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