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, bdev_dma_alignment(bdev), 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, errno_to_blk_status(ret)); 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 bio_endio_status(bio, errno_to_blk_status(ret)); 243 break; 244 } 245 } 246 247 if (is_read) { 248 if (dio->flags & DIO_SHOULD_DIRTY) 249 bio_set_pages_dirty(bio); 250 } else { 251 task_io_account_write(bio->bi_iter.bi_size); 252 } 253 dio->size += bio->bi_iter.bi_size; 254 pos += bio->bi_iter.bi_size; 255 256 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); 257 if (!nr_pages) { 258 submit_bio(bio); 259 break; 260 } 261 atomic_inc(&dio->ref); 262 submit_bio(bio); 263 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL); 264 } 265 266 blk_finish_plug(&plug); 267 268 if (!is_sync) 269 return -EIOCBQUEUED; 270 271 for (;;) { 272 set_current_state(TASK_UNINTERRUPTIBLE); 273 if (!READ_ONCE(dio->waiter)) 274 break; 275 blk_io_schedule(); 276 } 277 __set_current_state(TASK_RUNNING); 278 279 if (!ret) 280 ret = blk_status_to_errno(dio->bio.bi_status); 281 if (likely(!ret)) 282 ret = dio->size; 283 284 bio_put(&dio->bio); 285 return ret; 286 fail: 287 bio_release_pages(bio, false); 288 bio_clear_flag(bio, BIO_REFFED); 289 bio_put(bio); 290 blk_finish_plug(&plug); 291 return ret; 292 } 293 294 static void blkdev_bio_end_io_async(struct bio *bio) 295 { 296 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); 297 struct kiocb *iocb = dio->iocb; 298 ssize_t ret; 299 300 WRITE_ONCE(iocb->private, NULL); 301 302 if (likely(!bio->bi_status)) { 303 ret = dio->size; 304 iocb->ki_pos += ret; 305 } else { 306 ret = blk_status_to_errno(bio->bi_status); 307 } 308 309 if (bio_integrity(bio)) 310 bio_integrity_unmap_user(bio); 311 312 iocb->ki_complete(iocb, ret); 313 314 if (dio->flags & DIO_SHOULD_DIRTY) { 315 bio_check_pages_dirty(bio); 316 } else { 317 bio_release_pages(bio, false); 318 bio_put(bio); 319 } 320 } 321 322 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, 323 struct iov_iter *iter, 324 struct block_device *bdev, 325 unsigned int nr_pages) 326 { 327 bool is_read = iov_iter_rw(iter) == READ; 328 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 329 struct blkdev_dio *dio; 330 struct bio *bio; 331 loff_t pos = iocb->ki_pos; 332 int ret = 0; 333 334 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 335 &blkdev_dio_pool); 336 dio = container_of(bio, struct blkdev_dio, bio); 337 dio->flags = 0; 338 dio->iocb = iocb; 339 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 340 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 341 bio->bi_write_stream = iocb->ki_write_stream; 342 bio->bi_end_io = blkdev_bio_end_io_async; 343 bio->bi_ioprio = iocb->ki_ioprio; 344 345 /* 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 if (!bio_iov_iter_set(bio, iter)) { 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 /* HIPRI needs private as bio; HAS_METADATA keeps it as uio_meta */ 406 if ((iocb->ki_flags & IOCB_HIPRI) && 407 (iocb->ki_flags & IOCB_HAS_METADATA)) 408 return -EOPNOTSUPP; 409 410 if (iov_iter_rw(iter) == WRITE) { 411 u16 max_write_streams = bdev_max_write_streams(bdev); 412 413 if (iocb->ki_write_stream) { 414 if (iocb->ki_write_stream > max_write_streams) 415 return -EINVAL; 416 } else if (max_write_streams) { 417 enum rw_hint write_hint = 418 file_inode(iocb->ki_filp)->i_write_hint; 419 420 /* 421 * Just use the write hint as write stream for block 422 * device writes. This assumes no file system is 423 * mounted that would use the streams differently. 424 */ 425 if (write_hint <= max_write_streams) 426 iocb->ki_write_stream = write_hint; 427 } 428 } 429 430 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); 431 if (likely(nr_pages <= BIO_MAX_VECS && 432 !(iocb->ki_flags & IOCB_HAS_METADATA))) { 433 if (is_sync_kiocb(iocb)) 434 return __blkdev_direct_IO_simple(iocb, iter, bdev, 435 nr_pages); 436 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages); 437 } else if (iocb->ki_flags & IOCB_ATOMIC) { 438 return -EINVAL; 439 } 440 return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages)); 441 } 442 443 static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 444 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 445 { 446 struct block_device *bdev = I_BDEV(inode); 447 loff_t isize = i_size_read(inode); 448 449 if (offset >= isize) 450 return -EIO; 451 452 iomap->bdev = bdev; 453 iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev)); 454 iomap->type = IOMAP_MAPPED; 455 iomap->addr = iomap->offset; 456 iomap->length = isize - iomap->offset; 457 iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */ 458 return 0; 459 } 460 461 static DEFINE_IOMAP_ITER_NEXT(blkdev_iomap_next, blkdev_iomap_begin); 462 463 static const struct iomap_ops blkdev_iomap_ops = { 464 .iomap_next = blkdev_iomap_next, 465 }; 466 467 #ifdef CONFIG_BUFFER_HEAD 468 static int blkdev_get_block(struct inode *inode, sector_t iblock, 469 struct buffer_head *bh, int create) 470 { 471 bh->b_bdev = I_BDEV(inode); 472 bh->b_blocknr = iblock; 473 set_buffer_mapped(bh); 474 return 0; 475 } 476 477 /* 478 * We cannot call mpage_writepages() as it does not take the buffer lock. 479 * We must use block_write_full_folio() directly which holds the buffer 480 * lock. The buffer lock provides the synchronisation with writeback 481 * that filesystems rely on when they use the blockdev's mapping. 482 */ 483 static int blkdev_writepages(struct address_space *mapping, 484 struct writeback_control *wbc) 485 { 486 struct folio *folio = NULL; 487 struct blk_plug plug; 488 int err; 489 490 blk_start_plug(&plug); 491 while ((folio = writeback_iter(mapping, wbc, folio, &err))) 492 err = block_write_full_folio(folio, wbc, blkdev_get_block); 493 blk_finish_plug(&plug); 494 495 return err; 496 } 497 498 static int blkdev_read_folio(struct file *file, struct folio *folio) 499 { 500 return block_read_full_folio(folio, blkdev_get_block); 501 } 502 503 static void blkdev_readahead(struct readahead_control *rac) 504 { 505 mpage_readahead(rac, blkdev_get_block); 506 } 507 508 const struct address_space_operations def_blk_aops = { 509 .dirty_folio = block_dirty_folio, 510 .invalidate_folio = block_invalidate_folio, 511 .read_folio = blkdev_read_folio, 512 .readahead = blkdev_readahead, 513 .writepages = blkdev_writepages, 514 .migrate_folio = buffer_migrate_folio_norefs, 515 .is_dirty_writeback = buffer_check_dirty_writeback, 516 }; 517 #else /* CONFIG_BUFFER_HEAD */ 518 static int blkdev_read_folio(struct file *file, struct folio *folio) 519 { 520 iomap_bio_read_folio(folio, &blkdev_iomap_ops); 521 return 0; 522 } 523 524 static void blkdev_readahead(struct readahead_control *rac) 525 { 526 iomap_bio_readahead(rac, &blkdev_iomap_ops); 527 } 528 529 static ssize_t blkdev_writeback_range(struct iomap_writepage_ctx *wpc, 530 struct folio *folio, u64 offset, unsigned int len, u64 end_pos) 531 { 532 loff_t isize = i_size_read(wpc->inode); 533 534 if (WARN_ON_ONCE(offset >= isize)) 535 return -EIO; 536 537 if (offset < wpc->iomap.offset || 538 offset >= wpc->iomap.offset + wpc->iomap.length) { 539 int error; 540 541 error = blkdev_iomap_begin(wpc->inode, offset, isize - offset, 542 IOMAP_WRITE, &wpc->iomap, NULL); 543 if (error) 544 return error; 545 } 546 547 return iomap_add_to_ioend(wpc, folio, offset, end_pos, len); 548 } 549 550 static const struct iomap_writeback_ops blkdev_writeback_ops = { 551 .writeback_range = blkdev_writeback_range, 552 .writeback_submit = iomap_ioend_writeback_submit, 553 }; 554 555 static int blkdev_writepages(struct address_space *mapping, 556 struct writeback_control *wbc) 557 { 558 struct iomap_writepage_ctx wpc = { 559 .inode = mapping->host, 560 .wbc = wbc, 561 .ops = &blkdev_writeback_ops 562 }; 563 564 return iomap_writepages(&wpc); 565 } 566 567 const struct address_space_operations def_blk_aops = { 568 .dirty_folio = filemap_dirty_folio, 569 .release_folio = iomap_release_folio, 570 .invalidate_folio = iomap_invalidate_folio, 571 .read_folio = blkdev_read_folio, 572 .readahead = blkdev_readahead, 573 .writepages = blkdev_writepages, 574 .is_partially_uptodate = iomap_is_partially_uptodate, 575 .error_remove_folio = generic_error_remove_folio, 576 .migrate_folio = filemap_migrate_folio, 577 }; 578 #endif /* CONFIG_BUFFER_HEAD */ 579 580 /* 581 * for a block special file file_inode(file)->i_size is zero 582 * so we compute the size by hand (just as in block_read/write above) 583 */ 584 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) 585 { 586 struct inode *bd_inode = bdev_file_inode(file); 587 loff_t retval; 588 589 inode_lock(bd_inode); 590 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); 591 inode_unlock(bd_inode); 592 return retval; 593 } 594 595 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, 596 int datasync) 597 { 598 struct block_device *bdev = I_BDEV(filp->f_mapping->host); 599 int error; 600 601 error = file_write_and_wait_range(filp, start, end); 602 if (error) 603 return error; 604 605 /* 606 * There is no need to serialise calls to blkdev_issue_flush with 607 * i_mutex and doing so causes performance issues with concurrent 608 * O_SYNC writers to a block device. 609 */ 610 error = blkdev_issue_flush(bdev); 611 if (error == -EOPNOTSUPP) 612 error = 0; 613 614 return error; 615 } 616 617 /** 618 * file_to_blk_mode - get block open flags from file flags 619 * @file: file whose open flags should be converted 620 * 621 * Look at file open flags and generate corresponding block open flags from 622 * them. The function works both for file just being open (e.g. during ->open 623 * callback) and for file that is already open. This is actually non-trivial 624 * (see comment in the function). 625 */ 626 blk_mode_t file_to_blk_mode(struct file *file) 627 { 628 blk_mode_t mode = 0; 629 630 if (file->f_mode & FMODE_READ) 631 mode |= BLK_OPEN_READ; 632 if (file->f_mode & FMODE_WRITE) 633 mode |= BLK_OPEN_WRITE; 634 /* 635 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data 636 * to determine whether the open was exclusive for already open files. 637 */ 638 if (file->private_data) 639 mode |= BLK_OPEN_EXCL; 640 else if (file->f_flags & O_EXCL) 641 mode |= BLK_OPEN_EXCL; 642 if (file->f_flags & O_NDELAY) 643 mode |= BLK_OPEN_NDELAY; 644 645 /* 646 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy 647 * driver has historically allowed ioctls as if the file was opened for 648 * writing, but does not allow and actual reads or writes. 649 */ 650 if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY)) 651 mode |= BLK_OPEN_WRITE_IOCTL; 652 653 return mode; 654 } 655 656 static int blkdev_open(struct inode *inode, struct file *filp) 657 { 658 struct block_device *bdev; 659 blk_mode_t mode; 660 int ret; 661 662 mode = file_to_blk_mode(filp); 663 /* Use the file as the holder. */ 664 if (mode & BLK_OPEN_EXCL) 665 filp->private_data = filp; 666 ret = bdev_permission(inode->i_rdev, mode, filp->private_data); 667 if (ret) 668 return ret; 669 670 bdev = blkdev_get_no_open(inode->i_rdev, true); 671 if (!bdev) 672 return -ENXIO; 673 674 if (bdev_can_atomic_write(bdev)) 675 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE; 676 if (blk_get_integrity(bdev->bd_disk)) 677 filp->f_mode |= FMODE_HAS_METADATA; 678 679 ret = bdev_open(bdev, mode, filp->private_data, NULL, filp); 680 if (ret) 681 blkdev_put_no_open(bdev); 682 return ret; 683 } 684 685 static int blkdev_release(struct inode *inode, struct file *filp) 686 { 687 bdev_release(filp); 688 return 0; 689 } 690 691 static ssize_t 692 blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from) 693 { 694 size_t count = iov_iter_count(from); 695 ssize_t written; 696 697 written = kiocb_invalidate_pages(iocb, count); 698 if (written) { 699 if (written == -EBUSY) 700 return 0; 701 return written; 702 } 703 704 written = blkdev_direct_IO(iocb, from); 705 if (written > 0) { 706 kiocb_invalidate_post_direct_write(iocb, count); 707 iocb->ki_pos += written; 708 count -= written; 709 } 710 if (written != -EIOCBQUEUED) 711 iov_iter_revert(from, count - iov_iter_count(from)); 712 return written; 713 } 714 715 static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from) 716 { 717 return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL, 718 NULL); 719 } 720 721 /* 722 * Write data to the block device. Only intended for the block device itself 723 * and the raw driver which basically is a fake block device. 724 * 725 * Does not take i_mutex for the write and thus is not for general purpose 726 * use. 727 */ 728 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) 729 { 730 struct file *file = iocb->ki_filp; 731 struct inode *bd_inode = bdev_file_inode(file); 732 struct block_device *bdev = I_BDEV(bd_inode); 733 bool atomic = iocb->ki_flags & IOCB_ATOMIC; 734 loff_t size = bdev_nr_bytes(bdev); 735 size_t shorted = 0; 736 ssize_t ret; 737 738 if (bdev_read_only(bdev)) 739 return -EPERM; 740 741 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) 742 return -ETXTBSY; 743 744 if (!iov_iter_count(from)) 745 return 0; 746 747 if (iocb->ki_pos >= size) 748 return -ENOSPC; 749 750 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) 751 return -EOPNOTSUPP; 752 753 if (atomic) { 754 ret = generic_atomic_write_valid(iocb, from); 755 if (ret) 756 return ret; 757 } 758 759 size -= iocb->ki_pos; 760 if (iov_iter_count(from) > size) { 761 if (atomic) 762 return -EINVAL; 763 shorted = iov_iter_count(from) - size; 764 iov_iter_truncate(from, size); 765 } 766 767 ret = file_update_time(file); 768 if (ret) 769 return ret; 770 771 if (iocb->ki_flags & IOCB_DIRECT) { 772 ret = blkdev_direct_write(iocb, from); 773 if (ret >= 0 && iov_iter_count(from)) 774 ret = direct_write_fallback(iocb, from, ret, 775 blkdev_buffered_write(iocb, from)); 776 } else { 777 /* 778 * Take i_rwsem and invalidate_lock to avoid racing with 779 * set_blocksize changing i_blkbits/folio order and punching 780 * out the pagecache. 781 */ 782 inode_lock_shared(bd_inode); 783 ret = blkdev_buffered_write(iocb, from); 784 inode_unlock_shared(bd_inode); 785 } 786 787 if (ret > 0) 788 ret = generic_write_sync(iocb, ret); 789 iov_iter_reexpand(from, iov_iter_count(from) + shorted); 790 return ret; 791 } 792 793 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) 794 { 795 struct inode *bd_inode = bdev_file_inode(iocb->ki_filp); 796 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 797 loff_t size = bdev_nr_bytes(bdev); 798 loff_t pos = iocb->ki_pos; 799 size_t shorted = 0; 800 ssize_t ret = 0; 801 size_t count; 802 803 if (unlikely(pos + iov_iter_count(to) > size)) { 804 if (pos >= size) 805 return 0; 806 size -= pos; 807 shorted = iov_iter_count(to) - size; 808 iov_iter_truncate(to, size); 809 } 810 811 count = iov_iter_count(to); 812 if (!count) 813 goto reexpand; /* skip atime */ 814 815 if (iocb->ki_flags & IOCB_DIRECT) { 816 ret = kiocb_write_and_wait(iocb, count); 817 if (ret < 0) 818 goto reexpand; 819 file_accessed(iocb->ki_filp); 820 821 ret = blkdev_direct_IO(iocb, to); 822 if (ret > 0) { 823 iocb->ki_pos += ret; 824 count -= ret; 825 } 826 if (ret != -EIOCBQUEUED) 827 iov_iter_revert(to, count - iov_iter_count(to)); 828 if (ret < 0 || !count) 829 goto reexpand; 830 } 831 832 /* 833 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize 834 * changing i_blkbits/folio order and punching out the pagecache. 835 */ 836 inode_lock_shared(bd_inode); 837 ret = filemap_read(iocb, to, ret); 838 inode_unlock_shared(bd_inode); 839 840 reexpand: 841 if (unlikely(shorted)) 842 iov_iter_reexpand(to, iov_iter_count(to) + shorted); 843 return ret; 844 } 845 846 #define BLKDEV_FALLOC_FL_SUPPORTED \ 847 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 848 FALLOC_FL_ZERO_RANGE | FALLOC_FL_WRITE_ZEROES) 849 850 static long blkdev_fallocate(struct file *file, int mode, loff_t start, 851 loff_t len) 852 { 853 struct inode *inode = bdev_file_inode(file); 854 struct block_device *bdev = I_BDEV(inode); 855 loff_t end = start + len - 1; 856 loff_t isize; 857 unsigned int flags; 858 int error; 859 860 /* Fail if we don't recognize the flags. */ 861 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) 862 return -EOPNOTSUPP; 863 /* 864 * Don't allow writing zeroes if the device does not enable the 865 * unmap write zeroes operation. 866 */ 867 if ((mode & FALLOC_FL_WRITE_ZEROES) && 868 !bdev_write_zeroes_unmap_sectors(bdev)) 869 return -EOPNOTSUPP; 870 871 /* Don't go off the end of the device. */ 872 isize = bdev_nr_bytes(bdev); 873 if (start >= isize) 874 return -EINVAL; 875 if (end >= isize) { 876 if (mode & FALLOC_FL_KEEP_SIZE) { 877 len = isize - start; 878 end = start + len - 1; 879 } else 880 return -EINVAL; 881 } 882 883 /* 884 * Don't allow IO that isn't aligned to logical block size. 885 */ 886 if ((start | len) & (bdev_logical_block_size(bdev) - 1)) 887 return -EINVAL; 888 889 inode_lock(inode); 890 filemap_invalidate_lock(inode->i_mapping); 891 892 switch (mode) { 893 case FALLOC_FL_ZERO_RANGE: 894 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: 895 flags = BLKDEV_ZERO_NOUNMAP; 896 break; 897 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: 898 flags = BLKDEV_ZERO_NOFALLBACK; 899 break; 900 case FALLOC_FL_WRITE_ZEROES: 901 flags = 0; 902 break; 903 default: 904 error = -EOPNOTSUPP; 905 goto fail; 906 } 907 908 /* 909 * Invalidate the page cache, including dirty pages, for valid 910 * de-allocate mode calls to fallocate(). 911 */ 912 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end); 913 if (error) 914 goto fail; 915 916 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 917 len >> SECTOR_SHIFT, GFP_KERNEL, flags); 918 fail: 919 filemap_invalidate_unlock(inode->i_mapping); 920 inode_unlock(inode); 921 return error; 922 } 923 924 static int blkdev_mmap_prepare(struct vm_area_desc *desc) 925 { 926 struct file *file = desc->file; 927 928 if (bdev_read_only(I_BDEV(bdev_file_inode(file)))) 929 return generic_file_readonly_mmap_prepare(desc); 930 931 return generic_file_mmap_prepare(desc); 932 } 933 934 const struct file_operations def_blk_fops = { 935 .open = blkdev_open, 936 .release = blkdev_release, 937 .llseek = blkdev_llseek, 938 .read_iter = blkdev_read_iter, 939 .write_iter = blkdev_write_iter, 940 .iopoll = iocb_bio_iopoll, 941 .mmap_prepare = blkdev_mmap_prepare, 942 .fsync = blkdev_fsync, 943 .unlocked_ioctl = blkdev_ioctl, 944 #ifdef CONFIG_COMPAT 945 .compat_ioctl = compat_blkdev_ioctl, 946 #endif 947 .splice_read = filemap_splice_read, 948 .splice_write = iter_file_splice_write, 949 .fallocate = blkdev_fallocate, 950 .uring_cmd = blkdev_uring_cmd, 951 .fop_flags = FOP_BUFFER_RASYNC | FOP_DONTCACHE, 952 }; 953 954 static __init int blkdev_init(void) 955 { 956 return bioset_init(&blkdev_dio_pool, 4, 957 offsetof(struct blkdev_dio, bio), 958 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); 959 } 960 module_init(blkdev_init); 961