1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/kernel.h> 3 #include <linux/errno.h> 4 #include <linux/fs.h> 5 #include <linux/file.h> 6 #include <linux/blk-mq.h> 7 #include <linux/mm.h> 8 #include <linux/slab.h> 9 #include <linux/fsnotify.h> 10 #include <linux/poll.h> 11 #include <linux/nospec.h> 12 #include <linux/compat.h> 13 #include <linux/io_uring.h> 14 15 #include <uapi/linux/io_uring.h> 16 17 #include "io_uring.h" 18 #include "opdef.h" 19 #include "kbuf.h" 20 #include "rsrc.h" 21 #include "rw.h" 22 23 struct io_rw { 24 /* NOTE: kiocb has the file as the first member, so don't do it here */ 25 struct kiocb kiocb; 26 u64 addr; 27 u32 len; 28 rwf_t flags; 29 }; 30 31 static inline bool io_file_supports_nowait(struct io_kiocb *req) 32 { 33 return req->flags & REQ_F_SUPPORT_NOWAIT; 34 } 35 36 #ifdef CONFIG_COMPAT 37 static int io_iov_compat_buffer_select_prep(struct io_rw *rw) 38 { 39 struct compat_iovec __user *uiov; 40 compat_ssize_t clen; 41 42 uiov = u64_to_user_ptr(rw->addr); 43 if (!access_ok(uiov, sizeof(*uiov))) 44 return -EFAULT; 45 if (__get_user(clen, &uiov->iov_len)) 46 return -EFAULT; 47 if (clen < 0) 48 return -EINVAL; 49 50 rw->len = clen; 51 return 0; 52 } 53 #endif 54 55 static int io_iov_buffer_select_prep(struct io_kiocb *req) 56 { 57 struct iovec __user *uiov; 58 struct iovec iov; 59 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 60 61 if (rw->len != 1) 62 return -EINVAL; 63 64 #ifdef CONFIG_COMPAT 65 if (req->ctx->compat) 66 return io_iov_compat_buffer_select_prep(rw); 67 #endif 68 69 uiov = u64_to_user_ptr(rw->addr); 70 if (copy_from_user(&iov, uiov, sizeof(*uiov))) 71 return -EFAULT; 72 rw->len = iov.iov_len; 73 return 0; 74 } 75 76 int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) 77 { 78 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 79 unsigned ioprio; 80 int ret; 81 82 rw->kiocb.ki_pos = READ_ONCE(sqe->off); 83 /* used for fixed read/write too - just read unconditionally */ 84 req->buf_index = READ_ONCE(sqe->buf_index); 85 86 if (req->opcode == IORING_OP_READ_FIXED || 87 req->opcode == IORING_OP_WRITE_FIXED) { 88 struct io_ring_ctx *ctx = req->ctx; 89 u16 index; 90 91 if (unlikely(req->buf_index >= ctx->nr_user_bufs)) 92 return -EFAULT; 93 index = array_index_nospec(req->buf_index, ctx->nr_user_bufs); 94 req->imu = ctx->user_bufs[index]; 95 io_req_set_rsrc_node(req, ctx, 0); 96 } 97 98 ioprio = READ_ONCE(sqe->ioprio); 99 if (ioprio) { 100 ret = ioprio_check_cap(ioprio); 101 if (ret) 102 return ret; 103 104 rw->kiocb.ki_ioprio = ioprio; 105 } else { 106 rw->kiocb.ki_ioprio = get_current_ioprio(); 107 } 108 rw->kiocb.dio_complete = NULL; 109 110 rw->addr = READ_ONCE(sqe->addr); 111 rw->len = READ_ONCE(sqe->len); 112 rw->flags = READ_ONCE(sqe->rw_flags); 113 114 /* Have to do this validation here, as this is in io_read() rw->len might 115 * have chanaged due to buffer selection 116 */ 117 if (req->opcode == IORING_OP_READV && req->flags & REQ_F_BUFFER_SELECT) { 118 ret = io_iov_buffer_select_prep(req); 119 if (ret) 120 return ret; 121 } 122 123 return 0; 124 } 125 126 void io_readv_writev_cleanup(struct io_kiocb *req) 127 { 128 struct io_async_rw *io = req->async_data; 129 130 kfree(io->free_iovec); 131 } 132 133 static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) 134 { 135 switch (ret) { 136 case -EIOCBQUEUED: 137 break; 138 case -ERESTARTSYS: 139 case -ERESTARTNOINTR: 140 case -ERESTARTNOHAND: 141 case -ERESTART_RESTARTBLOCK: 142 /* 143 * We can't just restart the syscall, since previously 144 * submitted sqes may already be in progress. Just fail this 145 * IO with EINTR. 146 */ 147 ret = -EINTR; 148 fallthrough; 149 default: 150 kiocb->ki_complete(kiocb, ret); 151 } 152 } 153 154 static inline loff_t *io_kiocb_update_pos(struct io_kiocb *req) 155 { 156 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 157 158 if (rw->kiocb.ki_pos != -1) 159 return &rw->kiocb.ki_pos; 160 161 if (!(req->file->f_mode & FMODE_STREAM)) { 162 req->flags |= REQ_F_CUR_POS; 163 rw->kiocb.ki_pos = req->file->f_pos; 164 return &rw->kiocb.ki_pos; 165 } 166 167 rw->kiocb.ki_pos = 0; 168 return NULL; 169 } 170 171 static void io_req_task_queue_reissue(struct io_kiocb *req) 172 { 173 req->io_task_work.func = io_queue_iowq; 174 io_req_task_work_add(req); 175 } 176 177 #ifdef CONFIG_BLOCK 178 static bool io_resubmit_prep(struct io_kiocb *req) 179 { 180 struct io_async_rw *io = req->async_data; 181 182 if (!req_has_async_data(req)) 183 return !io_req_prep_async(req); 184 iov_iter_restore(&io->s.iter, &io->s.iter_state); 185 return true; 186 } 187 188 static bool io_rw_should_reissue(struct io_kiocb *req) 189 { 190 umode_t mode = file_inode(req->file)->i_mode; 191 struct io_ring_ctx *ctx = req->ctx; 192 193 if (!S_ISBLK(mode) && !S_ISREG(mode)) 194 return false; 195 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && 196 !(ctx->flags & IORING_SETUP_IOPOLL))) 197 return false; 198 /* 199 * If ref is dying, we might be running poll reap from the exit work. 200 * Don't attempt to reissue from that path, just let it fail with 201 * -EAGAIN. 202 */ 203 if (percpu_ref_is_dying(&ctx->refs)) 204 return false; 205 /* 206 * Play it safe and assume not safe to re-import and reissue if we're 207 * not in the original thread group (or in task context). 208 */ 209 if (!same_thread_group(req->task, current) || !in_task()) 210 return false; 211 return true; 212 } 213 #else 214 static bool io_resubmit_prep(struct io_kiocb *req) 215 { 216 return false; 217 } 218 static bool io_rw_should_reissue(struct io_kiocb *req) 219 { 220 return false; 221 } 222 #endif 223 224 static void io_req_end_write(struct io_kiocb *req) 225 { 226 if (req->flags & REQ_F_ISREG) { 227 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 228 229 kiocb_end_write(&rw->kiocb); 230 } 231 } 232 233 /* 234 * Trigger the notifications after having done some IO, and finish the write 235 * accounting, if any. 236 */ 237 static void io_req_io_end(struct io_kiocb *req) 238 { 239 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 240 241 if (rw->kiocb.ki_flags & IOCB_WRITE) { 242 io_req_end_write(req); 243 fsnotify_modify(req->file); 244 } else { 245 fsnotify_access(req->file); 246 } 247 } 248 249 static bool __io_complete_rw_common(struct io_kiocb *req, long res) 250 { 251 if (unlikely(res != req->cqe.res)) { 252 if ((res == -EAGAIN || res == -EOPNOTSUPP) && 253 io_rw_should_reissue(req)) { 254 /* 255 * Reissue will start accounting again, finish the 256 * current cycle. 257 */ 258 io_req_io_end(req); 259 req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO; 260 return true; 261 } 262 req_set_fail(req); 263 req->cqe.res = res; 264 } 265 return false; 266 } 267 268 static inline int io_fixup_rw_res(struct io_kiocb *req, long res) 269 { 270 struct io_async_rw *io = req->async_data; 271 272 /* add previously done IO, if any */ 273 if (req_has_async_data(req) && io->bytes_done > 0) { 274 if (res < 0) 275 res = io->bytes_done; 276 else 277 res += io->bytes_done; 278 } 279 return res; 280 } 281 282 void io_req_rw_complete(struct io_kiocb *req, struct io_tw_state *ts) 283 { 284 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 285 struct kiocb *kiocb = &rw->kiocb; 286 287 if ((kiocb->ki_flags & IOCB_DIO_CALLER_COMP) && kiocb->dio_complete) { 288 long res = kiocb->dio_complete(rw->kiocb.private); 289 290 io_req_set_res(req, io_fixup_rw_res(req, res), 0); 291 } 292 293 io_req_io_end(req); 294 295 if (req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING)) { 296 unsigned issue_flags = ts->locked ? 0 : IO_URING_F_UNLOCKED; 297 298 req->cqe.flags |= io_put_kbuf(req, issue_flags); 299 } 300 io_req_task_complete(req, ts); 301 } 302 303 static void io_complete_rw(struct kiocb *kiocb, long res) 304 { 305 struct io_rw *rw = container_of(kiocb, struct io_rw, kiocb); 306 struct io_kiocb *req = cmd_to_io_kiocb(rw); 307 308 if (!kiocb->dio_complete || !(kiocb->ki_flags & IOCB_DIO_CALLER_COMP)) { 309 if (__io_complete_rw_common(req, res)) 310 return; 311 io_req_set_res(req, io_fixup_rw_res(req, res), 0); 312 } 313 req->io_task_work.func = io_req_rw_complete; 314 __io_req_task_work_add(req, IOU_F_TWQ_LAZY_WAKE); 315 } 316 317 static void io_complete_rw_iopoll(struct kiocb *kiocb, long res) 318 { 319 struct io_rw *rw = container_of(kiocb, struct io_rw, kiocb); 320 struct io_kiocb *req = cmd_to_io_kiocb(rw); 321 322 if (kiocb->ki_flags & IOCB_WRITE) 323 io_req_end_write(req); 324 if (unlikely(res != req->cqe.res)) { 325 if (res == -EAGAIN && io_rw_should_reissue(req)) { 326 req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO; 327 return; 328 } 329 req->cqe.res = res; 330 } 331 332 /* order with io_iopoll_complete() checking ->iopoll_completed */ 333 smp_store_release(&req->iopoll_completed, 1); 334 } 335 336 static int kiocb_done(struct io_kiocb *req, ssize_t ret, 337 unsigned int issue_flags) 338 { 339 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 340 unsigned final_ret = io_fixup_rw_res(req, ret); 341 342 if (ret >= 0 && req->flags & REQ_F_CUR_POS) 343 req->file->f_pos = rw->kiocb.ki_pos; 344 if (ret >= 0 && (rw->kiocb.ki_complete == io_complete_rw)) { 345 if (!__io_complete_rw_common(req, ret)) { 346 /* 347 * Safe to call io_end from here as we're inline 348 * from the submission path. 349 */ 350 io_req_io_end(req); 351 io_req_set_res(req, final_ret, 352 io_put_kbuf(req, issue_flags)); 353 return IOU_OK; 354 } 355 } else { 356 io_rw_done(&rw->kiocb, ret); 357 } 358 359 if (req->flags & REQ_F_REISSUE) { 360 req->flags &= ~REQ_F_REISSUE; 361 if (io_resubmit_prep(req)) 362 io_req_task_queue_reissue(req); 363 else 364 io_req_task_queue_fail(req, final_ret); 365 } 366 return IOU_ISSUE_SKIP_COMPLETE; 367 } 368 369 static struct iovec *__io_import_iovec(int ddir, struct io_kiocb *req, 370 struct io_rw_state *s, 371 unsigned int issue_flags) 372 { 373 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 374 struct iov_iter *iter = &s->iter; 375 u8 opcode = req->opcode; 376 struct iovec *iovec; 377 void __user *buf; 378 size_t sqe_len; 379 ssize_t ret; 380 381 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { 382 ret = io_import_fixed(ddir, iter, req->imu, rw->addr, rw->len); 383 if (ret) 384 return ERR_PTR(ret); 385 return NULL; 386 } 387 388 buf = u64_to_user_ptr(rw->addr); 389 sqe_len = rw->len; 390 391 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE || 392 (req->flags & REQ_F_BUFFER_SELECT)) { 393 if (io_do_buffer_select(req)) { 394 buf = io_buffer_select(req, &sqe_len, issue_flags); 395 if (!buf) 396 return ERR_PTR(-ENOBUFS); 397 rw->addr = (unsigned long) buf; 398 rw->len = sqe_len; 399 } 400 401 ret = import_ubuf(ddir, buf, sqe_len, iter); 402 if (ret) 403 return ERR_PTR(ret); 404 return NULL; 405 } 406 407 iovec = s->fast_iov; 408 ret = __import_iovec(ddir, buf, sqe_len, UIO_FASTIOV, &iovec, iter, 409 req->ctx->compat); 410 if (unlikely(ret < 0)) 411 return ERR_PTR(ret); 412 return iovec; 413 } 414 415 static inline int io_import_iovec(int rw, struct io_kiocb *req, 416 struct iovec **iovec, struct io_rw_state *s, 417 unsigned int issue_flags) 418 { 419 *iovec = __io_import_iovec(rw, req, s, issue_flags); 420 if (IS_ERR(*iovec)) 421 return PTR_ERR(*iovec); 422 423 iov_iter_save_state(&s->iter, &s->iter_state); 424 return 0; 425 } 426 427 static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) 428 { 429 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; 430 } 431 432 /* 433 * For files that don't have ->read_iter() and ->write_iter(), handle them 434 * by looping over ->read() or ->write() manually. 435 */ 436 static ssize_t loop_rw_iter(int ddir, struct io_rw *rw, struct iov_iter *iter) 437 { 438 struct kiocb *kiocb = &rw->kiocb; 439 struct file *file = kiocb->ki_filp; 440 ssize_t ret = 0; 441 loff_t *ppos; 442 443 /* 444 * Don't support polled IO through this interface, and we can't 445 * support non-blocking either. For the latter, this just causes 446 * the kiocb to be handled from an async context. 447 */ 448 if (kiocb->ki_flags & IOCB_HIPRI) 449 return -EOPNOTSUPP; 450 if ((kiocb->ki_flags & IOCB_NOWAIT) && 451 !(kiocb->ki_filp->f_flags & O_NONBLOCK)) 452 return -EAGAIN; 453 454 ppos = io_kiocb_ppos(kiocb); 455 456 while (iov_iter_count(iter)) { 457 void __user *addr; 458 size_t len; 459 ssize_t nr; 460 461 if (iter_is_ubuf(iter)) { 462 addr = iter->ubuf + iter->iov_offset; 463 len = iov_iter_count(iter); 464 } else if (!iov_iter_is_bvec(iter)) { 465 addr = iter_iov_addr(iter); 466 len = iter_iov_len(iter); 467 } else { 468 addr = u64_to_user_ptr(rw->addr); 469 len = rw->len; 470 } 471 472 if (ddir == READ) 473 nr = file->f_op->read(file, addr, len, ppos); 474 else 475 nr = file->f_op->write(file, addr, len, ppos); 476 477 if (nr < 0) { 478 if (!ret) 479 ret = nr; 480 break; 481 } 482 ret += nr; 483 if (!iov_iter_is_bvec(iter)) { 484 iov_iter_advance(iter, nr); 485 } else { 486 rw->addr += nr; 487 rw->len -= nr; 488 if (!rw->len) 489 break; 490 } 491 if (nr != len) 492 break; 493 } 494 495 return ret; 496 } 497 498 static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, 499 const struct iovec *fast_iov, struct iov_iter *iter) 500 { 501 struct io_async_rw *io = req->async_data; 502 503 memcpy(&io->s.iter, iter, sizeof(*iter)); 504 io->free_iovec = iovec; 505 io->bytes_done = 0; 506 /* can only be fixed buffers, no need to do anything */ 507 if (iov_iter_is_bvec(iter) || iter_is_ubuf(iter)) 508 return; 509 if (!iovec) { 510 unsigned iov_off = 0; 511 512 io->s.iter.__iov = io->s.fast_iov; 513 if (iter->__iov != fast_iov) { 514 iov_off = iter_iov(iter) - fast_iov; 515 io->s.iter.__iov += iov_off; 516 } 517 if (io->s.fast_iov != fast_iov) 518 memcpy(io->s.fast_iov + iov_off, fast_iov + iov_off, 519 sizeof(struct iovec) * iter->nr_segs); 520 } else { 521 req->flags |= REQ_F_NEED_CLEANUP; 522 } 523 } 524 525 static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, 526 struct io_rw_state *s, bool force) 527 { 528 if (!force && !io_cold_defs[req->opcode].prep_async) 529 return 0; 530 if (!req_has_async_data(req)) { 531 struct io_async_rw *iorw; 532 533 if (io_alloc_async_data(req)) { 534 kfree(iovec); 535 return -ENOMEM; 536 } 537 538 io_req_map_rw(req, iovec, s->fast_iov, &s->iter); 539 iorw = req->async_data; 540 /* we've copied and mapped the iter, ensure state is saved */ 541 iov_iter_save_state(&iorw->s.iter, &iorw->s.iter_state); 542 } 543 return 0; 544 } 545 546 static inline int io_rw_prep_async(struct io_kiocb *req, int rw) 547 { 548 struct io_async_rw *iorw = req->async_data; 549 struct iovec *iov; 550 int ret; 551 552 /* submission path, ->uring_lock should already be taken */ 553 ret = io_import_iovec(rw, req, &iov, &iorw->s, 0); 554 if (unlikely(ret < 0)) 555 return ret; 556 557 iorw->bytes_done = 0; 558 iorw->free_iovec = iov; 559 if (iov) 560 req->flags |= REQ_F_NEED_CLEANUP; 561 return 0; 562 } 563 564 int io_readv_prep_async(struct io_kiocb *req) 565 { 566 return io_rw_prep_async(req, ITER_DEST); 567 } 568 569 int io_writev_prep_async(struct io_kiocb *req) 570 { 571 return io_rw_prep_async(req, ITER_SOURCE); 572 } 573 574 /* 575 * This is our waitqueue callback handler, registered through __folio_lock_async() 576 * when we initially tried to do the IO with the iocb armed our waitqueue. 577 * This gets called when the page is unlocked, and we generally expect that to 578 * happen when the page IO is completed and the page is now uptodate. This will 579 * queue a task_work based retry of the operation, attempting to copy the data 580 * again. If the latter fails because the page was NOT uptodate, then we will 581 * do a thread based blocking retry of the operation. That's the unexpected 582 * slow path. 583 */ 584 static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, 585 int sync, void *arg) 586 { 587 struct wait_page_queue *wpq; 588 struct io_kiocb *req = wait->private; 589 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 590 struct wait_page_key *key = arg; 591 592 wpq = container_of(wait, struct wait_page_queue, wait); 593 594 if (!wake_page_match(wpq, key)) 595 return 0; 596 597 rw->kiocb.ki_flags &= ~IOCB_WAITQ; 598 list_del_init(&wait->entry); 599 io_req_task_queue(req); 600 return 1; 601 } 602 603 /* 604 * This controls whether a given IO request should be armed for async page 605 * based retry. If we return false here, the request is handed to the async 606 * worker threads for retry. If we're doing buffered reads on a regular file, 607 * we prepare a private wait_page_queue entry and retry the operation. This 608 * will either succeed because the page is now uptodate and unlocked, or it 609 * will register a callback when the page is unlocked at IO completion. Through 610 * that callback, io_uring uses task_work to setup a retry of the operation. 611 * That retry will attempt the buffered read again. The retry will generally 612 * succeed, or in rare cases where it fails, we then fall back to using the 613 * async worker threads for a blocking retry. 614 */ 615 static bool io_rw_should_retry(struct io_kiocb *req) 616 { 617 struct io_async_rw *io = req->async_data; 618 struct wait_page_queue *wait = &io->wpq; 619 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 620 struct kiocb *kiocb = &rw->kiocb; 621 622 /* never retry for NOWAIT, we just complete with -EAGAIN */ 623 if (req->flags & REQ_F_NOWAIT) 624 return false; 625 626 /* Only for buffered IO */ 627 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) 628 return false; 629 630 /* 631 * just use poll if we can, and don't attempt if the fs doesn't 632 * support callback based unlocks 633 */ 634 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) 635 return false; 636 637 wait->wait.func = io_async_buf_func; 638 wait->wait.private = req; 639 wait->wait.flags = 0; 640 INIT_LIST_HEAD(&wait->wait.entry); 641 kiocb->ki_flags |= IOCB_WAITQ; 642 kiocb->ki_flags &= ~IOCB_NOWAIT; 643 kiocb->ki_waitq = wait; 644 return true; 645 } 646 647 static inline int io_iter_do_read(struct io_rw *rw, struct iov_iter *iter) 648 { 649 struct file *file = rw->kiocb.ki_filp; 650 651 if (likely(file->f_op->read_iter)) 652 return call_read_iter(file, &rw->kiocb, iter); 653 else if (file->f_op->read) 654 return loop_rw_iter(READ, rw, iter); 655 else 656 return -EINVAL; 657 } 658 659 static bool need_complete_io(struct io_kiocb *req) 660 { 661 return req->flags & REQ_F_ISREG || 662 S_ISBLK(file_inode(req->file)->i_mode); 663 } 664 665 static int io_rw_init_file(struct io_kiocb *req, fmode_t mode) 666 { 667 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 668 struct kiocb *kiocb = &rw->kiocb; 669 struct io_ring_ctx *ctx = req->ctx; 670 struct file *file = req->file; 671 int ret; 672 673 if (unlikely(!file || !(file->f_mode & mode))) 674 return -EBADF; 675 676 if (!(req->flags & REQ_F_FIXED_FILE)) 677 req->flags |= io_file_get_flags(file); 678 679 kiocb->ki_flags = file->f_iocb_flags; 680 ret = kiocb_set_rw_flags(kiocb, rw->flags); 681 if (unlikely(ret)) 682 return ret; 683 kiocb->ki_flags |= IOCB_ALLOC_CACHE; 684 685 /* 686 * If the file is marked O_NONBLOCK, still allow retry for it if it 687 * supports async. Otherwise it's impossible to use O_NONBLOCK files 688 * reliably. If not, or it IOCB_NOWAIT is set, don't retry. 689 */ 690 if ((kiocb->ki_flags & IOCB_NOWAIT) || 691 ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req))) 692 req->flags |= REQ_F_NOWAIT; 693 694 if (ctx->flags & IORING_SETUP_IOPOLL) { 695 if (!(kiocb->ki_flags & IOCB_DIRECT) || !file->f_op->iopoll) 696 return -EOPNOTSUPP; 697 698 kiocb->private = NULL; 699 kiocb->ki_flags |= IOCB_HIPRI; 700 kiocb->ki_complete = io_complete_rw_iopoll; 701 req->iopoll_completed = 0; 702 } else { 703 if (kiocb->ki_flags & IOCB_HIPRI) 704 return -EINVAL; 705 kiocb->ki_complete = io_complete_rw; 706 } 707 708 return 0; 709 } 710 711 int io_read(struct io_kiocb *req, unsigned int issue_flags) 712 { 713 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 714 struct io_rw_state __s, *s = &__s; 715 struct iovec *iovec; 716 struct kiocb *kiocb = &rw->kiocb; 717 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 718 struct io_async_rw *io; 719 ssize_t ret, ret2; 720 loff_t *ppos; 721 722 if (!req_has_async_data(req)) { 723 ret = io_import_iovec(ITER_DEST, req, &iovec, s, issue_flags); 724 if (unlikely(ret < 0)) 725 return ret; 726 } else { 727 io = req->async_data; 728 s = &io->s; 729 730 /* 731 * Safe and required to re-import if we're using provided 732 * buffers, as we dropped the selected one before retry. 733 */ 734 if (io_do_buffer_select(req)) { 735 ret = io_import_iovec(ITER_DEST, req, &iovec, s, issue_flags); 736 if (unlikely(ret < 0)) 737 return ret; 738 } 739 740 /* 741 * We come here from an earlier attempt, restore our state to 742 * match in case it doesn't. It's cheap enough that we don't 743 * need to make this conditional. 744 */ 745 iov_iter_restore(&s->iter, &s->iter_state); 746 iovec = NULL; 747 } 748 ret = io_rw_init_file(req, FMODE_READ); 749 if (unlikely(ret)) { 750 kfree(iovec); 751 return ret; 752 } 753 req->cqe.res = iov_iter_count(&s->iter); 754 755 if (force_nonblock) { 756 /* If the file doesn't support async, just async punt */ 757 if (unlikely(!io_file_supports_nowait(req))) { 758 ret = io_setup_async_rw(req, iovec, s, true); 759 return ret ?: -EAGAIN; 760 } 761 kiocb->ki_flags |= IOCB_NOWAIT; 762 } else { 763 /* Ensure we clear previously set non-block flag */ 764 kiocb->ki_flags &= ~IOCB_NOWAIT; 765 } 766 767 ppos = io_kiocb_update_pos(req); 768 769 ret = rw_verify_area(READ, req->file, ppos, req->cqe.res); 770 if (unlikely(ret)) { 771 kfree(iovec); 772 return ret; 773 } 774 775 ret = io_iter_do_read(rw, &s->iter); 776 777 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { 778 req->flags &= ~REQ_F_REISSUE; 779 /* if we can poll, just do that */ 780 if (req->opcode == IORING_OP_READ && file_can_poll(req->file)) 781 return -EAGAIN; 782 /* IOPOLL retry should happen for io-wq threads */ 783 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) 784 goto done; 785 /* no retry on NONBLOCK nor RWF_NOWAIT */ 786 if (req->flags & REQ_F_NOWAIT) 787 goto done; 788 ret = 0; 789 } else if (ret == -EIOCBQUEUED) { 790 if (iovec) 791 kfree(iovec); 792 return IOU_ISSUE_SKIP_COMPLETE; 793 } else if (ret == req->cqe.res || ret <= 0 || !force_nonblock || 794 (req->flags & REQ_F_NOWAIT) || !need_complete_io(req)) { 795 /* read all, failed, already did sync or don't want to retry */ 796 goto done; 797 } 798 799 /* 800 * Don't depend on the iter state matching what was consumed, or being 801 * untouched in case of error. Restore it and we'll advance it 802 * manually if we need to. 803 */ 804 iov_iter_restore(&s->iter, &s->iter_state); 805 806 ret2 = io_setup_async_rw(req, iovec, s, true); 807 iovec = NULL; 808 if (ret2) { 809 ret = ret > 0 ? ret : ret2; 810 goto done; 811 } 812 813 io = req->async_data; 814 s = &io->s; 815 /* 816 * Now use our persistent iterator and state, if we aren't already. 817 * We've restored and mapped the iter to match. 818 */ 819 820 do { 821 /* 822 * We end up here because of a partial read, either from 823 * above or inside this loop. Advance the iter by the bytes 824 * that were consumed. 825 */ 826 iov_iter_advance(&s->iter, ret); 827 if (!iov_iter_count(&s->iter)) 828 break; 829 io->bytes_done += ret; 830 iov_iter_save_state(&s->iter, &s->iter_state); 831 832 /* if we can retry, do so with the callbacks armed */ 833 if (!io_rw_should_retry(req)) { 834 kiocb->ki_flags &= ~IOCB_WAITQ; 835 return -EAGAIN; 836 } 837 838 req->cqe.res = iov_iter_count(&s->iter); 839 /* 840 * Now retry read with the IOCB_WAITQ parts set in the iocb. If 841 * we get -EIOCBQUEUED, then we'll get a notification when the 842 * desired page gets unlocked. We can also get a partial read 843 * here, and if we do, then just retry at the new offset. 844 */ 845 ret = io_iter_do_read(rw, &s->iter); 846 if (ret == -EIOCBQUEUED) 847 return IOU_ISSUE_SKIP_COMPLETE; 848 /* we got some bytes, but not all. retry. */ 849 kiocb->ki_flags &= ~IOCB_WAITQ; 850 iov_iter_restore(&s->iter, &s->iter_state); 851 } while (ret > 0); 852 done: 853 /* it's faster to check here then delegate to kfree */ 854 if (iovec) 855 kfree(iovec); 856 return kiocb_done(req, ret, issue_flags); 857 } 858 859 int io_write(struct io_kiocb *req, unsigned int issue_flags) 860 { 861 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 862 struct io_rw_state __s, *s = &__s; 863 struct iovec *iovec; 864 struct kiocb *kiocb = &rw->kiocb; 865 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; 866 ssize_t ret, ret2; 867 loff_t *ppos; 868 869 if (!req_has_async_data(req)) { 870 ret = io_import_iovec(ITER_SOURCE, req, &iovec, s, issue_flags); 871 if (unlikely(ret < 0)) 872 return ret; 873 } else { 874 struct io_async_rw *io = req->async_data; 875 876 s = &io->s; 877 iov_iter_restore(&s->iter, &s->iter_state); 878 iovec = NULL; 879 } 880 ret = io_rw_init_file(req, FMODE_WRITE); 881 if (unlikely(ret)) { 882 kfree(iovec); 883 return ret; 884 } 885 req->cqe.res = iov_iter_count(&s->iter); 886 887 if (force_nonblock) { 888 /* If the file doesn't support async, just async punt */ 889 if (unlikely(!io_file_supports_nowait(req))) 890 goto copy_iov; 891 892 /* File path supports NOWAIT for non-direct_IO only for block devices. */ 893 if (!(kiocb->ki_flags & IOCB_DIRECT) && 894 !(kiocb->ki_filp->f_mode & FMODE_BUF_WASYNC) && 895 (req->flags & REQ_F_ISREG)) 896 goto copy_iov; 897 898 kiocb->ki_flags |= IOCB_NOWAIT; 899 } else { 900 /* Ensure we clear previously set non-block flag */ 901 kiocb->ki_flags &= ~IOCB_NOWAIT; 902 } 903 904 ppos = io_kiocb_update_pos(req); 905 906 ret = rw_verify_area(WRITE, req->file, ppos, req->cqe.res); 907 if (unlikely(ret)) { 908 kfree(iovec); 909 return ret; 910 } 911 912 if (req->flags & REQ_F_ISREG) 913 kiocb_start_write(kiocb); 914 kiocb->ki_flags |= IOCB_WRITE; 915 916 if (likely(req->file->f_op->write_iter)) 917 ret2 = call_write_iter(req->file, kiocb, &s->iter); 918 else if (req->file->f_op->write) 919 ret2 = loop_rw_iter(WRITE, rw, &s->iter); 920 else 921 ret2 = -EINVAL; 922 923 if (req->flags & REQ_F_REISSUE) { 924 req->flags &= ~REQ_F_REISSUE; 925 ret2 = -EAGAIN; 926 } 927 928 /* 929 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just 930 * retry them without IOCB_NOWAIT. 931 */ 932 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) 933 ret2 = -EAGAIN; 934 /* no retry on NONBLOCK nor RWF_NOWAIT */ 935 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) 936 goto done; 937 if (!force_nonblock || ret2 != -EAGAIN) { 938 /* IOPOLL retry should happen for io-wq threads */ 939 if (ret2 == -EAGAIN && (req->ctx->flags & IORING_SETUP_IOPOLL)) 940 goto copy_iov; 941 942 if (ret2 != req->cqe.res && ret2 >= 0 && need_complete_io(req)) { 943 struct io_async_rw *io; 944 945 trace_io_uring_short_write(req->ctx, kiocb->ki_pos - ret2, 946 req->cqe.res, ret2); 947 948 /* This is a partial write. The file pos has already been 949 * updated, setup the async struct to complete the request 950 * in the worker. Also update bytes_done to account for 951 * the bytes already written. 952 */ 953 iov_iter_save_state(&s->iter, &s->iter_state); 954 ret = io_setup_async_rw(req, iovec, s, true); 955 956 io = req->async_data; 957 if (io) 958 io->bytes_done += ret2; 959 960 if (kiocb->ki_flags & IOCB_WRITE) 961 io_req_end_write(req); 962 return ret ? ret : -EAGAIN; 963 } 964 done: 965 ret = kiocb_done(req, ret2, issue_flags); 966 } else { 967 copy_iov: 968 iov_iter_restore(&s->iter, &s->iter_state); 969 ret = io_setup_async_rw(req, iovec, s, false); 970 if (!ret) { 971 if (kiocb->ki_flags & IOCB_WRITE) 972 io_req_end_write(req); 973 return -EAGAIN; 974 } 975 return ret; 976 } 977 /* it's reportedly faster than delegating the null check to kfree() */ 978 if (iovec) 979 kfree(iovec); 980 return ret; 981 } 982 983 void io_rw_fail(struct io_kiocb *req) 984 { 985 int res; 986 987 res = io_fixup_rw_res(req, req->cqe.res); 988 io_req_set_res(req, res, req->cqe.flags); 989 } 990 991 int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin) 992 { 993 struct io_wq_work_node *pos, *start, *prev; 994 unsigned int poll_flags = 0; 995 DEFINE_IO_COMP_BATCH(iob); 996 int nr_events = 0; 997 998 /* 999 * Only spin for completions if we don't have multiple devices hanging 1000 * off our complete list. 1001 */ 1002 if (ctx->poll_multi_queue || force_nonspin) 1003 poll_flags |= BLK_POLL_ONESHOT; 1004 1005 wq_list_for_each(pos, start, &ctx->iopoll_list) { 1006 struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); 1007 struct file *file = req->file; 1008 int ret; 1009 1010 /* 1011 * Move completed and retryable entries to our local lists. 1012 * If we find a request that requires polling, break out 1013 * and complete those lists first, if we have entries there. 1014 */ 1015 if (READ_ONCE(req->iopoll_completed)) 1016 break; 1017 1018 if (req->opcode == IORING_OP_URING_CMD) { 1019 struct io_uring_cmd *ioucmd; 1020 1021 ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 1022 ret = file->f_op->uring_cmd_iopoll(ioucmd, &iob, 1023 poll_flags); 1024 } else { 1025 struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw); 1026 1027 ret = file->f_op->iopoll(&rw->kiocb, &iob, poll_flags); 1028 } 1029 if (unlikely(ret < 0)) 1030 return ret; 1031 else if (ret) 1032 poll_flags |= BLK_POLL_ONESHOT; 1033 1034 /* iopoll may have completed current req */ 1035 if (!rq_list_empty(iob.req_list) || 1036 READ_ONCE(req->iopoll_completed)) 1037 break; 1038 } 1039 1040 if (!rq_list_empty(iob.req_list)) 1041 iob.complete(&iob); 1042 else if (!pos) 1043 return 0; 1044 1045 prev = start; 1046 wq_list_for_each_resume(pos, prev) { 1047 struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); 1048 1049 /* order with io_complete_rw_iopoll(), e.g. ->result updates */ 1050 if (!smp_load_acquire(&req->iopoll_completed)) 1051 break; 1052 nr_events++; 1053 req->cqe.flags = io_put_kbuf(req, 0); 1054 } 1055 if (unlikely(!nr_events)) 1056 return 0; 1057 1058 pos = start ? start->next : ctx->iopoll_list.first; 1059 wq_list_cut(&ctx->iopoll_list, prev, start); 1060 1061 if (WARN_ON_ONCE(!wq_list_empty(&ctx->submit_state.compl_reqs))) 1062 return 0; 1063 ctx->submit_state.compl_reqs.first = pos; 1064 __io_submit_flush_completions(ctx); 1065 return nr_events; 1066 } 1067