net/core/skmsg.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
   3
   4 #include <linux/skmsg.h>
   5 #include <linux/skbuff.h>
   6 #include <linux/scatterlist.h>
   7
   8 #include <net/sock.h>
   9 #include <net/tcp.h>
  10
  11 static bool sk_msg_try_coalesce_ok(struct sk_msg *msg, int elem_first_coalesce)
  12 {
  13         if (msg->sg.end > msg->sg.start &&
  14             elem_first_coalesce < msg->sg.end)
  15                 return true;
  16
  17         if (msg->sg.end < msg->sg.start &&
  18             (elem_first_coalesce > msg->sg.start ||
  19              elem_first_coalesce < msg->sg.end))
  20                 return true;
  21
  22         return false;
  23 }
  24
  25 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
  26                  int elem_first_coalesce)
  27 {
  28         struct page_frag *pfrag = sk_page_frag(sk);
  29         int ret = 0;
  30
  31         len -= msg->sg.size;
  32         while (len > 0) {
  33                 struct scatterlist *sge;
  34                 u32 orig_offset;
  35                 int use, i;
  36
  37                 if (!sk_page_frag_refill(sk, pfrag))
  38                         return -ENOMEM;
  39
  40                 orig_offset = pfrag->offset;
  41                 use = min_t(int, len, pfrag->size - orig_offset);
  42                 if (!sk_wmem_schedule(sk, use))
  43                         return -ENOMEM;
  44
  45                 i = msg->sg.end;
  46                 sk_msg_iter_var_prev(i);
  47                 sge = &msg->sg.data[i];
  48
  49                 if (sk_msg_try_coalesce_ok(msg, elem_first_coalesce) &&
  50                     sg_page(sge) == pfrag->page &&
  51                     sge->offset + sge->length == orig_offset) {
  52                         sge->length += use;
  53                 } else {
  54                         if (sk_msg_full(msg)) {
  55                                 ret = -ENOSPC;
  56                                 break;
  57                         }
  58
  59                         sge = &msg->sg.data[msg->sg.end];
  60                         sg_unmark_end(sge);
  61                         sg_set_page(sge, pfrag->page, use, orig_offset);
  62                         get_page(pfrag->page);
  63                         sk_msg_iter_next(msg, end);
  64                 }
  65
  66                 sk_mem_charge(sk, use);
  67                 msg->sg.size += use;
  68                 pfrag->offset += use;
  69                 len -= use;
  70         }
  71
  72         return ret;
  73 }
  74 EXPORT_SYMBOL_GPL(sk_msg_alloc);
  75
  76 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
  77                  u32 off, u32 len)
  78 {
  79         int i = src->sg.start;
  80         struct scatterlist *sge = sk_msg_elem(src, i);
  81         u32 sge_len, sge_off;
  82
  83         if (sk_msg_full(dst))
  84                 return -ENOSPC;
  85
  86         while (off) {
  87                 if (sge->length > off)
  88                         break;
  89                 off -= sge->length;
  90                 sk_msg_iter_var_next(i);
  91                 if (i == src->sg.end && off)
  92                         return -ENOSPC;
  93                 sge = sk_msg_elem(src, i);
  94         }
  95
  96         while (len) {
  97                 if (sk_msg_full(dst))
  98                         return -ENOSPC;
  99
 100                 sge_len = sge->length - off;
 101                 sge_off = sge->offset + off;
 102                 if (sge_len > len)
 103                         sge_len = len;
 104                 off = 0;
 105                 len -= sge_len;
 106                 sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off);
 107                 sk_mem_charge(sk, sge_len);
 108                 sk_msg_iter_var_next(i);
 109                 if (i == src->sg.end && len)
 110                         return -ENOSPC;
 111                 sge = sk_msg_elem(src, i);
 112         }
 113
 114         return 0;
 115 }
 116 EXPORT_SYMBOL_GPL(sk_msg_clone);
 117
 118 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes)
 119 {
 120         int i = msg->sg.start;
 121
 122         do {
 123                 struct scatterlist *sge = sk_msg_elem(msg, i);
 124
 125                 if (bytes < sge->length) {
 126                         sge->length -= bytes;
 127                         sge->offset += bytes;
 128                         sk_mem_uncharge(sk, bytes);
 129                         break;
 130                 }
 131
 132                 sk_mem_uncharge(sk, sge->length);
 133                 bytes -= sge->length;
 134                 sge->length = 0;
 135                 sge->offset = 0;
 136                 sk_msg_iter_var_next(i);
 137         } while (bytes && i != msg->sg.end);
 138         msg->sg.start = i;
 139 }
 140 EXPORT_SYMBOL_GPL(sk_msg_return_zero);
 141
 142 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes)
 143 {
 144         int i = msg->sg.start;
 145
 146         do {
 147                 struct scatterlist *sge = &msg->sg.data[i];
 148                 int uncharge = (bytes < sge->length) ? bytes : sge->length;
 149
 150                 sk_mem_uncharge(sk, uncharge);
 151                 bytes -= uncharge;
 152                 sk_msg_iter_var_next(i);
 153         } while (i != msg->sg.end);
 154 }
 155 EXPORT_SYMBOL_GPL(sk_msg_return);
 156
 157 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i,
 158                             bool charge)
 159 {
 160         struct scatterlist *sge = sk_msg_elem(msg, i);
 161         u32 len = sge->length;
 162
 163         if (charge)
 164                 sk_mem_uncharge(sk, len);
 165         if (!msg->skb)
 166                 put_page(sg_page(sge));
 167         memset(sge, 0, sizeof(*sge));
 168         return len;
 169 }
 170
 171 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i,
 172                          bool charge)
 173 {
 174         struct scatterlist *sge = sk_msg_elem(msg, i);
 175         int freed = 0;
 176
 177         while (msg->sg.size) {
 178                 msg->sg.size -= sge->length;
 179                 freed += sk_msg_free_elem(sk, msg, i, charge);
 180                 sk_msg_iter_var_next(i);
 181                 sk_msg_check_to_free(msg, i, msg->sg.size);
 182                 sge = sk_msg_elem(msg, i);
 183         }
 184         if (msg->skb)
 185                 consume_skb(msg->skb);
 186         sk_msg_init(msg);
 187         return freed;
 188 }
 189
 190 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg)
 191 {
 192         return __sk_msg_free(sk, msg, msg->sg.start, false);
 193 }
 194 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge);
 195
 196 int sk_msg_free(struct sock *sk, struct sk_msg *msg)
 197 {
 198         return __sk_msg_free(sk, msg, msg->sg.start, true);
 199 }
 200 EXPORT_SYMBOL_GPL(sk_msg_free);
 201
 202 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg,
 203                                   u32 bytes, bool charge)
 204 {
 205         struct scatterlist *sge;
 206         u32 i = msg->sg.start;
 207
 208         while (bytes) {
 209                 sge = sk_msg_elem(msg, i);
 210                 if (!sge->length)
 211                         break;
 212                 if (bytes < sge->length) {
 213                         if (charge)
 214                                 sk_mem_uncharge(sk, bytes);
 215                         sge->length -= bytes;
 216                         sge->offset += bytes;
 217                         msg->sg.size -= bytes;
 218                         break;
 219                 }
 220
 221                 msg->sg.size -= sge->length;
 222                 bytes -= sge->length;
 223                 sk_msg_free_elem(sk, msg, i, charge);
 224                 sk_msg_iter_var_next(i);
 225                 sk_msg_check_to_free(msg, i, bytes);
 226         }
 227         msg->sg.start = i;
 228 }
 229
 230 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes)
 231 {
 232         __sk_msg_free_partial(sk, msg, bytes, true);
 233 }
 234 EXPORT_SYMBOL_GPL(sk_msg_free_partial);
 235
 236 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
 237                                   u32 bytes)
 238 {
 239         __sk_msg_free_partial(sk, msg, bytes, false);
 240 }
 241
 242 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len)
 243 {
 244         int trim = msg->sg.size - len;
 245         u32 i = msg->sg.end;
 246
 247         if (trim <= 0) {
 248                 WARN_ON(trim < 0);
 249                 return;
 250         }
 251
 252         sk_msg_iter_var_prev(i);
 253         msg->sg.size = len;
 254         while (msg->sg.data[i].length &&
 255                trim >= msg->sg.data[i].length) {
 256                 trim -= msg->sg.data[i].length;
 257                 sk_msg_free_elem(sk, msg, i, true);
 258                 sk_msg_iter_var_prev(i);
 259                 if (!trim)
 260                         goto out;
 261         }
 262
 263         msg->sg.data[i].length -= trim;
 264         sk_mem_uncharge(sk, trim);
 265 out:
 266         /* If we trim data before curr pointer update copybreak and current
 267          * so that any future copy operations start at new copy location.
 268          * However trimed data that has not yet been used in a copy op
 269          * does not require an update.
 270          */
 271         if (msg->sg.curr >= i) {
 272                 msg->sg.curr = i;
 273                 msg->sg.copybreak = msg->sg.data[i].length;
 274         }
 275         sk_msg_iter_var_next(i);
 276         msg->sg.end = i;
 277 }
 278 EXPORT_SYMBOL_GPL(sk_msg_trim);
 279
 280 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
 281                               struct sk_msg *msg, u32 bytes)
 282 {
 283         int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg);
 284         const int to_max_pages = MAX_MSG_FRAGS;
 285         struct page *pages[MAX_MSG_FRAGS];
 286         ssize_t orig, copied, use, offset;
 287
 288         orig = msg->sg.size;
 289         while (bytes > 0) {
 290                 i = 0;
 291                 maxpages = to_max_pages - num_elems;
 292                 if (maxpages == 0) {
 293                         ret = -EFAULT;
 294                         goto out;
 295                 }
 296
 297                 copied = iov_iter_get_pages(from, pages, bytes, maxpages,
 298                                             &offset);
 299                 if (copied <= 0) {
 300                         ret = -EFAULT;
 301                         goto out;
 302                 }
 303
 304                 iov_iter_advance(from, copied);
 305                 bytes -= copied;
 306                 msg->sg.size += copied;
 307
 308                 while (copied) {
 309                         use = min_t(int, copied, PAGE_SIZE - offset);
 310                         sg_set_page(&msg->sg.data[msg->sg.end],
 311                                     pages[i], use, offset);
 312                         sg_unmark_end(&msg->sg.data[msg->sg.end]);
 313                         sk_mem_charge(sk, use);
 314
 315                         offset = 0;
 316                         copied -= use;
 317                         sk_msg_iter_next(msg, end);
 318                         num_elems++;
 319                         i++;
 320                 }
 321                 /* When zerocopy is mixed with sk_msg_*copy* operations we
 322                  * may have a copybreak set in this case clear and prefer
 323                  * zerocopy remainder when possible.
 324                  */
 325                 msg->sg.copybreak = 0;
 326                 msg->sg.curr = msg->sg.end;
 327         }
 328 out:
 329         /* Revert iov_iter updates, msg will need to use 'trim' later if it
 330          * also needs to be cleared.
 331          */
 332         if (ret)
 333                 iov_iter_revert(from, msg->sg.size - orig);
 334         return ret;
 335 }
 336 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter);
 337
 338 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 339                              struct sk_msg *msg, u32 bytes)
 340 {
 341         int ret = -ENOSPC, i = msg->sg.curr;
 342         struct scatterlist *sge;
 343         u32 copy, buf_size;
 344         void *to;
 345
 346         do {
 347                 sge = sk_msg_elem(msg, i);
 348                 /* This is possible if a trim operation shrunk the buffer */
 349                 if (msg->sg.copybreak >= sge->length) {
 350                         msg->sg.copybreak = 0;
 351                         sk_msg_iter_var_next(i);
 352                         if (i == msg->sg.end)
 353                                 break;
 354                         sge = sk_msg_elem(msg, i);
 355                 }
 356
 357                 buf_size = sge->length - msg->sg.copybreak;
 358                 copy = (buf_size > bytes) ? bytes : buf_size;
 359                 to = sg_virt(sge) + msg->sg.copybreak;
 360                 msg->sg.copybreak += copy;
 361                 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY)
 362                         ret = copy_from_iter_nocache(to, copy, from);
 363                 else
 364                         ret = copy_from_iter(to, copy, from);
 365                 if (ret != copy) {
 366                         ret = -EFAULT;
 367                         goto out;
 368                 }
 369                 bytes -= copy;
 370                 if (!bytes)
 371                         break;
 372                 msg->sg.copybreak = 0;
 373                 sk_msg_iter_var_next(i);
 374         } while (i != msg->sg.end);
 375 out:
 376         msg->sg.curr = i;
 377         return ret;
 378 }
 379 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
 380
 381 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb)
 382 {
 383         struct sock *sk = psock->sk;
 384         int copied = 0, num_sge;
 385         struct sk_msg *msg;
 386
 387         msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC);
 388         if (unlikely(!msg))
 389                 return -EAGAIN;
 390         if (!sk_rmem_schedule(sk, skb, skb->len)) {
 391                 kfree(msg);
 392                 return -EAGAIN;
 393         }
 394
 395         sk_msg_init(msg);
 396         num_sge = skb_to_sgvec(skb, msg->sg.data, 0, skb->len);
 397         if (unlikely(num_sge < 0)) {
 398                 kfree(msg);
 399                 return num_sge;
 400         }
 401
 402         sk_mem_charge(sk, skb->len);
 403         copied = skb->len;
 404         msg->sg.start = 0;
 405         msg->sg.end = num_sge == MAX_MSG_FRAGS ? 0 : num_sge;
 406         msg->skb = skb;
 407
 408         sk_psock_queue_msg(psock, msg);
 409         sk_psock_data_ready(sk, psock);
 410         return copied;
 411 }
 412
 413 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,
 414                                u32 off, u32 len, bool ingress)
 415 {
 416         if (ingress)
 417                 return sk_psock_skb_ingress(psock, skb);
 418         else
 419                 return skb_send_sock_locked(psock->sk, skb, off, len);
 420 }
 421
 422 static void sk_psock_backlog(struct work_struct *work)
 423 {
 424         struct sk_psock *psock = container_of(work, struct sk_psock, work);
 425         struct sk_psock_work_state *state = &psock->work_state;
 426         struct sk_buff *skb;
 427         bool ingress;
 428         u32 len, off;
 429         int ret;
 430
 431         /* Lock sock to avoid losing sk_socket during loop. */
 432         lock_sock(psock->sk);
 433         if (state->skb) {
 434                 skb = state->skb;
 435                 len = state->len;
 436                 off = state->off;
 437                 state->skb = NULL;
 438                 goto start;
 439         }
 440
 441         while ((skb = skb_dequeue(&psock->ingress_skb))) {
 442                 len = skb->len;
 443                 off = 0;
 444 start:
 445                 ingress = tcp_skb_bpf_ingress(skb);
 446                 do {
 447                         ret = -EIO;
 448                         if (likely(psock->sk->sk_socket))
 449                                 ret = sk_psock_handle_skb(psock, skb, off,
 450                                                           len, ingress);
 451                         if (ret <= 0) {
 452                                 if (ret == -EAGAIN) {
 453                                         state->skb = skb;
 454                                         state->len = len;
 455                                         state->off = off;
 456                                         goto end;
 457                                 }
 458                                 /* Hard errors break pipe and stop xmit. */
 459                                 sk_psock_report_error(psock, ret ? -ret : EPIPE);
 460                                 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
 461                                 kfree_skb(skb);
 462                                 goto end;
 463                         }
 464                         off += ret;
 465                         len -= ret;
 466                 } while (len);
 467
 468                 if (!ingress)
 469                         kfree_skb(skb);
 470         }
 471 end:
 472         release_sock(psock->sk);
 473 }
 474
 475 struct sk_psock *sk_psock_init(struct sock *sk, int node)
 476 {
 477         struct sk_psock *psock = kzalloc_node(sizeof(*psock),
 478                                               GFP_ATOMIC | __GFP_NOWARN,
 479                                               node);
 480         if (!psock)
 481                 return NULL;
 482
 483         psock->sk = sk;
 484         psock->eval =  __SK_NONE;
 485
 486         INIT_LIST_HEAD(&psock->link);
 487         spin_lock_init(&psock->link_lock);
 488
 489         INIT_WORK(&psock->work, sk_psock_backlog);
 490         INIT_LIST_HEAD(&psock->ingress_msg);
 491         skb_queue_head_init(&psock->ingress_skb);
 492
 493         sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED);
 494         refcount_set(&psock->refcnt, 1);
 495
 496         rcu_assign_sk_user_data(sk, psock);
 497         sock_hold(sk);
 498
 499         return psock;
 500 }
 501 EXPORT_SYMBOL_GPL(sk_psock_init);
 502
 503 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock)
 504 {
 505         struct sk_psock_link *link;
 506
 507         spin_lock_bh(&psock->link_lock);
 508         link = list_first_entry_or_null(&psock->link, struct sk_psock_link,
 509                                         list);
 510         if (link)
 511                 list_del(&link->list);
 512         spin_unlock_bh(&psock->link_lock);
 513         return link;
 514 }
 515
 516 void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
 517 {
 518         struct sk_msg *msg, *tmp;
 519
 520         list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) {
 521                 list_del(&msg->list);
 522                 sk_msg_free(psock->sk, msg);
 523                 kfree(msg);
 524         }
 525 }
 526
 527 static void sk_psock_zap_ingress(struct sk_psock *psock)
 528 {
 529         __skb_queue_purge(&psock->ingress_skb);
 530         __sk_psock_purge_ingress_msg(psock);
 531 }
 532
 533 static void sk_psock_link_destroy(struct sk_psock *psock)
 534 {
 535         struct sk_psock_link *link, *tmp;
 536
 537         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 538                 list_del(&link->list);
 539                 sk_psock_free_link(link);
 540         }
 541 }
 542
 543 static void sk_psock_destroy_deferred(struct work_struct *gc)
 544 {
 545         struct sk_psock *psock = container_of(gc, struct sk_psock, gc);
 546
 547         /* No sk_callback_lock since already detached. */
 548         if (psock->parser.enabled)
 549                 strp_done(&psock->parser.strp);
 550
 551         cancel_work_sync(&psock->work);
 552
 553         psock_progs_drop(&psock->progs);
 554
 555         sk_psock_link_destroy(psock);
 556         sk_psock_cork_free(psock);
 557         sk_psock_zap_ingress(psock);
 558
 559         if (psock->sk_redir)
 560                 sock_put(psock->sk_redir);
 561         sock_put(psock->sk);
 562         kfree(psock);
 563 }
 564
 565 void sk_psock_destroy(struct rcu_head *rcu)
 566 {
 567         struct sk_psock *psock = container_of(rcu, struct sk_psock, rcu);
 568
 569         INIT_WORK(&psock->gc, sk_psock_destroy_deferred);
 570         schedule_work(&psock->gc);
 571 }
 572 EXPORT_SYMBOL_GPL(sk_psock_destroy);
 573
 574 void sk_psock_drop(struct sock *sk, struct sk_psock *psock)
 575 {
 576         rcu_assign_sk_user_data(sk, NULL);
 577         sk_psock_cork_free(psock);
 578         sk_psock_zap_ingress(psock);
 579         sk_psock_restore_proto(sk, psock);
 580
 581         write_lock_bh(&sk->sk_callback_lock);
 582         if (psock->progs.skb_parser)
 583                 sk_psock_stop_strp(sk, psock);
 584         write_unlock_bh(&sk->sk_callback_lock);
 585         sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
 586
 587         call_rcu(&psock->rcu, sk_psock_destroy);
 588 }
 589 EXPORT_SYMBOL_GPL(sk_psock_drop);
 590
 591 static int sk_psock_map_verd(int verdict, bool redir)
 592 {
 593         switch (verdict) {
 594         case SK_PASS:
 595                 return redir ? __SK_REDIRECT : __SK_PASS;
 596         case SK_DROP:
 597         default:
 598                 break;
 599         }
 600
 601         return __SK_DROP;
 602 }
 603
 604 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
 605                          struct sk_msg *msg)
 606 {
 607         struct bpf_prog *prog;
 608         int ret;
 609
 610         preempt_disable();
 611         rcu_read_lock();
 612         prog = READ_ONCE(psock->progs.msg_parser);
 613         if (unlikely(!prog)) {
 614                 ret = __SK_PASS;
 615                 goto out;
 616         }
 617
 618         sk_msg_compute_data_pointers(msg);
 619         msg->sk = sk;
 620         ret = BPF_PROG_RUN(prog, msg);
 621         ret = sk_psock_map_verd(ret, msg->sk_redir);
 622         psock->apply_bytes = msg->apply_bytes;
 623         if (ret == __SK_REDIRECT) {
 624                 if (psock->sk_redir)
 625                         sock_put(psock->sk_redir);
 626                 psock->sk_redir = msg->sk_redir;
 627                 if (!psock->sk_redir) {
 628                         ret = __SK_DROP;
 629                         goto out;
 630                 }
 631                 sock_hold(psock->sk_redir);
 632         }
 633 out:
 634         rcu_read_unlock();
 635         preempt_enable();
 636         return ret;
 637 }
 638 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);
 639
 640 static int sk_psock_bpf_run(struct sk_psock *psock, struct bpf_prog *prog,
 641                             struct sk_buff *skb)
 642 {
 643         int ret;
 644
 645         skb->sk = psock->sk;
 646         bpf_compute_data_end_sk_skb(skb);
 647         preempt_disable();
 648         ret = BPF_PROG_RUN(prog, skb);
 649         preempt_enable();
 650         /* strparser clones the skb before handing it to a upper layer,
 651          * meaning skb_orphan has been called. We NULL sk on the way out
 652          * to ensure we don't trigger a BUG_ON() in skb/sk operations
 653          * later and because we are not charging the memory of this skb
 654          * to any socket yet.
 655          */
 656         skb->sk = NULL;
 657         return ret;
 658 }
 659
 660 static struct sk_psock *sk_psock_from_strp(struct strparser *strp)
 661 {
 662         struct sk_psock_parser *parser;
 663
 664         parser = container_of(strp, struct sk_psock_parser, strp);
 665         return container_of(parser, struct sk_psock, parser);
 666 }
 667
 668 static void sk_psock_verdict_apply(struct sk_psock *psock,
 669                                    struct sk_buff *skb, int verdict)
 670 {
 671         struct sk_psock *psock_other;
 672         struct sock *sk_other;
 673         bool ingress;
 674
 675         switch (verdict) {
 676         case __SK_PASS:
 677                 sk_other = psock->sk;
 678                 if (sock_flag(sk_other, SOCK_DEAD) ||
 679                     !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
 680                         goto out_free;
 681                 }
 682                 if (atomic_read(&sk_other->sk_rmem_alloc) <=
 683                     sk_other->sk_rcvbuf) {
 684                         struct tcp_skb_cb *tcp = TCP_SKB_CB(skb);
 685
 686                         tcp->bpf.flags |= BPF_F_INGRESS;
 687                         skb_queue_tail(&psock->ingress_skb, skb);
 688                         schedule_work(&psock->work);
 689                         break;
 690                 }
 691                 goto out_free;
 692         case __SK_REDIRECT:
 693                 sk_other = tcp_skb_bpf_redirect_fetch(skb);
 694                 if (unlikely(!sk_other))
 695                         goto out_free;
 696                 psock_other = sk_psock(sk_other);
 697                 if (!psock_other || sock_flag(sk_other, SOCK_DEAD) ||
 698                     !sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED))
 699                         goto out_free;
 700                 ingress = tcp_skb_bpf_ingress(skb);
 701                 if ((!ingress && sock_writeable(sk_other)) ||
 702                     (ingress &&
 703                      atomic_read(&sk_other->sk_rmem_alloc) <=
 704                      sk_other->sk_rcvbuf)) {
 705                         if (!ingress)
 706                                 skb_set_owner_w(skb, sk_other);
 707                         skb_queue_tail(&psock_other->ingress_skb, skb);
 708                         schedule_work(&psock_other->work);
 709                         break;
 710                 }
 711                 /* fall-through */
 712         case __SK_DROP:
 713                 /* fall-through */
 714         default:
 715 out_free:
 716                 kfree_skb(skb);
 717         }
 718 }
 719
 720 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb)
 721 {
 722         struct sk_psock *psock = sk_psock_from_strp(strp);
 723         struct bpf_prog *prog;
 724         int ret = __SK_DROP;
 725
 726         rcu_read_lock();
 727         prog = READ_ONCE(psock->progs.skb_verdict);
 728         if (likely(prog)) {
 729                 skb_orphan(skb);
 730                 tcp_skb_bpf_redirect_clear(skb);
 731                 ret = sk_psock_bpf_run(psock, prog, skb);
 732                 ret = sk_psock_map_verd(ret, tcp_skb_bpf_redirect_fetch(skb));
 733         }
 734         rcu_read_unlock();
 735         sk_psock_verdict_apply(psock, skb, ret);
 736 }
 737
 738 static int sk_psock_strp_read_done(struct strparser *strp, int err)
 739 {
 740         return err;
 741 }
 742
 743 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb)
 744 {
 745         struct sk_psock *psock = sk_psock_from_strp(strp);
 746         struct bpf_prog *prog;
 747         int ret = skb->len;
 748
 749         rcu_read_lock();
 750         prog = READ_ONCE(psock->progs.skb_parser);
 751         if (likely(prog))
 752                 ret = sk_psock_bpf_run(psock, prog, skb);
 753         rcu_read_unlock();
 754         return ret;
 755 }
 756
 757 /* Called with socket lock held. */
 758 static void sk_psock_strp_data_ready(struct sock *sk)
 759 {
 760         struct sk_psock *psock;
 761
 762         rcu_read_lock();
 763         psock = sk_psock(sk);
 764         if (likely(psock)) {
 765                 write_lock_bh(&sk->sk_callback_lock);
 766                 strp_data_ready(&psock->parser.strp);
 767                 write_unlock_bh(&sk->sk_callback_lock);
 768         }
 769         rcu_read_unlock();
 770 }
 771
 772 static void sk_psock_write_space(struct sock *sk)
 773 {
 774         struct sk_psock *psock;
 775         void (*write_space)(struct sock *sk);
 776
 777         rcu_read_lock();
 778         psock = sk_psock(sk);
 779         if (likely(psock && sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)))
 780                 schedule_work(&psock->work);
 781         write_space = psock->saved_write_space;
 782         rcu_read_unlock();
 783         write_space(sk);
 784 }
 785
 786 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
 787 {
 788         static const struct strp_callbacks cb = {
 789                 .rcv_msg        = sk_psock_strp_read,
 790                 .read_sock_done = sk_psock_strp_read_done,
 791                 .parse_msg      = sk_psock_strp_parse,
 792         };
 793
 794         psock->parser.enabled = false;
 795         return strp_init(&psock->parser.strp, sk, &cb);
 796 }
 797
 798 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
 799 {
 800         struct sk_psock_parser *parser = &psock->parser;
 801
 802         if (parser->enabled)
 803                 return;
 804
 805         parser->saved_data_ready = sk->sk_data_ready;
 806         sk->sk_data_ready = sk_psock_strp_data_ready;
 807         sk->sk_write_space = sk_psock_write_space;
 808         parser->enabled = true;
 809 }
 810
 811 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
 812 {
 813         struct sk_psock_parser *parser = &psock->parser;
 814
 815         if (!parser->enabled)
 816                 return;
 817
 818         sk->sk_data_ready = parser->saved_data_ready;
 819         parser->saved_data_ready = NULL;
 820         strp_stop(&parser->strp);
 821         parser->enabled = false;
 822 }