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 "blk.h" 19 20 static inline struct inode *bdev_file_inode(struct file *file) 21 { 22 return file->f_mapping->host; 23 } 24 25 static int blkdev_get_block(struct inode *inode, sector_t iblock, 26 struct buffer_head *bh, int create) 27 { 28 bh->b_bdev = I_BDEV(inode); 29 bh->b_blocknr = iblock; 30 set_buffer_mapped(bh); 31 return 0; 32 } 33 34 static unsigned int dio_bio_write_op(struct kiocb *iocb) 35 { 36 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; 37 38 /* avoid the need for a I/O completion work item */ 39 if (iocb->ki_flags & IOCB_DSYNC) 40 op |= REQ_FUA; 41 return op; 42 } 43 44 #define DIO_INLINE_BIO_VECS 4 45 46 static void blkdev_bio_end_io_simple(struct bio *bio) 47 { 48 struct task_struct *waiter = bio->bi_private; 49 50 WRITE_ONCE(bio->bi_private, NULL); 51 blk_wake_io_task(waiter); 52 } 53 54 static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb, 55 struct iov_iter *iter, unsigned int nr_pages) 56 { 57 struct block_device *bdev = iocb->ki_filp->private_data; 58 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; 59 loff_t pos = iocb->ki_pos; 60 bool should_dirty = false; 61 struct bio bio; 62 ssize_t ret; 63 64 if ((pos | iov_iter_alignment(iter)) & 65 (bdev_logical_block_size(bdev) - 1)) 66 return -EINVAL; 67 68 if (nr_pages <= DIO_INLINE_BIO_VECS) 69 vecs = inline_vecs; 70 else { 71 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), 72 GFP_KERNEL); 73 if (!vecs) 74 return -ENOMEM; 75 } 76 77 bio_init(&bio, vecs, nr_pages); 78 bio_set_dev(&bio, bdev); 79 bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT; 80 bio.bi_write_hint = iocb->ki_hint; 81 bio.bi_private = current; 82 bio.bi_end_io = blkdev_bio_end_io_simple; 83 bio.bi_ioprio = iocb->ki_ioprio; 84 85 ret = bio_iov_iter_get_pages(&bio, iter); 86 if (unlikely(ret)) 87 goto out; 88 ret = bio.bi_iter.bi_size; 89 90 if (iov_iter_rw(iter) == READ) { 91 bio.bi_opf = REQ_OP_READ; 92 if (iter_is_iovec(iter)) 93 should_dirty = true; 94 } else { 95 bio.bi_opf = dio_bio_write_op(iocb); 96 task_io_account_write(ret); 97 } 98 if (iocb->ki_flags & IOCB_NOWAIT) 99 bio.bi_opf |= REQ_NOWAIT; 100 if (iocb->ki_flags & IOCB_HIPRI) 101 bio_set_polled(&bio, iocb); 102 103 submit_bio(&bio); 104 for (;;) { 105 set_current_state(TASK_UNINTERRUPTIBLE); 106 if (!READ_ONCE(bio.bi_private)) 107 break; 108 if (!(iocb->ki_flags & IOCB_HIPRI) || !bio_poll(&bio, NULL, 0)) 109 blk_io_schedule(); 110 } 111 __set_current_state(TASK_RUNNING); 112 113 bio_release_pages(&bio, should_dirty); 114 if (unlikely(bio.bi_status)) 115 ret = blk_status_to_errno(bio.bi_status); 116 117 out: 118 if (vecs != inline_vecs) 119 kfree(vecs); 120 121 bio_uninit(&bio); 122 123 return ret; 124 } 125 126 enum { 127 DIO_SHOULD_DIRTY = 1, 128 DIO_IS_SYNC = 2, 129 }; 130 131 struct blkdev_dio { 132 union { 133 struct kiocb *iocb; 134 struct task_struct *waiter; 135 }; 136 size_t size; 137 atomic_t ref; 138 unsigned int flags; 139 struct bio bio ____cacheline_aligned_in_smp; 140 }; 141 142 static struct bio_set blkdev_dio_pool; 143 144 static void blkdev_bio_end_io(struct bio *bio) 145 { 146 struct blkdev_dio *dio = bio->bi_private; 147 bool should_dirty = dio->flags & DIO_SHOULD_DIRTY; 148 149 if (bio->bi_status && !dio->bio.bi_status) 150 dio->bio.bi_status = bio->bi_status; 151 152 if (atomic_dec_and_test(&dio->ref)) { 153 if (!(dio->flags & DIO_IS_SYNC)) { 154 struct kiocb *iocb = dio->iocb; 155 ssize_t ret; 156 157 WRITE_ONCE(iocb->private, NULL); 158 159 if (likely(!dio->bio.bi_status)) { 160 ret = dio->size; 161 iocb->ki_pos += ret; 162 } else { 163 ret = blk_status_to_errno(dio->bio.bi_status); 164 } 165 166 dio->iocb->ki_complete(iocb, ret); 167 bio_put(&dio->bio); 168 } else { 169 struct task_struct *waiter = dio->waiter; 170 171 WRITE_ONCE(dio->waiter, NULL); 172 blk_wake_io_task(waiter); 173 } 174 } 175 176 if (should_dirty) { 177 bio_check_pages_dirty(bio); 178 } else { 179 bio_release_pages(bio, false); 180 bio_put(bio); 181 } 182 } 183 184 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, 185 unsigned int nr_pages) 186 { 187 struct block_device *bdev = iocb->ki_filp->private_data; 188 struct blk_plug plug; 189 struct blkdev_dio *dio; 190 struct bio *bio; 191 bool is_read = (iov_iter_rw(iter) == READ), is_sync; 192 loff_t pos = iocb->ki_pos; 193 int ret = 0; 194 195 if ((pos | iov_iter_alignment(iter)) & 196 (bdev_logical_block_size(bdev) - 1)) 197 return -EINVAL; 198 199 bio = bio_alloc_kiocb(iocb, nr_pages, &blkdev_dio_pool); 200 201 dio = container_of(bio, struct blkdev_dio, bio); 202 atomic_set(&dio->ref, 1); 203 /* 204 * Grab an extra reference to ensure the dio structure which is embedded 205 * into the first bio stays around. 206 */ 207 bio_get(bio); 208 209 is_sync = is_sync_kiocb(iocb); 210 if (is_sync) { 211 dio->flags = DIO_IS_SYNC; 212 dio->waiter = current; 213 } else { 214 dio->flags = 0; 215 dio->iocb = iocb; 216 } 217 218 dio->size = 0; 219 if (is_read && iter_is_iovec(iter)) 220 dio->flags |= DIO_SHOULD_DIRTY; 221 222 blk_start_plug(&plug); 223 224 for (;;) { 225 bio_set_dev(bio, bdev); 226 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 227 bio->bi_write_hint = iocb->ki_hint; 228 bio->bi_private = dio; 229 bio->bi_end_io = blkdev_bio_end_io; 230 bio->bi_ioprio = iocb->ki_ioprio; 231 232 ret = bio_iov_iter_get_pages(bio, iter); 233 if (unlikely(ret)) { 234 bio->bi_status = BLK_STS_IOERR; 235 bio_endio(bio); 236 break; 237 } 238 239 if (is_read) { 240 bio->bi_opf = REQ_OP_READ; 241 if (dio->flags & DIO_SHOULD_DIRTY) 242 bio_set_pages_dirty(bio); 243 } else { 244 bio->bi_opf = dio_bio_write_op(iocb); 245 task_io_account_write(bio->bi_iter.bi_size); 246 } 247 if (iocb->ki_flags & IOCB_NOWAIT) 248 bio->bi_opf |= REQ_NOWAIT; 249 250 dio->size += bio->bi_iter.bi_size; 251 pos += bio->bi_iter.bi_size; 252 253 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); 254 if (!nr_pages) { 255 submit_bio(bio); 256 break; 257 } 258 atomic_inc(&dio->ref); 259 submit_bio(bio); 260 bio = bio_alloc(GFP_KERNEL, nr_pages); 261 } 262 263 blk_finish_plug(&plug); 264 265 if (!is_sync) 266 return -EIOCBQUEUED; 267 268 for (;;) { 269 set_current_state(TASK_UNINTERRUPTIBLE); 270 if (!READ_ONCE(dio->waiter)) 271 break; 272 blk_io_schedule(); 273 } 274 __set_current_state(TASK_RUNNING); 275 276 if (!ret) 277 ret = blk_status_to_errno(dio->bio.bi_status); 278 if (likely(!ret)) 279 ret = dio->size; 280 281 bio_put(&dio->bio); 282 return ret; 283 } 284 285 static void blkdev_bio_end_io_async(struct bio *bio) 286 { 287 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); 288 struct kiocb *iocb = dio->iocb; 289 ssize_t ret; 290 291 if (likely(!bio->bi_status)) { 292 ret = dio->size; 293 iocb->ki_pos += ret; 294 } else { 295 ret = blk_status_to_errno(bio->bi_status); 296 } 297 298 iocb->ki_complete(iocb, ret); 299 300 if (dio->flags & DIO_SHOULD_DIRTY) { 301 bio_check_pages_dirty(bio); 302 } else { 303 bio_release_pages(bio, false); 304 bio_put(bio); 305 } 306 } 307 308 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, 309 struct iov_iter *iter, 310 unsigned int nr_pages) 311 { 312 struct block_device *bdev = iocb->ki_filp->private_data; 313 struct blkdev_dio *dio; 314 struct bio *bio; 315 loff_t pos = iocb->ki_pos; 316 int ret = 0; 317 318 if ((pos | iov_iter_alignment(iter)) & 319 (bdev_logical_block_size(bdev) - 1)) 320 return -EINVAL; 321 322 bio = bio_alloc_kiocb(iocb, nr_pages, &blkdev_dio_pool); 323 dio = container_of(bio, struct blkdev_dio, bio); 324 dio->flags = 0; 325 dio->iocb = iocb; 326 bio_set_dev(bio, bdev); 327 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; 328 bio->bi_write_hint = iocb->ki_hint; 329 bio->bi_end_io = blkdev_bio_end_io_async; 330 bio->bi_ioprio = iocb->ki_ioprio; 331 332 if (iov_iter_is_bvec(iter)) { 333 /* 334 * Users don't rely on the iterator being in any particular 335 * state for async I/O returning -EIOCBQUEUED, hence we can 336 * avoid expensive iov_iter_advance(). Bypass 337 * bio_iov_iter_get_pages() and set the bvec directly. 338 */ 339 bio_iov_bvec_set(bio, iter); 340 } else { 341 ret = bio_iov_iter_get_pages(bio, iter); 342 if (unlikely(ret)) { 343 bio->bi_status = BLK_STS_IOERR; 344 bio_endio(bio); 345 return ret; 346 } 347 } 348 dio->size = bio->bi_iter.bi_size; 349 350 if (iov_iter_rw(iter) == READ) { 351 bio->bi_opf = REQ_OP_READ; 352 if (iter_is_iovec(iter)) { 353 dio->flags |= DIO_SHOULD_DIRTY; 354 bio_set_pages_dirty(bio); 355 } 356 } else { 357 bio->bi_opf = dio_bio_write_op(iocb); 358 task_io_account_write(bio->bi_iter.bi_size); 359 } 360 361 if (iocb->ki_flags & IOCB_HIPRI) { 362 bio->bi_opf |= REQ_POLLED | REQ_NOWAIT; 363 submit_bio(bio); 364 WRITE_ONCE(iocb->private, bio); 365 } else { 366 if (iocb->ki_flags & IOCB_NOWAIT) 367 bio->bi_opf |= REQ_NOWAIT; 368 submit_bio(bio); 369 } 370 return -EIOCBQUEUED; 371 } 372 373 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 374 { 375 unsigned int nr_pages; 376 377 if (!iov_iter_count(iter)) 378 return 0; 379 380 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); 381 if (likely(nr_pages <= BIO_MAX_VECS)) { 382 if (is_sync_kiocb(iocb)) 383 return __blkdev_direct_IO_simple(iocb, iter, nr_pages); 384 return __blkdev_direct_IO_async(iocb, iter, nr_pages); 385 } 386 return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages)); 387 } 388 389 static int blkdev_writepage(struct page *page, struct writeback_control *wbc) 390 { 391 return block_write_full_page(page, blkdev_get_block, wbc); 392 } 393 394 static int blkdev_readpage(struct file * file, struct page * page) 395 { 396 return block_read_full_page(page, blkdev_get_block); 397 } 398 399 static void blkdev_readahead(struct readahead_control *rac) 400 { 401 mpage_readahead(rac, blkdev_get_block); 402 } 403 404 static int blkdev_write_begin(struct file *file, struct address_space *mapping, 405 loff_t pos, unsigned len, unsigned flags, struct page **pagep, 406 void **fsdata) 407 { 408 return block_write_begin(mapping, pos, len, flags, pagep, 409 blkdev_get_block); 410 } 411 412 static int blkdev_write_end(struct file *file, struct address_space *mapping, 413 loff_t pos, unsigned len, unsigned copied, struct page *page, 414 void *fsdata) 415 { 416 int ret; 417 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); 418 419 unlock_page(page); 420 put_page(page); 421 422 return ret; 423 } 424 425 static int blkdev_writepages(struct address_space *mapping, 426 struct writeback_control *wbc) 427 { 428 return generic_writepages(mapping, wbc); 429 } 430 431 const struct address_space_operations def_blk_aops = { 432 .set_page_dirty = __set_page_dirty_buffers, 433 .readpage = blkdev_readpage, 434 .readahead = blkdev_readahead, 435 .writepage = blkdev_writepage, 436 .write_begin = blkdev_write_begin, 437 .write_end = blkdev_write_end, 438 .writepages = blkdev_writepages, 439 .direct_IO = blkdev_direct_IO, 440 .migratepage = buffer_migrate_page_norefs, 441 .is_dirty_writeback = buffer_check_dirty_writeback, 442 }; 443 444 /* 445 * for a block special file file_inode(file)->i_size is zero 446 * so we compute the size by hand (just as in block_read/write above) 447 */ 448 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) 449 { 450 struct inode *bd_inode = bdev_file_inode(file); 451 loff_t retval; 452 453 inode_lock(bd_inode); 454 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); 455 inode_unlock(bd_inode); 456 return retval; 457 } 458 459 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, 460 int datasync) 461 { 462 struct block_device *bdev = filp->private_data; 463 int error; 464 465 error = file_write_and_wait_range(filp, start, end); 466 if (error) 467 return error; 468 469 /* 470 * There is no need to serialise calls to blkdev_issue_flush with 471 * i_mutex and doing so causes performance issues with concurrent 472 * O_SYNC writers to a block device. 473 */ 474 error = blkdev_issue_flush(bdev); 475 if (error == -EOPNOTSUPP) 476 error = 0; 477 478 return error; 479 } 480 481 static int blkdev_open(struct inode *inode, struct file *filp) 482 { 483 struct block_device *bdev; 484 485 /* 486 * Preserve backwards compatibility and allow large file access 487 * even if userspace doesn't ask for it explicitly. Some mkfs 488 * binary needs it. We might want to drop this workaround 489 * during an unstable branch. 490 */ 491 filp->f_flags |= O_LARGEFILE; 492 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC; 493 494 if (filp->f_flags & O_NDELAY) 495 filp->f_mode |= FMODE_NDELAY; 496 if (filp->f_flags & O_EXCL) 497 filp->f_mode |= FMODE_EXCL; 498 if ((filp->f_flags & O_ACCMODE) == 3) 499 filp->f_mode |= FMODE_WRITE_IOCTL; 500 501 bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp); 502 if (IS_ERR(bdev)) 503 return PTR_ERR(bdev); 504 505 filp->private_data = bdev; 506 filp->f_mapping = bdev->bd_inode->i_mapping; 507 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); 508 return 0; 509 } 510 511 static int blkdev_close(struct inode *inode, struct file *filp) 512 { 513 struct block_device *bdev = filp->private_data; 514 515 blkdev_put(bdev, filp->f_mode); 516 return 0; 517 } 518 519 /* 520 * Write data to the block device. Only intended for the block device itself 521 * and the raw driver which basically is a fake block device. 522 * 523 * Does not take i_mutex for the write and thus is not for general purpose 524 * use. 525 */ 526 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) 527 { 528 struct block_device *bdev = iocb->ki_filp->private_data; 529 struct inode *bd_inode = bdev->bd_inode; 530 loff_t size = bdev_nr_bytes(bdev); 531 struct blk_plug plug; 532 size_t shorted = 0; 533 ssize_t ret; 534 535 if (bdev_read_only(bdev)) 536 return -EPERM; 537 538 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) 539 return -ETXTBSY; 540 541 if (!iov_iter_count(from)) 542 return 0; 543 544 if (iocb->ki_pos >= size) 545 return -ENOSPC; 546 547 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) 548 return -EOPNOTSUPP; 549 550 size -= iocb->ki_pos; 551 if (iov_iter_count(from) > size) { 552 shorted = iov_iter_count(from) - size; 553 iov_iter_truncate(from, size); 554 } 555 556 blk_start_plug(&plug); 557 ret = __generic_file_write_iter(iocb, from); 558 if (ret > 0) 559 ret = generic_write_sync(iocb, ret); 560 iov_iter_reexpand(from, iov_iter_count(from) + shorted); 561 blk_finish_plug(&plug); 562 return ret; 563 } 564 565 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) 566 { 567 struct block_device *bdev = iocb->ki_filp->private_data; 568 loff_t size = bdev_nr_bytes(bdev); 569 loff_t pos = iocb->ki_pos; 570 size_t shorted = 0; 571 ssize_t ret; 572 573 if (unlikely(pos + iov_iter_count(to) > size)) { 574 if (pos >= size) 575 return 0; 576 size -= pos; 577 if (iov_iter_count(to) > size) { 578 shorted = iov_iter_count(to) - size; 579 iov_iter_truncate(to, size); 580 } 581 } 582 583 ret = generic_file_read_iter(iocb, to); 584 585 if (unlikely(shorted)) 586 iov_iter_reexpand(to, iov_iter_count(to) + shorted); 587 return ret; 588 } 589 590 #define BLKDEV_FALLOC_FL_SUPPORTED \ 591 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ 592 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) 593 594 static long blkdev_fallocate(struct file *file, int mode, loff_t start, 595 loff_t len) 596 { 597 struct inode *inode = bdev_file_inode(file); 598 struct block_device *bdev = I_BDEV(inode); 599 loff_t end = start + len - 1; 600 loff_t isize; 601 int error; 602 603 /* Fail if we don't recognize the flags. */ 604 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) 605 return -EOPNOTSUPP; 606 607 /* Don't go off the end of the device. */ 608 isize = bdev_nr_bytes(bdev); 609 if (start >= isize) 610 return -EINVAL; 611 if (end >= isize) { 612 if (mode & FALLOC_FL_KEEP_SIZE) { 613 len = isize - start; 614 end = start + len - 1; 615 } else 616 return -EINVAL; 617 } 618 619 /* 620 * Don't allow IO that isn't aligned to logical block size. 621 */ 622 if ((start | len) & (bdev_logical_block_size(bdev) - 1)) 623 return -EINVAL; 624 625 filemap_invalidate_lock(inode->i_mapping); 626 627 /* Invalidate the page cache, including dirty pages. */ 628 error = truncate_bdev_range(bdev, file->f_mode, start, end); 629 if (error) 630 goto fail; 631 632 switch (mode) { 633 case FALLOC_FL_ZERO_RANGE: 634 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: 635 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 636 len >> SECTOR_SHIFT, GFP_KERNEL, 637 BLKDEV_ZERO_NOUNMAP); 638 break; 639 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: 640 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, 641 len >> SECTOR_SHIFT, GFP_KERNEL, 642 BLKDEV_ZERO_NOFALLBACK); 643 break; 644 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: 645 error = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, 646 len >> SECTOR_SHIFT, GFP_KERNEL, 0); 647 break; 648 default: 649 error = -EOPNOTSUPP; 650 } 651 652 fail: 653 filemap_invalidate_unlock(inode->i_mapping); 654 return error; 655 } 656 657 const struct file_operations def_blk_fops = { 658 .open = blkdev_open, 659 .release = blkdev_close, 660 .llseek = blkdev_llseek, 661 .read_iter = blkdev_read_iter, 662 .write_iter = blkdev_write_iter, 663 .iopoll = iocb_bio_iopoll, 664 .mmap = generic_file_mmap, 665 .fsync = blkdev_fsync, 666 .unlocked_ioctl = blkdev_ioctl, 667 #ifdef CONFIG_COMPAT 668 .compat_ioctl = compat_blkdev_ioctl, 669 #endif 670 .splice_read = generic_file_splice_read, 671 .splice_write = iter_file_splice_write, 672 .fallocate = blkdev_fallocate, 673 }; 674 675 static __init int blkdev_init(void) 676 { 677 return bioset_init(&blkdev_dio_pool, 4, 678 offsetof(struct blkdev_dio, bio), 679 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); 680 } 681 module_init(blkdev_init); 682