1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/kernel.h> 3 #include <linux/errno.h> 4 #include <linux/file.h> 5 #include <linux/slab.h> 6 #include <linux/net.h> 7 #include <linux/compat.h> 8 #include <net/compat.h> 9 #include <linux/io_uring.h> 10 11 #include <uapi/linux/io_uring.h> 12 13 #include "filetable.h" 14 #include "io_uring.h" 15 #include "kbuf.h" 16 #include "alloc_cache.h" 17 #include "net.h" 18 #include "notif.h" 19 #include "rsrc.h" 20 #include "zcrx.h" 21 22 struct io_shutdown { 23 struct file *file; 24 int how; 25 }; 26 27 struct io_accept { 28 struct file *file; 29 struct sockaddr __user *addr; 30 int __user *addr_len; 31 int flags; 32 int iou_flags; 33 u32 file_slot; 34 unsigned long nofile; 35 }; 36 37 struct io_socket { 38 struct file *file; 39 int domain; 40 int type; 41 int protocol; 42 int flags; 43 u32 file_slot; 44 unsigned long nofile; 45 }; 46 47 struct io_connect { 48 struct file *file; 49 struct sockaddr __user *addr; 50 int addr_len; 51 bool in_progress; 52 bool seen_econnaborted; 53 }; 54 55 struct io_bind { 56 struct file *file; 57 int addr_len; 58 }; 59 60 struct io_listen { 61 struct file *file; 62 int backlog; 63 }; 64 65 struct io_sr_msg { 66 struct file *file; 67 union { 68 struct compat_msghdr __user *umsg_compat; 69 struct user_msghdr __user *umsg; 70 void __user *buf; 71 }; 72 int len; 73 unsigned done_io; 74 unsigned msg_flags; 75 unsigned nr_multishot_loops; 76 u16 flags; 77 /* initialised and used only by !msg send variants */ 78 u16 buf_group; 79 /* per-invocation mshot limit */ 80 unsigned mshot_len; 81 /* overall mshot byte limit */ 82 unsigned mshot_total_len; 83 void __user *msg_control; 84 /* used only for send zerocopy */ 85 struct io_kiocb *notif; 86 }; 87 88 /* 89 * The UAPI flags are the lower 8 bits, as that's all sqe->ioprio will hold 90 * anyway. Use the upper 8 bits for internal uses. 91 */ 92 enum sr_retry_flags { 93 IORING_RECV_RETRY = (1U << 15), 94 IORING_RECV_PARTIAL_MAP = (1U << 14), 95 IORING_RECV_MSHOT_CAP = (1U << 13), 96 IORING_RECV_MSHOT_LIM = (1U << 12), 97 IORING_RECV_MSHOT_DONE = (1U << 11), 98 99 IORING_RECV_RETRY_CLEAR = IORING_RECV_RETRY | IORING_RECV_PARTIAL_MAP, 100 IORING_RECV_NO_RETRY = IORING_RECV_RETRY | IORING_RECV_PARTIAL_MAP | 101 IORING_RECV_MSHOT_CAP | IORING_RECV_MSHOT_DONE, 102 }; 103 104 /* 105 * Number of times we'll try and do receives if there's more data. If we 106 * exceed this limit, then add us to the back of the queue and retry from 107 * there. This helps fairness between flooding clients. 108 */ 109 #define MULTISHOT_MAX_RETRY 32 110 111 struct io_recvzc { 112 struct file *file; 113 u16 flags; 114 u32 len; 115 struct io_zcrx_ifq *ifq; 116 }; 117 118 static int io_sg_from_iter_iovec(struct sk_buff *skb, 119 struct iov_iter *from, size_t length); 120 static int io_sg_from_iter(struct sk_buff *skb, 121 struct iov_iter *from, size_t length); 122 123 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 124 { 125 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown); 126 127 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags || 128 sqe->buf_index || sqe->splice_fd_in)) 129 return -EINVAL; 130 131 shutdown->how = READ_ONCE(sqe->len); 132 req->flags |= REQ_F_FORCE_ASYNC; 133 return 0; 134 } 135 136 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) 137 { 138 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown); 139 struct socket *sock; 140 int ret; 141 142 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK); 143 144 sock = sock_from_file(req->file); 145 if (unlikely(!sock)) 146 return -ENOTSOCK; 147 148 ret = __sys_shutdown_sock(sock, shutdown->how); 149 io_req_set_res(req, ret, 0); 150 return IOU_COMPLETE; 151 } 152 153 static bool io_net_retry(struct socket *sock, int flags) 154 { 155 if (!(flags & MSG_WAITALL)) 156 return false; 157 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET; 158 } 159 160 static void io_netmsg_iovec_free(struct io_async_msghdr *kmsg) 161 { 162 if (kmsg->vec.iovec) 163 io_vec_free(&kmsg->vec); 164 } 165 166 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags) 167 { 168 struct io_async_msghdr *hdr = req->async_data; 169 170 /* can't recycle, ensure we free the iovec if we have one */ 171 if (unlikely(issue_flags & IO_URING_F_UNLOCKED)) { 172 io_netmsg_iovec_free(hdr); 173 return; 174 } 175 176 /* Let normal cleanup path reap it if we fail adding to the cache */ 177 io_alloc_cache_vec_kasan(&hdr->vec); 178 if (hdr->vec.nr > IO_VEC_CACHE_SOFT_CAP) 179 io_vec_free(&hdr->vec); 180 181 if (io_alloc_cache_put(&req->ctx->netmsg_cache, hdr)) 182 io_req_async_data_clear(req, REQ_F_NEED_CLEANUP); 183 } 184 185 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req) 186 { 187 struct io_ring_ctx *ctx = req->ctx; 188 struct io_async_msghdr *hdr; 189 190 hdr = io_uring_alloc_async_data(&ctx->netmsg_cache, req); 191 if (!hdr) 192 return NULL; 193 194 /* If the async data was cached, we might have an iov cached inside. */ 195 if (hdr->vec.iovec) 196 req->flags |= REQ_F_NEED_CLEANUP; 197 return hdr; 198 } 199 200 static inline void io_mshot_prep_retry(struct io_kiocb *req, 201 struct io_async_msghdr *kmsg) 202 { 203 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 204 205 req->flags &= ~REQ_F_BL_EMPTY; 206 sr->done_io = 0; 207 sr->flags &= ~IORING_RECV_RETRY_CLEAR; 208 sr->len = sr->mshot_len; 209 } 210 211 static int io_net_import_vec(struct io_kiocb *req, struct io_async_msghdr *iomsg, 212 const struct iovec __user *uiov, unsigned uvec_seg, 213 int ddir) 214 { 215 struct iovec *iov; 216 int ret, nr_segs; 217 218 if (iomsg->vec.iovec) { 219 nr_segs = iomsg->vec.nr; 220 iov = iomsg->vec.iovec; 221 } else { 222 nr_segs = 1; 223 iov = &iomsg->fast_iov; 224 } 225 226 ret = __import_iovec(ddir, uiov, uvec_seg, nr_segs, &iov, 227 &iomsg->msg.msg_iter, io_is_compat(req->ctx)); 228 if (unlikely(ret < 0)) 229 return ret; 230 231 if (iov) { 232 req->flags |= REQ_F_NEED_CLEANUP; 233 io_vec_reset_iovec(&iomsg->vec, iov, iomsg->msg.msg_iter.nr_segs); 234 } 235 return 0; 236 } 237 238 static int io_compat_msg_copy_hdr(struct io_kiocb *req, 239 struct io_async_msghdr *iomsg, 240 struct compat_msghdr *msg, int ddir, 241 struct sockaddr __user **save_addr) 242 { 243 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 244 struct compat_iovec __user *uiov; 245 int ret; 246 247 if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg))) 248 return -EFAULT; 249 250 ret = __get_compat_msghdr(&iomsg->msg, msg, save_addr); 251 if (ret) 252 return ret; 253 254 uiov = compat_ptr(msg->msg_iov); 255 if (req->flags & REQ_F_BUFFER_SELECT) { 256 if (msg->msg_iovlen == 0) { 257 sr->len = 0; 258 } else if (msg->msg_iovlen > 1) { 259 return -EINVAL; 260 } else { 261 struct compat_iovec tmp_iov; 262 263 if (copy_from_user(&tmp_iov, uiov, sizeof(tmp_iov))) 264 return -EFAULT; 265 sr->len = tmp_iov.iov_len; 266 } 267 } 268 return 0; 269 } 270 271 static int io_copy_msghdr_from_user(struct user_msghdr *msg, 272 struct user_msghdr __user *umsg) 273 { 274 if (!user_access_begin(umsg, sizeof(*umsg))) 275 return -EFAULT; 276 unsafe_get_user(msg->msg_name, &umsg->msg_name, ua_end); 277 unsafe_get_user(msg->msg_namelen, &umsg->msg_namelen, ua_end); 278 unsafe_get_user(msg->msg_iov, &umsg->msg_iov, ua_end); 279 unsafe_get_user(msg->msg_iovlen, &umsg->msg_iovlen, ua_end); 280 unsafe_get_user(msg->msg_control, &umsg->msg_control, ua_end); 281 unsafe_get_user(msg->msg_controllen, &umsg->msg_controllen, ua_end); 282 user_access_end(); 283 return 0; 284 ua_end: 285 user_access_end(); 286 return -EFAULT; 287 } 288 289 static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg, 290 struct user_msghdr *msg, int ddir, 291 struct sockaddr __user **save_addr) 292 { 293 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 294 struct user_msghdr __user *umsg = sr->umsg; 295 int ret; 296 297 iomsg->msg.msg_name = &iomsg->addr; 298 iomsg->msg.msg_iter.nr_segs = 0; 299 300 if (io_is_compat(req->ctx)) { 301 struct compat_msghdr cmsg; 302 303 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ddir, save_addr); 304 if (ret) 305 return ret; 306 307 memset(msg, 0, sizeof(*msg)); 308 msg->msg_namelen = cmsg.msg_namelen; 309 msg->msg_controllen = cmsg.msg_controllen; 310 msg->msg_iov = compat_ptr(cmsg.msg_iov); 311 msg->msg_iovlen = cmsg.msg_iovlen; 312 return 0; 313 } 314 315 ret = io_copy_msghdr_from_user(msg, umsg); 316 if (unlikely(ret)) 317 return ret; 318 319 msg->msg_flags = 0; 320 321 ret = __copy_msghdr(&iomsg->msg, msg, save_addr); 322 if (ret) 323 return ret; 324 325 if (req->flags & REQ_F_BUFFER_SELECT) { 326 if (msg->msg_iovlen == 0) { 327 sr->len = 0; 328 } else if (msg->msg_iovlen > 1) { 329 return -EINVAL; 330 } else { 331 struct iovec __user *uiov = msg->msg_iov; 332 struct iovec tmp_iov; 333 334 if (copy_from_user(&tmp_iov, uiov, sizeof(tmp_iov))) 335 return -EFAULT; 336 sr->len = tmp_iov.iov_len; 337 } 338 } 339 return 0; 340 } 341 342 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req) 343 { 344 struct io_async_msghdr *io = req->async_data; 345 346 io_netmsg_iovec_free(io); 347 } 348 349 static int io_send_setup(struct io_kiocb *req, const struct io_uring_sqe *sqe) 350 { 351 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 352 struct io_async_msghdr *kmsg = req->async_data; 353 void __user *addr; 354 u16 addr_len; 355 int ret; 356 357 sr->buf = u64_to_user_ptr(READ_ONCE(sqe->addr)); 358 359 if (READ_ONCE(sqe->__pad3[0])) 360 return -EINVAL; 361 362 kmsg->msg.msg_name = NULL; 363 kmsg->msg.msg_namelen = 0; 364 kmsg->msg.msg_control = NULL; 365 kmsg->msg.msg_controllen = 0; 366 kmsg->msg.msg_ubuf = NULL; 367 368 addr = u64_to_user_ptr(READ_ONCE(sqe->addr2)); 369 addr_len = READ_ONCE(sqe->addr_len); 370 if (addr) { 371 ret = move_addr_to_kernel(addr, addr_len, &kmsg->addr); 372 if (unlikely(ret < 0)) 373 return ret; 374 kmsg->msg.msg_name = &kmsg->addr; 375 kmsg->msg.msg_namelen = addr_len; 376 } 377 if (sr->flags & IORING_RECVSEND_FIXED_BUF) { 378 if (!(sr->flags & IORING_SEND_VECTORIZED)) { 379 req->flags |= REQ_F_IMPORT_BUFFER; 380 return 0; 381 } 382 383 kmsg->msg.msg_iter.nr_segs = sr->len; 384 return io_prep_reg_iovec(req, &kmsg->vec, sr->buf, sr->len); 385 } 386 if (req->flags & REQ_F_BUFFER_SELECT) 387 return 0; 388 389 if (sr->flags & IORING_SEND_VECTORIZED) 390 return io_net_import_vec(req, kmsg, sr->buf, sr->len, ITER_SOURCE); 391 392 return import_ubuf(ITER_SOURCE, sr->buf, sr->len, &kmsg->msg.msg_iter); 393 } 394 395 static int io_sendmsg_setup(struct io_kiocb *req, const struct io_uring_sqe *sqe) 396 { 397 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 398 struct io_async_msghdr *kmsg = req->async_data; 399 struct user_msghdr msg; 400 int ret; 401 402 sr->flags |= IORING_SEND_VECTORIZED; 403 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); 404 ret = io_msg_copy_hdr(req, kmsg, &msg, ITER_SOURCE, NULL); 405 if (unlikely(ret)) 406 return ret; 407 /* save msg_control as sys_sendmsg() overwrites it */ 408 sr->msg_control = kmsg->msg.msg_control_user; 409 410 if (sr->flags & IORING_RECVSEND_FIXED_BUF) { 411 kmsg->msg.msg_iter.nr_segs = msg.msg_iovlen; 412 return io_prep_reg_iovec(req, &kmsg->vec, msg.msg_iov, 413 msg.msg_iovlen); 414 } 415 if (req->flags & REQ_F_BUFFER_SELECT) 416 return 0; 417 return io_net_import_vec(req, kmsg, msg.msg_iov, msg.msg_iovlen, ITER_SOURCE); 418 } 419 420 #define SENDMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_BUNDLE | IORING_SEND_VECTORIZED) 421 422 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 423 { 424 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 425 426 sr->done_io = 0; 427 sr->len = READ_ONCE(sqe->len); 428 if (unlikely(sr->len < 0)) 429 return -EINVAL; 430 sr->flags = READ_ONCE(sqe->ioprio); 431 if (sr->flags & ~SENDMSG_FLAGS) 432 return -EINVAL; 433 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; 434 if (sr->msg_flags & MSG_DONTWAIT) 435 req->flags |= REQ_F_NOWAIT; 436 if (req->flags & REQ_F_BUFFER_SELECT) 437 sr->buf_group = req->buf_index; 438 if (sr->flags & IORING_RECVSEND_BUNDLE) { 439 if (req->opcode == IORING_OP_SENDMSG) 440 return -EINVAL; 441 sr->msg_flags |= MSG_WAITALL; 442 req->flags |= REQ_F_MULTISHOT; 443 } 444 445 if (io_is_compat(req->ctx)) 446 sr->msg_flags |= MSG_CMSG_COMPAT; 447 448 if (unlikely(!io_msg_alloc_async(req))) 449 return -ENOMEM; 450 if (req->opcode != IORING_OP_SENDMSG) 451 return io_send_setup(req, sqe); 452 if (unlikely(sqe->addr2 || sqe->file_index)) 453 return -EINVAL; 454 return io_sendmsg_setup(req, sqe); 455 } 456 457 static void io_req_msg_cleanup(struct io_kiocb *req, 458 unsigned int issue_flags) 459 { 460 io_netmsg_recycle(req, issue_flags); 461 } 462 463 /* 464 * For bundle completions, we need to figure out how many segments we consumed. 465 * A bundle could be using a single ITER_UBUF if that's all we mapped, or it 466 * could be using an ITER_IOVEC. If the latter, then if we consumed all of 467 * the segments, then it's a trivial questiont o answer. If we have residual 468 * data in the iter, then loop the segments to figure out how much we 469 * transferred. 470 */ 471 static int io_bundle_nbufs(struct io_async_msghdr *kmsg, int ret) 472 { 473 struct iovec *iov; 474 int nbufs; 475 476 /* no data is always zero segments, and a ubuf is always 1 segment */ 477 if (ret <= 0) 478 return 0; 479 if (iter_is_ubuf(&kmsg->msg.msg_iter)) 480 return 1; 481 482 iov = kmsg->vec.iovec; 483 if (!iov) 484 iov = &kmsg->fast_iov; 485 486 /* if all data was transferred, it's basic pointer math */ 487 if (!iov_iter_count(&kmsg->msg.msg_iter)) 488 return iter_iov(&kmsg->msg.msg_iter) - iov; 489 490 /* short transfer, count segments */ 491 nbufs = 0; 492 do { 493 int this_len = min_t(int, iov[nbufs].iov_len, ret); 494 495 nbufs++; 496 ret -= this_len; 497 } while (ret); 498 499 return nbufs; 500 } 501 502 static int io_net_kbuf_recyle(struct io_kiocb *req, struct io_buffer_list *bl, 503 struct io_async_msghdr *kmsg, int len) 504 { 505 req->flags |= REQ_F_BL_NO_RECYCLE; 506 if (req->flags & REQ_F_BUFFERS_COMMIT) 507 io_kbuf_commit(req, bl, len, io_bundle_nbufs(kmsg, len)); 508 return IOU_RETRY; 509 } 510 511 static inline bool io_send_finish(struct io_kiocb *req, 512 struct io_async_msghdr *kmsg, 513 struct io_br_sel *sel) 514 { 515 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 516 bool bundle_finished = sel->val <= 0; 517 unsigned int cflags; 518 519 if (!(sr->flags & IORING_RECVSEND_BUNDLE)) { 520 cflags = io_put_kbuf(req, sel->val, sel->buf_list); 521 goto finish; 522 } 523 524 cflags = io_put_kbufs(req, sel->val, sel->buf_list, io_bundle_nbufs(kmsg, sel->val)); 525 526 /* 527 * Don't start new bundles if the buffer list is empty, or if the 528 * current operation needed to go through polling to complete. 529 */ 530 if (bundle_finished || req->flags & (REQ_F_BL_EMPTY | REQ_F_POLLED)) 531 goto finish; 532 533 /* 534 * Fill CQE for this receive and see if we should keep trying to 535 * receive from this socket. 536 */ 537 if (io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) { 538 io_mshot_prep_retry(req, kmsg); 539 return false; 540 } 541 542 /* Otherwise stop bundle and use the current result. */ 543 finish: 544 io_req_set_res(req, sel->val, cflags); 545 sel->val = IOU_COMPLETE; 546 return true; 547 } 548 549 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) 550 { 551 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 552 struct io_async_msghdr *kmsg = req->async_data; 553 struct socket *sock; 554 unsigned flags; 555 int min_ret = 0; 556 int ret; 557 558 sock = sock_from_file(req->file); 559 if (unlikely(!sock)) 560 return -ENOTSOCK; 561 562 if (!(req->flags & REQ_F_POLLED) && 563 (sr->flags & IORING_RECVSEND_POLL_FIRST)) 564 return -EAGAIN; 565 566 flags = sr->msg_flags; 567 if (issue_flags & IO_URING_F_NONBLOCK) 568 flags |= MSG_DONTWAIT; 569 if (flags & MSG_WAITALL) 570 min_ret = iov_iter_count(&kmsg->msg.msg_iter); 571 572 kmsg->msg.msg_control_user = sr->msg_control; 573 574 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); 575 576 if (ret < min_ret) { 577 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) 578 return -EAGAIN; 579 if (ret > 0 && io_net_retry(sock, flags)) { 580 kmsg->msg.msg_controllen = 0; 581 kmsg->msg.msg_control = NULL; 582 sr->done_io += ret; 583 return -EAGAIN; 584 } 585 if (ret == -ERESTARTSYS) 586 ret = -EINTR; 587 req_set_fail(req); 588 } 589 io_req_msg_cleanup(req, issue_flags); 590 if (ret >= 0) 591 ret += sr->done_io; 592 else if (sr->done_io) 593 ret = sr->done_io; 594 io_req_set_res(req, ret, 0); 595 return IOU_COMPLETE; 596 } 597 598 static int io_send_select_buffer(struct io_kiocb *req, unsigned int issue_flags, 599 struct io_br_sel *sel, struct io_async_msghdr *kmsg) 600 { 601 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 602 struct buf_sel_arg arg = { 603 .iovs = &kmsg->fast_iov, 604 .max_len = min_not_zero(sr->len, INT_MAX), 605 .nr_iovs = 1, 606 .buf_group = sr->buf_group, 607 }; 608 int ret; 609 610 if (kmsg->vec.iovec) { 611 arg.nr_iovs = kmsg->vec.nr; 612 arg.iovs = kmsg->vec.iovec; 613 arg.mode = KBUF_MODE_FREE; 614 } 615 616 if (!(sr->flags & IORING_RECVSEND_BUNDLE)) 617 arg.nr_iovs = 1; 618 else 619 arg.mode |= KBUF_MODE_EXPAND; 620 621 ret = io_buffers_select(req, &arg, sel, issue_flags); 622 if (unlikely(ret < 0)) 623 return ret; 624 625 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) { 626 kmsg->vec.nr = ret; 627 kmsg->vec.iovec = arg.iovs; 628 req->flags |= REQ_F_NEED_CLEANUP; 629 } 630 sr->len = arg.out_len; 631 632 if (ret == 1) { 633 sr->buf = arg.iovs[0].iov_base; 634 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, 635 &kmsg->msg.msg_iter); 636 if (unlikely(ret)) 637 return ret; 638 } else { 639 iov_iter_init(&kmsg->msg.msg_iter, ITER_SOURCE, 640 arg.iovs, ret, arg.out_len); 641 } 642 643 return 0; 644 } 645 646 int io_send(struct io_kiocb *req, unsigned int issue_flags) 647 { 648 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 649 struct io_async_msghdr *kmsg = req->async_data; 650 struct io_br_sel sel = { }; 651 struct socket *sock; 652 unsigned flags; 653 int min_ret = 0; 654 int ret; 655 656 sock = sock_from_file(req->file); 657 if (unlikely(!sock)) 658 return -ENOTSOCK; 659 660 if (!(req->flags & REQ_F_POLLED) && 661 (sr->flags & IORING_RECVSEND_POLL_FIRST)) 662 return -EAGAIN; 663 664 flags = sr->msg_flags; 665 if (issue_flags & IO_URING_F_NONBLOCK) 666 flags |= MSG_DONTWAIT; 667 668 retry_bundle: 669 sel.buf_list = NULL; 670 if (io_do_buffer_select(req)) { 671 ret = io_send_select_buffer(req, issue_flags, &sel, kmsg); 672 if (ret) 673 return ret; 674 } 675 676 /* 677 * If MSG_WAITALL is set, or this is a bundle send, then we need 678 * the full amount. If just bundle is set, if we do a short send 679 * then we complete the bundle sequence rather than continue on. 680 */ 681 if (flags & MSG_WAITALL || sr->flags & IORING_RECVSEND_BUNDLE) 682 min_ret = iov_iter_count(&kmsg->msg.msg_iter); 683 684 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS; 685 kmsg->msg.msg_flags = flags; 686 ret = sock_sendmsg(sock, &kmsg->msg); 687 if (ret < min_ret) { 688 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) 689 return -EAGAIN; 690 691 if (ret > 0 && io_net_retry(sock, flags)) { 692 sr->len -= ret; 693 sr->buf += ret; 694 sr->done_io += ret; 695 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret); 696 } 697 if (ret == -ERESTARTSYS) 698 ret = -EINTR; 699 req_set_fail(req); 700 } 701 if (ret >= 0) 702 ret += sr->done_io; 703 else if (sr->done_io) 704 ret = sr->done_io; 705 706 sel.val = ret; 707 if (!io_send_finish(req, kmsg, &sel)) 708 goto retry_bundle; 709 710 io_req_msg_cleanup(req, issue_flags); 711 return sel.val; 712 } 713 714 static int io_recvmsg_mshot_prep(struct io_kiocb *req, 715 struct io_async_msghdr *iomsg, 716 int namelen, size_t controllen) 717 { 718 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) == 719 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) { 720 int hdr; 721 722 if (unlikely(namelen < 0)) 723 return -EOVERFLOW; 724 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out), 725 namelen, &hdr)) 726 return -EOVERFLOW; 727 if (check_add_overflow(hdr, controllen, &hdr)) 728 return -EOVERFLOW; 729 730 iomsg->namelen = namelen; 731 iomsg->controllen = controllen; 732 return 0; 733 } 734 735 return 0; 736 } 737 738 static int io_recvmsg_copy_hdr(struct io_kiocb *req, 739 struct io_async_msghdr *iomsg) 740 { 741 struct user_msghdr msg; 742 int ret; 743 744 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST, &iomsg->uaddr); 745 if (unlikely(ret)) 746 return ret; 747 748 if (!(req->flags & REQ_F_BUFFER_SELECT)) { 749 ret = io_net_import_vec(req, iomsg, msg.msg_iov, msg.msg_iovlen, 750 ITER_DEST); 751 if (unlikely(ret)) 752 return ret; 753 } 754 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen, 755 msg.msg_controllen); 756 } 757 758 static int io_recvmsg_prep_setup(struct io_kiocb *req) 759 { 760 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 761 struct io_async_msghdr *kmsg; 762 763 kmsg = io_msg_alloc_async(req); 764 if (unlikely(!kmsg)) 765 return -ENOMEM; 766 767 if (req->opcode == IORING_OP_RECV) { 768 kmsg->msg.msg_name = NULL; 769 kmsg->msg.msg_namelen = 0; 770 kmsg->msg.msg_inq = 0; 771 kmsg->msg.msg_control = NULL; 772 kmsg->msg.msg_get_inq = 1; 773 kmsg->msg.msg_controllen = 0; 774 kmsg->msg.msg_ubuf = NULL; 775 776 if (req->flags & REQ_F_BUFFER_SELECT) 777 return 0; 778 return import_ubuf(ITER_DEST, sr->buf, sr->len, 779 &kmsg->msg.msg_iter); 780 } 781 782 return io_recvmsg_copy_hdr(req, kmsg); 783 } 784 785 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT | \ 786 IORING_RECVSEND_BUNDLE) 787 788 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 789 { 790 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 791 792 sr->done_io = 0; 793 794 if (unlikely(sqe->addr2)) 795 return -EINVAL; 796 797 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); 798 sr->len = READ_ONCE(sqe->len); 799 if (unlikely(sr->len < 0)) 800 return -EINVAL; 801 sr->flags = READ_ONCE(sqe->ioprio); 802 if (sr->flags & ~RECVMSG_FLAGS) 803 return -EINVAL; 804 sr->msg_flags = READ_ONCE(sqe->msg_flags); 805 if (sr->msg_flags & MSG_DONTWAIT) 806 req->flags |= REQ_F_NOWAIT; 807 if (sr->msg_flags & MSG_ERRQUEUE) 808 req->flags |= REQ_F_CLEAR_POLLIN; 809 if (req->flags & REQ_F_BUFFER_SELECT) 810 sr->buf_group = req->buf_index; 811 sr->mshot_total_len = sr->mshot_len = 0; 812 if (sr->flags & IORING_RECV_MULTISHOT) { 813 if (!(req->flags & REQ_F_BUFFER_SELECT)) 814 return -EINVAL; 815 if (sr->msg_flags & MSG_WAITALL) 816 return -EINVAL; 817 if (req->opcode == IORING_OP_RECV) { 818 sr->mshot_len = sr->len; 819 sr->mshot_total_len = READ_ONCE(sqe->optlen); 820 if (sr->mshot_total_len) 821 sr->flags |= IORING_RECV_MSHOT_LIM; 822 } else if (sqe->optlen) { 823 return -EINVAL; 824 } 825 req->flags |= REQ_F_APOLL_MULTISHOT; 826 } else if (sqe->optlen) { 827 return -EINVAL; 828 } 829 830 if (sr->flags & IORING_RECVSEND_BUNDLE) { 831 if (req->opcode == IORING_OP_RECVMSG) 832 return -EINVAL; 833 } 834 835 if (io_is_compat(req->ctx)) 836 sr->msg_flags |= MSG_CMSG_COMPAT; 837 838 sr->nr_multishot_loops = 0; 839 return io_recvmsg_prep_setup(req); 840 } 841 842 /* bits to clear in old and inherit in new cflags on bundle retry */ 843 #define CQE_F_MASK (IORING_CQE_F_SOCK_NONEMPTY|IORING_CQE_F_MORE) 844 845 /* 846 * Finishes io_recv and io_recvmsg. 847 * 848 * Returns true if it is actually finished, or false if it should run 849 * again (for multishot). 850 */ 851 static inline bool io_recv_finish(struct io_kiocb *req, 852 struct io_async_msghdr *kmsg, 853 struct io_br_sel *sel, bool mshot_finished, 854 unsigned issue_flags) 855 { 856 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 857 unsigned int cflags = 0; 858 859 if (kmsg->msg.msg_inq > 0) 860 cflags |= IORING_CQE_F_SOCK_NONEMPTY; 861 862 if (sel->val > 0 && sr->flags & IORING_RECV_MSHOT_LIM) { 863 /* 864 * If sr->len hits zero, the limit has been reached. Mark 865 * mshot as finished, and flag MSHOT_DONE as well to prevent 866 * a potential bundle from being retried. 867 */ 868 sr->mshot_total_len -= min_t(int, sel->val, sr->mshot_total_len); 869 if (!sr->mshot_total_len) { 870 sr->flags |= IORING_RECV_MSHOT_DONE; 871 mshot_finished = true; 872 } 873 } 874 875 if (sr->flags & IORING_RECVSEND_BUNDLE) { 876 size_t this_ret = sel->val - sr->done_io; 877 878 cflags |= io_put_kbufs(req, this_ret, sel->buf_list, io_bundle_nbufs(kmsg, this_ret)); 879 if (sr->flags & IORING_RECV_RETRY) 880 cflags = req->cqe.flags | (cflags & CQE_F_MASK); 881 if (sr->mshot_len && sel->val >= sr->mshot_len) 882 sr->flags |= IORING_RECV_MSHOT_CAP; 883 /* bundle with no more immediate buffers, we're done */ 884 if (req->flags & REQ_F_BL_EMPTY) 885 goto finish; 886 /* 887 * If more is available AND it was a full transfer, retry and 888 * append to this one 889 */ 890 if (!(sr->flags & IORING_RECV_NO_RETRY) && 891 kmsg->msg.msg_inq > 1 && this_ret > 0 && 892 !iov_iter_count(&kmsg->msg.msg_iter)) { 893 req->cqe.flags = cflags & ~CQE_F_MASK; 894 sr->len = kmsg->msg.msg_inq; 895 sr->done_io += this_ret; 896 sr->flags |= IORING_RECV_RETRY; 897 return false; 898 } 899 } else { 900 cflags |= io_put_kbuf(req, sel->val, sel->buf_list); 901 } 902 903 /* 904 * Fill CQE for this receive and see if we should keep trying to 905 * receive from this socket. 906 */ 907 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished && 908 io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) { 909 sel->val = IOU_RETRY; 910 io_mshot_prep_retry(req, kmsg); 911 /* Known not-empty or unknown state, retry */ 912 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || kmsg->msg.msg_inq < 0) { 913 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY && 914 !(sr->flags & IORING_RECV_MSHOT_CAP)) { 915 return false; 916 } 917 /* mshot retries exceeded, force a requeue */ 918 sr->nr_multishot_loops = 0; 919 sr->flags &= ~IORING_RECV_MSHOT_CAP; 920 if (issue_flags & IO_URING_F_MULTISHOT) 921 sel->val = IOU_REQUEUE; 922 } 923 return true; 924 } 925 926 /* Finish the request / stop multishot. */ 927 finish: 928 io_req_set_res(req, sel->val, cflags); 929 sel->val = IOU_COMPLETE; 930 io_req_msg_cleanup(req, issue_flags); 931 return true; 932 } 933 934 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg, 935 struct io_sr_msg *sr, void __user **buf, 936 size_t *len) 937 { 938 unsigned long ubuf = (unsigned long) *buf; 939 unsigned long hdr; 940 941 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen + 942 kmsg->controllen; 943 if (*len < hdr) 944 return -EFAULT; 945 946 if (kmsg->controllen) { 947 unsigned long control = ubuf + hdr - kmsg->controllen; 948 949 kmsg->msg.msg_control_user = (void __user *) control; 950 kmsg->msg.msg_controllen = kmsg->controllen; 951 } 952 953 sr->buf = *buf; /* stash for later copy */ 954 *buf = (void __user *) (ubuf + hdr); 955 kmsg->payloadlen = *len = *len - hdr; 956 return 0; 957 } 958 959 struct io_recvmsg_multishot_hdr { 960 struct io_uring_recvmsg_out msg; 961 struct sockaddr_storage addr; 962 }; 963 964 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io, 965 struct io_async_msghdr *kmsg, 966 unsigned int flags, bool *finished) 967 { 968 int err; 969 int copy_len; 970 struct io_recvmsg_multishot_hdr hdr; 971 972 if (kmsg->namelen) 973 kmsg->msg.msg_name = &hdr.addr; 974 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT); 975 kmsg->msg.msg_namelen = 0; 976 977 if (sock->file->f_flags & O_NONBLOCK) 978 flags |= MSG_DONTWAIT; 979 980 err = sock_recvmsg(sock, &kmsg->msg, flags); 981 *finished = err <= 0; 982 if (err < 0) 983 return err; 984 985 hdr.msg = (struct io_uring_recvmsg_out) { 986 .controllen = kmsg->controllen - kmsg->msg.msg_controllen, 987 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT 988 }; 989 990 hdr.msg.payloadlen = err; 991 if (err > kmsg->payloadlen) 992 err = kmsg->payloadlen; 993 994 copy_len = sizeof(struct io_uring_recvmsg_out); 995 if (kmsg->msg.msg_namelen > kmsg->namelen) 996 copy_len += kmsg->namelen; 997 else 998 copy_len += kmsg->msg.msg_namelen; 999 1000 /* 1001 * "fromlen shall refer to the value before truncation.." 1002 * 1003.1g 1003 */ 1004 hdr.msg.namelen = kmsg->msg.msg_namelen; 1005 1006 /* ensure that there is no gap between hdr and sockaddr_storage */ 1007 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) != 1008 sizeof(struct io_uring_recvmsg_out)); 1009 if (copy_to_user(io->buf, &hdr, copy_len)) { 1010 *finished = true; 1011 return -EFAULT; 1012 } 1013 1014 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen + 1015 kmsg->controllen + err; 1016 } 1017 1018 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) 1019 { 1020 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1021 struct io_async_msghdr *kmsg = req->async_data; 1022 struct io_br_sel sel = { }; 1023 struct socket *sock; 1024 unsigned flags; 1025 int ret, min_ret = 0; 1026 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 1027 bool mshot_finished = true; 1028 1029 sock = sock_from_file(req->file); 1030 if (unlikely(!sock)) 1031 return -ENOTSOCK; 1032 1033 if (!(req->flags & REQ_F_POLLED) && 1034 (sr->flags & IORING_RECVSEND_POLL_FIRST)) 1035 return -EAGAIN; 1036 1037 flags = sr->msg_flags; 1038 if (force_nonblock) 1039 flags |= MSG_DONTWAIT; 1040 1041 retry_multishot: 1042 sel.buf_list = NULL; 1043 if (io_do_buffer_select(req)) { 1044 size_t len = sr->len; 1045 1046 sel = io_buffer_select(req, &len, sr->buf_group, issue_flags); 1047 if (!sel.addr) 1048 return -ENOBUFS; 1049 1050 if (req->flags & REQ_F_APOLL_MULTISHOT) { 1051 ret = io_recvmsg_prep_multishot(kmsg, sr, &sel.addr, &len); 1052 if (ret) { 1053 io_kbuf_recycle(req, sel.buf_list, issue_flags); 1054 return ret; 1055 } 1056 } 1057 1058 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, sel.addr, len); 1059 } 1060 1061 kmsg->msg.msg_get_inq = 1; 1062 kmsg->msg.msg_inq = -1; 1063 if (req->flags & REQ_F_APOLL_MULTISHOT) { 1064 ret = io_recvmsg_multishot(sock, sr, kmsg, flags, 1065 &mshot_finished); 1066 } else { 1067 /* disable partial retry for recvmsg with cmsg attached */ 1068 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen) 1069 min_ret = iov_iter_count(&kmsg->msg.msg_iter); 1070 1071 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg, 1072 kmsg->uaddr, flags); 1073 } 1074 1075 if (ret < min_ret) { 1076 if (ret == -EAGAIN && force_nonblock) { 1077 io_kbuf_recycle(req, sel.buf_list, issue_flags); 1078 return IOU_RETRY; 1079 } 1080 if (ret > 0 && io_net_retry(sock, flags)) { 1081 sr->done_io += ret; 1082 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret); 1083 } 1084 if (ret == -ERESTARTSYS) 1085 ret = -EINTR; 1086 req_set_fail(req); 1087 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { 1088 req_set_fail(req); 1089 } 1090 1091 if (ret > 0) 1092 ret += sr->done_io; 1093 else if (sr->done_io) 1094 ret = sr->done_io; 1095 else 1096 io_kbuf_recycle(req, sel.buf_list, issue_flags); 1097 1098 sel.val = ret; 1099 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags)) 1100 goto retry_multishot; 1101 1102 return sel.val; 1103 } 1104 1105 static int io_recv_buf_select(struct io_kiocb *req, struct io_async_msghdr *kmsg, 1106 struct io_br_sel *sel, unsigned int issue_flags) 1107 { 1108 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1109 int ret; 1110 1111 /* 1112 * If the ring isn't locked, then don't use the peek interface 1113 * to grab multiple buffers as we will lock/unlock between 1114 * this selection and posting the buffers. 1115 */ 1116 if (!(issue_flags & IO_URING_F_UNLOCKED) && 1117 sr->flags & IORING_RECVSEND_BUNDLE) { 1118 struct buf_sel_arg arg = { 1119 .iovs = &kmsg->fast_iov, 1120 .nr_iovs = 1, 1121 .mode = KBUF_MODE_EXPAND, 1122 .buf_group = sr->buf_group, 1123 }; 1124 1125 if (kmsg->vec.iovec) { 1126 arg.nr_iovs = kmsg->vec.nr; 1127 arg.iovs = kmsg->vec.iovec; 1128 arg.mode |= KBUF_MODE_FREE; 1129 } 1130 1131 if (sel->val) 1132 arg.max_len = sel->val; 1133 else if (kmsg->msg.msg_inq > 1) 1134 arg.max_len = min_not_zero(sel->val, (ssize_t) kmsg->msg.msg_inq); 1135 1136 /* if mshot limited, ensure we don't go over */ 1137 if (sr->flags & IORING_RECV_MSHOT_LIM) 1138 arg.max_len = min_not_zero(arg.max_len, sr->mshot_total_len); 1139 ret = io_buffers_peek(req, &arg, sel); 1140 if (unlikely(ret < 0)) 1141 return ret; 1142 1143 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) { 1144 kmsg->vec.nr = ret; 1145 kmsg->vec.iovec = arg.iovs; 1146 req->flags |= REQ_F_NEED_CLEANUP; 1147 } 1148 if (arg.partial_map) 1149 sr->flags |= IORING_RECV_PARTIAL_MAP; 1150 1151 /* special case 1 vec, can be a fast path */ 1152 if (ret == 1) { 1153 sr->buf = arg.iovs[0].iov_base; 1154 sr->len = arg.iovs[0].iov_len; 1155 goto map_ubuf; 1156 } 1157 iov_iter_init(&kmsg->msg.msg_iter, ITER_DEST, arg.iovs, ret, 1158 arg.out_len); 1159 } else { 1160 size_t len = sel->val; 1161 1162 *sel = io_buffer_select(req, &len, sr->buf_group, issue_flags); 1163 if (!sel->addr) 1164 return -ENOBUFS; 1165 sr->buf = sel->addr; 1166 sr->len = len; 1167 map_ubuf: 1168 ret = import_ubuf(ITER_DEST, sr->buf, sr->len, 1169 &kmsg->msg.msg_iter); 1170 if (unlikely(ret)) 1171 return ret; 1172 } 1173 1174 return 0; 1175 } 1176 1177 int io_recv(struct io_kiocb *req, unsigned int issue_flags) 1178 { 1179 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1180 struct io_async_msghdr *kmsg = req->async_data; 1181 struct io_br_sel sel; 1182 struct socket *sock; 1183 unsigned flags; 1184 int ret, min_ret = 0; 1185 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 1186 bool mshot_finished; 1187 1188 if (!(req->flags & REQ_F_POLLED) && 1189 (sr->flags & IORING_RECVSEND_POLL_FIRST)) 1190 return -EAGAIN; 1191 1192 sock = sock_from_file(req->file); 1193 if (unlikely(!sock)) 1194 return -ENOTSOCK; 1195 1196 flags = sr->msg_flags; 1197 if (force_nonblock) 1198 flags |= MSG_DONTWAIT; 1199 1200 retry_multishot: 1201 sel.buf_list = NULL; 1202 if (io_do_buffer_select(req)) { 1203 sel.val = sr->len; 1204 ret = io_recv_buf_select(req, kmsg, &sel, issue_flags); 1205 if (unlikely(ret < 0)) { 1206 kmsg->msg.msg_inq = -1; 1207 goto out_free; 1208 } 1209 sr->buf = NULL; 1210 } 1211 1212 kmsg->msg.msg_flags = 0; 1213 kmsg->msg.msg_inq = -1; 1214 1215 if (flags & MSG_WAITALL) 1216 min_ret = iov_iter_count(&kmsg->msg.msg_iter); 1217 1218 ret = sock_recvmsg(sock, &kmsg->msg, flags); 1219 if (ret < min_ret) { 1220 if (ret == -EAGAIN && force_nonblock) { 1221 io_kbuf_recycle(req, sel.buf_list, issue_flags); 1222 return IOU_RETRY; 1223 } 1224 if (ret > 0 && io_net_retry(sock, flags)) { 1225 sr->len -= ret; 1226 sr->buf += ret; 1227 sr->done_io += ret; 1228 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret); 1229 } 1230 if (ret == -ERESTARTSYS) 1231 ret = -EINTR; 1232 req_set_fail(req); 1233 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { 1234 out_free: 1235 req_set_fail(req); 1236 } 1237 1238 mshot_finished = ret <= 0; 1239 if (ret > 0) 1240 ret += sr->done_io; 1241 else if (sr->done_io) 1242 ret = sr->done_io; 1243 else 1244 io_kbuf_recycle(req, sel.buf_list, issue_flags); 1245 1246 sel.val = ret; 1247 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags)) 1248 goto retry_multishot; 1249 1250 return sel.val; 1251 } 1252 1253 int io_recvzc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1254 { 1255 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc); 1256 unsigned ifq_idx; 1257 1258 if (unlikely(sqe->addr2 || sqe->addr || sqe->addr3)) 1259 return -EINVAL; 1260 1261 ifq_idx = READ_ONCE(sqe->zcrx_ifq_idx); 1262 zc->ifq = xa_load(&req->ctx->zcrx_ctxs, ifq_idx); 1263 if (!zc->ifq) 1264 return -EINVAL; 1265 1266 zc->len = READ_ONCE(sqe->len); 1267 zc->flags = READ_ONCE(sqe->ioprio); 1268 if (READ_ONCE(sqe->msg_flags)) 1269 return -EINVAL; 1270 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)) 1271 return -EINVAL; 1272 /* multishot required */ 1273 if (!(zc->flags & IORING_RECV_MULTISHOT)) 1274 return -EINVAL; 1275 /* All data completions are posted as aux CQEs. */ 1276 req->flags |= REQ_F_APOLL_MULTISHOT; 1277 1278 return 0; 1279 } 1280 1281 int io_recvzc(struct io_kiocb *req, unsigned int issue_flags) 1282 { 1283 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc); 1284 struct socket *sock; 1285 unsigned int len; 1286 int ret; 1287 1288 if (!(req->flags & REQ_F_POLLED) && 1289 (zc->flags & IORING_RECVSEND_POLL_FIRST)) 1290 return -EAGAIN; 1291 1292 sock = sock_from_file(req->file); 1293 if (unlikely(!sock)) 1294 return -ENOTSOCK; 1295 1296 len = zc->len; 1297 ret = io_zcrx_recv(req, zc->ifq, sock, 0, issue_flags, &zc->len); 1298 if (len && zc->len == 0) { 1299 io_req_set_res(req, 0, 0); 1300 1301 return IOU_COMPLETE; 1302 } 1303 if (unlikely(ret <= 0) && ret != -EAGAIN) { 1304 if (ret == -ERESTARTSYS) 1305 ret = -EINTR; 1306 if (ret == IOU_REQUEUE) 1307 return IOU_REQUEUE; 1308 1309 req_set_fail(req); 1310 io_req_set_res(req, ret, 0); 1311 return IOU_COMPLETE; 1312 } 1313 return IOU_RETRY; 1314 } 1315 1316 void io_send_zc_cleanup(struct io_kiocb *req) 1317 { 1318 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg); 1319 struct io_async_msghdr *io = req->async_data; 1320 1321 if (req_has_async_data(req)) 1322 io_netmsg_iovec_free(io); 1323 if (zc->notif) { 1324 io_notif_flush(zc->notif); 1325 zc->notif = NULL; 1326 } 1327 } 1328 1329 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF) 1330 #define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE | \ 1331 IORING_SEND_VECTORIZED) 1332 1333 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1334 { 1335 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg); 1336 struct io_ring_ctx *ctx = req->ctx; 1337 struct io_async_msghdr *iomsg; 1338 struct io_kiocb *notif; 1339 u64 user_data; 1340 int ret; 1341 1342 zc->done_io = 0; 1343 1344 if (unlikely(READ_ONCE(sqe->__pad2[0]))) 1345 return -EINVAL; 1346 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */ 1347 if (req->flags & REQ_F_CQE_SKIP) 1348 return -EINVAL; 1349 1350 notif = zc->notif = io_alloc_notif(ctx); 1351 if (!notif) 1352 return -ENOMEM; 1353 user_data = READ_ONCE(sqe->addr3); 1354 if (!user_data) 1355 user_data = req->cqe.user_data; 1356 1357 notif->cqe.user_data = user_data; 1358 notif->cqe.res = 0; 1359 notif->cqe.flags = IORING_CQE_F_NOTIF; 1360 req->flags |= REQ_F_NEED_CLEANUP | REQ_F_POLL_NO_LAZY; 1361 1362 zc->flags = READ_ONCE(sqe->ioprio); 1363 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) { 1364 if (zc->flags & ~IO_ZC_FLAGS_VALID) 1365 return -EINVAL; 1366 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) { 1367 struct io_notif_data *nd = io_notif_to_data(notif); 1368 1369 nd->zc_report = true; 1370 nd->zc_used = false; 1371 nd->zc_copied = false; 1372 } 1373 } 1374 1375 zc->len = READ_ONCE(sqe->len); 1376 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL | MSG_ZEROCOPY; 1377 req->buf_index = READ_ONCE(sqe->buf_index); 1378 if (zc->msg_flags & MSG_DONTWAIT) 1379 req->flags |= REQ_F_NOWAIT; 1380 1381 if (io_is_compat(ctx)) 1382 zc->msg_flags |= MSG_CMSG_COMPAT; 1383 1384 iomsg = io_msg_alloc_async(req); 1385 if (unlikely(!iomsg)) 1386 return -ENOMEM; 1387 1388 if (req->opcode == IORING_OP_SEND_ZC) { 1389 ret = io_send_setup(req, sqe); 1390 } else { 1391 if (unlikely(sqe->addr2 || sqe->file_index)) 1392 return -EINVAL; 1393 ret = io_sendmsg_setup(req, sqe); 1394 } 1395 if (unlikely(ret)) 1396 return ret; 1397 1398 if (!(zc->flags & IORING_RECVSEND_FIXED_BUF)) { 1399 iomsg->msg.sg_from_iter = io_sg_from_iter_iovec; 1400 return io_notif_account_mem(zc->notif, iomsg->msg.msg_iter.count); 1401 } 1402 iomsg->msg.sg_from_iter = io_sg_from_iter; 1403 return 0; 1404 } 1405 1406 static int io_sg_from_iter_iovec(struct sk_buff *skb, 1407 struct iov_iter *from, size_t length) 1408 { 1409 skb_zcopy_downgrade_managed(skb); 1410 return zerocopy_fill_skb_from_iter(skb, from, length); 1411 } 1412 1413 static int io_sg_from_iter(struct sk_buff *skb, 1414 struct iov_iter *from, size_t length) 1415 { 1416 struct skb_shared_info *shinfo = skb_shinfo(skb); 1417 int frag = shinfo->nr_frags; 1418 int ret = 0; 1419 struct bvec_iter bi; 1420 ssize_t copied = 0; 1421 unsigned long truesize = 0; 1422 1423 if (!frag) 1424 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS; 1425 else if (unlikely(!skb_zcopy_managed(skb))) 1426 return zerocopy_fill_skb_from_iter(skb, from, length); 1427 1428 bi.bi_size = min(from->count, length); 1429 bi.bi_bvec_done = from->iov_offset; 1430 bi.bi_idx = 0; 1431 1432 while (bi.bi_size && frag < MAX_SKB_FRAGS) { 1433 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi); 1434 1435 copied += v.bv_len; 1436 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset); 1437 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page, 1438 v.bv_offset, v.bv_len); 1439 bvec_iter_advance_single(from->bvec, &bi, v.bv_len); 1440 } 1441 if (bi.bi_size) 1442 ret = -EMSGSIZE; 1443 1444 shinfo->nr_frags = frag; 1445 from->bvec += bi.bi_idx; 1446 from->nr_segs -= bi.bi_idx; 1447 from->count -= copied; 1448 from->iov_offset = bi.bi_bvec_done; 1449 1450 skb->data_len += copied; 1451 skb->len += copied; 1452 skb->truesize += truesize; 1453 return ret; 1454 } 1455 1456 static int io_send_zc_import(struct io_kiocb *req, 1457 struct io_async_msghdr *kmsg, 1458 unsigned int issue_flags) 1459 { 1460 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1461 struct io_kiocb *notif = sr->notif; 1462 int ret; 1463 1464 WARN_ON_ONCE(!(sr->flags & IORING_RECVSEND_FIXED_BUF)); 1465 1466 notif->buf_index = req->buf_index; 1467 1468 if (!(sr->flags & IORING_SEND_VECTORIZED)) { 1469 ret = io_import_reg_buf(notif, &kmsg->msg.msg_iter, 1470 (u64)(uintptr_t)sr->buf, sr->len, 1471 ITER_SOURCE, issue_flags); 1472 } else { 1473 unsigned uvec_segs = kmsg->msg.msg_iter.nr_segs; 1474 1475 ret = io_import_reg_vec(ITER_SOURCE, &kmsg->msg.msg_iter, 1476 notif, &kmsg->vec, uvec_segs, 1477 issue_flags); 1478 } 1479 1480 if (unlikely(ret)) 1481 return ret; 1482 req->flags &= ~REQ_F_IMPORT_BUFFER; 1483 return 0; 1484 } 1485 1486 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags) 1487 { 1488 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1489 struct io_async_msghdr *kmsg = req->async_data; 1490 struct socket *sock; 1491 unsigned msg_flags; 1492 int ret, min_ret = 0; 1493 1494 sock = sock_from_file(req->file); 1495 if (unlikely(!sock)) 1496 return -ENOTSOCK; 1497 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags)) 1498 return -EOPNOTSUPP; 1499 if (!(req->flags & REQ_F_POLLED) && 1500 (sr->flags & IORING_RECVSEND_POLL_FIRST)) 1501 return -EAGAIN; 1502 1503 if (req->flags & REQ_F_IMPORT_BUFFER) { 1504 ret = io_send_zc_import(req, kmsg, issue_flags); 1505 if (unlikely(ret)) 1506 return ret; 1507 } 1508 1509 msg_flags = sr->msg_flags; 1510 if (issue_flags & IO_URING_F_NONBLOCK) 1511 msg_flags |= MSG_DONTWAIT; 1512 if (msg_flags & MSG_WAITALL) 1513 min_ret = iov_iter_count(&kmsg->msg.msg_iter); 1514 1515 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg; 1516 1517 if (req->opcode == IORING_OP_SEND_ZC) { 1518 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS; 1519 kmsg->msg.msg_flags = msg_flags; 1520 ret = sock_sendmsg(sock, &kmsg->msg); 1521 } else { 1522 kmsg->msg.msg_control_user = sr->msg_control; 1523 ret = __sys_sendmsg_sock(sock, &kmsg->msg, msg_flags); 1524 } 1525 1526 if (unlikely(ret < min_ret)) { 1527 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) 1528 return -EAGAIN; 1529 1530 if (ret > 0 && io_net_retry(sock, sr->msg_flags)) { 1531 sr->done_io += ret; 1532 return -EAGAIN; 1533 } 1534 if (ret == -ERESTARTSYS) 1535 ret = -EINTR; 1536 req_set_fail(req); 1537 } 1538 1539 if (ret >= 0) 1540 ret += sr->done_io; 1541 else if (sr->done_io) 1542 ret = sr->done_io; 1543 1544 /* 1545 * If we're in io-wq we can't rely on tw ordering guarantees, defer 1546 * flushing notif to io_send_zc_cleanup() 1547 */ 1548 if (!(issue_flags & IO_URING_F_UNLOCKED)) { 1549 io_notif_flush(sr->notif); 1550 sr->notif = NULL; 1551 io_req_msg_cleanup(req, 0); 1552 } 1553 io_req_set_res(req, ret, IORING_CQE_F_MORE); 1554 return IOU_COMPLETE; 1555 } 1556 1557 void io_sendrecv_fail(struct io_kiocb *req) 1558 { 1559 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg); 1560 1561 if (sr->done_io) 1562 req->cqe.res = sr->done_io; 1563 1564 if ((req->flags & REQ_F_NEED_CLEANUP) && 1565 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC)) 1566 req->cqe.flags |= IORING_CQE_F_MORE; 1567 } 1568 1569 #define ACCEPT_FLAGS (IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT | \ 1570 IORING_ACCEPT_POLL_FIRST) 1571 1572 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1573 { 1574 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept); 1575 1576 if (sqe->len || sqe->buf_index) 1577 return -EINVAL; 1578 1579 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); 1580 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); 1581 accept->flags = READ_ONCE(sqe->accept_flags); 1582 accept->nofile = rlimit(RLIMIT_NOFILE); 1583 accept->iou_flags = READ_ONCE(sqe->ioprio); 1584 if (accept->iou_flags & ~ACCEPT_FLAGS) 1585 return -EINVAL; 1586 1587 accept->file_slot = READ_ONCE(sqe->file_index); 1588 if (accept->file_slot) { 1589 if (accept->flags & SOCK_CLOEXEC) 1590 return -EINVAL; 1591 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT && 1592 accept->file_slot != IORING_FILE_INDEX_ALLOC) 1593 return -EINVAL; 1594 } 1595 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) 1596 return -EINVAL; 1597 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) 1598 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; 1599 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT) 1600 req->flags |= REQ_F_APOLL_MULTISHOT; 1601 if (accept->iou_flags & IORING_ACCEPT_DONTWAIT) 1602 req->flags |= REQ_F_NOWAIT; 1603 return 0; 1604 } 1605 1606 int io_accept(struct io_kiocb *req, unsigned int issue_flags) 1607 { 1608 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept); 1609 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 1610 bool fixed = !!accept->file_slot; 1611 struct proto_accept_arg arg = { 1612 .flags = force_nonblock ? O_NONBLOCK : 0, 1613 }; 1614 struct file *file; 1615 unsigned cflags; 1616 int ret, fd; 1617 1618 if (!(req->flags & REQ_F_POLLED) && 1619 accept->iou_flags & IORING_ACCEPT_POLL_FIRST) 1620 return -EAGAIN; 1621 1622 retry: 1623 if (!fixed) { 1624 fd = __get_unused_fd_flags(accept->flags, accept->nofile); 1625 if (unlikely(fd < 0)) 1626 return fd; 1627 } 1628 arg.err = 0; 1629 arg.is_empty = -1; 1630 file = do_accept(req->file, &arg, accept->addr, accept->addr_len, 1631 accept->flags); 1632 if (IS_ERR(file)) { 1633 if (!fixed) 1634 put_unused_fd(fd); 1635 ret = PTR_ERR(file); 1636 if (ret == -EAGAIN && force_nonblock && 1637 !(accept->iou_flags & IORING_ACCEPT_DONTWAIT)) 1638 return IOU_RETRY; 1639 1640 if (ret == -ERESTARTSYS) 1641 ret = -EINTR; 1642 } else if (!fixed) { 1643 fd_install(fd, file); 1644 ret = fd; 1645 } else { 1646 ret = io_fixed_fd_install(req, issue_flags, file, 1647 accept->file_slot); 1648 } 1649 1650 cflags = 0; 1651 if (!arg.is_empty) 1652 cflags |= IORING_CQE_F_SOCK_NONEMPTY; 1653 1654 if (ret >= 0 && (req->flags & REQ_F_APOLL_MULTISHOT) && 1655 io_req_post_cqe(req, ret, cflags | IORING_CQE_F_MORE)) { 1656 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || arg.is_empty == -1) 1657 goto retry; 1658 return IOU_RETRY; 1659 } 1660 1661 io_req_set_res(req, ret, cflags); 1662 if (ret < 0) 1663 req_set_fail(req); 1664 return IOU_COMPLETE; 1665 } 1666 1667 void io_socket_bpf_populate(struct io_uring_bpf_ctx *bctx, struct io_kiocb *req) 1668 { 1669 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket); 1670 1671 bctx->socket.family = sock->domain; 1672 bctx->socket.type = sock->type; 1673 bctx->socket.protocol = sock->protocol; 1674 } 1675 1676 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1677 { 1678 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket); 1679 1680 if (sqe->addr || sqe->rw_flags || sqe->buf_index) 1681 return -EINVAL; 1682 1683 sock->domain = READ_ONCE(sqe->fd); 1684 sock->type = READ_ONCE(sqe->off); 1685 sock->protocol = READ_ONCE(sqe->len); 1686 sock->file_slot = READ_ONCE(sqe->file_index); 1687 sock->nofile = rlimit(RLIMIT_NOFILE); 1688 1689 sock->flags = sock->type & ~SOCK_TYPE_MASK; 1690 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC)) 1691 return -EINVAL; 1692 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) 1693 return -EINVAL; 1694 return 0; 1695 } 1696 1697 int io_socket(struct io_kiocb *req, unsigned int issue_flags) 1698 { 1699 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket); 1700 bool fixed = !!sock->file_slot; 1701 struct file *file; 1702 int ret, fd; 1703 1704 if (!fixed) { 1705 fd = __get_unused_fd_flags(sock->flags, sock->nofile); 1706 if (unlikely(fd < 0)) 1707 return fd; 1708 } 1709 file = __sys_socket_file(sock->domain, sock->type, sock->protocol); 1710 if (IS_ERR(file)) { 1711 if (!fixed) 1712 put_unused_fd(fd); 1713 ret = PTR_ERR(file); 1714 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) 1715 return -EAGAIN; 1716 if (ret == -ERESTARTSYS) 1717 ret = -EINTR; 1718 req_set_fail(req); 1719 } else if (!fixed) { 1720 fd_install(fd, file); 1721 ret = fd; 1722 } else { 1723 ret = io_fixed_fd_install(req, issue_flags, file, 1724 sock->file_slot); 1725 } 1726 io_req_set_res(req, ret, 0); 1727 return IOU_COMPLETE; 1728 } 1729 1730 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1731 { 1732 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect); 1733 struct io_async_msghdr *io; 1734 1735 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in) 1736 return -EINVAL; 1737 1738 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); 1739 conn->addr_len = READ_ONCE(sqe->addr2); 1740 conn->in_progress = conn->seen_econnaborted = false; 1741 1742 io = io_msg_alloc_async(req); 1743 if (unlikely(!io)) 1744 return -ENOMEM; 1745 1746 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->addr); 1747 } 1748 1749 int io_connect(struct io_kiocb *req, unsigned int issue_flags) 1750 { 1751 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect); 1752 struct io_async_msghdr *io = req->async_data; 1753 unsigned file_flags; 1754 int ret; 1755 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 1756 1757 if (connect->in_progress) { 1758 struct poll_table_struct pt = { ._key = EPOLLERR }; 1759 1760 if (vfs_poll(req->file, &pt) & EPOLLERR) 1761 goto get_sock_err; 1762 } 1763 1764 file_flags = force_nonblock ? O_NONBLOCK : 0; 1765 1766 ret = __sys_connect_file(req->file, &io->addr, connect->addr_len, 1767 file_flags); 1768 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED) 1769 && force_nonblock) { 1770 if (ret == -EINPROGRESS) { 1771 connect->in_progress = true; 1772 } else if (ret == -ECONNABORTED) { 1773 if (connect->seen_econnaborted) 1774 goto out; 1775 connect->seen_econnaborted = true; 1776 } 1777 return -EAGAIN; 1778 } 1779 if (connect->in_progress) { 1780 /* 1781 * At least bluetooth will return -EBADFD on a re-connect 1782 * attempt, and it's (supposedly) also valid to get -EISCONN 1783 * which means the previous result is good. For both of these, 1784 * grab the sock_error() and use that for the completion. 1785 */ 1786 if (ret == -EBADFD || ret == -EISCONN) { 1787 get_sock_err: 1788 ret = sock_error(sock_from_file(req->file)->sk); 1789 } 1790 } 1791 if (ret == -ERESTARTSYS) 1792 ret = -EINTR; 1793 out: 1794 if (ret < 0) 1795 req_set_fail(req); 1796 io_req_msg_cleanup(req, issue_flags); 1797 io_req_set_res(req, ret, 0); 1798 return IOU_COMPLETE; 1799 } 1800 1801 int io_bind_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1802 { 1803 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind); 1804 struct sockaddr __user *uaddr; 1805 struct io_async_msghdr *io; 1806 1807 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in) 1808 return -EINVAL; 1809 1810 uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr)); 1811 bind->addr_len = READ_ONCE(sqe->addr2); 1812 1813 io = io_msg_alloc_async(req); 1814 if (unlikely(!io)) 1815 return -ENOMEM; 1816 return move_addr_to_kernel(uaddr, bind->addr_len, &io->addr); 1817 } 1818 1819 int io_bind(struct io_kiocb *req, unsigned int issue_flags) 1820 { 1821 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind); 1822 struct io_async_msghdr *io = req->async_data; 1823 struct socket *sock; 1824 int ret; 1825 1826 sock = sock_from_file(req->file); 1827 if (unlikely(!sock)) 1828 return -ENOTSOCK; 1829 1830 ret = __sys_bind_socket(sock, &io->addr, bind->addr_len); 1831 if (ret < 0) 1832 req_set_fail(req); 1833 io_req_set_res(req, ret, 0); 1834 return 0; 1835 } 1836 1837 int io_listen_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 1838 { 1839 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen); 1840 1841 if (sqe->addr || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in || sqe->addr2) 1842 return -EINVAL; 1843 1844 listen->backlog = READ_ONCE(sqe->len); 1845 return 0; 1846 } 1847 1848 int io_listen(struct io_kiocb *req, unsigned int issue_flags) 1849 { 1850 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen); 1851 struct socket *sock; 1852 int ret; 1853 1854 sock = sock_from_file(req->file); 1855 if (unlikely(!sock)) 1856 return -ENOTSOCK; 1857 1858 ret = __sys_listen_socket(sock, listen->backlog); 1859 if (ret < 0) 1860 req_set_fail(req); 1861 io_req_set_res(req, ret, 0); 1862 return 0; 1863 } 1864 1865 void io_netmsg_cache_free(const void *entry) 1866 { 1867 struct io_async_msghdr *kmsg = (struct io_async_msghdr *) entry; 1868 1869 io_vec_free(&kmsg->vec); 1870 kfree(kmsg); 1871 } 1872