1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1991, 1992 Linus Torvalds 4 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE 5 * Copyright (C) 2016 - 2020 Christoph Hellwig 6 */ 7 #include <linux/init.h> 8 #include <linux/mm.h> 9 #include <linux/blkdev.h> 10 #include <linux/blk-integrity.h> 11 #include <linux/buffer_head.h> 12 #include <linux/mpage.h> 13 #include <linux/uio.h> 14 #include <linux/namei.h> 15 #include <linux/task_io_accounting_ops.h> 16 #include <linux/falloc.h> 17 #include <linux/suspend.h> 18 #include <linux/fs.h> 19 #include <linux/iomap.h> 20 #include <linux/module.h> 21 #include <linux/io_uring/cmd.h> 22 #include "blk.h" 23 24 static inline struct inode *bdev_file_inode(struct file *file) 25 { 26 return file->f_mapping->host; 27 } 28 29 static blk_opf_t dio_bio_write_op(struct kiocb *iocb) 30 { 31 blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; 32 33 /* avoid the need for a I/O completion work item */ 34 if (iocb_is_dsync(iocb)) 35 opf |= REQ_FUA; 36 return opf; 37 } 38 39 static bool blkdev_dio_invalid(struct block_device *bdev, struct kiocb *iocb, 40 struct iov_iter *iter) 41 { 42 return (iocb->ki_pos | iov_iter_count(iter)) & 43 (bdev_logical_block_size(bdev) - 1); 44 } 45 46 static inline int blkdev_iov_iter_get_pages(struct bio *bio, 47 struct iov_iter *iter, struct block_device *bdev) 48 { 49 return bio_iov_iter_get_pages(bio, iter, 50 bdev_logical_block_size(bdev) - 1); 51 } 52 53 #define DIO_INLINE_BIO_VECS 4 54 55 static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb, 56 struct iov_iter *iter, struct block_device *bdev, 57 unsigned int nr_pages) 58 { 59 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; 60 loff_t pos = iocb->ki_pos; 61 bool should_dirty = false; 62 struct bio bio; 63 ssize_t ret; 64 65 if (nr_pages <= DIO_INLINE_BIO_VECS) 66 vecs = inline_vecs; 67 else { 68 vecs = kmalloc_objs(struct bio_vec, nr_pages); 69 if (!vecs) 70 return -ENOMEM; 71 } 72 73 if (iov_iter_rw(iter) == READ) { 74 bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ); 75 if (user_backed_iter(iter)) 76 should_dirty = true; 77 } else { 78 bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb)); 79 } 80 bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT; 81 bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 82 bio.bi_write_stream = iocb->ki_write_stream; 83 bio.bi_ioprio = iocb->ki_ioprio; 84 if (iocb->ki_flags & IOCB_ATOMIC) 85 bio.bi_opf |= REQ_ATOMIC; 86 87 ret = blkdev_iov_iter_get_pages(&bio, iter, bdev); 88 if (unlikely(ret)) 89 goto out; 90 ret = bio.bi_iter.bi_size; 91 92 if (iov_iter_rw(iter) == WRITE) 93 task_io_account_write(ret); 94 95 if (iocb->ki_flags & IOCB_NOWAIT) 96 bio.bi_opf |= REQ_NOWAIT; 97 98 submit_bio_wait(&bio); 99 100 bio_release_pages(&bio, should_dirty); 101 if (unlikely(bio.bi_status)) 102 ret = blk_status_to_errno(bio.bi_status); 103 104 out: 105 if (vecs != inline_vecs) 106 kfree(vecs); 107 108 bio_uninit(&bio); 109 110 return ret; 111 } 112 113 enum { 114 DIO_SHOULD_DIRTY = 1, 115 DIO_IS_SYNC = 2, 116 }; 117 118 struct blkdev_dio { 119 union { 120 struct kiocb *iocb; 121 struct task_struct *waiter; 122 }; 123 size_t size; 124 atomic_t ref; 125 unsigned int flags; 126 struct bio bio ____cacheline_aligned_in_smp; 127 }; 128 129 static struct bio_set blkdev_dio_pool; 130 131 static void blkdev_bio_end_io(struct bio *bio) 132 { 133 struct blkdev_dio *dio = bio->bi_private; 134 bool should_dirty = dio->flags & DIO_SHOULD_DIRTY; 135 bool is_sync = dio->flags & DIO_IS_SYNC; 136 137 if (bio->bi_status && !dio->bio.bi_status) 138 dio->bio.bi_status = bio->bi_status; 139 140 if (bio_integrity(bio)) 141 bio_integrity_unmap_user(bio); 142 143 if (atomic_dec_and_test(&dio->ref)) { 144 if (!is_sync) { 145 struct kiocb *iocb = dio->iocb; 146 ssize_t ret; 147 148 WRITE_ONCE(iocb->private, NULL); 149 150 if (likely(!dio->bio.bi_status)) { 151 ret = dio->size; 152 iocb->ki_pos += ret; 153 } else { 154 ret = blk_status_to_errno(dio->bio.bi_status); 155 } 156 157 dio->iocb->ki_complete(iocb, ret); 158 bio_put(&dio->bio); 159 } else { 160 struct task_struct *waiter = dio->waiter; 161 162 WRITE_ONCE(dio->waiter, NULL); 163 blk_wake_io_task(waiter); 164 } 165 } 166 167 if (should_dirty) { 168 bio_check_pages_dirty(bio); 169 } else { 170 bio_release_pages(bio, false); 171 bio_put(bio); 172 } 173 } 174 175 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, 176 struct block_device *bdev, unsigned int nr_pages) 177 { 178 struct blk_plug plug; 179 struct blkdev_dio *dio; 180 struct bio *bio; 181 bool is_read = (iov_iter_rw(iter) == READ), is_sync; 182 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 183 loff_t pos = iocb->ki_pos; 184 int ret = 0; 185 186 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 187 &blkdev_dio_pool); 188 dio = container_of(bio, struct blkdev_dio, bio); 189 atomic_set(&dio->ref, 1); 190 /* 191 * Grab an extra reference to ensure the dio structure which is embedded 192 * into the first bio stays around. 193 */ 194 bio_get(bio); 195 196 is_sync = is_sync_kiocb(iocb); 197 if (is_sync) { 198 dio->flags = DIO_IS_SYNC; 199 dio->waiter = current; 200 } else { 201 dio->flags = 0; 202 dio->iocb = iocb; 203 } 204 205 dio->size = 0; 206 if (is_read && user_backed_iter(iter)) 207 dio->flags |= DIO_SHOULD_DIRTY; 208 209 blk_start_plug(&plug); 210 211 for (;;) { 212 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 213 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 214 bio->bi_write_stream = iocb->ki_write_stream; 215 bio->bi_private = dio; 216 bio->bi_end_io = blkdev_bio_end_io; 217 bio->bi_ioprio = iocb->ki_ioprio; 218 219 ret = blkdev_iov_iter_get_pages(bio, iter, bdev); 220 if (unlikely(ret)) { 221 bio->bi_status = BLK_STS_IOERR; 222 bio_endio(bio); 223 break; 224 } 225 if (iocb->ki_flags & IOCB_NOWAIT) { 226 /* 227 * This is nonblocking IO, and we need to allocate 228 * another bio if we have data left to map. As we 229 * cannot guarantee that one of the sub bios will not 230 * fail getting issued FOR NOWAIT and as error results 231 * are coalesced across all of them, be safe and ask for 232 * a retry of this from blocking context. 233 */ 234 if (unlikely(iov_iter_count(iter))) { 235 ret = -EAGAIN; 236 goto fail; 237 } 238 bio->bi_opf |= REQ_NOWAIT; 239 } 240 if (iocb->ki_flags & IOCB_HAS_METADATA) { 241 ret = bio_integrity_map_iter(bio, iocb->private); 242 if (unlikely(ret)) 243 goto fail; 244 } 245 246 if (is_read) { 247 if (dio->flags & DIO_SHOULD_DIRTY) 248 bio_set_pages_dirty(bio); 249 } else { 250 task_io_account_write(bio->bi_iter.bi_size); 251 } 252 dio->size += bio->bi_iter.bi_size; 253 pos += bio->bi_iter.bi_size; 254 255 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); 256 if (!nr_pages) { 257 submit_bio(bio); 258 break; 259 } 260 atomic_inc(&dio->ref); 261 submit_bio(bio); 262 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL); 263 } 264 265 blk_finish_plug(&plug); 266 267 if (!is_sync) 268 return -EIOCBQUEUED; 269 270 for (;;) { 271 set_current_state(TASK_UNINTERRUPTIBLE); 272 if (!READ_ONCE(dio->waiter)) 273 break; 274 blk_io_schedule(); 275 } 276 __set_current_state(TASK_RUNNING); 277 278 if (!ret) 279 ret = blk_status_to_errno(dio->bio.bi_status); 280 if (likely(!ret)) 281 ret = dio->size; 282 283 bio_put(&dio->bio); 284 return ret; 285 fail: 286 bio_release_pages(bio, false); 287 bio_clear_flag(bio, BIO_REFFED); 288 bio_put(bio); 289 blk_finish_plug(&plug); 290 return ret; 291 } 292 293 static void blkdev_bio_end_io_async(struct bio *bio) 294 { 295 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); 296 struct kiocb *iocb = dio->iocb; 297 ssize_t ret; 298 299 WRITE_ONCE(iocb->private, NULL); 300 301 if (likely(!bio->bi_status)) { 302 ret = dio->size; 303 iocb->ki_pos += ret; 304 } else { 305 ret = blk_status_to_errno(bio->bi_status); 306 } 307 308 if (bio_integrity(bio)) 309 bio_integrity_unmap_user(bio); 310 311 iocb->ki_complete(iocb, ret); 312 313 if (dio->flags & DIO_SHOULD_DIRTY) { 314 bio_check_pages_dirty(bio); 315 } else { 316 bio_release_pages(bio, false); 317 bio_put(bio); 318 } 319 } 320 321 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, 322 struct iov_iter *iter, 323 struct block_device *bdev, 324 unsigned int nr_pages) 325 { 326 bool is_read = iov_iter_rw(iter) == READ; 327 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 328 struct blkdev_dio *dio; 329 struct bio *bio; 330 loff_t pos = iocb->ki_pos; 331 int ret = 0; 332 333 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 334 &blkdev_dio_pool); 335 dio = container_of(bio, struct blkdev_dio, bio); 336 dio->flags = 0; 337 dio->iocb = iocb; 338 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 339 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 340 bio->bi_write_stream = iocb->ki_write_stream; 341 bio->bi_end_io = blkdev_bio_end_io_async; 342 bio->bi_ioprio = iocb->ki_ioprio; 343 344 if (iov_iter_is_bvec(iter)) { 345 /* 346 * Users don't rely on the iterator being in any particular 347 * state for async I/O returning -EIOCBQUEUED, hence we can 348 * avoid expensive iov_iter_advance(). Bypass 349 * bio_iov_iter_get_pages() and set the bvec directly. 350 */ 351 bio_iov_bvec_set(bio, iter); 352 } else { 353 ret = blkdev_iov_iter_get_pages(bio, iter, bdev); 354 if (unlikely(ret)) 355 goto out_bio_put; 356 } 357 dio->size = bio->bi_iter.bi_size; 358 359 if (is_read) { 360 if (user_backed_iter(iter)) { 361 dio->flags |= DIO_SHOULD_DIRTY; 362 bio_set_pages_dirty(bio); 363 } 364 } else { 365 task_io_account_write(bio->bi_iter.bi_size); 366 } 367 368 if (iocb->ki_flags & IOCB_HAS_METADATA) { 369 ret = bio_integrity_map_iter(bio, iocb->private); 370 WRITE_ONCE(iocb->private, NULL); 371 if (unlikely(ret)) 372 goto out_bio_put; 373 } 374 375 if (iocb->ki_flags & IOCB_ATOMIC) 376 bio->bi_opf |= REQ_ATOMIC; 377 378 if (iocb->ki_flags & IOCB_NOWAIT) 379 bio->bi_opf |= REQ_NOWAIT; 380 381 if (iocb->ki_flags & IOCB_HIPRI) { 382 bio->bi_opf |= REQ_POLLED; 383 submit_bio(bio); 384 WRITE_ONCE(iocb->private, bio); 385 } else { 386 submit_bio(bio); 387 } 388 return -EIOCBQUEUED; 389 390 out_bio_put: 391 bio_put(bio); 392 return ret; 393 } 394 395 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 396 { 397 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 398 unsigned int nr_pages; 399 400 if (!iov_iter_count(iter)) 401 return 0; 402 403 if (blkdev_dio_invalid(bdev, iocb, iter)) 404 return -EINVAL; 405 406 if (iov_iter_rw(iter) == WRITE) { 407 u16 max_write_streams = bdev_max_write_streams(bdev); 408 409 if (iocb->ki_write_stream) { 410 if (iocb->ki_write_stream > max_write_streams) 411 return -EINVAL; 412 } else if (max_write_streams) { 413 enum rw_hint write_hint = 414 file_inode(iocb->ki_filp)->i_write_hint; 415 416 /* 417 * Just use the write hint as write stream for block 418 * device writes. This assumes no file system is 419 * mounted that would use the streams differently. 420 */ 421 if (write_hint <= max_write_streams) 422 iocb->ki_write_stream = write_hint; 423 } 424 } 425 426 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); 427 if (likely(nr_pages <= BIO_MAX_VECS && 428 !(iocb->ki_flags & IOCB_HAS_METADATA))) { 429 if (is_sync_kiocb(iocb)) 430 return __blkdev_direct_IO_simple(iocb, iter, bdev, 431 nr_pages); 432 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages); 433 } else if (iocb->ki_flags & IOCB_ATOMIC) { 434 return -EINVAL; 435 } 436 return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages)); 437 } 438 439 static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 440 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 441 { 442 struct block_device *bdev = I_BDEV(inode); 443 loff_t isize = i_size_read(inode); 444 445 if (offset >= isize) 446 return -EIO; 447 448 iomap->bdev = bdev; 449 iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev)); 450 iomap->type = IOMAP_MAPPED; 451 iomap->addr = iomap->offset; 452 iomap->length = isize - iomap->offset; 453 iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */ 454 return 0; 455 } 456 457 static const struct iomap_ops blkdev_iomap_ops = { 458 .iomap_begin = blkdev_iomap_begin, 459 }; 460 461 #ifdef CONFIG_BUFFER_HEAD 462 static int blkdev_get_block(struct inode *inode, sector_t iblock, 463 struct buffer_head *bh, int create) 464 { 465 bh->b_bdev = I_BDEV(inode); 466 bh->b_blocknr = iblock; 467 set_buffer_mapped(bh); 468 return 0; 469 } 470 471 /* 472 * We cannot call mpage_writepages() as it does not take the buffer lock. 473 * We must use block_write_full_folio() directly which holds the buffer 474 * lock. The buffer lock provides the synchronisation with writeback 475 * that filesystems rely on when they use the blockdev's mapping. 476 */ 477 static int blkdev_writepages(struct address_space *mapping, 478 struct writeback_control *wbc) 479 { 480 struct folio *folio = NULL; 481 struct blk_plug plug; 482 int err; 483 484 blk_start_plug(&plug); 485 while ((folio = writeback_iter(mapping, wbc, folio, &err))) 486 err = block_write_full_folio(folio, wbc, blkdev_get_block); 487 blk_finish_plug(&plug); 488 489 return err; 490 } 491 492 static int blkdev_read_folio(struct file *file, struct folio *folio) 493 { 494 return block_read_full_folio(folio, blkdev_get_block); 495 } 496 497 static void blkdev_readahead(struct readahead_control *rac) 498 { 499 mpage_readahead(rac, blkdev_get_block); 500 } 501 502 const struct address_space_operations def_blk_aops = { 503 .dirty_folio = block_dirty_folio, 504 .invalidate_folio = block_invalidate_folio, 505 .read_folio = blkdev_read_folio, 506 .readahead = blkdev_readahead, 507 .writepages = blkdev_writepages, 508 .migrate_folio = buffer_migrate_folio_norefs, 509 .is_dirty_writeback = buffer_check_dirty_writeback, 510 }; 511 #else /* CONFIG_BUFFER_HEAD */ 512 static int blkdev_read_folio(struct file *file, struct folio *folio) 513 { 514 iomap_bio_read_folio(folio, &blkdev_iomap_ops); 515 return 0; 516 } 517 518 static void blkdev_readahead(struct readahead_control *rac) 519 { 520 iomap_bio_readahead(rac, &blkdev_iomap_ops); 521 } 522 523 static ssize_t blkdev_writeback_range(struct iomap_writepage_ctx *wpc, 524 struct folio *folio, u64 offset, unsigned int len, u64 end_pos) 525 { 526 loff_t isize = i_size_read(wpc->inode); 527 528 if (WARN_ON_ONCE(offset >= isize)) 529 return -EIO; 530 531 if (offset < wpc->iomap.offset || 532 offset >= wpc->iomap.offset + wpc->iomap.length) { 533 int error; 534 535 error = blkdev_iomap_begin(wpc->inode, offset, isize - offset, 536 IOMAP_WRITE, &wpc->iomap, NULL); 537 if (error) 538 return error; 539 } 540 541 return iomap_add_to_ioend(wpc, folio, offset, end_pos, len); 542 } 543 544 static const struct iomap_writeback_ops blkdev_writeback_ops = { 545 .writeback_range = blkdev_writeback_range, 546 .writeback_submit = iomap_ioend_writeback_submit, 547 }; 548 549 static int blkdev_writepages(struct address_space *mapping, 550 struct writeback_control *wbc) 551 { 552 struct iomap_writepage_ctx wpc = { 553 .inode = mapping->host, 554 .wbc = wbc, 555 .ops = &blkdev_writeback_ops 556 }; 557 558 return iomap_writepages(&wpc); 559 } 560 561 const struct address_space_operations def_blk_aops = { 562 .dirty_folio = filemap_dirty_folio, 563 .release_folio = iomap_release_folio, 564 .invalidate_folio = iomap_invalidate_folio, 565 .read_folio = blkdev_read_folio, 566 .readahead = blkdev_readahead, 567 .writepages = blkdev_writepages, 568 .is_partially_uptodate = iomap_is_partially_uptodate, 569 .error_remove_folio = generic_error_remove_folio, 570 .migrate_folio = filemap_migrate_folio, 571 }; 572 #endif /* CONFIG_BUFFER_HEAD */ 573 574 /* 575 * for a block special file file_inode(file)->i_size is zero 576 * so we compute the size by hand (just as in block_read/write above) 577 */ 578 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) 579 { 580 struct inode *bd_inode = bdev_file_inode(file); 581 loff_t retval; 582 583 inode_lock(bd_inode); 584 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); 585 inode_unlock(bd_inode); 586 return retval; 587 } 588 589 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, 590 int datasync) 591 { 592 struct block_device *bdev = I_BDEV(filp->f_mapping->host); 593 int error; 594 595 error = file_write_and_wait_range(filp, start, end); 596 if (error) 597 return error; 598 599 /* 600 * There is no need to serialise calls to blkdev_issue_flush with 601 * i_mutex and doing so causes performance issues with concurrent 602 * O_SYNC writers to a block device. 603 */ 604 error = blkdev_issue_flush(bdev); 605 if (error == -EOPNOTSUPP) 606 error = 0; 607 608 return error; 609 } 610 611 /** 612 * file_to_blk_mode - get block open flags from file flags 613 * @file: file whose open flags should be converted 614 * 615 * Look at file open flags and generate corresponding block open flags from 616 * them. The function works both for file just being open (e.g. during ->open 617 * callback) and for file that is already open. This is actually non-trivial 618 * (see comment in the function). 619 */ 620 blk_mode_t file_to_blk_mode(struct file *file) 621 { 622 blk_mode_t mode = 0; 623 624 if (file->f_mode & FMODE_READ) 625 mode |= BLK_OPEN_READ; 626 if (file->f_mode & FMODE_WRITE) 627 mode |= BLK_OPEN_WRITE; 628 /* 629 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data 630 * to determine whether the open was exclusive for already open files. 631 */ 632 if (file->private_data) 633 mode |= BLK_OPEN_EXCL; 634 else if (file->f_flags & O_EXCL) 635 mode |= BLK_OPEN_EXCL; 636 if (file->f_flags & O_NDELAY) 637 mode |= BLK_OPEN_NDELAY; 638 639 /* 640 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy 641 * driver has historically allowed ioctls as if the file was opened for 642 * writing, but does not allow and actual reads or writes. 643 */ 644 if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY)) 645 mode |= BLK_OPEN_WRITE_IOCTL; 646 647 return mode; 648 } 649 650 static int blkdev_open(struct inode *inode, struct file *filp) 651 { 652 struct block_device *bdev; 653 blk_mode_t mode; 654 int ret; 655 656 mode = file_to_blk_mode(filp); 657 /* Use the file as the holder. */ 658 if (mode & BLK_OPEN_EXCL) 659 filp->private_data = filp; 660 ret = bdev_permission(inode->i_rdev, mode, filp->private_data); 661 if (ret) 662 return ret; 663 664 bdev = blkdev_get_no_open(inode->i_rdev, true); 665 if (!bdev) 666 return -ENXIO; 667 668 if (bdev_can_atomic_write(bdev)) 669 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE; 670 if (blk_get_integrity(bdev->bd_disk)) 671 filp->f_mode |= FMODE_HAS_METADATA; 672 673 ret = bdev_open(bdev, mode, filp->private_data, NULL, filp); 674 if (ret) 675 blkdev_put_no_open(bdev); 676 return ret; 677 } 678 679 static int blkdev_release(struct inode *inode, struct file *filp) 680 { 681 bdev_release(filp); 682 return 0; 683 } 684 685 static ssize_t 686 blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from) 687 { 688 size_t count = iov_iter_count(from); 689 ssize_t written; 690 691 written = kiocb_invalidate_pages(iocb, count); 692 if (written) { 693 if (written == -EBUSY) 694 return 0; 695 return written; 696 } 697 698 written = blkdev_direct_IO(iocb, from); 699 if (written > 0) { 700 kiocb_invalidate_post_direct_write(iocb, count); 701 iocb->ki_pos += written; 702 count -= written; 703 } 704 if (written != -EIOCBQUEUED) 705 iov_iter_revert(from, count - iov_iter_count(from)); 706 return written; 707 } 708 709 static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from) 710 { 711 return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL, 712 NULL); 713 } 714 715 /* 716 * Write data to the block device. Only intended for the block device itself 717 * and the raw driver which basically is a fake block device. 718 * 719 * Does not take i_mutex for the write and thus is not for general purpose 720 * use. 721 */ 722 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) 723 { 724 struct file *file = iocb->ki_filp; 725 struct inode *bd_inode = bdev_file_inode(file); 726 struct block_device *bdev = I_BDEV(bd_inode); 727 bool atomic = iocb->ki_flags & IOCB_ATOMIC; 728 loff_t size = bdev_nr_bytes(bdev); 729 size_t shorted = 0; 730 ssize_t ret; 731 732 if (bdev_read_only(bdev)) 733 return -EPERM; 734 735 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) 736 return -ETXTBSY; 737 738 if (!iov_iter_count(from)) 739 return 0; 740 741 if (iocb->ki_pos >= size) 742 return -ENOSPC; 743 744 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) 745 return -EOPNOTSUPP; 746 747 if (atomic) { 748 ret = generic_atomic_write_valid(iocb, from); 749 if (ret) 750 return ret; 751 } 752 753 size -= iocb->ki_pos; 754 if (iov_iter_count(from) > size) { 755 if (atomic) 756 return -EINVAL; 757 shorted = iov_iter_count(from) - size; 758 iov_iter_truncate(from, size); 759 } 760 761 ret = file_update_time(file); 762 if (ret) 763 return ret; 764 765 if (iocb->ki_flags & IOCB_DIRECT) { 766 ret = blkdev_direct_write(iocb, from); 767 if (ret >= 0 && iov_iter_count(from)) 768 ret = direct_write_fallback(iocb, from, ret, 769 blkdev_buffered_write(iocb, from)); 770 } else { 771 /* 772 * Take i_rwsem and invalidate_lock to avoid racing with 773 * set_blocksize changing i_blkbits/folio order and punching 774 * out the pagecache. 775 */ 776 inode_lock_shared(bd_inode); 777 ret = blkdev_buffered_write(iocb, from); 778 inode_unlock_shared(bd_inode); 779 } 780 781 if (ret > 0) 782 ret = generic_write_sync(iocb, ret); 783 iov_iter_reexpand(from, iov_iter_count(from) + shorted); 784 return ret; 785 } 786 787 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) 788 { 789 struct inode *bd_inode = bdev_file_inode(iocb->ki_filp); 790 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 791 loff_t size = bdev_nr_bytes(bdev); 792 loff_t pos = iocb->ki_pos; 793 size_t shorted = 0; 794 ssize_t ret = 0; 795 size_t count; 796 797 if (unlikely(pos + iov_iter_count(to) > size)) { 798 if (pos >= size) 799 return 0; 800 size -= pos; 801 shorted = iov_iter_count(to) - size; 802 iov_iter_truncate(to, size); 803 } 804 805 count = iov_iter_count(to); 806 if (!count) 807 goto reexpand; /* skip atime */ 808 809 if (iocb->ki_flags & IOCB_DIRECT) { 810 ret = kiocb_write_and_wait(iocb, count); 811 if (ret < 0) 812 goto reexpand; 813 file_accessed(iocb->ki_filp); 814 815 ret = blkdev_direct_IO(iocb, to); 816 if (ret > 0) { 817 iocb->ki_pos += ret; 818 count -= ret; 819 } 820 if (ret != -EIOCBQUEUED) 821 iov_iter_revert(to, count - iov_iter_count(to)); 822 if (ret < 0 || !count) 823 goto reexpand; 824 } 825 826 /* 827 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize 828 * changing i_blkbits/folio order and punching out the pagecache. 829 */ 830 inode_lock_shared(bd_inode); 831 ret = filemap_read(iocb, to, ret); 832 inode_unlock_shared(bd_inode); 833 834 reexpand: 835 if (unlikely(shorted)) 836 iov_iter_reexpand(to, iov_iter_count(to) + shorted); 837 return ret; 838 } 839 840 #define BLKDEV_FALLOC_FL_SUPPORTED \ 841 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 842 FALLOC_FL_ZERO_RANGE | FALLOC_FL_WRITE_ZEROES) 843 844 static long blkdev_fallocate(struct file *file, int mode, loff_t start, 845 loff_t len) 846 { 847 struct inode *inode = bdev_file_inode(file); 848 struct block_device *bdev = I_BDEV(inode); 849 loff_t end = start + len - 1; 850 loff_t isize; 851 unsigned int flags; 852 int error; 853 854 /* Fail if we don't recognize the flags. */ 855 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) 856 return -EOPNOTSUPP; 857 /* 858 * Don't allow writing zeroes if the device does not enable the 859 * unmap write zeroes operation. 860 */ 861 if ((mode & FALLOC_FL_WRITE_ZEROES) && 862 !bdev_write_zeroes_unmap_sectors(bdev)) 863 return -EOPNOTSUPP; 864 865 /* Don't go off the end of the device. */ 866 isize = bdev_nr_bytes(bdev); 867 if (start >= isize) 868 return -EINVAL; 869 if (end >= isize) { 870 if (mode & FALLOC_FL_KEEP_SIZE) { 871 len = isize - start; 872 end = start + len - 1; 873 } else 874 return -EINVAL; 875 } 876 877 /* 878 * Don't allow IO that isn't aligned to logical block size. 879 */ 880 if ((start | len) & (bdev_logical_block_size(bdev) - 1)) 881 return -EINVAL; 882 883 inode_lock(inode); 884 filemap_invalidate_lock(inode->i_mapping); 885 886 switch (mode) { 887 case FALLOC_FL_ZERO_RANGE: 888 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: 889 flags = BLKDEV_ZERO_NOUNMAP; 890 break; 891 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: 892 flags = BLKDEV_ZERO_NOFALLBACK; 893 break; 894 case FALLOC_FL_WRITE_ZEROES: 895 flags = 0; 896 break; 897 default: 898 error = -EOPNOTSUPP; 899 goto fail; 900 } 901 902 /* 903 * Invalidate the page cache, including dirty pages, for valid 904 * de-allocate mode calls to fallocate(). 905 */ 906 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end); 907 if (error) 908 goto fail; 909 910 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 911 len >> SECTOR_SHIFT, GFP_KERNEL, flags); 912 fail: 913 filemap_invalidate_unlock(inode->i_mapping); 914 inode_unlock(inode); 915 return error; 916 } 917 918 static int blkdev_mmap_prepare(struct vm_area_desc *desc) 919 { 920 struct file *file = desc->file; 921 922 if (bdev_read_only(I_BDEV(bdev_file_inode(file)))) 923 return generic_file_readonly_mmap_prepare(desc); 924 925 return generic_file_mmap_prepare(desc); 926 } 927 928 const struct file_operations def_blk_fops = { 929 .open = blkdev_open, 930 .release = blkdev_release, 931 .llseek = blkdev_llseek, 932 .read_iter = blkdev_read_iter, 933 .write_iter = blkdev_write_iter, 934 .iopoll = iocb_bio_iopoll, 935 .mmap_prepare = blkdev_mmap_prepare, 936 .fsync = blkdev_fsync, 937 .unlocked_ioctl = blkdev_ioctl, 938 #ifdef CONFIG_COMPAT 939 .compat_ioctl = compat_blkdev_ioctl, 940 #endif 941 .splice_read = filemap_splice_read, 942 .splice_write = iter_file_splice_write, 943 .fallocate = blkdev_fallocate, 944 .uring_cmd = blkdev_uring_cmd, 945 .fop_flags = FOP_BUFFER_RASYNC, 946 }; 947 948 static __init int blkdev_init(void) 949 { 950 return bioset_init(&blkdev_dio_pool, 4, 951 offsetof(struct blkdev_dio, bio), 952 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); 953 } 954 module_init(blkdev_init); 955