls_decrypt_arg darg;
2194
2195		err = tls_rx_rec_wait(sk, NULL, flags & SPLICE_F_NONBLOCK,
2196				      true);
2197		if (err <= 0)
2198			goto splice_read_end;
 
2199
2200		memset(&darg.inargs, 0, sizeof(darg.inargs));
2201
2202		err = tls_rx_one_record(sk, NULL, &darg);
2203		if (err < 0) {
2204			tls_err_abort(sk, -EBADMSG);
2205			goto splice_read_end;
2206		}
2207
2208		tls_rx_rec_done(ctx);
2209		skb = darg.skb;
2210	}
2211
2212	rxm = strp_msg(skb);
2213	tlm = tls_msg(skb);
2214
2215	/* splice does not support reading control messages */
2216	if (tlm->control != TLS_RECORD_TYPE_DATA) {
2217		err = -EINVAL;
2218		goto splice_requeue;
2219	}
2220
2221	chunk = min_t(unsigned int, rxm->full_len, len);
2222	copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2223	if (copied < 0)
2224		goto splice_requeue;
2225
2226	if (chunk < rxm->full_len) {
2227		rxm->offset += len;
2228		rxm->full_len -= len;
2229		goto splice_requeue;
2230	}
2231
2232	consume_skb(skb);
2233
2234splice_read_end:
2235	tls_rx_reader_unlock(sk, ctx);
2236	return copied ? : err;
2237
2238splice_requeue:
2239	__skb_queue_head(&ctx->rx_list, skb);
2240	goto splice_read_end;
2241}
2242
2243int tls_sw_read_sock(struct sock *sk, read_descriptor_t *desc,
2244		     sk_read_actor_t read_actor)
2245{
2246	struct tls_context *tls_ctx = tls_get_ctx(sk);
2247	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2248	struct tls_prot_info *prot = &tls_ctx->prot_info;
2249	struct strp_msg *rxm = NULL;
2250	struct sk_buff *skb = NULL;
2251	struct sk_psock *psock;
2252	size_t flushed_at = 0;
2253	bool released = true;
2254	struct tls_msg *tlm;
2255	ssize_t copied = 0;
2256	ssize_t decrypted;
2257	int err, used;
2258
2259	psock = sk_psock_get(sk);
2260	if (psock) {
2261		sk_psock_put(sk, psock);
2262		return -EINVAL;
2263	}
2264	err = tls_rx_reader_acquire(sk, ctx, true);
2265	if (err < 0)
2266		return err;
2267
2268	/* If crypto failed the connection is broken */
2269	err = ctx->async_wait.err;
2270	if (err)
2271		goto read_sock_end;
2272
2273	decrypted = 0;
2274	do {
2275		if (!skb_queue_empty(&ctx->rx_list)) {
2276			skb = __skb_dequeue(&ctx->rx_list);
2277			rxm = strp_msg(skb);
2278			tlm = tls_msg(skb);
2279		} else {
2280			struct tls_decrypt_arg darg;
2281
2282			err = tls_rx_rec_wait(sk, NULL, true, released);
2283			if (err <= 0)
2284				goto read_sock_end;
2285
2286			memset(&darg.inargs, 0, sizeof(darg.inargs));
2287
2288			err = tls_rx_one_record(sk, NULL, &darg);
2289			if (err < 0) {
2290				tls_err_abort(sk, -EBADMSG);
2291				goto read_sock_end;
2292			}
2293
2294			released = tls_read_flush_backlog(sk, prot, INT_MAX,
2295							  0, decrypted,
2296							  &flushed_at);
2297			skb = darg.skb;
2298			rxm = strp_msg(skb);
2299			tlm = tls_msg(skb);
2300			decrypted += rxm->full_len;
2301
2302			tls_rx_rec_done(ctx);
2303		}
2304
2305		/* read_sock does not support reading control messages */
2306		if (tlm->control != TLS_RECORD_TYPE_DATA) {
2307			err = -EINVAL;
2308			goto read_sock_requeue;
2309		}
2310
2311		used = read_actor(desc, skb, rxm->offset, rxm->full_len);
2312		if (used <= 0) {
2313			if (!copied)
2314				err = used;
2315			goto read_sock_requeue;
2316		}
2317		copied += used;
2318		if (used < rxm->full_len) {
2319			rxm->offset += used;
2320			rxm->full_len -= used;
2321			if (!desc->count)
2322				goto read_sock_requeue;
2323		} else {
2324			consume_skb(skb);
2325			if (!desc->count)
2326				skb = NULL;
2327		}
2328	} while (skb);
2329
2330read_sock_end:
2331	tls_rx_reader_release(sk, ctx);
2332	return copied ? : err;
2333
2334read_sock_requeue:
2335	__skb_queue_head(&ctx->rx_list, skb);
2336	goto read_sock_end;
2337}
2338
2339bool tls_sw_sock_is_readable(struct sock *sk)
2340{
2341	struct tls_context *tls_ctx = tls_get_ctx(sk);
2342	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2343	bool ingress_empty = true;
2344	struct sk_psock *psock;
2345
2346	rcu_read_lock();
2347	psock = sk_psock(sk);
2348	if (psock)
2349		ingress_empty = list_empty(&psock->ingress_msg);
2350	rcu_read_unlock();
2351
2352	return !ingress_empty || tls_strp_msg_ready(ctx) ||
2353		!skb_queue_empty(&ctx->rx_list);
2354}
2355
2356int tls_rx_msg_size(struct tls_strparser *strp, struct sk_buff *skb)
2357{
2358	struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
 
2359	struct tls_prot_info *prot = &tls_ctx->prot_info;
2360	char header[TLS_HEADER_SIZE + TLS_MAX_IV_SIZE];
 
2361	size_t cipher_overhead;
2362	size_t data_len = 0;
2363	int ret;
2364
2365	/* Verify that we have a full TLS header, or wait for more data */
2366	if (strp->stm.offset + prot->prepend_size > skb->len)
2367		return 0;
2368
2369	/* Sanity-check size of on-stack buffer. */
2370	if (WARN_ON(prot->prepend_size > sizeof(header))) {
2371		ret = -EINVAL;
2372		goto read_failure;
2373	}
2374
2375	/* Linearize header to local buffer */
2376	ret = skb_copy_bits(skb, strp->stm.offset, header, prot->prepend_size);
 
2377	if (ret < 0)
2378		goto read_failure;
2379
2380	strp->mark = header[0];
2381
2382	data_len = ((header[4] & 0xFF) | (header[3] << 8));
2383
2384	cipher_overhead = prot->tag_size;
2385	if (prot->version != TLS_1_3_VERSION &&
2386	    prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2387		cipher_overhead += prot->iv_size;
2388
2389	if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2390	    prot->tail_size) {
2391		ret = -EMSGSIZE;
2392		goto read_failure;
2393	}
2394	if (data_len < cipher_overhead) {
2395		ret = -EBADMSG;
2396		goto read_failure;
2397	}
2398
2399	/* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2400	if (header[1] != TLS_1_2_VERSION_MINOR ||
2401	    header[2] != TLS_1_2_VERSION_MAJOR) {
2402		ret = -EINVAL;
2403		goto read_failure;
2404	}
2405
2406	tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2407				     TCP_SKB_CB(skb)->seq + strp->stm.offset);
2408	return data_len + TLS_HEADER_SIZE;
2409
2410read_failure:
2411	tls_err_abort(strp->sk, ret);
2412
2413	return ret;
2414}
2415
2416void tls_rx_msg_ready(struct tls_strparser *strp)
2417{
2418	struct tls_sw_context_rx *ctx;
 
 
 
 
 
 
2419
2420	ctx = container_of(strp, struct tls_sw_context_rx, strp);
2421	ctx->saved_data_ready(strp->sk);
2422}
2423
2424static void tls_data_ready(struct sock *sk)
2425{
2426	struct tls_context *tls_ctx = tls_get_ctx(sk);
2427	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2428	struct sk_psock *psock;
2429	gfp_t alloc_save;
2430
2431	trace_sk_data_ready(sk);
2432
2433	alloc_save = sk->sk_allocation;
2434	sk->sk_allocation = GFP_ATOMIC;
2435	tls_strp_data_ready(&ctx->strp);
2436	sk->sk_allocation = alloc_save;
2437
2438	psock = sk_psock_get(sk);
2439	if (psock) {
2440		if (!list_empty(&psock->ingress_msg))
2441			ctx->saved_data_ready(sk);
2442		sk_psock_put(sk, psock);
2443	}
2444}
2445
2446void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2447{
2448	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2449
2450	set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2451	set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2452	cancel_delayed_work_sync(&ctx->tx_work.work);
2453}
2454
2455void tls_sw_release_resources_tx(struct sock *sk)
2456{
2457	struct tls_context *tls_ctx = tls_get_ctx(sk);
2458	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2459	struct tls_rec *rec, *tmp;
 
2460
2461	/* Wait for any pending async encryptions to complete */
2462	tls_encrypt_async_wait(ctx);
 
 
 
 
 
 
2463
2464	tls_tx_records(sk, -1);
2465
2466	/* Free up un-sent records in tx_list. First, free
2467	 * the partially sent record if any at head of tx_list.
2468	 */
2469	if (tls_ctx->partially_sent_record) {
2470		tls_free_partial_record(sk, tls_ctx);
2471		rec = list_first_entry(&ctx->tx_list,
2472				       struct tls_rec, list);
2473		list_del(&rec->list);
2474		sk_msg_free(sk, &rec->msg_plaintext);
2475		kfree(rec);
2476	}
2477
2478	list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2479		list_del(&rec->list);
2480		sk_msg_free(sk, &rec->msg_encrypted);
2481		sk_msg_free(sk, &rec->msg_plaintext);
2482		kfree(rec);
2483	}
2484
2485	crypto_free_aead(ctx->aead_send);
2486	tls_free_open_rec(sk);
2487}
2488
2489void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2490{
2491	struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2492
2493	kfree(ctx);
2494}
2495
2496void tls_sw_release_resources_rx(struct sock *sk)
2497{
2498	struct tls_context *tls_ctx = tls_get_ctx(sk);
2499	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2500
 
 
 
2501	if (ctx->aead_recv) {
2502		__skb_queue_purge(&ctx->rx_list);
 
 
2503		crypto_free_aead(ctx->aead_recv);
2504		tls_strp_stop(&ctx->strp);
2505		/* If tls_sw_strparser_arm() was not called (cleanup paths)
2506		 * we still want to tls_strp_stop(), but sk->sk_data_ready was
2507		 * never swapped.
2508		 */
2509		if (ctx->saved_data_ready) {
2510			write_lock_bh(&sk->sk_callback_lock);
2511			sk->sk_data_ready = ctx->saved_data_ready;
2512			write_unlock_bh(&sk->sk_callback_lock);
2513		}
2514	}
2515}
2516
2517void tls_sw_strparser_done(struct tls_context *tls_ctx)
2518{
2519	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2520
2521	tls_strp_done(&ctx->strp);
2522}
2523
2524void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2525{
2526	struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2527
2528	kfree(ctx);
2529}
2530
2531void tls_sw_free_resources_rx(struct sock *sk)
2532{
2533	struct tls_context *tls_ctx = tls_get_ctx(sk);
2534
2535	tls_sw_release_resources_rx(sk);
2536	tls_sw_free_ctx_rx(tls_ctx);
2537}
2538
2539/* The work handler to transmitt the encrypted records in tx_list */
2540static void tx_work_handler(struct work_struct *work)
2541{
2542	struct delayed_work *delayed_work = to_delayed_work(work);
2543	struct tx_work *tx_work = container_of(delayed_work,
2544					       struct tx_work, work);
2545	struct sock *sk = tx_work->sk;
2546	struct tls_context *tls_ctx = tls_get_ctx(sk);
2547	struct tls_sw_context_tx *ctx;
2548
2549	if (unlikely(!tls_ctx))
2550		return;
2551
2552	ctx = tls_sw_ctx_tx(tls_ctx);
2553	if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2554		return;
2555
2556	if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2557		return;
2558
2559	if (mutex_trylock(&tls_ctx->tx_lock)) {
2560		lock_sock(sk);
2561		tls_tx_records(sk, -1);
2562		release_sock(sk);
2563		mutex_unlock(&tls_ctx->tx_lock);
2564	} else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2565		/* Someone is holding the tx_lock, they will likely run Tx
2566		 * and cancel the work on their way out of the lock section.
2567		 * Schedule a long delay just in case.
2568		 */
2569		schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2570	}
2571}
2572
2573static bool tls_is_tx_ready(struct tls_sw_context_tx *ctx)
2574{
2575	struct tls_rec *rec;
2576
2577	rec = list_first_entry_or_null(&ctx->tx_list, struct tls_rec, list);
2578	if (!rec)
2579		return false;
2580
2581	return READ_ONCE(rec->tx_ready);
2582}
2583
2584void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2585{
2586	struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2587
2588	/* Schedule the transmission if tx list is ready */
2589	if (tls_is_tx_ready(tx_ctx) &&
2590	    !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2591		schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2592}
2593
2594void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2595{
2596	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2597
2598	write_lock_bh(&sk->sk_callback_lock);
2599	rx_ctx->saved_data_ready = sk->sk_data_ready;
2600	sk->sk_data_ready = tls_data_ready;
2601	write_unlock_bh(&sk->sk_callback_lock);
2602}
2603
2604void tls_update_rx_zc_capable(struct tls_context *tls_ctx)
2605{
2606	struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2607
2608	rx_ctx->zc_capable = tls_ctx->rx_no_pad ||
2609		tls_ctx->prot_info.version != TLS_1_3_VERSION;
2610}
2611
2612static struct tls_sw_context_tx *init_ctx_tx(struct tls_context *ctx, struct sock *sk)
2613{
2614	struct tls_sw_context_tx *sw_ctx_tx;
2615
2616	if (!ctx->priv_ctx_tx) {
2617		sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2618		if (!sw_ctx_tx)
2619			return NULL;
2620	} else {
2621		sw_ctx_tx = ctx->priv_ctx_tx;
2622	}
2623
2624	crypto_init_wait(&sw_ctx_tx->async_wait);
2625	atomic_set(&sw_ctx_tx->encrypt_pending, 1);
2626	INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2627	INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2628	sw_ctx_tx->tx_work.sk = sk;
2629
2630	return sw_ctx_tx;
2631}
2632
2633static struct tls_sw_context_rx *init_ctx_rx(struct tls_context *ctx)
2634{
2635	struct tls_sw_context_rx *sw_ctx_rx;
2636
2637	if (!ctx->priv_ctx_rx) {
2638		sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2639		if (!sw_ctx_rx)
2640			return NULL;
2641	} else {
2642		sw_ctx_rx = ctx->priv_ctx_rx;
2643	}
2644
2645	crypto_init_wait(&sw_ctx_rx->async_wait);
2646	atomic_set(&sw_ctx_rx->decrypt_pending, 1);
2647	init_waitqueue_head(&sw_ctx_rx->wq);
2648	skb_queue_head_init(&sw_ctx_rx->rx_list);
2649	skb_queue_head_init(&sw_ctx_rx->async_hold);
2650
2651	return sw_ctx_rx;
2652}
2653
2654int init_prot_info(struct tls_prot_info *prot,
2655		   const struct tls_crypto_info *crypto_info,
2656		   const struct tls_cipher_desc *cipher_desc)
2657{
2658	u16 nonce_size = cipher_desc->nonce;
2659
2660	if (crypto_info->version == TLS_1_3_VERSION) {
2661		nonce_size = 0;
2662		prot->aad_size = TLS_HEADER_SIZE;
2663		prot->tail_size = 1;
2664	} else {
2665		prot->aad_size = TLS_AAD_SPACE_SIZE;
2666		prot->tail_size = 0;
2667	}
2668
2669	/* Sanity-check the sizes for stack allocations. */
2670	if (nonce_size > TLS_MAX_IV_SIZE || prot->aad_size > TLS_MAX_AAD_SIZE)
2671		return -EINVAL;
2672
2673	prot->version = crypto_info->version;
2674	prot->cipher_type = crypto_info->cipher_type;
2675	prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2676	prot->tag_size = cipher_desc->tag;
2677	prot->overhead_size = prot->prepend_size + prot->tag_size + prot->tail_size;
2678	prot->iv_size = cipher_desc->iv;
2679	prot->salt_size = cipher_desc->salt;
2680	prot->rec_seq_size = cipher_desc->rec_seq;
2681
2682	return 0;
2683}
2684
2685int tls_set_sw_offload(struct sock *sk, int tx)
2686{
 
 
 
 
 
 
 
2687	struct tls_sw_context_tx *sw_ctx_tx = NULL;
2688	struct tls_sw_context_rx *sw_ctx_rx = NULL;
2689	const struct tls_cipher_desc *cipher_desc;
2690	struct tls_crypto_info *crypto_info;
2691	char *iv, *rec_seq, *key, *salt;
2692	struct cipher_context *cctx;
2693	struct tls_prot_info *prot;
2694	struct crypto_aead **aead;
2695	struct tls_context *ctx;
 
2696	struct crypto_tfm *tfm;
 
 
2697	int rc = 0;
2698
2699	ctx = tls_get_ctx(sk);
2700	prot = &ctx->prot_info;
 
 
2701
2702	if (tx) {
2703		ctx->priv_ctx_tx = init_ctx_tx(ctx, sk);
2704		if (!ctx->priv_ctx_tx)
2705			return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2706
2707		sw_ctx_tx = ctx->priv_ctx_tx;
 
 
2708		crypto_info = &ctx->crypto_send.info;
2709		cctx = &ctx->tx;
2710		aead = &sw_ctx_tx->aead_send;
 
 
 
2711	} else {
2712		ctx->priv_ctx_rx = init_ctx_rx(ctx);
2713		if (!ctx->priv_ctx_rx)
2714			return -ENOMEM;
2715
2716		sw_ctx_rx = ctx->priv_ctx_rx;
2717		crypto_info = &ctx->crypto_recv.info;
2718		cctx = &ctx->rx;
 
2719		aead = &sw_ctx_rx->aead_recv;
2720	}
2721
2722	cipher_desc = get_cipher_desc(crypto_info->cipher_type);
2723	if (!cipher_desc) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2724		rc = -EINVAL;
2725		goto free_priv;
2726	}
2727
2728	rc = init_prot_info(prot, crypto_info, cipher_desc);
2729	if (rc)
 
 
2730		goto free_priv;
 
2731
2732	iv = crypto_info_iv(crypto_info, cipher_desc);
2733	key = crypto_info_key(crypto_info, cipher_desc);
2734	salt = crypto_info_salt(crypto_info, cipher_desc);
2735	rec_seq = crypto_info_rec_seq(crypto_info, cipher_desc);
2736
2737	memcpy(cctx->iv, salt, cipher_desc->salt);
2738	memcpy(cctx->iv + cipher_desc->salt, iv, cipher_desc->iv);
2739	memcpy(cctx->rec_seq, rec_seq, cipher_desc->rec_seq);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2740
2741	if (!*aead) {
2742		*aead = crypto_alloc_aead(cipher_desc->cipher_name, 0, 0);
2743		if (IS_ERR(*aead)) {
2744			rc = PTR_ERR(*aead);
2745			*aead = NULL;
2746			goto free_priv;
2747		}
2748	}
2749
2750	ctx->push_pending_record = tls_sw_push_pending_record;
2751
2752	rc = crypto_aead_setkey(*aead, key, cipher_desc->key);
 
2753	if (rc)
2754		goto free_aead;
2755
2756	rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2757	if (rc)
2758		goto free_aead;
2759
2760	if (sw_ctx_rx) {
2761		tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2762
2763		tls_update_rx_zc_capable(ctx);
2764		sw_ctx_rx->async_capable =
2765			crypto_info->version != TLS_1_3_VERSION &&
2766			!!(tfm->__crt_alg->cra_flags & CRYPTO_ALG_ASYNC);
 
 
 
 
 
 
 
2767
2768		rc = tls_strp_init(&sw_ctx_rx->strp, sk);
2769		if (rc)
2770			goto free_aead;
2771	}
2772
2773	goto out;
2774
2775free_aead:
2776	crypto_free_aead(*aead);
2777	*aead = NULL;
 
 
 
 
 
 
2778free_priv:
2779	if (tx) {
2780		kfree(ctx->priv_ctx_tx);
2781		ctx->priv_ctx_tx = NULL;
2782	} else {
2783		kfree(ctx->priv_ctx_rx);
2784		ctx->priv_ctx_rx = NULL;
2785	}
2786out:
2787	return rc;
2788}