drivers/crypto/chelsio/chtls/chtls_io.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2018 Chelsio Communications, Inc.
   4  *
   5  * Written by: Atul Gupta (atul.gupta@chelsio.com)
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/list.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/timer.h>
  13 #include <linux/notifier.h>
  14 #include <linux/inetdevice.h>
  15 #include <linux/ip.h>
  16 #include <linux/tcp.h>
  17 #include <linux/sched/signal.h>
  18 #include <net/tcp.h>
  19 #include <net/busy_poll.h>
  20 #include <crypto/aes.h>
  21
  22 #include "chtls.h"
  23 #include "chtls_cm.h"
  24
  25 static bool is_tls_tx(struct chtls_sock *csk)
  26 {
  27         return csk->tlshws.txkey >= 0;
  28 }
  29
  30 static bool is_tls_rx(struct chtls_sock *csk)
  31 {
  32         return csk->tlshws.rxkey >= 0;
  33 }
  34
  35 static int data_sgl_len(const struct sk_buff *skb)
  36 {
  37         unsigned int cnt;
  38
  39         cnt = skb_shinfo(skb)->nr_frags;
  40         return sgl_len(cnt) * 8;
  41 }
  42
  43 static int nos_ivs(struct sock *sk, unsigned int size)
  44 {
  45         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
  46
  47         return DIV_ROUND_UP(size, csk->tlshws.mfs);
  48 }
  49
  50 static int set_ivs_imm(struct sock *sk, const struct sk_buff *skb)
  51 {
  52         int ivs_size = nos_ivs(sk, skb->len) * CIPHER_BLOCK_SIZE;
  53         int hlen = TLS_WR_CPL_LEN + data_sgl_len(skb);
  54
  55         if ((hlen + KEY_ON_MEM_SZ + ivs_size) <
  56             MAX_IMM_OFLD_TX_DATA_WR_LEN) {
  57                 ULP_SKB_CB(skb)->ulp.tls.iv = 1;
  58                 return 1;
  59         }
  60         ULP_SKB_CB(skb)->ulp.tls.iv = 0;
  61         return 0;
  62 }
  63
  64 static int max_ivs_size(struct sock *sk, int size)
  65 {
  66         return nos_ivs(sk, size) * CIPHER_BLOCK_SIZE;
  67 }
  68
  69 static int ivs_size(struct sock *sk, const struct sk_buff *skb)
  70 {
  71         return set_ivs_imm(sk, skb) ? (nos_ivs(sk, skb->len) *
  72                  CIPHER_BLOCK_SIZE) : 0;
  73 }
  74
  75 static int flowc_wr_credits(int nparams, int *flowclenp)
  76 {
  77         int flowclen16, flowclen;
  78
  79         flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
  80         flowclen16 = DIV_ROUND_UP(flowclen, 16);
  81         flowclen = flowclen16 * 16;
  82
  83         if (flowclenp)
  84                 *flowclenp = flowclen;
  85
  86         return flowclen16;
  87 }
  88
  89 static struct sk_buff *create_flowc_wr_skb(struct sock *sk,
  90                                            struct fw_flowc_wr *flowc,
  91                                            int flowclen)
  92 {
  93         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
  94         struct sk_buff *skb;
  95
  96         skb = alloc_skb(flowclen, GFP_ATOMIC);
  97         if (!skb)
  98                 return NULL;
  99
 100         memcpy(__skb_put(skb, flowclen), flowc, flowclen);
 101         skb_set_queue_mapping(skb, (csk->txq_idx << 1) | CPL_PRIORITY_DATA);
 102
 103         return skb;
 104 }
 105
 106 static int send_flowc_wr(struct sock *sk, struct fw_flowc_wr *flowc,
 107                          int flowclen)
 108 {
 109         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 110         struct tcp_sock *tp = tcp_sk(sk);
 111         struct sk_buff *skb;
 112         int flowclen16;
 113         int ret;
 114
 115         flowclen16 = flowclen / 16;
 116
 117         if (csk_flag(sk, CSK_TX_DATA_SENT)) {
 118                 skb = create_flowc_wr_skb(sk, flowc, flowclen);
 119                 if (!skb)
 120                         return -ENOMEM;
 121
 122                 skb_entail(sk, skb,
 123                            ULPCB_FLAG_NO_HDR | ULPCB_FLAG_NO_APPEND);
 124                 return 0;
 125         }
 126
 127         ret = cxgb4_immdata_send(csk->egress_dev,
 128                                  csk->txq_idx,
 129                                  flowc, flowclen);
 130         if (!ret)
 131                 return flowclen16;
 132         skb = create_flowc_wr_skb(sk, flowc, flowclen);
 133         if (!skb)
 134                 return -ENOMEM;
 135         send_or_defer(sk, tp, skb, 0);
 136         return flowclen16;
 137 }
 138
 139 static u8 tcp_state_to_flowc_state(u8 state)
 140 {
 141         switch (state) {
 142         case TCP_ESTABLISHED:
 143                 return FW_FLOWC_MNEM_TCPSTATE_ESTABLISHED;
 144         case TCP_CLOSE_WAIT:
 145                 return FW_FLOWC_MNEM_TCPSTATE_CLOSEWAIT;
 146         case TCP_FIN_WAIT1:
 147                 return FW_FLOWC_MNEM_TCPSTATE_FINWAIT1;
 148         case TCP_CLOSING:
 149                 return FW_FLOWC_MNEM_TCPSTATE_CLOSING;
 150         case TCP_LAST_ACK:
 151                 return FW_FLOWC_MNEM_TCPSTATE_LASTACK;
 152         case TCP_FIN_WAIT2:
 153                 return FW_FLOWC_MNEM_TCPSTATE_FINWAIT2;
 154         }
 155
 156         return FW_FLOWC_MNEM_TCPSTATE_ESTABLISHED;
 157 }
 158
 159 int send_tx_flowc_wr(struct sock *sk, int compl,
 160                      u32 snd_nxt, u32 rcv_nxt)
 161 {
 162         struct flowc_packed {
 163                 struct fw_flowc_wr fc;
 164                 struct fw_flowc_mnemval mnemval[FW_FLOWC_MNEM_MAX];
 165         } __packed sflowc;
 166         int nparams, paramidx, flowclen16, flowclen;
 167         struct fw_flowc_wr *flowc;
 168         struct chtls_sock *csk;
 169         struct tcp_sock *tp;
 170
 171         csk = rcu_dereference_sk_user_data(sk);
 172         tp = tcp_sk(sk);
 173         memset(&sflowc, 0, sizeof(sflowc));
 174         flowc = &sflowc.fc;
 175
 176 #define FLOWC_PARAM(__m, __v) \
 177         do { \
 178                 flowc->mnemval[paramidx].mnemonic = FW_FLOWC_MNEM_##__m; \
 179                 flowc->mnemval[paramidx].val = cpu_to_be32(__v); \
 180                 paramidx++; \
 181         } while (0)
 182
 183         paramidx = 0;
 184
 185         FLOWC_PARAM(PFNVFN, FW_PFVF_CMD_PFN_V(csk->cdev->lldi->pf));
 186         FLOWC_PARAM(CH, csk->tx_chan);
 187         FLOWC_PARAM(PORT, csk->tx_chan);
 188         FLOWC_PARAM(IQID, csk->rss_qid);
 189         FLOWC_PARAM(SNDNXT, tp->snd_nxt);
 190         FLOWC_PARAM(RCVNXT, tp->rcv_nxt);
 191         FLOWC_PARAM(SNDBUF, csk->sndbuf);
 192         FLOWC_PARAM(MSS, tp->mss_cache);
 193         FLOWC_PARAM(TCPSTATE, tcp_state_to_flowc_state(sk->sk_state));
 194
 195         if (SND_WSCALE(tp))
 196                 FLOWC_PARAM(RCV_SCALE, SND_WSCALE(tp));
 197
 198         if (csk->ulp_mode == ULP_MODE_TLS)
 199                 FLOWC_PARAM(ULD_MODE, ULP_MODE_TLS);
 200
 201         if (csk->tlshws.fcplenmax)
 202                 FLOWC_PARAM(TXDATAPLEN_MAX, csk->tlshws.fcplenmax);
 203
 204         nparams = paramidx;
 205 #undef FLOWC_PARAM
 206
 207         flowclen16 = flowc_wr_credits(nparams, &flowclen);
 208         flowc->op_to_nparams =
 209                 cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
 210                             FW_WR_COMPL_V(compl) |
 211                             FW_FLOWC_WR_NPARAMS_V(nparams));
 212         flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
 213                                           FW_WR_FLOWID_V(csk->tid));
 214
 215         return send_flowc_wr(sk, flowc, flowclen);
 216 }
 217
 218 /* Copy IVs to WR */
 219 static int tls_copy_ivs(struct sock *sk, struct sk_buff *skb)
 220
 221 {
 222         struct chtls_sock *csk;
 223         unsigned char *iv_loc;
 224         struct chtls_hws *hws;
 225         unsigned char *ivs;
 226         u16 number_of_ivs;
 227         struct page *page;
 228         int err = 0;
 229
 230         csk = rcu_dereference_sk_user_data(sk);
 231         hws = &csk->tlshws;
 232         number_of_ivs = nos_ivs(sk, skb->len);
 233
 234         if (number_of_ivs > MAX_IVS_PAGE) {
 235                 pr_warn("MAX IVs in PAGE exceeded %d\n", number_of_ivs);
 236                 return -ENOMEM;
 237         }
 238
 239         /* generate the  IVs */
 240         ivs = kmalloc_array(CIPHER_BLOCK_SIZE, number_of_ivs, GFP_ATOMIC);
 241         if (!ivs)
 242                 return -ENOMEM;
 243         get_random_bytes(ivs, number_of_ivs * CIPHER_BLOCK_SIZE);
 244
 245         if (skb_ulp_tls_iv_imm(skb)) {
 246                 /* send the IVs as immediate data in the WR */
 247                 iv_loc = (unsigned char *)__skb_push(skb, number_of_ivs *
 248                                                 CIPHER_BLOCK_SIZE);
 249                 if (iv_loc)
 250                         memcpy(iv_loc, ivs, number_of_ivs * CIPHER_BLOCK_SIZE);
 251
 252                 hws->ivsize = number_of_ivs * CIPHER_BLOCK_SIZE;
 253         } else {
 254                 /* Send the IVs as sgls */
 255                 /* Already accounted IV DSGL for credits */
 256                 skb_shinfo(skb)->nr_frags--;
 257                 page = alloc_pages(sk->sk_allocation | __GFP_COMP, 0);
 258                 if (!page) {
 259                         pr_info("%s : Page allocation for IVs failed\n",
 260                                 __func__);
 261                         err = -ENOMEM;
 262                         goto out;
 263                 }
 264                 memcpy(page_address(page), ivs, number_of_ivs *
 265                        CIPHER_BLOCK_SIZE);
 266                 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, page, 0,
 267                                    number_of_ivs * CIPHER_BLOCK_SIZE);
 268                 hws->ivsize = 0;
 269         }
 270 out:
 271         kfree(ivs);
 272         return err;
 273 }
 274
 275 /* Copy Key to WR */
 276 static void tls_copy_tx_key(struct sock *sk, struct sk_buff *skb)
 277 {
 278         struct ulptx_sc_memrd *sc_memrd;
 279         struct chtls_sock *csk;
 280         struct chtls_dev *cdev;
 281         struct ulptx_idata *sc;
 282         struct chtls_hws *hws;
 283         u32 immdlen;
 284         int kaddr;
 285
 286         csk = rcu_dereference_sk_user_data(sk);
 287         hws = &csk->tlshws;
 288         cdev = csk->cdev;
 289
 290         immdlen = sizeof(*sc) + sizeof(*sc_memrd);
 291         kaddr = keyid_to_addr(cdev->kmap.start, hws->txkey);
 292         sc = (struct ulptx_idata *)__skb_push(skb, immdlen);
 293         if (sc) {
 294                 sc->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_NOOP));
 295                 sc->len = htonl(0);
 296                 sc_memrd = (struct ulptx_sc_memrd *)(sc + 1);
 297                 sc_memrd->cmd_to_len =
 298                                 htonl(ULPTX_CMD_V(ULP_TX_SC_MEMRD) |
 299                                 ULP_TX_SC_MORE_V(1) |
 300                                 ULPTX_LEN16_V(hws->keylen >> 4));
 301                 sc_memrd->addr = htonl(kaddr);
 302         }
 303 }
 304
 305 static u64 tlstx_incr_seqnum(struct chtls_hws *hws)
 306 {
 307         return hws->tx_seq_no++;
 308 }
 309
 310 static bool is_sg_request(const struct sk_buff *skb)
 311 {
 312         return skb->peeked ||
 313                 (skb->len > MAX_IMM_ULPTX_WR_LEN);
 314 }
 315
 316 /*
 317  * Returns true if an sk_buff carries urgent data.
 318  */
 319 static bool skb_urgent(struct sk_buff *skb)
 320 {
 321         return ULP_SKB_CB(skb)->flags & ULPCB_FLAG_URG;
 322 }
 323
 324 /* TLS content type for CPL SFO */
 325 static unsigned char tls_content_type(unsigned char content_type)
 326 {
 327         switch (content_type) {
 328         case TLS_HDR_TYPE_CCS:
 329                 return CPL_TX_TLS_SFO_TYPE_CCS;
 330         case TLS_HDR_TYPE_ALERT:
 331                 return CPL_TX_TLS_SFO_TYPE_ALERT;
 332         case TLS_HDR_TYPE_HANDSHAKE:
 333                 return CPL_TX_TLS_SFO_TYPE_HANDSHAKE;
 334         case TLS_HDR_TYPE_HEARTBEAT:
 335                 return CPL_TX_TLS_SFO_TYPE_HEARTBEAT;
 336         }
 337         return CPL_TX_TLS_SFO_TYPE_DATA;
 338 }
 339
 340 static void tls_tx_data_wr(struct sock *sk, struct sk_buff *skb,
 341                            int dlen, int tls_immd, u32 credits,
 342                            int expn, int pdus)
 343 {
 344         struct fw_tlstx_data_wr *req_wr;
 345         struct cpl_tx_tls_sfo *req_cpl;
 346         unsigned int wr_ulp_mode_force;
 347         struct tls_scmd *updated_scmd;
 348         unsigned char data_type;
 349         struct chtls_sock *csk;
 350         struct net_device *dev;
 351         struct chtls_hws *hws;
 352         struct tls_scmd *scmd;
 353         struct adapter *adap;
 354         unsigned char *req;
 355         int immd_len;
 356         int iv_imm;
 357         int len;
 358
 359         csk = rcu_dereference_sk_user_data(sk);
 360         iv_imm = skb_ulp_tls_iv_imm(skb);
 361         dev = csk->egress_dev;
 362         adap = netdev2adap(dev);
 363         hws = &csk->tlshws;
 364         scmd = &hws->scmd;
 365         len = dlen + expn;
 366
 367         dlen = (dlen < hws->mfs) ? dlen : hws->mfs;
 368         atomic_inc(&adap->chcr_stats.tls_pdu_tx);
 369
 370         updated_scmd = scmd;
 371         updated_scmd->seqno_numivs &= 0xffffff80;
 372         updated_scmd->seqno_numivs |= SCMD_NUM_IVS_V(pdus);
 373         hws->scmd = *updated_scmd;
 374
 375         req = (unsigned char *)__skb_push(skb, sizeof(struct cpl_tx_tls_sfo));
 376         req_cpl = (struct cpl_tx_tls_sfo *)req;
 377         req = (unsigned char *)__skb_push(skb, (sizeof(struct
 378                                 fw_tlstx_data_wr)));
 379
 380         req_wr = (struct fw_tlstx_data_wr *)req;
 381         immd_len = (tls_immd ? dlen : 0);
 382         req_wr->op_to_immdlen =
 383                 htonl(FW_WR_OP_V(FW_TLSTX_DATA_WR) |
 384                 FW_TLSTX_DATA_WR_COMPL_V(1) |
 385                 FW_TLSTX_DATA_WR_IMMDLEN_V(immd_len));
 386         req_wr->flowid_len16 = htonl(FW_TLSTX_DATA_WR_FLOWID_V(csk->tid) |
 387                                      FW_TLSTX_DATA_WR_LEN16_V(credits));
 388         wr_ulp_mode_force = TX_ULP_MODE_V(ULP_MODE_TLS);
 389
 390         if (is_sg_request(skb))
 391                 wr_ulp_mode_force |= FW_OFLD_TX_DATA_WR_ALIGNPLD_F |
 392                         ((tcp_sk(sk)->nonagle & TCP_NAGLE_OFF) ? 0 :
 393                         FW_OFLD_TX_DATA_WR_SHOVE_F);
 394
 395         req_wr->lsodisable_to_flags =
 396                         htonl(TX_ULP_MODE_V(ULP_MODE_TLS) |
 397                               TX_URG_V(skb_urgent(skb)) |
 398                               T6_TX_FORCE_F | wr_ulp_mode_force |
 399                               TX_SHOVE_V((!csk_flag(sk, CSK_TX_MORE_DATA)) &&
 400                                          skb_queue_empty(&csk->txq)));
 401
 402         req_wr->ctxloc_to_exp =
 403                         htonl(FW_TLSTX_DATA_WR_NUMIVS_V(pdus) |
 404                               FW_TLSTX_DATA_WR_EXP_V(expn) |
 405                               FW_TLSTX_DATA_WR_CTXLOC_V(CHTLS_KEY_CONTEXT_DDR) |
 406                               FW_TLSTX_DATA_WR_IVDSGL_V(!iv_imm) |
 407                               FW_TLSTX_DATA_WR_KEYSIZE_V(hws->keylen >> 4));
 408
 409         /* Fill in the length */
 410         req_wr->plen = htonl(len);
 411         req_wr->mfs = htons(hws->mfs);
 412         req_wr->adjustedplen_pkd =
 413                 htons(FW_TLSTX_DATA_WR_ADJUSTEDPLEN_V(hws->adjustlen));
 414         req_wr->expinplenmax_pkd =
 415                 htons(FW_TLSTX_DATA_WR_EXPINPLENMAX_V(hws->expansion));
 416         req_wr->pdusinplenmax_pkd =
 417                 FW_TLSTX_DATA_WR_PDUSINPLENMAX_V(hws->pdus);
 418         req_wr->r10 = 0;
 419
 420         data_type = tls_content_type(ULP_SKB_CB(skb)->ulp.tls.type);
 421         req_cpl->op_to_seg_len = htonl(CPL_TX_TLS_SFO_OPCODE_V(CPL_TX_TLS_SFO) |
 422                                        CPL_TX_TLS_SFO_DATA_TYPE_V(data_type) |
 423                                        CPL_TX_TLS_SFO_CPL_LEN_V(2) |
 424                                        CPL_TX_TLS_SFO_SEG_LEN_V(dlen));
 425         req_cpl->pld_len = htonl(len - expn);
 426
 427         req_cpl->type_protover = htonl(CPL_TX_TLS_SFO_TYPE_V
 428                 ((data_type == CPL_TX_TLS_SFO_TYPE_HEARTBEAT) ?
 429                 TLS_HDR_TYPE_HEARTBEAT : 0) |
 430                 CPL_TX_TLS_SFO_PROTOVER_V(0));
 431
 432         /* create the s-command */
 433         req_cpl->r1_lo = 0;
 434         req_cpl->seqno_numivs  = cpu_to_be32(hws->scmd.seqno_numivs);
 435         req_cpl->ivgen_hdrlen = cpu_to_be32(hws->scmd.ivgen_hdrlen);
 436         req_cpl->scmd1 = cpu_to_be64(tlstx_incr_seqnum(hws));
 437 }
 438
 439 /*
 440  * Calculate the TLS data expansion size
 441  */
 442 static int chtls_expansion_size(struct sock *sk, int data_len,
 443                                 int fullpdu,
 444                                 unsigned short *pducnt)
 445 {
 446         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 447         struct chtls_hws *hws = &csk->tlshws;
 448         struct tls_scmd *scmd = &hws->scmd;
 449         int fragsize = hws->mfs;
 450         int expnsize = 0;
 451         int fragleft;
 452         int fragcnt;
 453         int expppdu;
 454
 455         if (SCMD_CIPH_MODE_G(scmd->seqno_numivs) ==
 456             SCMD_CIPH_MODE_AES_GCM) {
 457                 expppdu = GCM_TAG_SIZE + AEAD_EXPLICIT_DATA_SIZE +
 458                           TLS_HEADER_LENGTH;
 459
 460                 if (fullpdu) {
 461                         *pducnt = data_len / (expppdu + fragsize);
 462                         if (*pducnt > 32)
 463                                 *pducnt = 32;
 464                         else if (!*pducnt)
 465                                 *pducnt = 1;
 466                         expnsize = (*pducnt) * expppdu;
 467                         return expnsize;
 468                 }
 469                 fragcnt = (data_len / fragsize);
 470                 expnsize =  fragcnt * expppdu;
 471                 fragleft = data_len % fragsize;
 472                 if (fragleft > 0)
 473                         expnsize += expppdu;
 474         }
 475         return expnsize;
 476 }
 477
 478 /* WR with IV, KEY and CPL SFO added */
 479 static void make_tlstx_data_wr(struct sock *sk, struct sk_buff *skb,
 480                                int tls_tx_imm, int tls_len, u32 credits)
 481 {
 482         unsigned short pdus_per_ulp = 0;
 483         struct chtls_sock *csk;
 484         struct chtls_hws *hws;
 485         int expn_sz;
 486         int pdus;
 487
 488         csk = rcu_dereference_sk_user_data(sk);
 489         hws = &csk->tlshws;
 490         pdus = DIV_ROUND_UP(tls_len, hws->mfs);
 491         expn_sz = chtls_expansion_size(sk, tls_len, 0, NULL);
 492         if (!hws->compute) {
 493                 hws->expansion = chtls_expansion_size(sk,
 494                                                       hws->fcplenmax,
 495                                                       1, &pdus_per_ulp);
 496                 hws->pdus = pdus_per_ulp;
 497                 hws->adjustlen = hws->pdus *
 498                         ((hws->expansion / hws->pdus) + hws->mfs);
 499                 hws->compute = 1;
 500         }
 501         if (tls_copy_ivs(sk, skb))
 502                 return;
 503         tls_copy_tx_key(sk, skb);
 504         tls_tx_data_wr(sk, skb, tls_len, tls_tx_imm, credits, expn_sz, pdus);
 505         hws->tx_seq_no += (pdus - 1);
 506 }
 507
 508 static void make_tx_data_wr(struct sock *sk, struct sk_buff *skb,
 509                             unsigned int immdlen, int len,
 510                             u32 credits, u32 compl)
 511 {
 512         struct fw_ofld_tx_data_wr *req;
 513         unsigned int wr_ulp_mode_force;
 514         struct chtls_sock *csk;
 515         unsigned int opcode;
 516
 517         csk = rcu_dereference_sk_user_data(sk);
 518         opcode = FW_OFLD_TX_DATA_WR;
 519
 520         req = (struct fw_ofld_tx_data_wr *)__skb_push(skb, sizeof(*req));
 521         req->op_to_immdlen = htonl(WR_OP_V(opcode) |
 522                                 FW_WR_COMPL_V(compl) |
 523                                 FW_WR_IMMDLEN_V(immdlen));
 524         req->flowid_len16 = htonl(FW_WR_FLOWID_V(csk->tid) |
 525                                 FW_WR_LEN16_V(credits));
 526
 527         wr_ulp_mode_force = TX_ULP_MODE_V(csk->ulp_mode);
 528         if (is_sg_request(skb))
 529                 wr_ulp_mode_force |= FW_OFLD_TX_DATA_WR_ALIGNPLD_F |
 530                         ((tcp_sk(sk)->nonagle & TCP_NAGLE_OFF) ? 0 :
 531                                 FW_OFLD_TX_DATA_WR_SHOVE_F);
 532
 533         req->tunnel_to_proxy = htonl(wr_ulp_mode_force |
 534                         TX_URG_V(skb_urgent(skb)) |
 535                         TX_SHOVE_V((!csk_flag(sk, CSK_TX_MORE_DATA)) &&
 536                                    skb_queue_empty(&csk->txq)));
 537         req->plen = htonl(len);
 538 }
 539
 540 static int chtls_wr_size(struct chtls_sock *csk, const struct sk_buff *skb,
 541                          bool size)
 542 {
 543         int wr_size;
 544
 545         wr_size = TLS_WR_CPL_LEN;
 546         wr_size += KEY_ON_MEM_SZ;
 547         wr_size += ivs_size(csk->sk, skb);
 548
 549         if (size)
 550                 return wr_size;
 551
 552         /* frags counted for IV dsgl */
 553         if (!skb_ulp_tls_iv_imm(skb))
 554                 skb_shinfo(skb)->nr_frags++;
 555
 556         return wr_size;
 557 }
 558
 559 static bool is_ofld_imm(struct chtls_sock *csk, const struct sk_buff *skb)
 560 {
 561         int length = skb->len;
 562
 563         if (skb->peeked || skb->len > MAX_IMM_ULPTX_WR_LEN)
 564                 return false;
 565
 566         if (likely(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NEED_HDR)) {
 567                 /* Check TLS header len for Immediate */
 568                 if (csk->ulp_mode == ULP_MODE_TLS &&
 569                     skb_ulp_tls_inline(skb))
 570                         length += chtls_wr_size(csk, skb, true);
 571                 else
 572                         length += sizeof(struct fw_ofld_tx_data_wr);
 573
 574                 return length <= MAX_IMM_OFLD_TX_DATA_WR_LEN;
 575         }
 576         return true;
 577 }
 578
 579 static unsigned int calc_tx_flits(const struct sk_buff *skb,
 580                                   unsigned int immdlen)
 581 {
 582         unsigned int flits, cnt;
 583
 584         flits = immdlen / 8;   /* headers */
 585         cnt = skb_shinfo(skb)->nr_frags;
 586         if (skb_tail_pointer(skb) != skb_transport_header(skb))
 587                 cnt++;
 588         return flits + sgl_len(cnt);
 589 }
 590
 591 static void arp_failure_discard(void *handle, struct sk_buff *skb)
 592 {
 593         kfree_skb(skb);
 594 }
 595
 596 int chtls_push_frames(struct chtls_sock *csk, int comp)
 597 {
 598         struct chtls_hws *hws = &csk->tlshws;
 599         struct tcp_sock *tp;
 600         struct sk_buff *skb;
 601         int total_size = 0;
 602         struct sock *sk;
 603         int wr_size;
 604
 605         wr_size = sizeof(struct fw_ofld_tx_data_wr);
 606         sk = csk->sk;
 607         tp = tcp_sk(sk);
 608
 609         if (unlikely(sk_in_state(sk, TCPF_SYN_SENT | TCPF_CLOSE)))
 610                 return 0;
 611
 612         if (unlikely(csk_flag(sk, CSK_ABORT_SHUTDOWN)))
 613                 return 0;
 614
 615         while (csk->wr_credits && (skb = skb_peek(&csk->txq)) &&
 616                (!(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_HOLD) ||
 617                 skb_queue_len(&csk->txq) > 1)) {
 618                 unsigned int credit_len = skb->len;
 619                 unsigned int credits_needed;
 620                 unsigned int completion = 0;
 621                 int tls_len = skb->len;/* TLS data len before IV/key */
 622                 unsigned int immdlen;
 623                 int len = skb->len;    /* length [ulp bytes] inserted by hw */
 624                 int flowclen16 = 0;
 625                 int tls_tx_imm = 0;
 626
 627                 immdlen = skb->len;
 628                 if (!is_ofld_imm(csk, skb)) {
 629                         immdlen = skb_transport_offset(skb);
 630                         if (skb_ulp_tls_inline(skb))
 631                                 wr_size = chtls_wr_size(csk, skb, false);
 632                         credit_len = 8 * calc_tx_flits(skb, immdlen);
 633                 } else {
 634                         if (skb_ulp_tls_inline(skb)) {
 635                                 wr_size = chtls_wr_size(csk, skb, false);
 636                                 tls_tx_imm = 1;
 637                         }
 638                 }
 639                 if (likely(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NEED_HDR))
 640                         credit_len += wr_size;
 641                 credits_needed = DIV_ROUND_UP(credit_len, 16);
 642                 if (!csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
 643                         flowclen16 = send_tx_flowc_wr(sk, 1, tp->snd_nxt,
 644                                                       tp->rcv_nxt);
 645                         if (flowclen16 <= 0)
 646                                 break;
 647                         csk->wr_credits -= flowclen16;
 648                         csk->wr_unacked += flowclen16;
 649                         csk->wr_nondata += flowclen16;
 650                         csk_set_flag(csk, CSK_TX_DATA_SENT);
 651                 }
 652
 653                 if (csk->wr_credits < credits_needed) {
 654                         if (skb_ulp_tls_inline(skb) &&
 655                             !skb_ulp_tls_iv_imm(skb))
 656                                 skb_shinfo(skb)->nr_frags--;
 657                         break;
 658                 }
 659
 660                 __skb_unlink(skb, &csk->txq);
 661                 skb_set_queue_mapping(skb, (csk->txq_idx << 1) |
 662                                       CPL_PRIORITY_DATA);
 663                 if (hws->ofld)
 664                         hws->txqid = (skb->queue_mapping >> 1);
 665                 skb->csum = (__force __wsum)(credits_needed + csk->wr_nondata);
 666                 csk->wr_credits -= credits_needed;
 667                 csk->wr_unacked += credits_needed;
 668                 csk->wr_nondata = 0;
 669                 enqueue_wr(csk, skb);
 670
 671                 if (likely(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NEED_HDR)) {
 672                         if ((comp && csk->wr_unacked == credits_needed) ||
 673                             (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_COMPL) ||
 674                             csk->wr_unacked >= csk->wr_max_credits / 2) {
 675                                 completion = 1;
 676                                 csk->wr_unacked = 0;
 677                         }
 678                         if (skb_ulp_tls_inline(skb))
 679                                 make_tlstx_data_wr(sk, skb, tls_tx_imm,
 680                                                    tls_len, credits_needed);
 681                         else
 682                                 make_tx_data_wr(sk, skb, immdlen, len,
 683                                                 credits_needed, completion);
 684                         tp->snd_nxt += len;
 685                         tp->lsndtime = tcp_time_stamp(tp);
 686                         if (completion)
 687                                 ULP_SKB_CB(skb)->flags &= ~ULPCB_FLAG_NEED_HDR;
 688                 } else {
 689                         struct cpl_close_con_req *req = cplhdr(skb);
 690                         unsigned int cmd  = CPL_OPCODE_G(ntohl
 691                                              (OPCODE_TID(req)));
 692
 693                         if (cmd == CPL_CLOSE_CON_REQ)
 694                                 csk_set_flag(csk,
 695                                              CSK_CLOSE_CON_REQUESTED);
 696
 697                         if ((ULP_SKB_CB(skb)->flags & ULPCB_FLAG_COMPL) &&
 698                             (csk->wr_unacked >= csk->wr_max_credits / 2)) {
 699                                 req->wr.wr_hi |= htonl(FW_WR_COMPL_F);
 700                                 csk->wr_unacked = 0;
 701                         }
 702                 }
 703                 total_size += skb->truesize;
 704                 if (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_BARRIER)
 705                         csk_set_flag(csk, CSK_TX_WAIT_IDLE);
 706                 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
 707                 cxgb4_l2t_send(csk->egress_dev, skb, csk->l2t_entry);
 708         }
 709         sk->sk_wmem_queued -= total_size;
 710         return total_size;
 711 }
 712
 713 static void mark_urg(struct tcp_sock *tp, int flags,
 714                      struct sk_buff *skb)
 715 {
 716         if (unlikely(flags & MSG_OOB)) {
 717                 tp->snd_up = tp->write_seq;
 718                 ULP_SKB_CB(skb)->flags = ULPCB_FLAG_URG |
 719                                          ULPCB_FLAG_BARRIER |
 720                                          ULPCB_FLAG_NO_APPEND |
 721                                          ULPCB_FLAG_NEED_HDR;
 722         }
 723 }
 724
 725 /*
 726  * Returns true if a connection should send more data to TCP engine
 727  */
 728 static bool should_push(struct sock *sk)
 729 {
 730         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 731         struct chtls_dev *cdev = csk->cdev;
 732         struct tcp_sock *tp = tcp_sk(sk);
 733
 734         /*
 735          * If we've released our offload resources there's nothing to do ...
 736          */
 737         if (!cdev)
 738                 return false;
 739
 740         /*
 741          * If there aren't any work requests in flight, or there isn't enough
 742          * data in flight, or Nagle is off then send the current TX_DATA
 743          * otherwise hold it and wait to accumulate more data.
 744          */
 745         return csk->wr_credits == csk->wr_max_credits ||
 746                 (tp->nonagle & TCP_NAGLE_OFF);
 747 }
 748
 749 /*
 750  * Returns true if a TCP socket is corked.
 751  */
 752 static bool corked(const struct tcp_sock *tp, int flags)
 753 {
 754         return (flags & MSG_MORE) || (tp->nonagle & TCP_NAGLE_CORK);
 755 }
 756
 757 /*
 758  * Returns true if a send should try to push new data.
 759  */
 760 static bool send_should_push(struct sock *sk, int flags)
 761 {
 762         return should_push(sk) && !corked(tcp_sk(sk), flags);
 763 }
 764
 765 void chtls_tcp_push(struct sock *sk, int flags)
 766 {
 767         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 768         int qlen = skb_queue_len(&csk->txq);
 769
 770         if (likely(qlen)) {
 771                 struct sk_buff *skb = skb_peek_tail(&csk->txq);
 772                 struct tcp_sock *tp = tcp_sk(sk);
 773
 774                 mark_urg(tp, flags, skb);
 775
 776                 if (!(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND) &&
 777                     corked(tp, flags)) {
 778                         ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_HOLD;
 779                         return;
 780                 }
 781
 782                 ULP_SKB_CB(skb)->flags &= ~ULPCB_FLAG_HOLD;
 783                 if (qlen == 1 &&
 784                     ((ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND) ||
 785                      should_push(sk)))
 786                         chtls_push_frames(csk, 1);
 787         }
 788 }
 789
 790 /*
 791  * Calculate the size for a new send sk_buff.  It's maximum size so we can
 792  * pack lots of data into it, unless we plan to send it immediately, in which
 793  * case we size it more tightly.
 794  *
 795  * Note: we don't bother compensating for MSS < PAGE_SIZE because it doesn't
 796  * arise in normal cases and when it does we are just wasting memory.
 797  */
 798 static int select_size(struct sock *sk, int io_len, int flags, int len)
 799 {
 800         const int pgbreak = SKB_MAX_HEAD(len);
 801
 802         /*
 803          * If the data wouldn't fit in the main body anyway, put only the
 804          * header in the main body so it can use immediate data and place all
 805          * the payload in page fragments.
 806          */
 807         if (io_len > pgbreak)
 808                 return 0;
 809
 810         /*
 811          * If we will be accumulating payload get a large main body.
 812          */
 813         if (!send_should_push(sk, flags))
 814                 return pgbreak;
 815
 816         return io_len;
 817 }
 818
 819 void skb_entail(struct sock *sk, struct sk_buff *skb, int flags)
 820 {
 821         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 822         struct tcp_sock *tp = tcp_sk(sk);
 823
 824         ULP_SKB_CB(skb)->seq = tp->write_seq;
 825         ULP_SKB_CB(skb)->flags = flags;
 826         __skb_queue_tail(&csk->txq, skb);
 827         sk->sk_wmem_queued += skb->truesize;
 828
 829         if (TCP_PAGE(sk) && TCP_OFF(sk)) {
 830                 put_page(TCP_PAGE(sk));
 831                 TCP_PAGE(sk) = NULL;
 832                 TCP_OFF(sk) = 0;
 833         }
 834 }
 835
 836 static struct sk_buff *get_tx_skb(struct sock *sk, int size)
 837 {
 838         struct sk_buff *skb;
 839
 840         skb = alloc_skb(size + TX_HEADER_LEN, sk->sk_allocation);
 841         if (likely(skb)) {
 842                 skb_reserve(skb, TX_HEADER_LEN);
 843                 skb_entail(sk, skb, ULPCB_FLAG_NEED_HDR);
 844                 skb_reset_transport_header(skb);
 845         }
 846         return skb;
 847 }
 848
 849 static struct sk_buff *get_record_skb(struct sock *sk, int size, bool zcopy)
 850 {
 851         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 852         struct sk_buff *skb;
 853
 854         skb = alloc_skb(((zcopy ? 0 : size) + TX_TLSHDR_LEN +
 855                         KEY_ON_MEM_SZ + max_ivs_size(sk, size)),
 856                         sk->sk_allocation);
 857         if (likely(skb)) {
 858                 skb_reserve(skb, (TX_TLSHDR_LEN +
 859                             KEY_ON_MEM_SZ + max_ivs_size(sk, size)));
 860                 skb_entail(sk, skb, ULPCB_FLAG_NEED_HDR);
 861                 skb_reset_transport_header(skb);
 862                 ULP_SKB_CB(skb)->ulp.tls.ofld = 1;
 863                 ULP_SKB_CB(skb)->ulp.tls.type = csk->tlshws.type;
 864         }
 865         return skb;
 866 }
 867
 868 static void tx_skb_finalize(struct sk_buff *skb)
 869 {
 870         struct ulp_skb_cb *cb = ULP_SKB_CB(skb);
 871
 872         if (!(cb->flags & ULPCB_FLAG_NO_HDR))
 873                 cb->flags = ULPCB_FLAG_NEED_HDR;
 874         cb->flags |= ULPCB_FLAG_NO_APPEND;
 875 }
 876
 877 static void push_frames_if_head(struct sock *sk)
 878 {
 879         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 880
 881         if (skb_queue_len(&csk->txq) == 1)
 882                 chtls_push_frames(csk, 1);
 883 }
 884
 885 static int chtls_skb_copy_to_page_nocache(struct sock *sk,
 886                                           struct iov_iter *from,
 887                                           struct sk_buff *skb,
 888                                           struct page *page,
 889                                           int off, int copy)
 890 {
 891         int err;
 892
 893         err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) +
 894                                        off, copy, skb->len);
 895         if (err)
 896                 return err;
 897
 898         skb->len             += copy;
 899         skb->data_len        += copy;
 900         skb->truesize        += copy;
 901         sk->sk_wmem_queued   += copy;
 902         return 0;
 903 }
 904
 905 /* Read TLS header to find content type and data length */
 906 static int tls_header_read(struct tls_hdr *thdr, struct iov_iter *from)
 907 {
 908         if (copy_from_iter(thdr, sizeof(*thdr), from) != sizeof(*thdr))
 909                 return -EFAULT;
 910         return (__force int)cpu_to_be16(thdr->length);
 911 }
 912
 913 static int csk_mem_free(struct chtls_dev *cdev, struct sock *sk)
 914 {
 915         return (cdev->max_host_sndbuf - sk->sk_wmem_queued);
 916 }
 917
 918 static int csk_wait_memory(struct chtls_dev *cdev,
 919                            struct sock *sk, long *timeo_p)
 920 {
 921         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 922         int err = 0;
 923         long current_timeo;
 924         long vm_wait = 0;
 925         bool noblock;
 926
 927         current_timeo = *timeo_p;
 928         noblock = (*timeo_p ? false : true);
 929         if (csk_mem_free(cdev, sk)) {
 930                 current_timeo = (prandom_u32() % (HZ / 5)) + 2;
 931                 vm_wait = (prandom_u32() % (HZ / 5)) + 2;
 932         }
 933
 934         add_wait_queue(sk_sleep(sk), &wait);
 935         while (1) {
 936                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 937
 938                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 939                         goto do_error;
 940                 if (!*timeo_p) {
 941                         if (noblock)
 942                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 943                         goto do_nonblock;
 944                 }
 945                 if (signal_pending(current))
 946                         goto do_interrupted;
 947                 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 948                 if (csk_mem_free(cdev, sk) && !vm_wait)
 949                         break;
 950
 951                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 952                 sk->sk_write_pending++;
 953                 sk_wait_event(sk, &current_timeo, sk->sk_err ||
 954                               (sk->sk_shutdown & SEND_SHUTDOWN) ||
 955                               (csk_mem_free(cdev, sk) && !vm_wait), &wait);
 956                 sk->sk_write_pending--;
 957
 958                 if (vm_wait) {
 959                         vm_wait -= current_timeo;
 960                         current_timeo = *timeo_p;
 961                         if (current_timeo != MAX_SCHEDULE_TIMEOUT) {
 962                                 current_timeo -= vm_wait;
 963                                 if (current_timeo < 0)
 964                                         current_timeo = 0;
 965                         }
 966                         vm_wait = 0;
 967                 }
 968                 *timeo_p = current_timeo;
 969         }
 970 do_rm_wq:
 971         remove_wait_queue(sk_sleep(sk), &wait);
 972         return err;
 973 do_error:
 974         err = -EPIPE;
 975         goto do_rm_wq;
 976 do_nonblock:
 977         err = -EAGAIN;
 978         goto do_rm_wq;
 979 do_interrupted:
 980         err = sock_intr_errno(*timeo_p);
 981         goto do_rm_wq;
 982 }
 983
 984 int chtls_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 985 {
 986         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 987         struct chtls_dev *cdev = csk->cdev;
 988         struct tcp_sock *tp = tcp_sk(sk);
 989         struct sk_buff *skb;
 990         int mss, flags, err;
 991         int recordsz = 0;
 992         int copied = 0;
 993         long timeo;
 994
 995         lock_sock(sk);
 996         flags = msg->msg_flags;
 997         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 998
 999         if (!sk_in_state(sk, TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1000                 err = sk_stream_wait_connect(sk, &timeo);
1001                 if (err)
1002                         goto out_err;
1003         }
1004
1005         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1006         err = -EPIPE;
1007         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1008                 goto out_err;
1009
1010         mss = csk->mss;
1011         csk_set_flag(csk, CSK_TX_MORE_DATA);
1012
1013         while (msg_data_left(msg)) {
1014                 int copy = 0;
1015
1016                 skb = skb_peek_tail(&csk->txq);
1017                 if (skb) {
1018                         copy = mss - skb->len;
1019                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1020                 }
1021                 if (!csk_mem_free(cdev, sk))
1022                         goto wait_for_sndbuf;
1023
1024                 if (is_tls_tx(csk) && !csk->tlshws.txleft) {
1025                         struct tls_hdr hdr;
1026
1027                         recordsz = tls_header_read(&hdr, &msg->msg_iter);
1028                         size -= TLS_HEADER_LENGTH;
1029                         copied += TLS_HEADER_LENGTH;
1030                         csk->tlshws.txleft = recordsz;
1031                         csk->tlshws.type = hdr.type;
1032                         if (skb)
1033                                 ULP_SKB_CB(skb)->ulp.tls.type = hdr.type;
1034                 }
1035
1036                 if (!skb || (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND) ||
1037                     copy <= 0) {
1038 new_buf:
1039                         if (skb) {
1040                                 tx_skb_finalize(skb);
1041                                 push_frames_if_head(sk);
1042                         }
1043
1044                         if (is_tls_tx(csk)) {
1045                                 skb = get_record_skb(sk,
1046                                                      select_size(sk,
1047                                                                  recordsz,
1048                                                                  flags,
1049                                                                  TX_TLSHDR_LEN),
1050                                                                  false);
1051                         } else {
1052                                 skb = get_tx_skb(sk,
1053                                                  select_size(sk, size, flags,
1054                                                              TX_HEADER_LEN));
1055                         }
1056                         if (unlikely(!skb))
1057                                 goto wait_for_memory;
1058
1059                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1060                         copy = mss;
1061                 }
1062                 if (copy > size)
1063                         copy = size;
1064
1065                 if (skb_tailroom(skb) > 0) {
1066                         copy = min(copy, skb_tailroom(skb));
1067                         if (is_tls_tx(csk))
1068                                 copy = min_t(int, copy, csk->tlshws.txleft);
1069                         err = skb_add_data_nocache(sk, skb,
1070                                                    &msg->msg_iter, copy);
1071                         if (err)
1072                                 goto do_fault;
1073                 } else {
1074                         int i = skb_shinfo(skb)->nr_frags;
1075                         struct page *page = TCP_PAGE(sk);
1076                         int pg_size = PAGE_SIZE;
1077                         int off = TCP_OFF(sk);
1078                         bool merge;
1079
1080                         if (page)
1081                                 pg_size = page_size(page);
1082                         if (off < pg_size &&
1083                             skb_can_coalesce(skb, i, page, off)) {
1084                                 merge = 1;
1085                                 goto copy;
1086                         }
1087                         merge = 0;
1088                         if (i == (is_tls_tx(csk) ? (MAX_SKB_FRAGS - 1) :
1089                             MAX_SKB_FRAGS))
1090                                 goto new_buf;
1091
1092                         if (page && off == pg_size) {
1093                                 put_page(page);
1094                                 TCP_PAGE(sk) = page = NULL;
1095                                 pg_size = PAGE_SIZE;
1096                         }
1097
1098                         if (!page) {
1099                                 gfp_t gfp = sk->sk_allocation;
1100                                 int order = cdev->send_page_order;
1101
1102                                 if (order) {
1103                                         page = alloc_pages(gfp | __GFP_COMP |
1104                                                            __GFP_NOWARN |
1105                                                            __GFP_NORETRY,
1106                                                            order);
1107                                         if (page)
1108                                                 pg_size <<= order;
1109                                 }
1110                                 if (!page) {
1111                                         page = alloc_page(gfp);
1112                                         pg_size = PAGE_SIZE;
1113                                 }
1114                                 if (!page)
1115                                         goto wait_for_memory;
1116                                 off = 0;
1117                         }
1118 copy:
1119                         if (copy > pg_size - off)
1120                                 copy = pg_size - off;
1121                         if (is_tls_tx(csk))
1122                                 copy = min_t(int, copy, csk->tlshws.txleft);
1123
1124                         err = chtls_skb_copy_to_page_nocache(sk, &msg->msg_iter,
1125                                                              skb, page,
1126                                                              off, copy);
1127                         if (unlikely(err)) {
1128                                 if (!TCP_PAGE(sk)) {
1129                                         TCP_PAGE(sk) = page;
1130                                         TCP_OFF(sk) = 0;
1131                                 }
1132                                 goto do_fault;
1133                         }
1134                         /* Update the skb. */
1135                         if (merge) {
1136                                 skb_frag_size_add(
1137                                                 &skb_shinfo(skb)->frags[i - 1],
1138                                                 copy);
1139                         } else {
1140                                 skb_fill_page_desc(skb, i, page, off, copy);
1141                                 if (off + copy < pg_size) {
1142                                         /* space left keep page */
1143                                         get_page(page);
1144                                         TCP_PAGE(sk) = page;
1145                                 } else {
1146                                         TCP_PAGE(sk) = NULL;
1147                                 }
1148                         }
1149                         TCP_OFF(sk) = off + copy;
1150                 }
1151                 if (unlikely(skb->len == mss))
1152                         tx_skb_finalize(skb);
1153                 tp->write_seq += copy;
1154                 copied += copy;
1155                 size -= copy;
1156
1157                 if (is_tls_tx(csk))
1158                         csk->tlshws.txleft -= copy;
1159
1160                 if (corked(tp, flags) &&
1161                     (sk_stream_wspace(sk) < sk_stream_min_wspace(sk)))
1162                         ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_NO_APPEND;
1163
1164                 if (size == 0)
1165                         goto out;
1166
1167                 if (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND)
1168                         push_frames_if_head(sk);
1169                 continue;
1170 wait_for_sndbuf:
1171                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1172 wait_for_memory:
1173                 err = csk_wait_memory(cdev, sk, &timeo);
1174                 if (err)
1175                         goto do_error;
1176         }
1177 out:
1178         csk_reset_flag(csk, CSK_TX_MORE_DATA);
1179         if (copied)
1180                 chtls_tcp_push(sk, flags);
1181 done:
1182         release_sock(sk);
1183         return copied;
1184 do_fault:
1185         if (!skb->len) {
1186                 __skb_unlink(skb, &csk->txq);
1187                 sk->sk_wmem_queued -= skb->truesize;
1188                 __kfree_skb(skb);
1189         }
1190 do_error:
1191         if (copied)
1192                 goto out;
1193 out_err:
1194         if (csk_conn_inline(csk))
1195                 csk_reset_flag(csk, CSK_TX_MORE_DATA);
1196         copied = sk_stream_error(sk, flags, err);
1197         goto done;
1198 }
1199
1200 int chtls_sendpage(struct sock *sk, struct page *page,
1201                    int offset, size_t size, int flags)
1202 {
1203         struct chtls_sock *csk;
1204         struct chtls_dev *cdev;
1205         int mss, err, copied;
1206         struct tcp_sock *tp;
1207         long timeo;
1208
1209         tp = tcp_sk(sk);
1210         copied = 0;
1211         csk = rcu_dereference_sk_user_data(sk);
1212         cdev = csk->cdev;
1213         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1214
1215         err = sk_stream_wait_connect(sk, &timeo);
1216         if (!sk_in_state(sk, TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
1217             err != 0)
1218                 goto out_err;
1219
1220         mss = csk->mss;
1221         csk_set_flag(csk, CSK_TX_MORE_DATA);
1222
1223         while (size > 0) {
1224                 struct sk_buff *skb = skb_peek_tail(&csk->txq);
1225                 int copy, i;
1226
1227                 if (!skb || (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND) ||
1228                     (copy = mss - skb->len) <= 0) {
1229 new_buf:
1230                         if (!csk_mem_free(cdev, sk))
1231                                 goto wait_for_sndbuf;
1232
1233                         if (is_tls_tx(csk)) {
1234                                 skb = get_record_skb(sk,
1235                                                      select_size(sk, size,
1236                                                                  flags,
1237                                                                  TX_TLSHDR_LEN),
1238                                                      true);
1239                         } else {
1240                                 skb = get_tx_skb(sk, 0);
1241                         }
1242                         if (!skb)
1243                                 goto wait_for_memory;
1244                         copy = mss;
1245                 }
1246                 if (copy > size)
1247                         copy = size;
1248
1249                 i = skb_shinfo(skb)->nr_frags;
1250                 if (skb_can_coalesce(skb, i, page, offset)) {
1251                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1252                 } else if (i < MAX_SKB_FRAGS) {
1253                         get_page(page);
1254                         skb_fill_page_desc(skb, i, page, offset, copy);
1255                 } else {
1256                         tx_skb_finalize(skb);
1257                         push_frames_if_head(sk);
1258                         goto new_buf;
1259                 }
1260
1261                 skb->len += copy;
1262                 if (skb->len == mss)
1263                         tx_skb_finalize(skb);
1264                 skb->data_len += copy;
1265                 skb->truesize += copy;
1266                 sk->sk_wmem_queued += copy;
1267                 tp->write_seq += copy;
1268                 copied += copy;
1269                 offset += copy;
1270                 size -= copy;
1271
1272                 if (corked(tp, flags) &&
1273                     (sk_stream_wspace(sk) < sk_stream_min_wspace(sk)))
1274                         ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_NO_APPEND;
1275
1276                 if (!size)
1277                         break;
1278
1279                 if (unlikely(ULP_SKB_CB(skb)->flags & ULPCB_FLAG_NO_APPEND))
1280                         push_frames_if_head(sk);
1281                 continue;
1282 wait_for_sndbuf:
1283                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1284 wait_for_memory:
1285                 err = csk_wait_memory(cdev, sk, &timeo);
1286                 if (err)
1287                         goto do_error;
1288         }
1289 out:
1290         csk_reset_flag(csk, CSK_TX_MORE_DATA);
1291         if (copied)
1292                 chtls_tcp_push(sk, flags);
1293 done:
1294         release_sock(sk);
1295         return copied;
1296
1297 do_error:
1298         if (copied)
1299                 goto out;
1300
1301 out_err:
1302         if (csk_conn_inline(csk))
1303                 csk_reset_flag(csk, CSK_TX_MORE_DATA);
1304         copied = sk_stream_error(sk, flags, err);
1305         goto done;
1306 }
1307
1308 static void chtls_select_window(struct sock *sk)
1309 {
1310         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1311         struct tcp_sock *tp = tcp_sk(sk);
1312         unsigned int wnd = tp->rcv_wnd;
1313
1314         wnd = max_t(unsigned int, wnd, tcp_full_space(sk));
1315         wnd = max_t(unsigned int, MIN_RCV_WND, wnd);
1316
1317         if (wnd > MAX_RCV_WND)
1318                 wnd = MAX_RCV_WND;
1319
1320 /*
1321  * Check if we need to grow the receive window in response to an increase in
1322  * the socket's receive buffer size.  Some applications increase the buffer
1323  * size dynamically and rely on the window to grow accordingly.
1324  */
1325
1326         if (wnd > tp->rcv_wnd) {
1327                 tp->rcv_wup -= wnd - tp->rcv_wnd;
1328                 tp->rcv_wnd = wnd;
1329                 /* Mark the receive window as updated */
1330                 csk_reset_flag(csk, CSK_UPDATE_RCV_WND);
1331         }
1332 }
1333
1334 /*
1335  * Send RX credits through an RX_DATA_ACK CPL message.  We are permitted
1336  * to return without sending the message in case we cannot allocate
1337  * an sk_buff.  Returns the number of credits sent.
1338  */
1339 static u32 send_rx_credits(struct chtls_sock *csk, u32 credits)
1340 {
1341         struct cpl_rx_data_ack *req;
1342         struct sk_buff *skb;
1343
1344         skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
1345         if (!skb)
1346                 return 0;
1347         __skb_put(skb, sizeof(*req));
1348         req = (struct cpl_rx_data_ack *)skb->head;
1349
1350         set_wr_txq(skb, CPL_PRIORITY_ACK, csk->port_id);
1351         INIT_TP_WR(req, csk->tid);
1352         OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1353                                                     csk->tid));
1354         req->credit_dack = cpu_to_be32(RX_CREDITS_V(credits) |
1355                                        RX_FORCE_ACK_F);
1356         cxgb4_ofld_send(csk->cdev->ports[csk->port_id], skb);
1357         return credits;
1358 }
1359
1360 #define CREDIT_RETURN_STATE (TCPF_ESTABLISHED | \
1361                              TCPF_FIN_WAIT1 | \
1362                              TCPF_FIN_WAIT2)
1363
1364 /*
1365  * Called after some received data has been read.  It returns RX credits
1366  * to the HW for the amount of data processed.
1367  */
1368 static void chtls_cleanup_rbuf(struct sock *sk, int copied)
1369 {
1370         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1371         struct tcp_sock *tp;
1372         int must_send;
1373         u32 credits;
1374         u32 thres;
1375
1376         thres = 15 * 1024;
1377
1378         if (!sk_in_state(sk, CREDIT_RETURN_STATE))
1379                 return;
1380
1381         chtls_select_window(sk);
1382         tp = tcp_sk(sk);
1383         credits = tp->copied_seq - tp->rcv_wup;
1384         if (unlikely(!credits))
1385                 return;
1386
1387 /*
1388  * For coalescing to work effectively ensure the receive window has
1389  * at least 16KB left.
1390  */
1391         must_send = credits + 16384 >= tp->rcv_wnd;
1392
1393         if (must_send || credits >= thres)
1394                 tp->rcv_wup += send_rx_credits(csk, credits);
1395 }
1396
1397 static int chtls_pt_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1398                             int nonblock, int flags, int *addr_len)
1399 {
1400         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1401         struct chtls_hws *hws = &csk->tlshws;
1402         struct tcp_sock *tp = tcp_sk(sk);
1403         unsigned long avail;
1404         int buffers_freed;
1405         int copied = 0;
1406         int target;
1407         long timeo;
1408
1409         buffers_freed = 0;
1410
1411         timeo = sock_rcvtimeo(sk, nonblock);
1412         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1413
1414         if (unlikely(csk_flag(sk, CSK_UPDATE_RCV_WND)))
1415                 chtls_cleanup_rbuf(sk, copied);
1416
1417         do {
1418                 struct sk_buff *skb;
1419                 u32 offset = 0;
1420
1421                 if (unlikely(tp->urg_data &&
1422                              tp->urg_seq == tp->copied_seq)) {
1423                         if (copied)
1424                                 break;
1425                         if (signal_pending(current)) {
1426                                 copied = timeo ? sock_intr_errno(timeo) :
1427                                         -EAGAIN;
1428                                 break;
1429                         }
1430                 }
1431                 skb = skb_peek(&sk->sk_receive_queue);
1432                 if (skb)
1433                         goto found_ok_skb;
1434                 if (csk->wr_credits &&
1435                     skb_queue_len(&csk->txq) &&
1436                     chtls_push_frames(csk, csk->wr_credits ==
1437                                       csk->wr_max_credits))
1438                         sk->sk_write_space(sk);
1439
1440                 if (copied >= target && !READ_ONCE(sk->sk_backlog.tail))
1441                         break;
1442
1443                 if (copied) {
1444                         if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
1445                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1446                             signal_pending(current))
1447                                 break;
1448
1449                         if (!timeo)
1450                                 break;
1451                 } else {
1452                         if (sock_flag(sk, SOCK_DONE))
1453                                 break;
1454                         if (sk->sk_err) {
1455                                 copied = sock_error(sk);
1456                                 break;
1457                         }
1458                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1459                                 break;
1460                         if (sk->sk_state == TCP_CLOSE) {
1461                                 copied = -ENOTCONN;
1462                                 break;
1463                         }
1464                         if (!timeo) {
1465                                 copied = -EAGAIN;
1466                                 break;
1467                         }
1468                         if (signal_pending(current)) {
1469                                 copied = sock_intr_errno(timeo);
1470                                 break;
1471                         }
1472                 }
1473                 if (READ_ONCE(sk->sk_backlog.tail)) {
1474                         release_sock(sk);
1475                         lock_sock(sk);
1476                         chtls_cleanup_rbuf(sk, copied);
1477                         continue;
1478                 }
1479
1480                 if (copied >= target)
1481                         break;
1482                 chtls_cleanup_rbuf(sk, copied);
1483                 sk_wait_data(sk, &timeo, NULL);
1484                 continue;
1485 found_ok_skb:
1486                 if (!skb->len) {
1487                         skb_dst_set(skb, NULL);
1488                         __skb_unlink(skb, &sk->sk_receive_queue);
1489                         kfree_skb(skb);
1490
1491                         if (!copied && !timeo) {
1492                                 copied = -EAGAIN;
1493                                 break;
1494                         }
1495
1496                         if (copied < target) {
1497                                 release_sock(sk);
1498                                 lock_sock(sk);
1499                                 continue;
1500                         }
1501                         break;
1502                 }
1503                 offset = hws->copied_seq;
1504                 avail = skb->len - offset;
1505                 if (len < avail)
1506                         avail = len;
1507
1508                 if (unlikely(tp->urg_data)) {
1509                         u32 urg_offset = tp->urg_seq - tp->copied_seq;
1510
1511                         if (urg_offset < avail) {
1512                                 if (urg_offset) {
1513                                         avail = urg_offset;
1514                                 } else if (!sock_flag(sk, SOCK_URGINLINE)) {
1515                                         /* First byte is urgent, skip */
1516                                         tp->copied_seq++;
1517                                         offset++;
1518                                         avail--;
1519                                         if (!avail)
1520                                                 goto skip_copy;
1521                                 }
1522                         }
1523                 }
1524                 if (skb_copy_datagram_msg(skb, offset, msg, avail)) {
1525                         if (!copied) {
1526                                 copied = -EFAULT;
1527                                 break;
1528                         }
1529                 }
1530
1531                 copied += avail;
1532                 len -= avail;
1533                 hws->copied_seq += avail;
1534 skip_copy:
1535                 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq))
1536                         tp->urg_data = 0;
1537
1538                 if ((avail + offset) >= skb->len) {
1539                         if (ULP_SKB_CB(skb)->flags & ULPCB_FLAG_TLS_HDR) {
1540                                 tp->copied_seq += skb->len;
1541                                 hws->rcvpld = skb->hdr_len;
1542                         } else {
1543                                 tp->copied_seq += hws->rcvpld;
1544                         }
1545                         chtls_free_skb(sk, skb);
1546                         buffers_freed++;
1547                         hws->copied_seq = 0;
1548                         if (copied >= target &&
1549                             !skb_peek(&sk->sk_receive_queue))
1550                                 break;
1551                 }
1552         } while (len > 0);
1553
1554         if (buffers_freed)
1555                 chtls_cleanup_rbuf(sk, copied);
1556         release_sock(sk);
1557         return copied;
1558 }
1559
1560 /*
1561  * Peek at data in a socket's receive buffer.
1562  */
1563 static int peekmsg(struct sock *sk, struct msghdr *msg,
1564                    size_t len, int nonblock, int flags)
1565 {
1566         struct tcp_sock *tp = tcp_sk(sk);
1567         u32 peek_seq, offset;
1568         struct sk_buff *skb;
1569         int copied = 0;
1570         size_t avail;          /* amount of available data in current skb */
1571         long timeo;
1572
1573         lock_sock(sk);
1574         timeo = sock_rcvtimeo(sk, nonblock);
1575         peek_seq = tp->copied_seq;
1576
1577         do {
1578                 if (unlikely(tp->urg_data && tp->urg_seq == peek_seq)) {
1579                         if (copied)
1580                                 break;
1581                         if (signal_pending(current)) {
1582                                 copied = timeo ? sock_intr_errno(timeo) :
1583                                 -EAGAIN;
1584                                 break;
1585                         }
1586                 }
1587
1588                 skb_queue_walk(&sk->sk_receive_queue, skb) {
1589                         offset = peek_seq - ULP_SKB_CB(skb)->seq;
1590                         if (offset < skb->len)
1591                                 goto found_ok_skb;
1592                 }
1593
1594                 /* empty receive queue */
1595                 if (copied)
1596                         break;
1597                 if (sock_flag(sk, SOCK_DONE))
1598                         break;
1599                 if (sk->sk_err) {
1600                         copied = sock_error(sk);
1601                         break;
1602                 }
1603                 if (sk->sk_shutdown & RCV_SHUTDOWN)
1604                         break;
1605                 if (sk->sk_state == TCP_CLOSE) {
1606                         copied = -ENOTCONN;
1607                         break;
1608                 }
1609                 if (!timeo) {
1610                         copied = -EAGAIN;
1611                         break;
1612                 }
1613                 if (signal_pending(current)) {
1614                         copied = sock_intr_errno(timeo);
1615                         break;
1616                 }
1617
1618                 if (READ_ONCE(sk->sk_backlog.tail)) {
1619                         /* Do not sleep, just process backlog. */
1620                         release_sock(sk);
1621                         lock_sock(sk);
1622                 } else {
1623                         sk_wait_data(sk, &timeo, NULL);
1624                 }
1625
1626                 if (unlikely(peek_seq != tp->copied_seq)) {
1627                         if (net_ratelimit())
1628                                 pr_info("TCP(%s:%d), race in MSG_PEEK.\n",
1629                                         current->comm, current->pid);
1630                         peek_seq = tp->copied_seq;
1631                 }
1632                 continue;
1633
1634 found_ok_skb:
1635                 avail = skb->len - offset;
1636                 if (len < avail)
1637                         avail = len;
1638                 /*
1639                  * Do we have urgent data here?  We need to skip over the
1640                  * urgent byte.
1641                  */
1642                 if (unlikely(tp->urg_data)) {
1643                         u32 urg_offset = tp->urg_seq - peek_seq;
1644
1645                         if (urg_offset < avail) {
1646                                 /*
1647                                  * The amount of data we are preparing to copy
1648                                  * contains urgent data.
1649                                  */
1650                                 if (!urg_offset) { /* First byte is urgent */
1651                                         if (!sock_flag(sk, SOCK_URGINLINE)) {
1652                                                 peek_seq++;
1653                                                 offset++;
1654                                                 avail--;
1655                                         }
1656                                         if (!avail)
1657                                                 continue;
1658                                 } else {
1659                                         /* stop short of the urgent data */
1660                                         avail = urg_offset;
1661                                 }
1662                         }
1663                 }
1664
1665                 /*
1666                  * If MSG_TRUNC is specified the data is discarded.
1667                  */
1668                 if (likely(!(flags & MSG_TRUNC)))
1669                         if (skb_copy_datagram_msg(skb, offset, msg, len)) {
1670                                 if (!copied) {
1671                                         copied = -EFAULT;
1672                                         break;
1673                                 }
1674                         }
1675                 peek_seq += avail;
1676                 copied += avail;
1677                 len -= avail;
1678         } while (len > 0);
1679
1680         release_sock(sk);
1681         return copied;
1682 }
1683
1684 int chtls_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1685                   int nonblock, int flags, int *addr_len)
1686 {
1687         struct tcp_sock *tp = tcp_sk(sk);
1688         struct chtls_sock *csk;
1689         unsigned long avail;    /* amount of available data in current skb */
1690         int buffers_freed;
1691         int copied = 0;
1692         long timeo;
1693         int target;             /* Read at least this many bytes */
1694
1695         buffers_freed = 0;
1696
1697         if (unlikely(flags & MSG_OOB))
1698                 return tcp_prot.recvmsg(sk, msg, len, nonblock, flags,
1699                                         addr_len);
1700
1701         if (unlikely(flags & MSG_PEEK))
1702                 return peekmsg(sk, msg, len, nonblock, flags);
1703
1704         if (sk_can_busy_loop(sk) &&
1705             skb_queue_empty_lockless(&sk->sk_receive_queue) &&
1706             sk->sk_state == TCP_ESTABLISHED)
1707                 sk_busy_loop(sk, nonblock);
1708
1709         lock_sock(sk);
1710         csk = rcu_dereference_sk_user_data(sk);
1711
1712         if (is_tls_rx(csk))
1713                 return chtls_pt_recvmsg(sk, msg, len, nonblock,
1714                                         flags, addr_len);
1715
1716         timeo = sock_rcvtimeo(sk, nonblock);
1717         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1718
1719         if (unlikely(csk_flag(sk, CSK_UPDATE_RCV_WND)))
1720                 chtls_cleanup_rbuf(sk, copied);
1721
1722         do {
1723                 struct sk_buff *skb;
1724                 u32 offset;
1725
1726                 if (unlikely(tp->urg_data && tp->urg_seq == tp->copied_seq)) {
1727                         if (copied)
1728                                 break;
1729                         if (signal_pending(current)) {
1730                                 copied = timeo ? sock_intr_errno(timeo) :
1731                                         -EAGAIN;
1732                                 break;
1733                         }
1734                 }
1735
1736                 skb = skb_peek(&sk->sk_receive_queue);
1737                 if (skb)
1738                         goto found_ok_skb;
1739
1740                 if (csk->wr_credits &&
1741                     skb_queue_len(&csk->txq) &&
1742                     chtls_push_frames(csk, csk->wr_credits ==
1743                                       csk->wr_max_credits))
1744                         sk->sk_write_space(sk);
1745
1746                 if (copied >= target && !READ_ONCE(sk->sk_backlog.tail))
1747                         break;
1748
1749                 if (copied) {
1750                         if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
1751                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1752                             signal_pending(current))
1753                                 break;
1754                 } else {
1755                         if (sock_flag(sk, SOCK_DONE))
1756                                 break;
1757                         if (sk->sk_err) {
1758                                 copied = sock_error(sk);
1759                                 break;
1760                         }
1761                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1762                                 break;
1763                         if (sk->sk_state == TCP_CLOSE) {
1764                                 copied = -ENOTCONN;
1765                                 break;
1766                         }
1767                         if (!timeo) {
1768                                 copied = -EAGAIN;
1769                                 break;
1770                         }
1771                         if (signal_pending(current)) {
1772                                 copied = sock_intr_errno(timeo);
1773                                 break;
1774                         }
1775                 }
1776
1777                 if (READ_ONCE(sk->sk_backlog.tail)) {
1778                         release_sock(sk);
1779                         lock_sock(sk);
1780                         chtls_cleanup_rbuf(sk, copied);
1781                         continue;
1782                 }
1783
1784                 if (copied >= target)
1785                         break;
1786                 chtls_cleanup_rbuf(sk, copied);
1787                 sk_wait_data(sk, &timeo, NULL);
1788                 continue;
1789
1790 found_ok_skb:
1791                 if (!skb->len) {
1792                         chtls_kfree_skb(sk, skb);
1793                         if (!copied && !timeo) {
1794                                 copied = -EAGAIN;
1795                                 break;
1796                         }
1797
1798                         if (copied < target)
1799                                 continue;
1800
1801                         break;
1802                 }
1803
1804                 offset = tp->copied_seq - ULP_SKB_CB(skb)->seq;
1805                 avail = skb->len - offset;
1806                 if (len < avail)
1807                         avail = len;
1808
1809                 if (unlikely(tp->urg_data)) {
1810                         u32 urg_offset = tp->urg_seq - tp->copied_seq;
1811
1812                         if (urg_offset < avail) {
1813                                 if (urg_offset) {
1814                                         avail = urg_offset;
1815                                 } else if (!sock_flag(sk, SOCK_URGINLINE)) {
1816                                         tp->copied_seq++;
1817                                         offset++;
1818                                         avail--;
1819                                         if (!avail)
1820                                                 goto skip_copy;
1821                                 }
1822                         }
1823                 }
1824
1825                 if (likely(!(flags & MSG_TRUNC))) {
1826                         if (skb_copy_datagram_msg(skb, offset,
1827                                                   msg, avail)) {
1828                                 if (!copied) {
1829                                         copied = -EFAULT;
1830                                         break;
1831                                 }
1832                         }
1833                 }
1834
1835                 tp->copied_seq += avail;
1836                 copied += avail;
1837                 len -= avail;
1838
1839 skip_copy:
1840                 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq))
1841                         tp->urg_data = 0;
1842
1843                 if (avail + offset >= skb->len) {
1844                         if (likely(skb))
1845                                 chtls_free_skb(sk, skb);
1846                         buffers_freed++;
1847
1848                         if  (copied >= target &&
1849                              !skb_peek(&sk->sk_receive_queue))
1850                                 break;
1851                 }
1852         } while (len > 0);
1853
1854         if (buffers_freed)
1855                 chtls_cleanup_rbuf(sk, copied);
1856
1857         release_sock(sk);
1858         return copied;
1859 }