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_endio_status(bio, BLK_STS_IOERR); 222 break; 223 } 224 if (iocb->ki_flags & IOCB_NOWAIT) { 225 /* 226 * This is nonblocking IO, and we need to allocate 227 * another bio if we have data left to map. As we 228 * cannot guarantee that one of the sub bios will not 229 * fail getting issued FOR NOWAIT and as error results 230 * are coalesced across all of them, be safe and ask for 231 * a retry of this from blocking context. 232 */ 233 if (unlikely(iov_iter_count(iter))) { 234 ret = -EAGAIN; 235 goto fail; 236 } 237 bio->bi_opf |= REQ_NOWAIT; 238 } 239 if (iocb->ki_flags & IOCB_HAS_METADATA) { 240 ret = bio_integrity_map_iter(bio, iocb->private); 241 if (unlikely(ret)) 242 goto fail; 243 } 244 245 if (is_read) { 246 if (dio->flags & DIO_SHOULD_DIRTY) 247 bio_set_pages_dirty(bio); 248 } else { 249 task_io_account_write(bio->bi_iter.bi_size); 250 } 251 dio->size += bio->bi_iter.bi_size; 252 pos += bio->bi_iter.bi_size; 253 254 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); 255 if (!nr_pages) { 256 submit_bio(bio); 257 break; 258 } 259 atomic_inc(&dio->ref); 260 submit_bio(bio); 261 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL); 262 } 263 264 blk_finish_plug(&plug); 265 266 if (!is_sync) 267 return -EIOCBQUEUED; 268 269 for (;;) { 270 set_current_state(TASK_UNINTERRUPTIBLE); 271 if (!READ_ONCE(dio->waiter)) 272 break; 273 blk_io_schedule(); 274 } 275 __set_current_state(TASK_RUNNING); 276 277 if (!ret) 278 ret = blk_status_to_errno(dio->bio.bi_status); 279 if (likely(!ret)) 280 ret = dio->size; 281 282 bio_put(&dio->bio); 283 return ret; 284 fail: 285 bio_release_pages(bio, false); 286 bio_clear_flag(bio, BIO_REFFED); 287 bio_put(bio); 288 blk_finish_plug(&plug); 289 return ret; 290 } 291 292 static void blkdev_bio_end_io_async(struct bio *bio) 293 { 294 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); 295 struct kiocb *iocb = dio->iocb; 296 ssize_t ret; 297 298 WRITE_ONCE(iocb->private, NULL); 299 300 if (likely(!bio->bi_status)) { 301 ret = dio->size; 302 iocb->ki_pos += ret; 303 } else { 304 ret = blk_status_to_errno(bio->bi_status); 305 } 306 307 if (bio_integrity(bio)) 308 bio_integrity_unmap_user(bio); 309 310 iocb->ki_complete(iocb, ret); 311 312 if (dio->flags & DIO_SHOULD_DIRTY) { 313 bio_check_pages_dirty(bio); 314 } else { 315 bio_release_pages(bio, false); 316 bio_put(bio); 317 } 318 } 319 320 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, 321 struct iov_iter *iter, 322 struct block_device *bdev, 323 unsigned int nr_pages) 324 { 325 bool is_read = iov_iter_rw(iter) == READ; 326 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 327 struct blkdev_dio *dio; 328 struct bio *bio; 329 loff_t pos = iocb->ki_pos; 330 int ret = 0; 331 332 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 333 &blkdev_dio_pool); 334 dio = container_of(bio, struct blkdev_dio, bio); 335 dio->flags = 0; 336 dio->iocb = iocb; 337 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 338 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 339 bio->bi_write_stream = iocb->ki_write_stream; 340 bio->bi_end_io = blkdev_bio_end_io_async; 341 bio->bi_ioprio = iocb->ki_ioprio; 342 343 if (iov_iter_is_bvec(iter)) { 344 /* 345 * Users don't rely on the iterator being in any particular 346 * state for async I/O returning -EIOCBQUEUED, hence we can 347 * avoid expensive iov_iter_advance(). Bypass 348 * bio_iov_iter_get_pages() and set the bvec directly. 349 */ 350 bio_iov_bvec_set(bio, iter); 351 } else { 352 ret = blkdev_iov_iter_get_pages(bio, iter, bdev); 353 if (unlikely(ret)) 354 goto out_bio_put; 355 } 356 dio->size = bio->bi_iter.bi_size; 357 358 if (is_read) { 359 if (user_backed_iter(iter)) { 360 dio->flags |= DIO_SHOULD_DIRTY; 361 bio_set_pages_dirty(bio); 362 } 363 } else { 364 task_io_account_write(bio->bi_iter.bi_size); 365 } 366 367 if (iocb->ki_flags & IOCB_HAS_METADATA) { 368 ret = bio_integrity_map_iter(bio, iocb->private); 369 WRITE_ONCE(iocb->private, NULL); 370 if (unlikely(ret)) 371 goto out_bio_put; 372 } 373 374 if (iocb->ki_flags & IOCB_ATOMIC) 375 bio->bi_opf |= REQ_ATOMIC; 376 377 if (iocb->ki_flags & IOCB_NOWAIT) 378 bio->bi_opf |= REQ_NOWAIT; 379 380 if (iocb->ki_flags & IOCB_HIPRI) { 381 bio->bi_opf |= REQ_POLLED; 382 submit_bio(bio); 383 WRITE_ONCE(iocb->private, bio); 384 } else { 385 submit_bio(bio); 386 } 387 return -EIOCBQUEUED; 388 389 out_bio_put: 390 bio_put(bio); 391 return ret; 392 } 393 394 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 395 { 396 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 397 unsigned int nr_pages; 398 399 if (!iov_iter_count(iter)) 400 return 0; 401 402 if (blkdev_dio_invalid(bdev, iocb, iter)) 403 return -EINVAL; 404 405 if (iov_iter_rw(iter) == WRITE) { 406 u16 max_write_streams = bdev_max_write_streams(bdev); 407 408 if (iocb->ki_write_stream) { 409 if (iocb->ki_write_stream > max_write_streams) 410 return -EINVAL; 411 } else if (max_write_streams) { 412 enum rw_hint write_hint = 413 file_inode(iocb->ki_filp)->i_write_hint; 414 415 /* 416 * Just use the write hint as write stream for block 417 * device writes. This assumes no file system is 418 * mounted that would use the streams differently. 419 */ 420 if (write_hint <= max_write_streams) 421 iocb->ki_write_stream = write_hint; 422 } 423 } 424 425 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); 426 if (likely(nr_pages <= BIO_MAX_VECS && 427 !(iocb->ki_flags & IOCB_HAS_METADATA))) { 428 if (is_sync_kiocb(iocb)) 429 return __blkdev_direct_IO_simple(iocb, iter, bdev, 430 nr_pages); 431 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages); 432 } else if (iocb->ki_flags & IOCB_ATOMIC) { 433 return -EINVAL; 434 } 435 return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages)); 436 } 437 438 static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 439 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 440 { 441 struct block_device *bdev = I_BDEV(inode); 442 loff_t isize = i_size_read(inode); 443 444 if (offset >= isize) 445 return -EIO; 446 447 iomap->bdev = bdev; 448 iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev)); 449 iomap->type = IOMAP_MAPPED; 450 iomap->addr = iomap->offset; 451 iomap->length = isize - iomap->offset; 452 iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */ 453 return 0; 454 } 455 456 static DEFINE_IOMAP_ITER_NEXT(blkdev_iomap_next, blkdev_iomap_begin); 457 458 static const struct iomap_ops blkdev_iomap_ops = { 459 .iomap_next = blkdev_iomap_next, 460 }; 461 462 #ifdef CONFIG_BUFFER_HEAD 463 static int blkdev_get_block(struct inode *inode, sector_t iblock, 464 struct buffer_head *bh, int create) 465 { 466 bh->b_bdev = I_BDEV(inode); 467 bh->b_blocknr = iblock; 468 set_buffer_mapped(bh); 469 return 0; 470 } 471 472 /* 473 * We cannot call mpage_writepages() as it does not take the buffer lock. 474 * We must use block_write_full_folio() directly which holds the buffer 475 * lock. The buffer lock provides the synchronisation with writeback 476 * that filesystems rely on when they use the blockdev's mapping. 477 */ 478 static int blkdev_writepages(struct address_space *mapping, 479 struct writeback_control *wbc) 480 { 481 struct folio *folio = NULL; 482 struct blk_plug plug; 483 int err; 484 485 blk_start_plug(&plug); 486 while ((folio = writeback_iter(mapping, wbc, folio, &err))) 487 err = block_write_full_folio(folio, wbc, blkdev_get_block); 488 blk_finish_plug(&plug); 489 490 return err; 491 } 492 493 static int blkdev_read_folio(struct file *file, struct folio *folio) 494 { 495 return block_read_full_folio(folio, blkdev_get_block); 496 } 497 498 static void blkdev_readahead(struct readahead_control *rac) 499 { 500 mpage_readahead(rac, blkdev_get_block); 501 } 502 503 const struct address_space_operations def_blk_aops = { 504 .dirty_folio = block_dirty_folio, 505 .invalidate_folio = block_invalidate_folio, 506 .read_folio = blkdev_read_folio, 507 .readahead = blkdev_readahead, 508 .writepages = blkdev_writepages, 509 .migrate_folio = buffer_migrate_folio_norefs, 510 .is_dirty_writeback = buffer_check_dirty_writeback, 511 }; 512 #else /* CONFIG_BUFFER_HEAD */ 513 static int blkdev_read_folio(struct file *file, struct folio *folio) 514 { 515 iomap_bio_read_folio(folio, &blkdev_iomap_ops); 516 return 0; 517 } 518 519 static void blkdev_readahead(struct readahead_control *rac) 520 { 521 iomap_bio_readahead(rac, &blkdev_iomap_ops); 522 } 523 524 static ssize_t blkdev_writeback_range(struct iomap_writepage_ctx *wpc, 525 struct folio *folio, u64 offset, unsigned int len, u64 end_pos) 526 { 527 loff_t isize = i_size_read(wpc->inode); 528 529 if (WARN_ON_ONCE(offset >= isize)) 530 return -EIO; 531 532 if (offset < wpc->iomap.offset || 533 offset >= wpc->iomap.offset + wpc->iomap.length) { 534 int error; 535 536 error = blkdev_iomap_begin(wpc->inode, offset, isize - offset, 537 IOMAP_WRITE, &wpc->iomap, NULL); 538 if (error) 539 return error; 540 } 541 542 return iomap_add_to_ioend(wpc, folio, offset, end_pos, len); 543 } 544 545 static const struct iomap_writeback_ops blkdev_writeback_ops = { 546 .writeback_range = blkdev_writeback_range, 547 .writeback_submit = iomap_ioend_writeback_submit, 548 }; 549 550 static int blkdev_writepages(struct address_space *mapping, 551 struct writeback_control *wbc) 552 { 553 struct iomap_writepage_ctx wpc = { 554 .inode = mapping->host, 555 .wbc = wbc, 556 .ops = &blkdev_writeback_ops 557 }; 558 559 return iomap_writepages(&wpc); 560 } 561 562 const struct address_space_operations def_blk_aops = { 563 .dirty_folio = filemap_dirty_folio, 564 .release_folio = iomap_release_folio, 565 .invalidate_folio = iomap_invalidate_folio, 566 .read_folio = blkdev_read_folio, 567 .readahead = blkdev_readahead, 568 .writepages = blkdev_writepages, 569 .is_partially_uptodate = iomap_is_partially_uptodate, 570 .error_remove_folio = generic_error_remove_folio, 571 .migrate_folio = filemap_migrate_folio, 572 }; 573 #endif /* CONFIG_BUFFER_HEAD */ 574 575 /* 576 * for a block special file file_inode(file)->i_size is zero 577 * so we compute the size by hand (just as in block_read/write above) 578 */ 579 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) 580 { 581 struct inode *bd_inode = bdev_file_inode(file); 582 loff_t retval; 583 584 inode_lock(bd_inode); 585 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); 586 inode_unlock(bd_inode); 587 return retval; 588 } 589 590 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, 591 int datasync) 592 { 593 struct block_device *bdev = I_BDEV(filp->f_mapping->host); 594 int error; 595 596 error = file_write_and_wait_range(filp, start, end); 597 if (error) 598 return error; 599 600 /* 601 * There is no need to serialise calls to blkdev_issue_flush with 602 * i_mutex and doing so causes performance issues with concurrent 603 * O_SYNC writers to a block device. 604 */ 605 error = blkdev_issue_flush(bdev); 606 if (error == -EOPNOTSUPP) 607 error = 0; 608 609 return error; 610 } 611 612 /** 613 * file_to_blk_mode - get block open flags from file flags 614 * @file: file whose open flags should be converted 615 * 616 * Look at file open flags and generate corresponding block open flags from 617 * them. The function works both for file just being open (e.g. during ->open 618 * callback) and for file that is already open. This is actually non-trivial 619 * (see comment in the function). 620 */ 621 blk_mode_t file_to_blk_mode(struct file *file) 622 { 623 blk_mode_t mode = 0; 624 625 if (file->f_mode & FMODE_READ) 626 mode |= BLK_OPEN_READ; 627 if (file->f_mode & FMODE_WRITE) 628 mode |= BLK_OPEN_WRITE; 629 /* 630 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data 631 * to determine whether the open was exclusive for already open files. 632 */ 633 if (file->private_data) 634 mode |= BLK_OPEN_EXCL; 635 else if (file->f_flags & O_EXCL) 636 mode |= BLK_OPEN_EXCL; 637 if (file->f_flags & O_NDELAY) 638 mode |= BLK_OPEN_NDELAY; 639 640 /* 641 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy 642 * driver has historically allowed ioctls as if the file was opened for 643 * writing, but does not allow and actual reads or writes. 644 */ 645 if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY)) 646 mode |= BLK_OPEN_WRITE_IOCTL; 647 648 return mode; 649 } 650 651 static int blkdev_open(struct inode *inode, struct file *filp) 652 { 653 struct block_device *bdev; 654 blk_mode_t mode; 655 int ret; 656 657 mode = file_to_blk_mode(filp); 658 /* Use the file as the holder. */ 659 if (mode & BLK_OPEN_EXCL) 660 filp->private_data = filp; 661 ret = bdev_permission(inode->i_rdev, mode, filp->private_data); 662 if (ret) 663 return ret; 664 665 bdev = blkdev_get_no_open(inode->i_rdev, true); 666 if (!bdev) 667 return -ENXIO; 668 669 if (bdev_can_atomic_write(bdev)) 670 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE; 671 if (blk_get_integrity(bdev->bd_disk)) 672 filp->f_mode |= FMODE_HAS_METADATA; 673 674 ret = bdev_open(bdev, mode, filp->private_data, NULL, filp); 675 if (ret) 676 blkdev_put_no_open(bdev); 677 return ret; 678 } 679 680 static int blkdev_release(struct inode *inode, struct file *filp) 681 { 682 bdev_release(filp); 683 return 0; 684 } 685 686 static ssize_t 687 blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from) 688 { 689 size_t count = iov_iter_count(from); 690 ssize_t written; 691 692 written = kiocb_invalidate_pages(iocb, count); 693 if (written) { 694 if (written == -EBUSY) 695 return 0; 696 return written; 697 } 698 699 written = blkdev_direct_IO(iocb, from); 700 if (written > 0) { 701 kiocb_invalidate_post_direct_write(iocb, count); 702 iocb->ki_pos += written; 703 count -= written; 704 } 705 if (written != -EIOCBQUEUED) 706 iov_iter_revert(from, count - iov_iter_count(from)); 707 return written; 708 } 709 710 static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from) 711 { 712 return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL, 713 NULL); 714 } 715 716 /* 717 * Write data to the block device. Only intended for the block device itself 718 * and the raw driver which basically is a fake block device. 719 * 720 * Does not take i_mutex for the write and thus is not for general purpose 721 * use. 722 */ 723 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) 724 { 725 struct file *file = iocb->ki_filp; 726 struct inode *bd_inode = bdev_file_inode(file); 727 struct block_device *bdev = I_BDEV(bd_inode); 728 bool atomic = iocb->ki_flags & IOCB_ATOMIC; 729 loff_t size = bdev_nr_bytes(bdev); 730 size_t shorted = 0; 731 ssize_t ret; 732 733 if (bdev_read_only(bdev)) 734 return -EPERM; 735 736 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) 737 return -ETXTBSY; 738 739 if (!iov_iter_count(from)) 740 return 0; 741 742 if (iocb->ki_pos >= size) 743 return -ENOSPC; 744 745 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) 746 return -EOPNOTSUPP; 747 748 if (atomic) { 749 ret = generic_atomic_write_valid(iocb, from); 750 if (ret) 751 return ret; 752 } 753 754 size -= iocb->ki_pos; 755 if (iov_iter_count(from) > size) { 756 if (atomic) 757 return -EINVAL; 758 shorted = iov_iter_count(from) - size; 759 iov_iter_truncate(from, size); 760 } 761 762 ret = file_update_time(file); 763 if (ret) 764 return ret; 765 766 if (iocb->ki_flags & IOCB_DIRECT) { 767 ret = blkdev_direct_write(iocb, from); 768 if (ret >= 0 && iov_iter_count(from)) 769 ret = direct_write_fallback(iocb, from, ret, 770 blkdev_buffered_write(iocb, from)); 771 } else { 772 /* 773 * Take i_rwsem and invalidate_lock to avoid racing with 774 * set_blocksize changing i_blkbits/folio order and punching 775 * out the pagecache. 776 */ 777 inode_lock_shared(bd_inode); 778 ret = blkdev_buffered_write(iocb, from); 779 inode_unlock_shared(bd_inode); 780 } 781 782 if (ret > 0) 783 ret = generic_write_sync(iocb, ret); 784 iov_iter_reexpand(from, iov_iter_count(from) + shorted); 785 return ret; 786 } 787 788 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) 789 { 790 struct inode *bd_inode = bdev_file_inode(iocb->ki_filp); 791 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 792 loff_t size = bdev_nr_bytes(bdev); 793 loff_t pos = iocb->ki_pos; 794 size_t shorted = 0; 795 ssize_t ret = 0; 796 size_t count; 797 798 if (unlikely(pos + iov_iter_count(to) > size)) { 799 if (pos >= size) 800 return 0; 801 size -= pos; 802 shorted = iov_iter_count(to) - size; 803 iov_iter_truncate(to, size); 804 } 805 806 count = iov_iter_count(to); 807 if (!count) 808 goto reexpand; /* skip atime */ 809 810 if (iocb->ki_flags & IOCB_DIRECT) { 811 ret = kiocb_write_and_wait(iocb, count); 812 if (ret < 0) 813 goto reexpand; 814 file_accessed(iocb->ki_filp); 815 816 ret = blkdev_direct_IO(iocb, to); 817 if (ret > 0) { 818 iocb->ki_pos += ret; 819 count -= ret; 820 } 821 if (ret != -EIOCBQUEUED) 822 iov_iter_revert(to, count - iov_iter_count(to)); 823 if (ret < 0 || !count) 824 goto reexpand; 825 } 826 827 /* 828 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize 829 * changing i_blkbits/folio order and punching out the pagecache. 830 */ 831 inode_lock_shared(bd_inode); 832 ret = filemap_read(iocb, to, ret); 833 inode_unlock_shared(bd_inode); 834 835 reexpand: 836 if (unlikely(shorted)) 837 iov_iter_reexpand(to, iov_iter_count(to) + shorted); 838 return ret; 839 } 840 841 #define BLKDEV_FALLOC_FL_SUPPORTED \ 842 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 843 FALLOC_FL_ZERO_RANGE | FALLOC_FL_WRITE_ZEROES) 844 845 static long blkdev_fallocate(struct file *file, int mode, loff_t start, 846 loff_t len) 847 { 848 struct inode *inode = bdev_file_inode(file); 849 struct block_device *bdev = I_BDEV(inode); 850 loff_t end = start + len - 1; 851 loff_t isize; 852 unsigned int flags; 853 int error; 854 855 /* Fail if we don't recognize the flags. */ 856 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) 857 return -EOPNOTSUPP; 858 /* 859 * Don't allow writing zeroes if the device does not enable the 860 * unmap write zeroes operation. 861 */ 862 if ((mode & FALLOC_FL_WRITE_ZEROES) && 863 !bdev_write_zeroes_unmap_sectors(bdev)) 864 return -EOPNOTSUPP; 865 866 /* Don't go off the end of the device. */ 867 isize = bdev_nr_bytes(bdev); 868 if (start >= isize) 869 return -EINVAL; 870 if (end >= isize) { 871 if (mode & FALLOC_FL_KEEP_SIZE) { 872 len = isize - start; 873 end = start + len - 1; 874 } else 875 return -EINVAL; 876 } 877 878 /* 879 * Don't allow IO that isn't aligned to logical block size. 880 */ 881 if ((start | len) & (bdev_logical_block_size(bdev) - 1)) 882 return -EINVAL; 883 884 inode_lock(inode); 885 filemap_invalidate_lock(inode->i_mapping); 886 887 switch (mode) { 888 case FALLOC_FL_ZERO_RANGE: 889 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: 890 flags = BLKDEV_ZERO_NOUNMAP; 891 break; 892 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: 893 flags = BLKDEV_ZERO_NOFALLBACK; 894 break; 895 case FALLOC_FL_WRITE_ZEROES: 896 flags = 0; 897 break; 898 default: 899 error = -EOPNOTSUPP; 900 goto fail; 901 } 902 903 /* 904 * Invalidate the page cache, including dirty pages, for valid 905 * de-allocate mode calls to fallocate(). 906 */ 907 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end); 908 if (error) 909 goto fail; 910 911 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 912 len >> SECTOR_SHIFT, GFP_KERNEL, flags); 913 fail: 914 filemap_invalidate_unlock(inode->i_mapping); 915 inode_unlock(inode); 916 return error; 917 } 918 919 static int blkdev_mmap_prepare(struct vm_area_desc *desc) 920 { 921 struct file *file = desc->file; 922 923 if (bdev_read_only(I_BDEV(bdev_file_inode(file)))) 924 return generic_file_readonly_mmap_prepare(desc); 925 926 return generic_file_mmap_prepare(desc); 927 } 928 929 const struct file_operations def_blk_fops = { 930 .open = blkdev_open, 931 .release = blkdev_release, 932 .llseek = blkdev_llseek, 933 .read_iter = blkdev_read_iter, 934 .write_iter = blkdev_write_iter, 935 .iopoll = iocb_bio_iopoll, 936 .mmap_prepare = blkdev_mmap_prepare, 937 .fsync = blkdev_fsync, 938 .unlocked_ioctl = blkdev_ioctl, 939 #ifdef CONFIG_COMPAT 940 .compat_ioctl = compat_blkdev_ioctl, 941 #endif 942 .splice_read = filemap_splice_read, 943 .splice_write = iter_file_splice_write, 944 .fallocate = blkdev_fallocate, 945 .uring_cmd = blkdev_uring_cmd, 946 .fop_flags = FOP_BUFFER_RASYNC, 947 }; 948 949 static __init int blkdev_init(void) 950 { 951 return bioset_init(&blkdev_dio_pool, 4, 952 offsetof(struct blkdev_dio, bio), 953 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); 954 } 955 module_init(blkdev_init); 956