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 #include <net/tls.h> 11 12 static bool sk_msg_try_coalesce_ok(struct sk_msg *msg, int elem_first_coalesce) 13 { 14 if (msg->sg.end > msg->sg.start && 15 elem_first_coalesce < msg->sg.end) 16 return true; 17 18 if (msg->sg.end < msg->sg.start && 19 (elem_first_coalesce > msg->sg.start || 20 elem_first_coalesce < msg->sg.end)) 21 return true; 22 23 return false; 24 } 25 26 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len, 27 int elem_first_coalesce) 28 { 29 struct page_frag *pfrag = sk_page_frag(sk); 30 u32 osize = msg->sg.size; 31 int ret = 0; 32 33 len -= msg->sg.size; 34 while (len > 0) { 35 struct scatterlist *sge; 36 u32 orig_offset; 37 int use, i; 38 39 if (!sk_page_frag_refill(sk, pfrag)) { 40 ret = -ENOMEM; 41 goto msg_trim; 42 } 43 44 orig_offset = pfrag->offset; 45 use = min_t(int, len, pfrag->size - orig_offset); 46 if (!sk_wmem_schedule(sk, use)) { 47 ret = -ENOMEM; 48 goto msg_trim; 49 } 50 51 i = msg->sg.end; 52 sk_msg_iter_var_prev(i); 53 sge = &msg->sg.data[i]; 54 55 if (sk_msg_try_coalesce_ok(msg, elem_first_coalesce) && 56 sg_page(sge) == pfrag->page && 57 sge->offset + sge->length == orig_offset) { 58 sge->length += use; 59 } else { 60 if (sk_msg_full(msg)) { 61 ret = -ENOSPC; 62 break; 63 } 64 65 sge = &msg->sg.data[msg->sg.end]; 66 sg_unmark_end(sge); 67 sg_set_page(sge, pfrag->page, use, orig_offset); 68 get_page(pfrag->page); 69 sk_msg_iter_next(msg, end); 70 } 71 72 sk_mem_charge(sk, use); 73 msg->sg.size += use; 74 pfrag->offset += use; 75 len -= use; 76 } 77 78 return ret; 79 80 msg_trim: 81 sk_msg_trim(sk, msg, osize); 82 return ret; 83 } 84 EXPORT_SYMBOL_GPL(sk_msg_alloc); 85 86 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src, 87 u32 off, u32 len) 88 { 89 int i = src->sg.start; 90 struct scatterlist *sge = sk_msg_elem(src, i); 91 struct scatterlist *sgd = NULL; 92 u32 sge_len, sge_off; 93 94 while (off) { 95 if (sge->length > off) 96 break; 97 off -= sge->length; 98 sk_msg_iter_var_next(i); 99 if (i == src->sg.end && off) 100 return -ENOSPC; 101 sge = sk_msg_elem(src, i); 102 } 103 104 while (len) { 105 sge_len = sge->length - off; 106 if (sge_len > len) 107 sge_len = len; 108 109 if (dst->sg.end) 110 sgd = sk_msg_elem(dst, dst->sg.end - 1); 111 112 if (sgd && 113 (sg_page(sge) == sg_page(sgd)) && 114 (sg_virt(sge) + off == sg_virt(sgd) + sgd->length)) { 115 sgd->length += sge_len; 116 dst->sg.size += sge_len; 117 } else if (!sk_msg_full(dst)) { 118 sge_off = sge->offset + off; 119 sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off); 120 } else { 121 return -ENOSPC; 122 } 123 124 off = 0; 125 len -= sge_len; 126 sk_mem_charge(sk, sge_len); 127 sk_msg_iter_var_next(i); 128 if (i == src->sg.end && len) 129 return -ENOSPC; 130 sge = sk_msg_elem(src, i); 131 } 132 133 return 0; 134 } 135 EXPORT_SYMBOL_GPL(sk_msg_clone); 136 137 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes) 138 { 139 int i = msg->sg.start; 140 141 do { 142 struct scatterlist *sge = sk_msg_elem(msg, i); 143 144 if (bytes < sge->length) { 145 sge->length -= bytes; 146 sge->offset += bytes; 147 sk_mem_uncharge(sk, bytes); 148 break; 149 } 150 151 sk_mem_uncharge(sk, sge->length); 152 bytes -= sge->length; 153 sge->length = 0; 154 sge->offset = 0; 155 sk_msg_iter_var_next(i); 156 } while (bytes && i != msg->sg.end); 157 msg->sg.start = i; 158 } 159 EXPORT_SYMBOL_GPL(sk_msg_return_zero); 160 161 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes) 162 { 163 int i = msg->sg.start; 164 165 do { 166 struct scatterlist *sge = &msg->sg.data[i]; 167 int uncharge = (bytes < sge->length) ? bytes : sge->length; 168 169 sk_mem_uncharge(sk, uncharge); 170 bytes -= uncharge; 171 sk_msg_iter_var_next(i); 172 } while (i != msg->sg.end); 173 } 174 EXPORT_SYMBOL_GPL(sk_msg_return); 175 176 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i, 177 bool charge) 178 { 179 struct scatterlist *sge = sk_msg_elem(msg, i); 180 u32 len = sge->length; 181 182 /* When the skb owns the memory we free it from consume_skb path. */ 183 if (!msg->skb) { 184 if (charge) 185 sk_mem_uncharge(sk, len); 186 put_page(sg_page(sge)); 187 } 188 memset(sge, 0, sizeof(*sge)); 189 return len; 190 } 191 192 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i, 193 bool charge) 194 { 195 struct scatterlist *sge = sk_msg_elem(msg, i); 196 int freed = 0; 197 198 while (msg->sg.size) { 199 msg->sg.size -= sge->length; 200 freed += sk_msg_free_elem(sk, msg, i, charge); 201 sk_msg_iter_var_next(i); 202 sk_msg_check_to_free(msg, i, msg->sg.size); 203 sge = sk_msg_elem(msg, i); 204 } 205 consume_skb(msg->skb); 206 sk_msg_init(msg); 207 return freed; 208 } 209 210 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg) 211 { 212 return __sk_msg_free(sk, msg, msg->sg.start, false); 213 } 214 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge); 215 216 int sk_msg_free(struct sock *sk, struct sk_msg *msg) 217 { 218 return __sk_msg_free(sk, msg, msg->sg.start, true); 219 } 220 EXPORT_SYMBOL_GPL(sk_msg_free); 221 222 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, 223 u32 bytes, bool charge) 224 { 225 struct scatterlist *sge; 226 u32 i = msg->sg.start; 227 228 while (bytes) { 229 sge = sk_msg_elem(msg, i); 230 if (!sge->length) 231 break; 232 if (bytes < sge->length) { 233 if (charge) 234 sk_mem_uncharge(sk, bytes); 235 sge->length -= bytes; 236 sge->offset += bytes; 237 msg->sg.size -= bytes; 238 break; 239 } 240 241 msg->sg.size -= sge->length; 242 bytes -= sge->length; 243 sk_msg_free_elem(sk, msg, i, charge); 244 sk_msg_iter_var_next(i); 245 sk_msg_check_to_free(msg, i, bytes); 246 } 247 msg->sg.start = i; 248 } 249 250 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes) 251 { 252 __sk_msg_free_partial(sk, msg, bytes, true); 253 } 254 EXPORT_SYMBOL_GPL(sk_msg_free_partial); 255 256 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg, 257 u32 bytes) 258 { 259 __sk_msg_free_partial(sk, msg, bytes, false); 260 } 261 262 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len) 263 { 264 int trim = msg->sg.size - len; 265 u32 i = msg->sg.end; 266 267 if (trim <= 0) { 268 WARN_ON(trim < 0); 269 return; 270 } 271 272 sk_msg_iter_var_prev(i); 273 msg->sg.size = len; 274 while (msg->sg.data[i].length && 275 trim >= msg->sg.data[i].length) { 276 trim -= msg->sg.data[i].length; 277 sk_msg_free_elem(sk, msg, i, true); 278 sk_msg_iter_var_prev(i); 279 if (!trim) 280 goto out; 281 } 282 283 msg->sg.data[i].length -= trim; 284 sk_mem_uncharge(sk, trim); 285 /* Adjust copybreak if it falls into the trimmed part of last buf */ 286 if (msg->sg.curr == i && msg->sg.copybreak > msg->sg.data[i].length) 287 msg->sg.copybreak = msg->sg.data[i].length; 288 out: 289 sk_msg_iter_var_next(i); 290 msg->sg.end = i; 291 292 /* If we trim data a full sg elem before curr pointer update 293 * copybreak and current so that any future copy operations 294 * start at new copy location. 295 * However trimed data that has not yet been used in a copy op 296 * does not require an update. 297 */ 298 if (!msg->sg.size) { 299 msg->sg.curr = msg->sg.start; 300 msg->sg.copybreak = 0; 301 } else if (sk_msg_iter_dist(msg->sg.start, msg->sg.curr) >= 302 sk_msg_iter_dist(msg->sg.start, msg->sg.end)) { 303 sk_msg_iter_var_prev(i); 304 msg->sg.curr = i; 305 msg->sg.copybreak = msg->sg.data[i].length; 306 } 307 } 308 EXPORT_SYMBOL_GPL(sk_msg_trim); 309 310 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from, 311 struct sk_msg *msg, u32 bytes) 312 { 313 int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg); 314 const int to_max_pages = MAX_MSG_FRAGS; 315 struct page *pages[MAX_MSG_FRAGS]; 316 ssize_t orig, copied, use, offset; 317 318 orig = msg->sg.size; 319 while (bytes > 0) { 320 i = 0; 321 maxpages = to_max_pages - num_elems; 322 if (maxpages == 0) { 323 ret = -EFAULT; 324 goto out; 325 } 326 327 copied = iov_iter_get_pages(from, pages, bytes, maxpages, 328 &offset); 329 if (copied <= 0) { 330 ret = -EFAULT; 331 goto out; 332 } 333 334 iov_iter_advance(from, copied); 335 bytes -= copied; 336 msg->sg.size += copied; 337 338 while (copied) { 339 use = min_t(int, copied, PAGE_SIZE - offset); 340 sg_set_page(&msg->sg.data[msg->sg.end], 341 pages[i], use, offset); 342 sg_unmark_end(&msg->sg.data[msg->sg.end]); 343 sk_mem_charge(sk, use); 344 345 offset = 0; 346 copied -= use; 347 sk_msg_iter_next(msg, end); 348 num_elems++; 349 i++; 350 } 351 /* When zerocopy is mixed with sk_msg_*copy* operations we 352 * may have a copybreak set in this case clear and prefer 353 * zerocopy remainder when possible. 354 */ 355 msg->sg.copybreak = 0; 356 msg->sg.curr = msg->sg.end; 357 } 358 out: 359 /* Revert iov_iter updates, msg will need to use 'trim' later if it 360 * also needs to be cleared. 361 */ 362 if (ret) 363 iov_iter_revert(from, msg->sg.size - orig); 364 return ret; 365 } 366 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter); 367 368 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from, 369 struct sk_msg *msg, u32 bytes) 370 { 371 int ret = -ENOSPC, i = msg->sg.curr; 372 struct scatterlist *sge; 373 u32 copy, buf_size; 374 void *to; 375 376 do { 377 sge = sk_msg_elem(msg, i); 378 /* This is possible if a trim operation shrunk the buffer */ 379 if (msg->sg.copybreak >= sge->length) { 380 msg->sg.copybreak = 0; 381 sk_msg_iter_var_next(i); 382 if (i == msg->sg.end) 383 break; 384 sge = sk_msg_elem(msg, i); 385 } 386 387 buf_size = sge->length - msg->sg.copybreak; 388 copy = (buf_size > bytes) ? bytes : buf_size; 389 to = sg_virt(sge) + msg->sg.copybreak; 390 msg->sg.copybreak += copy; 391 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) 392 ret = copy_from_iter_nocache(to, copy, from); 393 else 394 ret = copy_from_iter(to, copy, from); 395 if (ret != copy) { 396 ret = -EFAULT; 397 goto out; 398 } 399 bytes -= copy; 400 if (!bytes) 401 break; 402 msg->sg.copybreak = 0; 403 sk_msg_iter_var_next(i); 404 } while (i != msg->sg.end); 405 out: 406 msg->sg.curr = i; 407 return ret; 408 } 409 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter); 410 411 /* Receive sk_msg from psock->ingress_msg to @msg. */ 412 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg, 413 int len, int flags) 414 { 415 struct iov_iter *iter = &msg->msg_iter; 416 int peek = flags & MSG_PEEK; 417 struct sk_msg *msg_rx; 418 int i, copied = 0; 419 420 msg_rx = sk_psock_peek_msg(psock); 421 while (copied != len) { 422 struct scatterlist *sge; 423 424 if (unlikely(!msg_rx)) 425 break; 426 427 i = msg_rx->sg.start; 428 do { 429 struct page *page; 430 int copy; 431 432 sge = sk_msg_elem(msg_rx, i); 433 copy = sge->length; 434 page = sg_page(sge); 435 if (copied + copy > len) 436 copy = len - copied; 437 copy = copy_page_to_iter(page, sge->offset, copy, iter); 438 if (!copy) 439 return copied ? copied : -EFAULT; 440 441 copied += copy; 442 if (likely(!peek)) { 443 sge->offset += copy; 444 sge->length -= copy; 445 if (!msg_rx->skb) 446 sk_mem_uncharge(sk, copy); 447 msg_rx->sg.size -= copy; 448 449 if (!sge->length) { 450 sk_msg_iter_var_next(i); 451 if (!msg_rx->skb) 452 put_page(page); 453 } 454 } else { 455 /* Lets not optimize peek case if copy_page_to_iter 456 * didn't copy the entire length lets just break. 457 */ 458 if (copy != sge->length) 459 return copied; 460 sk_msg_iter_var_next(i); 461 } 462 463 if (copied == len) 464 break; 465 } while (i != msg_rx->sg.end); 466 467 if (unlikely(peek)) { 468 msg_rx = sk_psock_next_msg(psock, msg_rx); 469 if (!msg_rx) 470 break; 471 continue; 472 } 473 474 msg_rx->sg.start = i; 475 if (!sge->length && msg_rx->sg.start == msg_rx->sg.end) { 476 msg_rx = sk_psock_dequeue_msg(psock); 477 kfree_sk_msg(msg_rx); 478 } 479 msg_rx = sk_psock_peek_msg(psock); 480 } 481 482 return copied; 483 } 484 EXPORT_SYMBOL_GPL(sk_msg_recvmsg); 485 486 bool sk_msg_is_readable(struct sock *sk) 487 { 488 struct sk_psock *psock; 489 bool empty = true; 490 491 rcu_read_lock(); 492 psock = sk_psock(sk); 493 if (likely(psock)) 494 empty = list_empty(&psock->ingress_msg); 495 rcu_read_unlock(); 496 return !empty; 497 } 498 EXPORT_SYMBOL_GPL(sk_msg_is_readable); 499 500 static struct sk_msg *sk_psock_create_ingress_msg(struct sock *sk, 501 struct sk_buff *skb) 502 { 503 struct sk_msg *msg; 504 505 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) 506 return NULL; 507 508 if (!sk_rmem_schedule(sk, skb, skb->truesize)) 509 return NULL; 510 511 msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_KERNEL); 512 if (unlikely(!msg)) 513 return NULL; 514 515 sk_msg_init(msg); 516 return msg; 517 } 518 519 static int sk_psock_skb_ingress_enqueue(struct sk_buff *skb, 520 u32 off, u32 len, 521 struct sk_psock *psock, 522 struct sock *sk, 523 struct sk_msg *msg) 524 { 525 int num_sge, copied; 526 527 num_sge = skb_to_sgvec(skb, msg->sg.data, off, len); 528 if (num_sge < 0) { 529 /* skb linearize may fail with ENOMEM, but lets simply try again 530 * later if this happens. Under memory pressure we don't want to 531 * drop the skb. We need to linearize the skb so that the mapping 532 * in skb_to_sgvec can not error. 533 */ 534 if (skb_linearize(skb)) 535 return -EAGAIN; 536 537 num_sge = skb_to_sgvec(skb, msg->sg.data, off, len); 538 if (unlikely(num_sge < 0)) 539 return num_sge; 540 } 541 542 copied = len; 543 msg->sg.start = 0; 544 msg->sg.size = copied; 545 msg->sg.end = num_sge; 546 msg->skb = skb; 547 548 sk_psock_queue_msg(psock, msg); 549 sk_psock_data_ready(sk, psock); 550 return copied; 551 } 552 553 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb, 554 u32 off, u32 len); 555 556 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb, 557 u32 off, u32 len) 558 { 559 struct sock *sk = psock->sk; 560 struct sk_msg *msg; 561 int err; 562 563 /* If we are receiving on the same sock skb->sk is already assigned, 564 * skip memory accounting and owner transition seeing it already set 565 * correctly. 566 */ 567 if (unlikely(skb->sk == sk)) 568 return sk_psock_skb_ingress_self(psock, skb, off, len); 569 msg = sk_psock_create_ingress_msg(sk, skb); 570 if (!msg) 571 return -EAGAIN; 572 573 /* This will transition ownership of the data from the socket where 574 * the BPF program was run initiating the redirect to the socket 575 * we will eventually receive this data on. The data will be released 576 * from skb_consume found in __tcp_bpf_recvmsg() after its been copied 577 * into user buffers. 578 */ 579 skb_set_owner_r(skb, sk); 580 err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg); 581 if (err < 0) 582 kfree(msg); 583 return err; 584 } 585 586 /* Puts an skb on the ingress queue of the socket already assigned to the 587 * skb. In this case we do not need to check memory limits or skb_set_owner_r 588 * because the skb is already accounted for here. 589 */ 590 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb, 591 u32 off, u32 len) 592 { 593 struct sk_msg *msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC); 594 struct sock *sk = psock->sk; 595 int err; 596 597 if (unlikely(!msg)) 598 return -EAGAIN; 599 sk_msg_init(msg); 600 skb_set_owner_r(skb, sk); 601 err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg); 602 if (err < 0) 603 kfree(msg); 604 return err; 605 } 606 607 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb, 608 u32 off, u32 len, bool ingress) 609 { 610 if (!ingress) { 611 if (!sock_writeable(psock->sk)) 612 return -EAGAIN; 613 return skb_send_sock(psock->sk, skb, off, len); 614 } 615 return sk_psock_skb_ingress(psock, skb, off, len); 616 } 617 618 static void sk_psock_skb_state(struct sk_psock *psock, 619 struct sk_psock_work_state *state, 620 struct sk_buff *skb, 621 int len, int off) 622 { 623 spin_lock_bh(&psock->ingress_lock); 624 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) { 625 state->skb = skb; 626 state->len = len; 627 state->off = off; 628 } else { 629 sock_drop(psock->sk, skb); 630 } 631 spin_unlock_bh(&psock->ingress_lock); 632 } 633 634 static void sk_psock_backlog(struct work_struct *work) 635 { 636 struct sk_psock *psock = container_of(work, struct sk_psock, work); 637 struct sk_psock_work_state *state = &psock->work_state; 638 struct sk_buff *skb = NULL; 639 bool ingress; 640 u32 len, off; 641 int ret; 642 643 mutex_lock(&psock->work_mutex); 644 if (unlikely(state->skb)) { 645 spin_lock_bh(&psock->ingress_lock); 646 skb = state->skb; 647 len = state->len; 648 off = state->off; 649 state->skb = NULL; 650 spin_unlock_bh(&psock->ingress_lock); 651 } 652 if (skb) 653 goto start; 654 655 while ((skb = skb_dequeue(&psock->ingress_skb))) { 656 len = skb->len; 657 off = 0; 658 if (skb_bpf_strparser(skb)) { 659 struct strp_msg *stm = strp_msg(skb); 660 661 off = stm->offset; 662 len = stm->full_len; 663 } 664 start: 665 ingress = skb_bpf_ingress(skb); 666 skb_bpf_redirect_clear(skb); 667 do { 668 ret = -EIO; 669 if (!sock_flag(psock->sk, SOCK_DEAD)) 670 ret = sk_psock_handle_skb(psock, skb, off, 671 len, ingress); 672 if (ret <= 0) { 673 if (ret == -EAGAIN) { 674 sk_psock_skb_state(psock, state, skb, 675 len, off); 676 goto end; 677 } 678 /* Hard errors break pipe and stop xmit. */ 679 sk_psock_report_error(psock, ret ? -ret : EPIPE); 680 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED); 681 sock_drop(psock->sk, skb); 682 goto end; 683 } 684 off += ret; 685 len -= ret; 686 } while (len); 687 688 if (!ingress) 689 kfree_skb(skb); 690 } 691 end: 692 mutex_unlock(&psock->work_mutex); 693 } 694 695 struct sk_psock *sk_psock_init(struct sock *sk, int node) 696 { 697 struct sk_psock *psock; 698 struct proto *prot; 699 700 write_lock_bh(&sk->sk_callback_lock); 701 702 if (sk_is_inet(sk) && inet_csk_has_ulp(sk)) { 703 psock = ERR_PTR(-EINVAL); 704 goto out; 705 } 706 707 if (sk->sk_user_data) { 708 psock = ERR_PTR(-EBUSY); 709 goto out; 710 } 711 712 psock = kzalloc_node(sizeof(*psock), GFP_ATOMIC | __GFP_NOWARN, node); 713 if (!psock) { 714 psock = ERR_PTR(-ENOMEM); 715 goto out; 716 } 717 718 prot = READ_ONCE(sk->sk_prot); 719 psock->sk = sk; 720 psock->eval = __SK_NONE; 721 psock->sk_proto = prot; 722 psock->saved_unhash = prot->unhash; 723 psock->saved_close = prot->close; 724 psock->saved_write_space = sk->sk_write_space; 725 726 INIT_LIST_HEAD(&psock->link); 727 spin_lock_init(&psock->link_lock); 728 729 INIT_WORK(&psock->work, sk_psock_backlog); 730 mutex_init(&psock->work_mutex); 731 INIT_LIST_HEAD(&psock->ingress_msg); 732 spin_lock_init(&psock->ingress_lock); 733 skb_queue_head_init(&psock->ingress_skb); 734 735 sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED); 736 refcount_set(&psock->refcnt, 1); 737 738 rcu_assign_sk_user_data_nocopy(sk, psock); 739 sock_hold(sk); 740 741 out: 742 write_unlock_bh(&sk->sk_callback_lock); 743 return psock; 744 } 745 EXPORT_SYMBOL_GPL(sk_psock_init); 746 747 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock) 748 { 749 struct sk_psock_link *link; 750 751 spin_lock_bh(&psock->link_lock); 752 link = list_first_entry_or_null(&psock->link, struct sk_psock_link, 753 list); 754 if (link) 755 list_del(&link->list); 756 spin_unlock_bh(&psock->link_lock); 757 return link; 758 } 759 760 static void __sk_psock_purge_ingress_msg(struct sk_psock *psock) 761 { 762 struct sk_msg *msg, *tmp; 763 764 list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) { 765 list_del(&msg->list); 766 sk_msg_free(psock->sk, msg); 767 kfree(msg); 768 } 769 } 770 771 static void __sk_psock_zap_ingress(struct sk_psock *psock) 772 { 773 struct sk_buff *skb; 774 775 while ((skb = skb_dequeue(&psock->ingress_skb)) != NULL) { 776 skb_bpf_redirect_clear(skb); 777 sock_drop(psock->sk, skb); 778 } 779 kfree_skb(psock->work_state.skb); 780 /* We null the skb here to ensure that calls to sk_psock_backlog 781 * do not pick up the free'd skb. 782 */ 783 psock->work_state.skb = NULL; 784 __sk_psock_purge_ingress_msg(psock); 785 } 786 787 static void sk_psock_link_destroy(struct sk_psock *psock) 788 { 789 struct sk_psock_link *link, *tmp; 790 791 list_for_each_entry_safe(link, tmp, &psock->link, list) { 792 list_del(&link->list); 793 sk_psock_free_link(link); 794 } 795 } 796 797 void sk_psock_stop(struct sk_psock *psock, bool wait) 798 { 799 spin_lock_bh(&psock->ingress_lock); 800 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED); 801 sk_psock_cork_free(psock); 802 __sk_psock_zap_ingress(psock); 803 spin_unlock_bh(&psock->ingress_lock); 804 805 if (wait) 806 cancel_work_sync(&psock->work); 807 } 808 809 static void sk_psock_done_strp(struct sk_psock *psock); 810 811 static void sk_psock_destroy(struct work_struct *work) 812 { 813 struct sk_psock *psock = container_of(to_rcu_work(work), 814 struct sk_psock, rwork); 815 /* No sk_callback_lock since already detached. */ 816 817 sk_psock_done_strp(psock); 818 819 cancel_work_sync(&psock->work); 820 mutex_destroy(&psock->work_mutex); 821 822 psock_progs_drop(&psock->progs); 823 824 sk_psock_link_destroy(psock); 825 sk_psock_cork_free(psock); 826 827 if (psock->sk_redir) 828 sock_put(psock->sk_redir); 829 sock_put(psock->sk); 830 kfree(psock); 831 } 832 833 void sk_psock_drop(struct sock *sk, struct sk_psock *psock) 834 { 835 write_lock_bh(&sk->sk_callback_lock); 836 sk_psock_restore_proto(sk, psock); 837 rcu_assign_sk_user_data(sk, NULL); 838 if (psock->progs.stream_parser) 839 sk_psock_stop_strp(sk, psock); 840 else if (psock->progs.stream_verdict || psock->progs.skb_verdict) 841 sk_psock_stop_verdict(sk, psock); 842 write_unlock_bh(&sk->sk_callback_lock); 843 844 sk_psock_stop(psock, false); 845 846 INIT_RCU_WORK(&psock->rwork, sk_psock_destroy); 847 queue_rcu_work(system_wq, &psock->rwork); 848 } 849 EXPORT_SYMBOL_GPL(sk_psock_drop); 850 851 static int sk_psock_map_verd(int verdict, bool redir) 852 { 853 switch (verdict) { 854 case SK_PASS: 855 return redir ? __SK_REDIRECT : __SK_PASS; 856 case SK_DROP: 857 default: 858 break; 859 } 860 861 return __SK_DROP; 862 } 863 864 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock, 865 struct sk_msg *msg) 866 { 867 struct bpf_prog *prog; 868 int ret; 869 870 rcu_read_lock(); 871 prog = READ_ONCE(psock->progs.msg_parser); 872 if (unlikely(!prog)) { 873 ret = __SK_PASS; 874 goto out; 875 } 876 877 sk_msg_compute_data_pointers(msg); 878 msg->sk = sk; 879 ret = bpf_prog_run_pin_on_cpu(prog, msg); 880 ret = sk_psock_map_verd(ret, msg->sk_redir); 881 psock->apply_bytes = msg->apply_bytes; 882 if (ret == __SK_REDIRECT) { 883 if (psock->sk_redir) 884 sock_put(psock->sk_redir); 885 psock->sk_redir = msg->sk_redir; 886 if (!psock->sk_redir) { 887 ret = __SK_DROP; 888 goto out; 889 } 890 sock_hold(psock->sk_redir); 891 } 892 out: 893 rcu_read_unlock(); 894 return ret; 895 } 896 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict); 897 898 static int sk_psock_skb_redirect(struct sk_psock *from, struct sk_buff *skb) 899 { 900 struct sk_psock *psock_other; 901 struct sock *sk_other; 902 903 sk_other = skb_bpf_redirect_fetch(skb); 904 /* This error is a buggy BPF program, it returned a redirect 905 * return code, but then didn't set a redirect interface. 906 */ 907 if (unlikely(!sk_other)) { 908 skb_bpf_redirect_clear(skb); 909 sock_drop(from->sk, skb); 910 return -EIO; 911 } 912 psock_other = sk_psock(sk_other); 913 /* This error indicates the socket is being torn down or had another 914 * error that caused the pipe to break. We can't send a packet on 915 * a socket that is in this state so we drop the skb. 916 */ 917 if (!psock_other || sock_flag(sk_other, SOCK_DEAD)) { 918 skb_bpf_redirect_clear(skb); 919 sock_drop(from->sk, skb); 920 return -EIO; 921 } 922 spin_lock_bh(&psock_other->ingress_lock); 923 if (!sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED)) { 924 spin_unlock_bh(&psock_other->ingress_lock); 925 skb_bpf_redirect_clear(skb); 926 sock_drop(from->sk, skb); 927 return -EIO; 928 } 929 930 skb_queue_tail(&psock_other->ingress_skb, skb); 931 schedule_work(&psock_other->work); 932 spin_unlock_bh(&psock_other->ingress_lock); 933 return 0; 934 } 935 936 static void sk_psock_tls_verdict_apply(struct sk_buff *skb, 937 struct sk_psock *from, int verdict) 938 { 939 switch (verdict) { 940 case __SK_REDIRECT: 941 sk_psock_skb_redirect(from, skb); 942 break; 943 case __SK_PASS: 944 case __SK_DROP: 945 default: 946 break; 947 } 948 } 949 950 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb) 951 { 952 struct bpf_prog *prog; 953 int ret = __SK_PASS; 954 955 rcu_read_lock(); 956 prog = READ_ONCE(psock->progs.stream_verdict); 957 if (likely(prog)) { 958 skb->sk = psock->sk; 959 skb_dst_drop(skb); 960 skb_bpf_redirect_clear(skb); 961 ret = bpf_prog_run_pin_on_cpu(prog, skb); 962 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb)); 963 skb->sk = NULL; 964 } 965 sk_psock_tls_verdict_apply(skb, psock, ret); 966 rcu_read_unlock(); 967 return ret; 968 } 969 EXPORT_SYMBOL_GPL(sk_psock_tls_strp_read); 970 971 static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb, 972 int verdict) 973 { 974 struct sock *sk_other; 975 int err = 0; 976 u32 len, off; 977 978 switch (verdict) { 979 case __SK_PASS: 980 err = -EIO; 981 sk_other = psock->sk; 982 if (sock_flag(sk_other, SOCK_DEAD) || 983 !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) { 984 skb_bpf_redirect_clear(skb); 985 goto out_free; 986 } 987 988 skb_bpf_set_ingress(skb); 989 990 /* If the queue is empty then we can submit directly 991 * into the msg queue. If its not empty we have to 992 * queue work otherwise we may get OOO data. Otherwise, 993 * if sk_psock_skb_ingress errors will be handled by 994 * retrying later from workqueue. 995 */ 996 if (skb_queue_empty(&psock->ingress_skb)) { 997 len = skb->len; 998 off = 0; 999 if (skb_bpf_strparser(skb)) { 1000 struct strp_msg *stm = strp_msg(skb); 1001 1002 off = stm->offset; 1003 len = stm->full_len; 1004 } 1005 err = sk_psock_skb_ingress_self(psock, skb, off, len); 1006 } 1007 if (err < 0) { 1008 spin_lock_bh(&psock->ingress_lock); 1009 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) { 1010 skb_queue_tail(&psock->ingress_skb, skb); 1011 schedule_work(&psock->work); 1012 err = 0; 1013 } 1014 spin_unlock_bh(&psock->ingress_lock); 1015 if (err < 0) { 1016 skb_bpf_redirect_clear(skb); 1017 goto out_free; 1018 } 1019 } 1020 break; 1021 case __SK_REDIRECT: 1022 err = sk_psock_skb_redirect(psock, skb); 1023 break; 1024 case __SK_DROP: 1025 default: 1026 out_free: 1027 sock_drop(psock->sk, skb); 1028 } 1029 1030 return err; 1031 } 1032 1033 static void sk_psock_write_space(struct sock *sk) 1034 { 1035 struct sk_psock *psock; 1036 void (*write_space)(struct sock *sk) = NULL; 1037 1038 rcu_read_lock(); 1039 psock = sk_psock(sk); 1040 if (likely(psock)) { 1041 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) 1042 schedule_work(&psock->work); 1043 write_space = psock->saved_write_space; 1044 } 1045 rcu_read_unlock(); 1046 if (write_space) 1047 write_space(sk); 1048 } 1049 1050 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER) 1051 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb) 1052 { 1053 struct sk_psock *psock; 1054 struct bpf_prog *prog; 1055 int ret = __SK_DROP; 1056 struct sock *sk; 1057 1058 rcu_read_lock(); 1059 sk = strp->sk; 1060 psock = sk_psock(sk); 1061 if (unlikely(!psock)) { 1062 sock_drop(sk, skb); 1063 goto out; 1064 } 1065 prog = READ_ONCE(psock->progs.stream_verdict); 1066 if (likely(prog)) { 1067 skb->sk = sk; 1068 skb_dst_drop(skb); 1069 skb_bpf_redirect_clear(skb); 1070 ret = bpf_prog_run_pin_on_cpu(prog, skb); 1071 if (ret == SK_PASS) 1072 skb_bpf_set_strparser(skb); 1073 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb)); 1074 skb->sk = NULL; 1075 } 1076 sk_psock_verdict_apply(psock, skb, ret); 1077 out: 1078 rcu_read_unlock(); 1079 } 1080 1081 static int sk_psock_strp_read_done(struct strparser *strp, int err) 1082 { 1083 return err; 1084 } 1085 1086 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb) 1087 { 1088 struct sk_psock *psock = container_of(strp, struct sk_psock, strp); 1089 struct bpf_prog *prog; 1090 int ret = skb->len; 1091 1092 rcu_read_lock(); 1093 prog = READ_ONCE(psock->progs.stream_parser); 1094 if (likely(prog)) { 1095 skb->sk = psock->sk; 1096 ret = bpf_prog_run_pin_on_cpu(prog, skb); 1097 skb->sk = NULL; 1098 } 1099 rcu_read_unlock(); 1100 return ret; 1101 } 1102 1103 /* Called with socket lock held. */ 1104 static void sk_psock_strp_data_ready(struct sock *sk) 1105 { 1106 struct sk_psock *psock; 1107 1108 rcu_read_lock(); 1109 psock = sk_psock(sk); 1110 if (likely(psock)) { 1111 if (tls_sw_has_ctx_rx(sk)) { 1112 psock->saved_data_ready(sk); 1113 } else { 1114 write_lock_bh(&sk->sk_callback_lock); 1115 strp_data_ready(&psock->strp); 1116 write_unlock_bh(&sk->sk_callback_lock); 1117 } 1118 } 1119 rcu_read_unlock(); 1120 } 1121 1122 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock) 1123 { 1124 static const struct strp_callbacks cb = { 1125 .rcv_msg = sk_psock_strp_read, 1126 .read_sock_done = sk_psock_strp_read_done, 1127 .parse_msg = sk_psock_strp_parse, 1128 }; 1129 1130 return strp_init(&psock->strp, sk, &cb); 1131 } 1132 1133 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock) 1134 { 1135 if (psock->saved_data_ready) 1136 return; 1137 1138 psock->saved_data_ready = sk->sk_data_ready; 1139 sk->sk_data_ready = sk_psock_strp_data_ready; 1140 sk->sk_write_space = sk_psock_write_space; 1141 } 1142 1143 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock) 1144 { 1145 psock_set_prog(&psock->progs.stream_parser, NULL); 1146 1147 if (!psock->saved_data_ready) 1148 return; 1149 1150 sk->sk_data_ready = psock->saved_data_ready; 1151 psock->saved_data_ready = NULL; 1152 strp_stop(&psock->strp); 1153 } 1154 1155 static void sk_psock_done_strp(struct sk_psock *psock) 1156 { 1157 /* Parser has been stopped */ 1158 if (psock->progs.stream_parser) 1159 strp_done(&psock->strp); 1160 } 1161 #else 1162 static void sk_psock_done_strp(struct sk_psock *psock) 1163 { 1164 } 1165 #endif /* CONFIG_BPF_STREAM_PARSER */ 1166 1167 static int sk_psock_verdict_recv(read_descriptor_t *desc, struct sk_buff *skb, 1168 unsigned int offset, size_t orig_len) 1169 { 1170 struct sock *sk = (struct sock *)desc->arg.data; 1171 struct sk_psock *psock; 1172 struct bpf_prog *prog; 1173 int ret = __SK_DROP; 1174 int len = orig_len; 1175 1176 /* clone here so sk_eat_skb() in tcp_read_sock does not drop our data */ 1177 skb = skb_clone(skb, GFP_ATOMIC); 1178 if (!skb) { 1179 desc->error = -ENOMEM; 1180 return 0; 1181 } 1182 1183 rcu_read_lock(); 1184 psock = sk_psock(sk); 1185 if (unlikely(!psock)) { 1186 len = 0; 1187 sock_drop(sk, skb); 1188 goto out; 1189 } 1190 prog = READ_ONCE(psock->progs.stream_verdict); 1191 if (!prog) 1192 prog = READ_ONCE(psock->progs.skb_verdict); 1193 if (likely(prog)) { 1194 skb->sk = sk; 1195 skb_dst_drop(skb); 1196 skb_bpf_redirect_clear(skb); 1197 ret = bpf_prog_run_pin_on_cpu(prog, skb); 1198 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb)); 1199 skb->sk = NULL; 1200 } 1201 if (sk_psock_verdict_apply(psock, skb, ret) < 0) 1202 len = 0; 1203 out: 1204 rcu_read_unlock(); 1205 return len; 1206 } 1207 1208 static void sk_psock_verdict_data_ready(struct sock *sk) 1209 { 1210 struct socket *sock = sk->sk_socket; 1211 read_descriptor_t desc; 1212 1213 if (unlikely(!sock || !sock->ops || !sock->ops->read_sock)) 1214 return; 1215 1216 desc.arg.data = sk; 1217 desc.error = 0; 1218 desc.count = 1; 1219 1220 sock->ops->read_sock(sk, &desc, sk_psock_verdict_recv); 1221 } 1222 1223 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock) 1224 { 1225 if (psock->saved_data_ready) 1226 return; 1227 1228 psock->saved_data_ready = sk->sk_data_ready; 1229 sk->sk_data_ready = sk_psock_verdict_data_ready; 1230 sk->sk_write_space = sk_psock_write_space; 1231 } 1232 1233 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock) 1234 { 1235 psock_set_prog(&psock->progs.stream_verdict, NULL); 1236 psock_set_prog(&psock->progs.skb_verdict, NULL); 1237 1238 if (!psock->saved_data_ready) 1239 return; 1240 1241 sk->sk_data_ready = psock->saved_data_ready; 1242 psock->saved_data_ready = NULL; 1243 } 1244