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         struct scatterlist *sgd = NULL;
  82         u32 sge_len, sge_off;
  83
  84         while (off) {
  85                 if (sge->length > off)
  86                         break;
  87                 off -= sge->length;
  88                 sk_msg_iter_var_next(i);
  89                 if (i == src->sg.end && off)
  90                         return -ENOSPC;
  91                 sge = sk_msg_elem(src, i);
  92         }
  93
  94         while (len) {
  95                 sge_len = sge->length - off;
  96                 if (sge_len > len)
  97                         sge_len = len;
  98
  99                 if (dst->sg.end)
 100                         sgd = sk_msg_elem(dst, dst->sg.end - 1);
 101
 102                 if (sgd &&
 103                     (sg_page(sge) == sg_page(sgd)) &&
 104                     (sg_virt(sge) + off == sg_virt(sgd) + sgd->length)) {
 105                         sgd->length += sge_len;
 106                         dst->sg.size += sge_len;
 107                 } else if (!sk_msg_full(dst)) {
 108                         sge_off = sge->offset + off;
 109                         sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off);
 110                 } else {
 111                         return -ENOSPC;
 112                 }
 113
 114                 off = 0;
 115                 len -= sge_len;
 116                 sk_mem_charge(sk, sge_len);
 117                 sk_msg_iter_var_next(i);
 118                 if (i == src->sg.end && len)
 119                         return -ENOSPC;
 120                 sge = sk_msg_elem(src, i);
 121         }
 122
 123         return 0;
 124 }
 125 EXPORT_SYMBOL_GPL(sk_msg_clone);
 126
 127 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes)
 128 {
 129         int i = msg->sg.start;
 130
 131         do {
 132                 struct scatterlist *sge = sk_msg_elem(msg, i);
 133
 134                 if (bytes < sge->length) {
 135                         sge->length -= bytes;
 136                         sge->offset += bytes;
 137                         sk_mem_uncharge(sk, bytes);
 138                         break;
 139                 }
 140
 141                 sk_mem_uncharge(sk, sge->length);
 142                 bytes -= sge->length;
 143                 sge->length = 0;
 144                 sge->offset = 0;
 145                 sk_msg_iter_var_next(i);
 146         } while (bytes && i != msg->sg.end);
 147         msg->sg.start = i;
 148 }
 149 EXPORT_SYMBOL_GPL(sk_msg_return_zero);
 150
 151 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes)
 152 {
 153         int i = msg->sg.start;
 154
 155         do {
 156                 struct scatterlist *sge = &msg->sg.data[i];
 157                 int uncharge = (bytes < sge->length) ? bytes : sge->length;
 158
 159                 sk_mem_uncharge(sk, uncharge);
 160                 bytes -= uncharge;
 161                 sk_msg_iter_var_next(i);
 162         } while (i != msg->sg.end);
 163 }
 164 EXPORT_SYMBOL_GPL(sk_msg_return);
 165
 166 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i,
 167                             bool charge)
 168 {
 169         struct scatterlist *sge = sk_msg_elem(msg, i);
 170         u32 len = sge->length;
 171
 172         if (charge)
 173                 sk_mem_uncharge(sk, len);
 174         if (!msg->skb)
 175                 put_page(sg_page(sge));
 176         memset(sge, 0, sizeof(*sge));
 177         return len;
 178 }
 179
 180 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i,
 181                          bool charge)
 182 {
 183         struct scatterlist *sge = sk_msg_elem(msg, i);
 184         int freed = 0;
 185
 186         while (msg->sg.size) {
 187                 msg->sg.size -= sge->length;
 188                 freed += sk_msg_free_elem(sk, msg, i, charge);
 189                 sk_msg_iter_var_next(i);
 190                 sk_msg_check_to_free(msg, i, msg->sg.size);
 191                 sge = sk_msg_elem(msg, i);
 192         }
 193         if (msg->skb)
 194                 consume_skb(msg->skb);
 195         sk_msg_init(msg);
 196         return freed;
 197 }
 198
 199 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg)
 200 {
 201         return __sk_msg_free(sk, msg, msg->sg.start, false);
 202 }
 203 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge);
 204
 205 int sk_msg_free(struct sock *sk, struct sk_msg *msg)
 206 {
 207         return __sk_msg_free(sk, msg, msg->sg.start, true);
 208 }
 209 EXPORT_SYMBOL_GPL(sk_msg_free);
 210
 211 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg,
 212                                   u32 bytes, bool charge)
 213 {
 214         struct scatterlist *sge;
 215         u32 i = msg->sg.start;
 216
 217         while (bytes) {
 218                 sge = sk_msg_elem(msg, i);
 219                 if (!sge->length)
 220                         break;
 221                 if (bytes < sge->length) {
 222                         if (charge)
 223                                 sk_mem_uncharge(sk, bytes);
 224                         sge->length -= bytes;
 225                         sge->offset += bytes;
 226                         msg->sg.size -= bytes;
 227                         break;
 228                 }
 229
 230                 msg->sg.size -= sge->length;
 231                 bytes -= sge->length;
 232                 sk_msg_free_elem(sk, msg, i, charge);
 233                 sk_msg_iter_var_next(i);
 234                 sk_msg_check_to_free(msg, i, bytes);
 235         }
 236         msg->sg.start = i;
 237 }
 238
 239 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes)
 240 {
 241         __sk_msg_free_partial(sk, msg, bytes, true);
 242 }
 243 EXPORT_SYMBOL_GPL(sk_msg_free_partial);
 244
 245 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
 246                                   u32 bytes)
 247 {
 248         __sk_msg_free_partial(sk, msg, bytes, false);
 249 }
 250
 251 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len)
 252 {
 253         int trim = msg->sg.size - len;
 254         u32 i = msg->sg.end;
 255
 256         if (trim <= 0) {
 257                 WARN_ON(trim < 0);
 258                 return;
 259         }
 260
 261         sk_msg_iter_var_prev(i);
 262         msg->sg.size = len;
 263         while (msg->sg.data[i].length &&
 264                trim >= msg->sg.data[i].length) {
 265                 trim -= msg->sg.data[i].length;
 266                 sk_msg_free_elem(sk, msg, i, true);
 267                 sk_msg_iter_var_prev(i);
 268                 if (!trim)
 269                         goto out;
 270         }
 271
 272         msg->sg.data[i].length -= trim;
 273         sk_mem_uncharge(sk, trim);
 274 out:
 275         /* If we trim data before curr pointer update copybreak and current
 276          * so that any future copy operations start at new copy location.
 277          * However trimed data that has not yet been used in a copy op
 278          * does not require an update.
 279          */
 280         if (msg->sg.curr >= i) {
 281                 msg->sg.curr = i;
 282                 msg->sg.copybreak = msg->sg.data[i].length;
 283         }
 284         sk_msg_iter_var_next(i);
 285         msg->sg.end = i;
 286 }
 287 EXPORT_SYMBOL_GPL(sk_msg_trim);
 288
 289 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
 290                               struct sk_msg *msg, u32 bytes)
 291 {
 292         int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg);
 293         const int to_max_pages = MAX_MSG_FRAGS;
 294         struct page *pages[MAX_MSG_FRAGS];
 295         ssize_t orig, copied, use, offset;
 296
 297         orig = msg->sg.size;
 298         while (bytes > 0) {
 299                 i = 0;
 300                 maxpages = to_max_pages - num_elems;
 301                 if (maxpages == 0) {
 302                         ret = -EFAULT;
 303                         goto out;
 304                 }
 305
 306                 copied = iov_iter_get_pages(from, pages, bytes, maxpages,
 307                                             &offset);
 308                 if (copied <= 0) {
 309                         ret = -EFAULT;
 310                         goto out;
 311                 }
 312
 313                 iov_iter_advance(from, copied);
 314                 bytes -= copied;
 315                 msg->sg.size += copied;
 316
 317                 while (copied) {
 318                         use = min_t(int, copied, PAGE_SIZE - offset);
 319                         sg_set_page(&msg->sg.data[msg->sg.end],
 320                                     pages[i], use, offset);
 321                         sg_unmark_end(&msg->sg.data[msg->sg.end]);
 322                         sk_mem_charge(sk, use);
 323
 324                         offset = 0;
 325                         copied -= use;
 326                         sk_msg_iter_next(msg, end);
 327                         num_elems++;
 328                         i++;
 329                 }
 330                 /* When zerocopy is mixed with sk_msg_*copy* operations we
 331                  * may have a copybreak set in this case clear and prefer
 332                  * zerocopy remainder when possible.
 333                  */
 334                 msg->sg.copybreak = 0;
 335                 msg->sg.curr = msg->sg.end;
 336         }
 337 out:
 338         /* Revert iov_iter updates, msg will need to use 'trim' later if it
 339          * also needs to be cleared.
 340          */
 341         if (ret)
 342                 iov_iter_revert(from, msg->sg.size - orig);
 343         return ret;
 344 }
 345 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter);
 346
 347 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
 348                              struct sk_msg *msg, u32 bytes)
 349 {
 350         int ret = -ENOSPC, i = msg->sg.curr;
 351         struct scatterlist *sge;
 352         u32 copy, buf_size;
 353         void *to;
 354
 355         do {
 356                 sge = sk_msg_elem(msg, i);
 357                 /* This is possible if a trim operation shrunk the buffer */
 358                 if (msg->sg.copybreak >= sge->length) {
 359                         msg->sg.copybreak = 0;
 360                         sk_msg_iter_var_next(i);
 361                         if (i == msg->sg.end)
 362                                 break;
 363                         sge = sk_msg_elem(msg, i);
 364                 }
 365
 366                 buf_size = sge->length - msg->sg.copybreak;
 367                 copy = (buf_size > bytes) ? bytes : buf_size;
 368                 to = sg_virt(sge) + msg->sg.copybreak;
 369                 msg->sg.copybreak += copy;
 370                 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY)
 371                         ret = copy_from_iter_nocache(to, copy, from);
 372                 else
 373                         ret = copy_from_iter(to, copy, from);
 374                 if (ret != copy) {
 375                         ret = -EFAULT;
 376                         goto out;
 377                 }
 378                 bytes -= copy;
 379                 if (!bytes)
 380                         break;
 381                 msg->sg.copybreak = 0;
 382                 sk_msg_iter_var_next(i);
 383         } while (i != msg->sg.end);
 384 out:
 385         msg->sg.curr = i;
 386         return ret;
 387 }
 388 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
 389
 390 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb)
 391 {
 392         struct sock *sk = psock->sk;
 393         int copied = 0, num_sge;
 394         struct sk_msg *msg;
 395
 396         msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC);
 397         if (unlikely(!msg))
 398                 return -EAGAIN;
 399         if (!sk_rmem_schedule(sk, skb, skb->len)) {
 400                 kfree(msg);
 401                 return -EAGAIN;
 402         }
 403
 404         sk_msg_init(msg);
 405         num_sge = skb_to_sgvec(skb, msg->sg.data, 0, skb->len);
 406         if (unlikely(num_sge < 0)) {
 407                 kfree(msg);
 408                 return num_sge;
 409         }
 410
 411         sk_mem_charge(sk, skb->len);
 412         copied = skb->len;
 413         msg->sg.start = 0;
 414         msg->sg.end = num_sge == MAX_MSG_FRAGS ? 0 : num_sge;
 415         msg->skb = skb;
 416
 417         sk_psock_queue_msg(psock, msg);
 418         sk_psock_data_ready(sk, psock);
 419         return copied;
 420 }
 421
 422 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,
 423                                u32 off, u32 len, bool ingress)
 424 {
 425         if (ingress)
 426                 return sk_psock_skb_ingress(psock, skb);
 427         else
 428                 return skb_send_sock_locked(psock->sk, skb, off, len);
 429 }
 430
 431 static void sk_psock_backlog(struct work_struct *work)
 432 {
 433         struct sk_psock *psock = container_of(work, struct sk_psock, work);
 434         struct sk_psock_work_state *state = &psock->work_state;
 435         struct sk_buff *skb;
 436         bool ingress;
 437         u32 len, off;
 438         int ret;
 439
 440         /* Lock sock to avoid losing sk_socket during loop. */
 441         lock_sock(psock->sk);
 442         if (state->skb) {
 443                 skb = state->skb;
 444                 len = state->len;
 445                 off = state->off;
 446                 state->skb = NULL;
 447                 goto start;
 448         }
 449
 450         while ((skb = skb_dequeue(&psock->ingress_skb))) {
 451                 len = skb->len;
 452                 off = 0;
 453 start:
 454                 ingress = tcp_skb_bpf_ingress(skb);
 455                 do {
 456                         ret = -EIO;
 457                         if (likely(psock->sk->sk_socket))
 458                                 ret = sk_psock_handle_skb(psock, skb, off,
 459                                                           len, ingress);
 460                         if (ret <= 0) {
 461                                 if (ret == -EAGAIN) {
 462                                         state->skb = skb;
 463                                         state->len = len;
 464                                         state->off = off;
 465                                         goto end;
 466                                 }
 467                                 /* Hard errors break pipe and stop xmit. */
 468                                 sk_psock_report_error(psock, ret ? -ret : EPIPE);
 469                                 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
 470                                 kfree_skb(skb);
 471                                 goto end;
 472                         }
 473                         off += ret;
 474                         len -= ret;
 475                 } while (len);
 476
 477                 if (!ingress)
 478                         kfree_skb(skb);
 479         }
 480 end:
 481         release_sock(psock->sk);
 482 }
 483
 484 struct sk_psock *sk_psock_init(struct sock *sk, int node)
 485 {
 486         struct sk_psock *psock = kzalloc_node(sizeof(*psock),
 487                                               GFP_ATOMIC | __GFP_NOWARN,
 488                                               node);
 489         if (!psock)
 490                 return NULL;
 491
 492         psock->sk = sk;
 493         psock->eval =  __SK_NONE;
 494
 495         INIT_LIST_HEAD(&psock->link);
 496         spin_lock_init(&psock->link_lock);
 497
 498         INIT_WORK(&psock->work, sk_psock_backlog);
 499         INIT_LIST_HEAD(&psock->ingress_msg);
 500         skb_queue_head_init(&psock->ingress_skb);
 501
 502         sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED);
 503         refcount_set(&psock->refcnt, 1);
 504
 505         rcu_assign_sk_user_data(sk, psock);
 506         sock_hold(sk);
 507
 508         return psock;
 509 }
 510 EXPORT_SYMBOL_GPL(sk_psock_init);
 511
 512 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock)
 513 {
 514         struct sk_psock_link *link;
 515
 516         spin_lock_bh(&psock->link_lock);
 517         link = list_first_entry_or_null(&psock->link, struct sk_psock_link,
 518                                         list);
 519         if (link)
 520                 list_del(&link->list);
 521         spin_unlock_bh(&psock->link_lock);
 522         return link;
 523 }
 524
 525 void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
 526 {
 527         struct sk_msg *msg, *tmp;
 528
 529         list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) {
 530                 list_del(&msg->list);
 531                 sk_msg_free(psock->sk, msg);
 532                 kfree(msg);
 533         }
 534 }
 535
 536 static void sk_psock_zap_ingress(struct sk_psock *psock)
 537 {
 538         __skb_queue_purge(&psock->ingress_skb);
 539         __sk_psock_purge_ingress_msg(psock);
 540 }
 541
 542 static void sk_psock_link_destroy(struct sk_psock *psock)
 543 {
 544         struct sk_psock_link *link, *tmp;
 545
 546         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 547                 list_del(&link->list);
 548                 sk_psock_free_link(link);
 549         }
 550 }
 551
 552 static void sk_psock_destroy_deferred(struct work_struct *gc)
 553 {
 554         struct sk_psock *psock = container_of(gc, struct sk_psock, gc);
 555
 556         /* No sk_callback_lock since already detached. */
 557         strp_stop(&psock->parser.strp);
 558         strp_done(&psock->parser.strp);
 559
 560         cancel_work_sync(&psock->work);
 561
 562         psock_progs_drop(&psock->progs);
 563
 564         sk_psock_link_destroy(psock);
 565         sk_psock_cork_free(psock);
 566         sk_psock_zap_ingress(psock);
 567
 568         if (psock->sk_redir)
 569                 sock_put(psock->sk_redir);
 570         sock_put(psock->sk);
 571         kfree(psock);
 572 }
 573
 574 void sk_psock_destroy(struct rcu_head *rcu)
 575 {
 576         struct sk_psock *psock = container_of(rcu, struct sk_psock, rcu);
 577
 578         INIT_WORK(&psock->gc, sk_psock_destroy_deferred);
 579         schedule_work(&psock->gc);
 580 }
 581 EXPORT_SYMBOL_GPL(sk_psock_destroy);
 582
 583 void sk_psock_drop(struct sock *sk, struct sk_psock *psock)
 584 {
 585         rcu_assign_sk_user_data(sk, NULL);
 586         sk_psock_cork_free(psock);
 587         sk_psock_zap_ingress(psock);
 588         sk_psock_restore_proto(sk, psock);
 589
 590         write_lock_bh(&sk->sk_callback_lock);
 591         if (psock->progs.skb_parser)
 592                 sk_psock_stop_strp(sk, psock);
 593         write_unlock_bh(&sk->sk_callback_lock);
 594         sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
 595
 596         call_rcu(&psock->rcu, sk_psock_destroy);
 597 }
 598 EXPORT_SYMBOL_GPL(sk_psock_drop);
 599
 600 static int sk_psock_map_verd(int verdict, bool redir)
 601 {
 602         switch (verdict) {
 603         case SK_PASS:
 604                 return redir ? __SK_REDIRECT : __SK_PASS;
 605         case SK_DROP:
 606         default:
 607                 break;
 608         }
 609
 610         return __SK_DROP;
 611 }
 612
 613 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
 614                          struct sk_msg *msg)
 615 {
 616         struct bpf_prog *prog;
 617         int ret;
 618
 619         preempt_disable();
 620         rcu_read_lock();
 621         prog = READ_ONCE(psock->progs.msg_parser);
 622         if (unlikely(!prog)) {
 623                 ret = __SK_PASS;
 624                 goto out;
 625         }
 626
 627         sk_msg_compute_data_pointers(msg);
 628         msg->sk = sk;
 629         ret = BPF_PROG_RUN(prog, msg);
 630         ret = sk_psock_map_verd(ret, msg->sk_redir);
 631         psock->apply_bytes = msg->apply_bytes;
 632         if (ret == __SK_REDIRECT) {
 633                 if (psock->sk_redir)
 634                         sock_put(psock->sk_redir);
 635                 psock->sk_redir = msg->sk_redir;
 636                 if (!psock->sk_redir) {
 637                         ret = __SK_DROP;
 638                         goto out;
 639                 }
 640                 sock_hold(psock->sk_redir);
 641         }
 642 out:
 643         rcu_read_unlock();
 644         preempt_enable();
 645         return ret;
 646 }
 647 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);
 648
 649 static int sk_psock_bpf_run(struct sk_psock *psock, struct bpf_prog *prog,
 650                             struct sk_buff *skb)
 651 {
 652         int ret;
 653
 654         skb->sk = psock->sk;
 655         bpf_compute_data_end_sk_skb(skb);
 656         preempt_disable();
 657         ret = BPF_PROG_RUN(prog, skb);
 658         preempt_enable();
 659         /* strparser clones the skb before handing it to a upper layer,
 660          * meaning skb_orphan has been called. We NULL sk on the way out
 661          * to ensure we don't trigger a BUG_ON() in skb/sk operations
 662          * later and because we are not charging the memory of this skb
 663          * to any socket yet.
 664          */
 665         skb->sk = NULL;
 666         return ret;
 667 }
 668
 669 static struct sk_psock *sk_psock_from_strp(struct strparser *strp)
 670 {
 671         struct sk_psock_parser *parser;
 672
 673         parser = container_of(strp, struct sk_psock_parser, strp);
 674         return container_of(parser, struct sk_psock, parser);
 675 }
 676
 677 static void sk_psock_verdict_apply(struct sk_psock *psock,
 678                                    struct sk_buff *skb, int verdict)
 679 {
 680         struct sk_psock *psock_other;
 681         struct sock *sk_other;
 682         bool ingress;
 683
 684         switch (verdict) {
 685         case __SK_PASS:
 686                 sk_other = psock->sk;
 687                 if (sock_flag(sk_other, SOCK_DEAD) ||
 688                     !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
 689                         goto out_free;
 690                 }
 691                 if (atomic_read(&sk_other->sk_rmem_alloc) <=
 692                     sk_other->sk_rcvbuf) {
 693                         struct tcp_skb_cb *tcp = TCP_SKB_CB(skb);
 694
 695                         tcp->bpf.flags |= BPF_F_INGRESS;
 696                         skb_queue_tail(&psock->ingress_skb, skb);
 697                         schedule_work(&psock->work);
 698                         break;
 699                 }
 700                 goto out_free;
 701         case __SK_REDIRECT:
 702                 sk_other = tcp_skb_bpf_redirect_fetch(skb);
 703                 if (unlikely(!sk_other))
 704                         goto out_free;
 705                 psock_other = sk_psock(sk_other);
 706                 if (!psock_other || sock_flag(sk_other, SOCK_DEAD) ||
 707                     !sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED))
 708                         goto out_free;
 709                 ingress = tcp_skb_bpf_ingress(skb);
 710                 if ((!ingress && sock_writeable(sk_other)) ||
 711                     (ingress &&
 712                      atomic_read(&sk_other->sk_rmem_alloc) <=
 713                      sk_other->sk_rcvbuf)) {
 714                         if (!ingress)
 715                                 skb_set_owner_w(skb, sk_other);
 716                         skb_queue_tail(&psock_other->ingress_skb, skb);
 717                         schedule_work(&psock_other->work);
 718                         break;
 719                 }
 720                 /* fall-through */
 721         case __SK_DROP:
 722                 /* fall-through */
 723         default:
 724 out_free:
 725                 kfree_skb(skb);
 726         }
 727 }
 728
 729 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb)
 730 {
 731         struct sk_psock *psock = sk_psock_from_strp(strp);
 732         struct bpf_prog *prog;
 733         int ret = __SK_DROP;
 734
 735         rcu_read_lock();
 736         prog = READ_ONCE(psock->progs.skb_verdict);
 737         if (likely(prog)) {
 738                 skb_orphan(skb);
 739                 tcp_skb_bpf_redirect_clear(skb);
 740                 ret = sk_psock_bpf_run(psock, prog, skb);
 741                 ret = sk_psock_map_verd(ret, tcp_skb_bpf_redirect_fetch(skb));
 742         }
 743         rcu_read_unlock();
 744         sk_psock_verdict_apply(psock, skb, ret);
 745 }
 746
 747 static int sk_psock_strp_read_done(struct strparser *strp, int err)
 748 {
 749         return err;
 750 }
 751
 752 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb)
 753 {
 754         struct sk_psock *psock = sk_psock_from_strp(strp);
 755         struct bpf_prog *prog;
 756         int ret = skb->len;
 757
 758         rcu_read_lock();
 759         prog = READ_ONCE(psock->progs.skb_parser);
 760         if (likely(prog))
 761                 ret = sk_psock_bpf_run(psock, prog, skb);
 762         rcu_read_unlock();
 763         return ret;
 764 }
 765
 766 /* Called with socket lock held. */
 767 static void sk_psock_strp_data_ready(struct sock *sk)
 768 {
 769         struct sk_psock *psock;
 770
 771         rcu_read_lock();
 772         psock = sk_psock(sk);
 773         if (likely(psock)) {
 774                 write_lock_bh(&sk->sk_callback_lock);
 775                 strp_data_ready(&psock->parser.strp);
 776                 write_unlock_bh(&sk->sk_callback_lock);
 777         }
 778         rcu_read_unlock();
 779 }
 780
 781 static void sk_psock_write_space(struct sock *sk)
 782 {
 783         struct sk_psock *psock;
 784         void (*write_space)(struct sock *sk);
 785
 786         rcu_read_lock();
 787         psock = sk_psock(sk);
 788         if (likely(psock && sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)))
 789                 schedule_work(&psock->work);
 790         write_space = psock->saved_write_space;
 791         rcu_read_unlock();
 792         write_space(sk);
 793 }
 794
 795 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
 796 {
 797         static const struct strp_callbacks cb = {
 798                 .rcv_msg        = sk_psock_strp_read,
 799                 .read_sock_done = sk_psock_strp_read_done,
 800                 .parse_msg      = sk_psock_strp_parse,
 801         };
 802
 803         psock->parser.enabled = false;
 804         return strp_init(&psock->parser.strp, sk, &cb);
 805 }
 806
 807 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
 808 {
 809         struct sk_psock_parser *parser = &psock->parser;
 810
 811         if (parser->enabled)
 812                 return;
 813
 814         parser->saved_data_ready = sk->sk_data_ready;
 815         sk->sk_data_ready = sk_psock_strp_data_ready;
 816         sk->sk_write_space = sk_psock_write_space;
 817         parser->enabled = true;
 818 }
 819
 820 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
 821 {
 822         struct sk_psock_parser *parser = &psock->parser;
 823
 824         if (!parser->enabled)
 825                 return;
 826
 827         sk->sk_data_ready = parser->saved_data_ready;
 828         parser->saved_data_ready = NULL;
 829         strp_stop(&parser->strp);
 830         parser->enabled = false;
 831 }