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/buffer_head.h> 11 #include <linux/mpage.h> 12 #include <linux/uio.h> 13 #include <linux/namei.h> 14 #include <linux/task_io_accounting_ops.h> 15 #include <linux/falloc.h> 16 #include <linux/suspend.h> 17 #include <linux/fs.h> 18 #include <linux/iomap.h> 19 #include <linux/module.h> 20 #include <linux/io_uring/cmd.h> 21 #include "blk.h" 22 23 static inline struct inode *bdev_file_inode(struct file *file) 24 { 25 return file->f_mapping->host; 26 } 27 28 static blk_opf_t dio_bio_write_op(struct kiocb *iocb) 29 { 30 blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; 31 32 /* avoid the need for a I/O completion work item */ 33 if (iocb_is_dsync(iocb)) 34 opf |= REQ_FUA; 35 return opf; 36 } 37 38 static bool blkdev_dio_invalid(struct block_device *bdev, struct kiocb *iocb, 39 struct iov_iter *iter) 40 { 41 return iocb->ki_pos & (bdev_logical_block_size(bdev) - 1) || 42 !bdev_iter_is_aligned(bdev, iter); 43 } 44 45 #define DIO_INLINE_BIO_VECS 4 46 47 static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb, 48 struct iov_iter *iter, struct block_device *bdev, 49 unsigned int nr_pages) 50 { 51 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; 52 loff_t pos = iocb->ki_pos; 53 bool should_dirty = false; 54 struct bio bio; 55 ssize_t ret; 56 57 WARN_ON_ONCE(iocb->ki_flags & IOCB_HAS_METADATA); 58 if (nr_pages <= DIO_INLINE_BIO_VECS) 59 vecs = inline_vecs; 60 else { 61 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), 62 GFP_KERNEL); 63 if (!vecs) 64 return -ENOMEM; 65 } 66 67 if (iov_iter_rw(iter) == READ) { 68 bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ); 69 if (user_backed_iter(iter)) 70 should_dirty = true; 71 } else { 72 bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb)); 73 } 74 bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT; 75 bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 76 bio.bi_write_stream = iocb->ki_write_stream; 77 bio.bi_ioprio = iocb->ki_ioprio; 78 if (iocb->ki_flags & IOCB_ATOMIC) 79 bio.bi_opf |= REQ_ATOMIC; 80 81 ret = bio_iov_iter_get_pages(&bio, iter); 82 if (unlikely(ret)) 83 goto out; 84 ret = bio.bi_iter.bi_size; 85 86 if (iov_iter_rw(iter) == WRITE) 87 task_io_account_write(ret); 88 89 if (iocb->ki_flags & IOCB_NOWAIT) 90 bio.bi_opf |= REQ_NOWAIT; 91 92 submit_bio_wait(&bio); 93 94 bio_release_pages(&bio, should_dirty); 95 if (unlikely(bio.bi_status)) 96 ret = blk_status_to_errno(bio.bi_status); 97 98 out: 99 if (vecs != inline_vecs) 100 kfree(vecs); 101 102 bio_uninit(&bio); 103 104 return ret; 105 } 106 107 enum { 108 DIO_SHOULD_DIRTY = 1, 109 DIO_IS_SYNC = 2, 110 }; 111 112 struct blkdev_dio { 113 union { 114 struct kiocb *iocb; 115 struct task_struct *waiter; 116 }; 117 size_t size; 118 atomic_t ref; 119 unsigned int flags; 120 struct bio bio ____cacheline_aligned_in_smp; 121 }; 122 123 static struct bio_set blkdev_dio_pool; 124 125 static void blkdev_bio_end_io(struct bio *bio) 126 { 127 struct blkdev_dio *dio = bio->bi_private; 128 bool should_dirty = dio->flags & DIO_SHOULD_DIRTY; 129 bool is_sync = dio->flags & DIO_IS_SYNC; 130 131 if (bio->bi_status && !dio->bio.bi_status) 132 dio->bio.bi_status = bio->bi_status; 133 134 if (!is_sync && (dio->iocb->ki_flags & IOCB_HAS_METADATA)) 135 bio_integrity_unmap_user(bio); 136 137 if (atomic_dec_and_test(&dio->ref)) { 138 if (!is_sync) { 139 struct kiocb *iocb = dio->iocb; 140 ssize_t ret; 141 142 WRITE_ONCE(iocb->private, NULL); 143 144 if (likely(!dio->bio.bi_status)) { 145 ret = dio->size; 146 iocb->ki_pos += ret; 147 } else { 148 ret = blk_status_to_errno(dio->bio.bi_status); 149 } 150 151 dio->iocb->ki_complete(iocb, ret); 152 bio_put(&dio->bio); 153 } else { 154 struct task_struct *waiter = dio->waiter; 155 156 WRITE_ONCE(dio->waiter, NULL); 157 blk_wake_io_task(waiter); 158 } 159 } 160 161 if (should_dirty) { 162 bio_check_pages_dirty(bio); 163 } else { 164 bio_release_pages(bio, false); 165 bio_put(bio); 166 } 167 } 168 169 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, 170 struct block_device *bdev, unsigned int nr_pages) 171 { 172 struct blk_plug plug; 173 struct blkdev_dio *dio; 174 struct bio *bio; 175 bool is_read = (iov_iter_rw(iter) == READ), is_sync; 176 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 177 loff_t pos = iocb->ki_pos; 178 int ret = 0; 179 180 if (iocb->ki_flags & IOCB_ALLOC_CACHE) 181 opf |= REQ_ALLOC_CACHE; 182 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 183 &blkdev_dio_pool); 184 dio = container_of(bio, struct blkdev_dio, bio); 185 atomic_set(&dio->ref, 1); 186 /* 187 * Grab an extra reference to ensure the dio structure which is embedded 188 * into the first bio stays around. 189 */ 190 bio_get(bio); 191 192 is_sync = is_sync_kiocb(iocb); 193 if (is_sync) { 194 dio->flags = DIO_IS_SYNC; 195 dio->waiter = current; 196 } else { 197 dio->flags = 0; 198 dio->iocb = iocb; 199 } 200 201 dio->size = 0; 202 if (is_read && user_backed_iter(iter)) 203 dio->flags |= DIO_SHOULD_DIRTY; 204 205 blk_start_plug(&plug); 206 207 for (;;) { 208 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 209 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 210 bio->bi_write_stream = iocb->ki_write_stream; 211 bio->bi_private = dio; 212 bio->bi_end_io = blkdev_bio_end_io; 213 bio->bi_ioprio = iocb->ki_ioprio; 214 215 ret = bio_iov_iter_get_pages(bio, iter); 216 if (unlikely(ret)) { 217 bio->bi_status = BLK_STS_IOERR; 218 bio_endio(bio); 219 break; 220 } 221 if (iocb->ki_flags & IOCB_NOWAIT) { 222 /* 223 * This is nonblocking IO, and we need to allocate 224 * another bio if we have data left to map. As we 225 * cannot guarantee that one of the sub bios will not 226 * fail getting issued FOR NOWAIT and as error results 227 * are coalesced across all of them, be safe and ask for 228 * a retry of this from blocking context. 229 */ 230 if (unlikely(iov_iter_count(iter))) { 231 ret = -EAGAIN; 232 goto fail; 233 } 234 bio->bi_opf |= REQ_NOWAIT; 235 } 236 if (!is_sync && (iocb->ki_flags & IOCB_HAS_METADATA)) { 237 ret = bio_integrity_map_iter(bio, iocb->private); 238 if (unlikely(ret)) 239 goto fail; 240 } 241 242 if (is_read) { 243 if (dio->flags & DIO_SHOULD_DIRTY) 244 bio_set_pages_dirty(bio); 245 } else { 246 task_io_account_write(bio->bi_iter.bi_size); 247 } 248 dio->size += bio->bi_iter.bi_size; 249 pos += bio->bi_iter.bi_size; 250 251 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); 252 if (!nr_pages) { 253 submit_bio(bio); 254 break; 255 } 256 atomic_inc(&dio->ref); 257 submit_bio(bio); 258 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL); 259 } 260 261 blk_finish_plug(&plug); 262 263 if (!is_sync) 264 return -EIOCBQUEUED; 265 266 for (;;) { 267 set_current_state(TASK_UNINTERRUPTIBLE); 268 if (!READ_ONCE(dio->waiter)) 269 break; 270 blk_io_schedule(); 271 } 272 __set_current_state(TASK_RUNNING); 273 274 if (!ret) 275 ret = blk_status_to_errno(dio->bio.bi_status); 276 if (likely(!ret)) 277 ret = dio->size; 278 279 bio_put(&dio->bio); 280 return ret; 281 fail: 282 bio_release_pages(bio, false); 283 bio_clear_flag(bio, BIO_REFFED); 284 bio_put(bio); 285 blk_finish_plug(&plug); 286 return ret; 287 } 288 289 static void blkdev_bio_end_io_async(struct bio *bio) 290 { 291 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); 292 struct kiocb *iocb = dio->iocb; 293 ssize_t ret; 294 295 WRITE_ONCE(iocb->private, NULL); 296 297 if (likely(!bio->bi_status)) { 298 ret = dio->size; 299 iocb->ki_pos += ret; 300 } else { 301 ret = blk_status_to_errno(bio->bi_status); 302 } 303 304 if (iocb->ki_flags & IOCB_HAS_METADATA) 305 bio_integrity_unmap_user(bio); 306 307 iocb->ki_complete(iocb, ret); 308 309 if (dio->flags & DIO_SHOULD_DIRTY) { 310 bio_check_pages_dirty(bio); 311 } else { 312 bio_release_pages(bio, false); 313 bio_put(bio); 314 } 315 } 316 317 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, 318 struct iov_iter *iter, 319 struct block_device *bdev, 320 unsigned int nr_pages) 321 { 322 bool is_read = iov_iter_rw(iter) == READ; 323 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); 324 struct blkdev_dio *dio; 325 struct bio *bio; 326 loff_t pos = iocb->ki_pos; 327 int ret = 0; 328 329 if (iocb->ki_flags & IOCB_ALLOC_CACHE) 330 opf |= REQ_ALLOC_CACHE; 331 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL, 332 &blkdev_dio_pool); 333 dio = container_of(bio, struct blkdev_dio, bio); 334 dio->flags = 0; 335 dio->iocb = iocb; 336 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 337 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint; 338 bio->bi_write_stream = iocb->ki_write_stream; 339 bio->bi_end_io = blkdev_bio_end_io_async; 340 bio->bi_ioprio = iocb->ki_ioprio; 341 342 if (iov_iter_is_bvec(iter)) { 343 /* 344 * Users don't rely on the iterator being in any particular 345 * state for async I/O returning -EIOCBQUEUED, hence we can 346 * avoid expensive iov_iter_advance(). Bypass 347 * bio_iov_iter_get_pages() and set the bvec directly. 348 */ 349 bio_iov_bvec_set(bio, iter); 350 } else { 351 ret = bio_iov_iter_get_pages(bio, iter); 352 if (unlikely(ret)) 353 goto out_bio_put; 354 } 355 dio->size = bio->bi_iter.bi_size; 356 357 if (is_read) { 358 if (user_backed_iter(iter)) { 359 dio->flags |= DIO_SHOULD_DIRTY; 360 bio_set_pages_dirty(bio); 361 } 362 } else { 363 task_io_account_write(bio->bi_iter.bi_size); 364 } 365 366 if (iocb->ki_flags & IOCB_HAS_METADATA) { 367 ret = bio_integrity_map_iter(bio, iocb->private); 368 WRITE_ONCE(iocb->private, NULL); 369 if (unlikely(ret)) 370 goto out_bio_put; 371 } 372 373 if (iocb->ki_flags & IOCB_ATOMIC) 374 bio->bi_opf |= REQ_ATOMIC; 375 376 if (iocb->ki_flags & IOCB_NOWAIT) 377 bio->bi_opf |= REQ_NOWAIT; 378 379 if (iocb->ki_flags & IOCB_HIPRI) { 380 bio->bi_opf |= REQ_POLLED; 381 submit_bio(bio); 382 WRITE_ONCE(iocb->private, bio); 383 } else { 384 submit_bio(bio); 385 } 386 return -EIOCBQUEUED; 387 388 out_bio_put: 389 bio_put(bio); 390 return ret; 391 } 392 393 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 394 { 395 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 396 unsigned int nr_pages; 397 398 if (!iov_iter_count(iter)) 399 return 0; 400 401 if (blkdev_dio_invalid(bdev, iocb, iter)) 402 return -EINVAL; 403 404 if (iov_iter_rw(iter) == WRITE) { 405 u16 max_write_streams = bdev_max_write_streams(bdev); 406 407 if (iocb->ki_write_stream) { 408 if (iocb->ki_write_stream > max_write_streams) 409 return -EINVAL; 410 } else if (max_write_streams) { 411 enum rw_hint write_hint = 412 file_inode(iocb->ki_filp)->i_write_hint; 413 414 /* 415 * Just use the write hint as write stream for block 416 * device writes. This assumes no file system is 417 * mounted that would use the streams differently. 418 */ 419 if (write_hint <= max_write_streams) 420 iocb->ki_write_stream = write_hint; 421 } 422 } 423 424 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); 425 if (likely(nr_pages <= BIO_MAX_VECS)) { 426 if (is_sync_kiocb(iocb)) 427 return __blkdev_direct_IO_simple(iocb, iter, bdev, 428 nr_pages); 429 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages); 430 } else if (iocb->ki_flags & IOCB_ATOMIC) { 431 return -EINVAL; 432 } 433 return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages)); 434 } 435 436 static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 437 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 438 { 439 struct block_device *bdev = I_BDEV(inode); 440 loff_t isize = i_size_read(inode); 441 442 if (offset >= isize) 443 return -EIO; 444 445 iomap->bdev = bdev; 446 iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev)); 447 iomap->type = IOMAP_MAPPED; 448 iomap->addr = iomap->offset; 449 iomap->length = isize - iomap->offset; 450 iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */ 451 return 0; 452 } 453 454 static const struct iomap_ops blkdev_iomap_ops = { 455 .iomap_begin = blkdev_iomap_begin, 456 }; 457 458 #ifdef CONFIG_BUFFER_HEAD 459 static int blkdev_get_block(struct inode *inode, sector_t iblock, 460 struct buffer_head *bh, int create) 461 { 462 bh->b_bdev = I_BDEV(inode); 463 bh->b_blocknr = iblock; 464 set_buffer_mapped(bh); 465 return 0; 466 } 467 468 /* 469 * We cannot call mpage_writepages() as it does not take the buffer lock. 470 * We must use block_write_full_folio() directly which holds the buffer 471 * lock. The buffer lock provides the synchronisation with writeback 472 * that filesystems rely on when they use the blockdev's mapping. 473 */ 474 static int blkdev_writepages(struct address_space *mapping, 475 struct writeback_control *wbc) 476 { 477 struct folio *folio = NULL; 478 struct blk_plug plug; 479 int err; 480 481 blk_start_plug(&plug); 482 while ((folio = writeback_iter(mapping, wbc, folio, &err))) 483 err = block_write_full_folio(folio, wbc, blkdev_get_block); 484 blk_finish_plug(&plug); 485 486 return err; 487 } 488 489 static int blkdev_read_folio(struct file *file, struct folio *folio) 490 { 491 return block_read_full_folio(folio, blkdev_get_block); 492 } 493 494 static void blkdev_readahead(struct readahead_control *rac) 495 { 496 mpage_readahead(rac, blkdev_get_block); 497 } 498 499 static int blkdev_write_begin(const struct kiocb *iocb, 500 struct address_space *mapping, loff_t pos, 501 unsigned len, struct folio **foliop, 502 void **fsdata) 503 { 504 return block_write_begin(mapping, pos, len, foliop, blkdev_get_block); 505 } 506 507 static int blkdev_write_end(const struct kiocb *iocb, 508 struct address_space *mapping, 509 loff_t pos, unsigned len, unsigned copied, 510 struct folio *folio, void *fsdata) 511 { 512 int ret; 513 ret = block_write_end(pos, len, copied, folio); 514 515 folio_unlock(folio); 516 folio_put(folio); 517 518 return ret; 519 } 520 521 const struct address_space_operations def_blk_aops = { 522 .dirty_folio = block_dirty_folio, 523 .invalidate_folio = block_invalidate_folio, 524 .read_folio = blkdev_read_folio, 525 .readahead = blkdev_readahead, 526 .writepages = blkdev_writepages, 527 .write_begin = blkdev_write_begin, 528 .write_end = blkdev_write_end, 529 .migrate_folio = buffer_migrate_folio_norefs, 530 .is_dirty_writeback = buffer_check_dirty_writeback, 531 }; 532 #else /* CONFIG_BUFFER_HEAD */ 533 static int blkdev_read_folio(struct file *file, struct folio *folio) 534 { 535 return iomap_read_folio(folio, &blkdev_iomap_ops); 536 } 537 538 static void blkdev_readahead(struct readahead_control *rac) 539 { 540 iomap_readahead(rac, &blkdev_iomap_ops); 541 } 542 543 static ssize_t blkdev_writeback_range(struct iomap_writepage_ctx *wpc, 544 struct folio *folio, u64 offset, unsigned int len, u64 end_pos) 545 { 546 loff_t isize = i_size_read(wpc->inode); 547 548 if (WARN_ON_ONCE(offset >= isize)) 549 return -EIO; 550 551 if (offset < wpc->iomap.offset || 552 offset >= wpc->iomap.offset + wpc->iomap.length) { 553 int error; 554 555 error = blkdev_iomap_begin(wpc->inode, offset, isize - offset, 556 IOMAP_WRITE, &wpc->iomap, NULL); 557 if (error) 558 return error; 559 } 560 561 return iomap_add_to_ioend(wpc, folio, offset, end_pos, len); 562 } 563 564 static const struct iomap_writeback_ops blkdev_writeback_ops = { 565 .writeback_range = blkdev_writeback_range, 566 .writeback_submit = iomap_ioend_writeback_submit, 567 }; 568 569 static int blkdev_writepages(struct address_space *mapping, 570 struct writeback_control *wbc) 571 { 572 struct iomap_writepage_ctx wpc = { 573 .inode = mapping->host, 574 .wbc = wbc, 575 .ops = &blkdev_writeback_ops 576 }; 577 578 return iomap_writepages(&wpc); 579 } 580 581 const struct address_space_operations def_blk_aops = { 582 .dirty_folio = filemap_dirty_folio, 583 .release_folio = iomap_release_folio, 584 .invalidate_folio = iomap_invalidate_folio, 585 .read_folio = blkdev_read_folio, 586 .readahead = blkdev_readahead, 587 .writepages = blkdev_writepages, 588 .is_partially_uptodate = iomap_is_partially_uptodate, 589 .error_remove_folio = generic_error_remove_folio, 590 .migrate_folio = filemap_migrate_folio, 591 }; 592 #endif /* CONFIG_BUFFER_HEAD */ 593 594 /* 595 * for a block special file file_inode(file)->i_size is zero 596 * so we compute the size by hand (just as in block_read/write above) 597 */ 598 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) 599 { 600 struct inode *bd_inode = bdev_file_inode(file); 601 loff_t retval; 602 603 inode_lock(bd_inode); 604 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); 605 inode_unlock(bd_inode); 606 return retval; 607 } 608 609 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, 610 int datasync) 611 { 612 struct block_device *bdev = I_BDEV(filp->f_mapping->host); 613 int error; 614 615 error = file_write_and_wait_range(filp, start, end); 616 if (error) 617 return error; 618 619 /* 620 * There is no need to serialise calls to blkdev_issue_flush with 621 * i_mutex and doing so causes performance issues with concurrent 622 * O_SYNC writers to a block device. 623 */ 624 error = blkdev_issue_flush(bdev); 625 if (error == -EOPNOTSUPP) 626 error = 0; 627 628 return error; 629 } 630 631 /** 632 * file_to_blk_mode - get block open flags from file flags 633 * @file: file whose open flags should be converted 634 * 635 * Look at file open flags and generate corresponding block open flags from 636 * them. The function works both for file just being open (e.g. during ->open 637 * callback) and for file that is already open. This is actually non-trivial 638 * (see comment in the function). 639 */ 640 blk_mode_t file_to_blk_mode(struct file *file) 641 { 642 blk_mode_t mode = 0; 643 644 if (file->f_mode & FMODE_READ) 645 mode |= BLK_OPEN_READ; 646 if (file->f_mode & FMODE_WRITE) 647 mode |= BLK_OPEN_WRITE; 648 /* 649 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data 650 * to determine whether the open was exclusive for already open files. 651 */ 652 if (file->private_data) 653 mode |= BLK_OPEN_EXCL; 654 else if (file->f_flags & O_EXCL) 655 mode |= BLK_OPEN_EXCL; 656 if (file->f_flags & O_NDELAY) 657 mode |= BLK_OPEN_NDELAY; 658 659 /* 660 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy 661 * driver has historically allowed ioctls as if the file was opened for 662 * writing, but does not allow and actual reads or writes. 663 */ 664 if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY)) 665 mode |= BLK_OPEN_WRITE_IOCTL; 666 667 return mode; 668 } 669 670 static int blkdev_open(struct inode *inode, struct file *filp) 671 { 672 struct block_device *bdev; 673 blk_mode_t mode; 674 int ret; 675 676 mode = file_to_blk_mode(filp); 677 /* Use the file as the holder. */ 678 if (mode & BLK_OPEN_EXCL) 679 filp->private_data = filp; 680 ret = bdev_permission(inode->i_rdev, mode, filp->private_data); 681 if (ret) 682 return ret; 683 684 bdev = blkdev_get_no_open(inode->i_rdev, true); 685 if (!bdev) 686 return -ENXIO; 687 688 if (bdev_can_atomic_write(bdev)) 689 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE; 690 691 ret = bdev_open(bdev, mode, filp->private_data, NULL, filp); 692 if (ret) 693 blkdev_put_no_open(bdev); 694 return ret; 695 } 696 697 static int blkdev_release(struct inode *inode, struct file *filp) 698 { 699 bdev_release(filp); 700 return 0; 701 } 702 703 static ssize_t 704 blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from) 705 { 706 size_t count = iov_iter_count(from); 707 ssize_t written; 708 709 written = kiocb_invalidate_pages(iocb, count); 710 if (written) { 711 if (written == -EBUSY) 712 return 0; 713 return written; 714 } 715 716 written = blkdev_direct_IO(iocb, from); 717 if (written > 0) { 718 kiocb_invalidate_post_direct_write(iocb, count); 719 iocb->ki_pos += written; 720 count -= written; 721 } 722 if (written != -EIOCBQUEUED) 723 iov_iter_revert(from, count - iov_iter_count(from)); 724 return written; 725 } 726 727 static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from) 728 { 729 return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL, 730 NULL); 731 } 732 733 /* 734 * Write data to the block device. Only intended for the block device itself 735 * and the raw driver which basically is a fake block device. 736 * 737 * Does not take i_mutex for the write and thus is not for general purpose 738 * use. 739 */ 740 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) 741 { 742 struct file *file = iocb->ki_filp; 743 struct inode *bd_inode = bdev_file_inode(file); 744 struct block_device *bdev = I_BDEV(bd_inode); 745 bool atomic = iocb->ki_flags & IOCB_ATOMIC; 746 loff_t size = bdev_nr_bytes(bdev); 747 size_t shorted = 0; 748 ssize_t ret; 749 750 if (bdev_read_only(bdev)) 751 return -EPERM; 752 753 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) 754 return -ETXTBSY; 755 756 if (!iov_iter_count(from)) 757 return 0; 758 759 if (iocb->ki_pos >= size) 760 return -ENOSPC; 761 762 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) 763 return -EOPNOTSUPP; 764 765 if (atomic) { 766 ret = generic_atomic_write_valid(iocb, from); 767 if (ret) 768 return ret; 769 } 770 771 size -= iocb->ki_pos; 772 if (iov_iter_count(from) > size) { 773 if (atomic) 774 return -EINVAL; 775 shorted = iov_iter_count(from) - size; 776 iov_iter_truncate(from, size); 777 } 778 779 ret = file_update_time(file); 780 if (ret) 781 return ret; 782 783 if (iocb->ki_flags & IOCB_DIRECT) { 784 ret = blkdev_direct_write(iocb, from); 785 if (ret >= 0 && iov_iter_count(from)) 786 ret = direct_write_fallback(iocb, from, ret, 787 blkdev_buffered_write(iocb, from)); 788 } else { 789 /* 790 * Take i_rwsem and invalidate_lock to avoid racing with 791 * set_blocksize changing i_blkbits/folio order and punching 792 * out the pagecache. 793 */ 794 inode_lock_shared(bd_inode); 795 ret = blkdev_buffered_write(iocb, from); 796 inode_unlock_shared(bd_inode); 797 } 798 799 if (ret > 0) 800 ret = generic_write_sync(iocb, ret); 801 iov_iter_reexpand(from, iov_iter_count(from) + shorted); 802 return ret; 803 } 804 805 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) 806 { 807 struct inode *bd_inode = bdev_file_inode(iocb->ki_filp); 808 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host); 809 loff_t size = bdev_nr_bytes(bdev); 810 loff_t pos = iocb->ki_pos; 811 size_t shorted = 0; 812 ssize_t ret = 0; 813 size_t count; 814 815 if (unlikely(pos + iov_iter_count(to) > size)) { 816 if (pos >= size) 817 return 0; 818 size -= pos; 819 shorted = iov_iter_count(to) - size; 820 iov_iter_truncate(to, size); 821 } 822 823 count = iov_iter_count(to); 824 if (!count) 825 goto reexpand; /* skip atime */ 826 827 if (iocb->ki_flags & IOCB_DIRECT) { 828 ret = kiocb_write_and_wait(iocb, count); 829 if (ret < 0) 830 goto reexpand; 831 file_accessed(iocb->ki_filp); 832 833 ret = blkdev_direct_IO(iocb, to); 834 if (ret > 0) { 835 iocb->ki_pos += ret; 836 count -= ret; 837 } 838 if (ret != -EIOCBQUEUED) 839 iov_iter_revert(to, count - iov_iter_count(to)); 840 if (ret < 0 || !count) 841 goto reexpand; 842 } 843 844 /* 845 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize 846 * changing i_blkbits/folio order and punching out the pagecache. 847 */ 848 inode_lock_shared(bd_inode); 849 ret = filemap_read(iocb, to, ret); 850 inode_unlock_shared(bd_inode); 851 852 reexpand: 853 if (unlikely(shorted)) 854 iov_iter_reexpand(to, iov_iter_count(to) + shorted); 855 return ret; 856 } 857 858 #define BLKDEV_FALLOC_FL_SUPPORTED \ 859 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 860 FALLOC_FL_ZERO_RANGE | FALLOC_FL_WRITE_ZEROES) 861 862 static long blkdev_fallocate(struct file *file, int mode, loff_t start, 863 loff_t len) 864 { 865 struct inode *inode = bdev_file_inode(file); 866 struct block_device *bdev = I_BDEV(inode); 867 loff_t end = start + len - 1; 868 loff_t isize; 869 unsigned int flags; 870 int error; 871 872 /* Fail if we don't recognize the flags. */ 873 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) 874 return -EOPNOTSUPP; 875 /* 876 * Don't allow writing zeroes if the device does not enable the 877 * unmap write zeroes operation. 878 */ 879 if ((mode & FALLOC_FL_WRITE_ZEROES) && 880 !bdev_write_zeroes_unmap_sectors(bdev)) 881 return -EOPNOTSUPP; 882 883 /* Don't go off the end of the device. */ 884 isize = bdev_nr_bytes(bdev); 885 if (start >= isize) 886 return -EINVAL; 887 if (end >= isize) { 888 if (mode & FALLOC_FL_KEEP_SIZE) { 889 len = isize - start; 890 end = start + len - 1; 891 } else 892 return -EINVAL; 893 } 894 895 /* 896 * Don't allow IO that isn't aligned to logical block size. 897 */ 898 if ((start | len) & (bdev_logical_block_size(bdev) - 1)) 899 return -EINVAL; 900 901 inode_lock(inode); 902 filemap_invalidate_lock(inode->i_mapping); 903 904 switch (mode) { 905 case FALLOC_FL_ZERO_RANGE: 906 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: 907 flags = BLKDEV_ZERO_NOUNMAP; 908 break; 909 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: 910 flags = BLKDEV_ZERO_NOFALLBACK; 911 break; 912 case FALLOC_FL_WRITE_ZEROES: 913 flags = 0; 914 break; 915 default: 916 error = -EOPNOTSUPP; 917 goto fail; 918 } 919 920 /* 921 * Invalidate the page cache, including dirty pages, for valid 922 * de-allocate mode calls to fallocate(). 923 */ 924 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end); 925 if (error) 926 goto fail; 927 928 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 929 len >> SECTOR_SHIFT, GFP_KERNEL, flags); 930 fail: 931 filemap_invalidate_unlock(inode->i_mapping); 932 inode_unlock(inode); 933 return error; 934 } 935 936 static int blkdev_mmap_prepare(struct vm_area_desc *desc) 937 { 938 struct file *file = desc->file; 939 940 if (bdev_read_only(I_BDEV(bdev_file_inode(file)))) 941 return generic_file_readonly_mmap_prepare(desc); 942 943 return generic_file_mmap_prepare(desc); 944 } 945 946 const struct file_operations def_blk_fops = { 947 .open = blkdev_open, 948 .release = blkdev_release, 949 .llseek = blkdev_llseek, 950 .read_iter = blkdev_read_iter, 951 .write_iter = blkdev_write_iter, 952 .iopoll = iocb_bio_iopoll, 953 .mmap_prepare = blkdev_mmap_prepare, 954 .fsync = blkdev_fsync, 955 .unlocked_ioctl = blkdev_ioctl, 956 #ifdef CONFIG_COMPAT 957 .compat_ioctl = compat_blkdev_ioctl, 958 #endif 959 .splice_read = filemap_splice_read, 960 .splice_write = iter_file_splice_write, 961 .fallocate = blkdev_fallocate, 962 .uring_cmd = blkdev_uring_cmd, 963 .fop_flags = FOP_BUFFER_RASYNC, 964 }; 965 966 static __init int blkdev_init(void) 967 { 968 return bioset_init(&blkdev_dio_pool, 4, 969 offsetof(struct blkdev_dio, bio), 970 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); 971 } 972 module_init(blkdev_init); 973