Line data Source code
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/module.h>
19 : #include "blk.h"
20 :
21 : static inline struct inode *bdev_file_inode(struct file *file)
22 : {
23 0 : return file->f_mapping->host;
24 : }
25 :
26 0 : static int blkdev_get_block(struct inode *inode, sector_t iblock,
27 : struct buffer_head *bh, int create)
28 : {
29 0 : bh->b_bdev = I_BDEV(inode);
30 0 : bh->b_blocknr = iblock;
31 0 : set_buffer_mapped(bh);
32 0 : return 0;
33 : }
34 :
35 : static blk_opf_t dio_bio_write_op(struct kiocb *iocb)
36 : {
37 0 : blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
38 :
39 : /* avoid the need for a I/O completion work item */
40 0 : if (iocb_is_dsync(iocb))
41 0 : opf |= REQ_FUA;
42 : return opf;
43 : }
44 :
45 0 : static bool blkdev_dio_unaligned(struct block_device *bdev, loff_t pos,
46 : struct iov_iter *iter)
47 : {
48 0 : return pos & (bdev_logical_block_size(bdev) - 1) ||
49 0 : !bdev_iter_is_aligned(bdev, iter);
50 : }
51 :
52 : #define DIO_INLINE_BIO_VECS 4
53 :
54 0 : static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
55 : struct iov_iter *iter, unsigned int nr_pages)
56 : {
57 0 : struct block_device *bdev = iocb->ki_filp->private_data;
58 : struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
59 0 : loff_t pos = iocb->ki_pos;
60 0 : bool should_dirty = false;
61 : struct bio bio;
62 : ssize_t ret;
63 :
64 0 : if (blkdev_dio_unaligned(bdev, pos, iter))
65 : return -EINVAL;
66 :
67 0 : if (nr_pages <= DIO_INLINE_BIO_VECS)
68 : vecs = inline_vecs;
69 : else {
70 0 : vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
71 : GFP_KERNEL);
72 0 : if (!vecs)
73 : return -ENOMEM;
74 : }
75 :
76 0 : if (iov_iter_rw(iter) == READ) {
77 0 : bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ);
78 0 : if (user_backed_iter(iter))
79 0 : should_dirty = true;
80 : } else {
81 0 : bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb));
82 : }
83 0 : bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
84 0 : bio.bi_ioprio = iocb->ki_ioprio;
85 :
86 0 : ret = bio_iov_iter_get_pages(&bio, iter);
87 0 : if (unlikely(ret))
88 : goto out;
89 0 : ret = bio.bi_iter.bi_size;
90 :
91 0 : if (iov_iter_rw(iter) == WRITE)
92 : task_io_account_write(ret);
93 :
94 0 : if (iocb->ki_flags & IOCB_NOWAIT)
95 0 : bio.bi_opf |= REQ_NOWAIT;
96 :
97 0 : submit_bio_wait(&bio);
98 :
99 0 : bio_release_pages(&bio, should_dirty);
100 0 : if (unlikely(bio.bi_status))
101 0 : ret = blk_status_to_errno(bio.bi_status);
102 :
103 : out:
104 0 : if (vecs != inline_vecs)
105 0 : kfree(vecs);
106 :
107 0 : bio_uninit(&bio);
108 :
109 0 : return ret;
110 : }
111 :
112 : enum {
113 : DIO_SHOULD_DIRTY = 1,
114 : DIO_IS_SYNC = 2,
115 : };
116 :
117 : struct blkdev_dio {
118 : union {
119 : struct kiocb *iocb;
120 : struct task_struct *waiter;
121 : };
122 : size_t size;
123 : atomic_t ref;
124 : unsigned int flags;
125 : struct bio bio ____cacheline_aligned_in_smp;
126 : };
127 :
128 : static struct bio_set blkdev_dio_pool;
129 :
130 0 : static void blkdev_bio_end_io(struct bio *bio)
131 : {
132 0 : struct blkdev_dio *dio = bio->bi_private;
133 0 : bool should_dirty = dio->flags & DIO_SHOULD_DIRTY;
134 :
135 0 : if (bio->bi_status && !dio->bio.bi_status)
136 0 : dio->bio.bi_status = bio->bi_status;
137 :
138 0 : if (atomic_dec_and_test(&dio->ref)) {
139 0 : if (!(dio->flags & DIO_IS_SYNC)) {
140 0 : struct kiocb *iocb = dio->iocb;
141 : ssize_t ret;
142 :
143 0 : WRITE_ONCE(iocb->private, NULL);
144 :
145 0 : if (likely(!dio->bio.bi_status)) {
146 0 : ret = dio->size;
147 0 : iocb->ki_pos += ret;
148 : } else {
149 0 : ret = blk_status_to_errno(dio->bio.bi_status);
150 : }
151 :
152 0 : dio->iocb->ki_complete(iocb, ret);
153 0 : bio_put(&dio->bio);
154 : } else {
155 0 : struct task_struct *waiter = dio->waiter;
156 :
157 0 : WRITE_ONCE(dio->waiter, NULL);
158 : blk_wake_io_task(waiter);
159 : }
160 : }
161 :
162 0 : if (should_dirty) {
163 0 : bio_check_pages_dirty(bio);
164 : } else {
165 0 : bio_release_pages(bio, false);
166 0 : bio_put(bio);
167 : }
168 0 : }
169 :
170 0 : static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
171 : unsigned int nr_pages)
172 : {
173 0 : struct block_device *bdev = iocb->ki_filp->private_data;
174 : struct blk_plug plug;
175 : struct blkdev_dio *dio;
176 : struct bio *bio;
177 0 : bool is_read = (iov_iter_rw(iter) == READ), is_sync;
178 0 : blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
179 0 : loff_t pos = iocb->ki_pos;
180 0 : int ret = 0;
181 :
182 0 : if (blkdev_dio_unaligned(bdev, pos, iter))
183 : return -EINVAL;
184 :
185 0 : if (iocb->ki_flags & IOCB_ALLOC_CACHE)
186 0 : opf |= REQ_ALLOC_CACHE;
187 0 : bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
188 : &blkdev_dio_pool);
189 0 : dio = container_of(bio, struct blkdev_dio, bio);
190 0 : atomic_set(&dio->ref, 1);
191 : /*
192 : * Grab an extra reference to ensure the dio structure which is embedded
193 : * into the first bio stays around.
194 : */
195 0 : bio_get(bio);
196 :
197 0 : is_sync = is_sync_kiocb(iocb);
198 0 : if (is_sync) {
199 0 : dio->flags = DIO_IS_SYNC;
200 0 : dio->waiter = current;
201 : } else {
202 0 : dio->flags = 0;
203 0 : dio->iocb = iocb;
204 : }
205 :
206 0 : dio->size = 0;
207 0 : if (is_read && user_backed_iter(iter))
208 0 : dio->flags |= DIO_SHOULD_DIRTY;
209 :
210 0 : blk_start_plug(&plug);
211 :
212 : for (;;) {
213 0 : bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
214 0 : bio->bi_private = dio;
215 0 : bio->bi_end_io = blkdev_bio_end_io;
216 0 : bio->bi_ioprio = iocb->ki_ioprio;
217 :
218 0 : ret = bio_iov_iter_get_pages(bio, iter);
219 0 : if (unlikely(ret)) {
220 0 : bio->bi_status = BLK_STS_IOERR;
221 0 : bio_endio(bio);
222 0 : break;
223 : }
224 0 : 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 0 : if (unlikely(iov_iter_count(iter))) {
234 0 : bio_release_pages(bio, false);
235 0 : bio_clear_flag(bio, BIO_REFFED);
236 0 : bio_put(bio);
237 0 : blk_finish_plug(&plug);
238 0 : return -EAGAIN;
239 : }
240 0 : bio->bi_opf |= REQ_NOWAIT;
241 : }
242 :
243 0 : if (is_read) {
244 0 : if (dio->flags & DIO_SHOULD_DIRTY)
245 0 : bio_set_pages_dirty(bio);
246 : } else {
247 : task_io_account_write(bio->bi_iter.bi_size);
248 : }
249 0 : dio->size += bio->bi_iter.bi_size;
250 0 : pos += bio->bi_iter.bi_size;
251 :
252 0 : nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
253 0 : if (!nr_pages) {
254 0 : submit_bio(bio);
255 0 : break;
256 : }
257 0 : atomic_inc(&dio->ref);
258 0 : submit_bio(bio);
259 0 : bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL);
260 : }
261 :
262 0 : blk_finish_plug(&plug);
263 :
264 0 : if (!is_sync)
265 : return -EIOCBQUEUED;
266 :
267 : for (;;) {
268 0 : set_current_state(TASK_UNINTERRUPTIBLE);
269 0 : if (!READ_ONCE(dio->waiter))
270 : break;
271 0 : blk_io_schedule();
272 : }
273 0 : __set_current_state(TASK_RUNNING);
274 :
275 0 : if (!ret)
276 0 : ret = blk_status_to_errno(dio->bio.bi_status);
277 0 : if (likely(!ret))
278 0 : ret = dio->size;
279 :
280 0 : bio_put(&dio->bio);
281 0 : return ret;
282 : }
283 :
284 0 : static void blkdev_bio_end_io_async(struct bio *bio)
285 : {
286 0 : struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio);
287 0 : struct kiocb *iocb = dio->iocb;
288 : ssize_t ret;
289 :
290 0 : WRITE_ONCE(iocb->private, NULL);
291 :
292 0 : if (likely(!bio->bi_status)) {
293 0 : ret = dio->size;
294 0 : iocb->ki_pos += ret;
295 : } else {
296 0 : ret = blk_status_to_errno(bio->bi_status);
297 : }
298 :
299 0 : iocb->ki_complete(iocb, ret);
300 :
301 0 : if (dio->flags & DIO_SHOULD_DIRTY) {
302 0 : bio_check_pages_dirty(bio);
303 : } else {
304 0 : bio_release_pages(bio, false);
305 0 : bio_put(bio);
306 : }
307 0 : }
308 :
309 0 : static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
310 : struct iov_iter *iter,
311 : unsigned int nr_pages)
312 : {
313 0 : struct block_device *bdev = iocb->ki_filp->private_data;
314 0 : bool is_read = iov_iter_rw(iter) == READ;
315 0 : blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
316 : struct blkdev_dio *dio;
317 : struct bio *bio;
318 0 : loff_t pos = iocb->ki_pos;
319 0 : int ret = 0;
320 :
321 0 : if (blkdev_dio_unaligned(bdev, pos, iter))
322 : return -EINVAL;
323 :
324 0 : if (iocb->ki_flags & IOCB_ALLOC_CACHE)
325 0 : opf |= REQ_ALLOC_CACHE;
326 0 : bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
327 : &blkdev_dio_pool);
328 0 : dio = container_of(bio, struct blkdev_dio, bio);
329 0 : dio->flags = 0;
330 0 : dio->iocb = iocb;
331 0 : bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
332 0 : bio->bi_end_io = blkdev_bio_end_io_async;
333 0 : bio->bi_ioprio = iocb->ki_ioprio;
334 :
335 0 : if (iov_iter_is_bvec(iter)) {
336 : /*
337 : * Users don't rely on the iterator being in any particular
338 : * state for async I/O returning -EIOCBQUEUED, hence we can
339 : * avoid expensive iov_iter_advance(). Bypass
340 : * bio_iov_iter_get_pages() and set the bvec directly.
341 : */
342 0 : bio_iov_bvec_set(bio, iter);
343 : } else {
344 0 : ret = bio_iov_iter_get_pages(bio, iter);
345 0 : if (unlikely(ret)) {
346 0 : bio_put(bio);
347 0 : return ret;
348 : }
349 : }
350 0 : dio->size = bio->bi_iter.bi_size;
351 :
352 0 : if (is_read) {
353 0 : if (user_backed_iter(iter)) {
354 0 : dio->flags |= DIO_SHOULD_DIRTY;
355 0 : bio_set_pages_dirty(bio);
356 : }
357 : } else {
358 : task_io_account_write(bio->bi_iter.bi_size);
359 : }
360 :
361 0 : if (iocb->ki_flags & IOCB_HIPRI) {
362 0 : bio->bi_opf |= REQ_POLLED | REQ_NOWAIT;
363 0 : submit_bio(bio);
364 0 : WRITE_ONCE(iocb->private, bio);
365 : } else {
366 0 : if (iocb->ki_flags & IOCB_NOWAIT)
367 0 : bio->bi_opf |= REQ_NOWAIT;
368 0 : submit_bio(bio);
369 : }
370 : return -EIOCBQUEUED;
371 : }
372 :
373 0 : static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
374 : {
375 : unsigned int nr_pages;
376 :
377 0 : if (!iov_iter_count(iter))
378 : return 0;
379 :
380 0 : nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
381 0 : if (likely(nr_pages <= BIO_MAX_VECS)) {
382 0 : if (is_sync_kiocb(iocb))
383 0 : return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
384 0 : return __blkdev_direct_IO_async(iocb, iter, nr_pages);
385 : }
386 0 : return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
387 : }
388 :
389 0 : static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
390 : {
391 0 : return block_write_full_page(page, blkdev_get_block, wbc);
392 : }
393 :
394 0 : static int blkdev_read_folio(struct file *file, struct folio *folio)
395 : {
396 0 : return block_read_full_folio(folio, blkdev_get_block);
397 : }
398 :
399 0 : static void blkdev_readahead(struct readahead_control *rac)
400 : {
401 0 : mpage_readahead(rac, blkdev_get_block);
402 0 : }
403 :
404 0 : static int blkdev_write_begin(struct file *file, struct address_space *mapping,
405 : loff_t pos, unsigned len, struct page **pagep, void **fsdata)
406 : {
407 0 : return block_write_begin(mapping, pos, len, pagep, blkdev_get_block);
408 : }
409 :
410 0 : static int blkdev_write_end(struct file *file, struct address_space *mapping,
411 : loff_t pos, unsigned len, unsigned copied, struct page *page,
412 : void *fsdata)
413 : {
414 : int ret;
415 0 : ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
416 :
417 0 : unlock_page(page);
418 0 : put_page(page);
419 :
420 0 : return ret;
421 : }
422 :
423 : const struct address_space_operations def_blk_aops = {
424 : .dirty_folio = block_dirty_folio,
425 : .invalidate_folio = block_invalidate_folio,
426 : .read_folio = blkdev_read_folio,
427 : .readahead = blkdev_readahead,
428 : .writepage = blkdev_writepage,
429 : .write_begin = blkdev_write_begin,
430 : .write_end = blkdev_write_end,
431 : .direct_IO = blkdev_direct_IO,
432 : .migrate_folio = buffer_migrate_folio_norefs,
433 : .is_dirty_writeback = buffer_check_dirty_writeback,
434 : };
435 :
436 : /*
437 : * for a block special file file_inode(file)->i_size is zero
438 : * so we compute the size by hand (just as in block_read/write above)
439 : */
440 0 : static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
441 : {
442 0 : struct inode *bd_inode = bdev_file_inode(file);
443 : loff_t retval;
444 :
445 0 : inode_lock(bd_inode);
446 0 : retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
447 0 : inode_unlock(bd_inode);
448 0 : return retval;
449 : }
450 :
451 0 : static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
452 : int datasync)
453 : {
454 0 : struct block_device *bdev = filp->private_data;
455 : int error;
456 :
457 0 : error = file_write_and_wait_range(filp, start, end);
458 0 : if (error)
459 : return error;
460 :
461 : /*
462 : * There is no need to serialise calls to blkdev_issue_flush with
463 : * i_mutex and doing so causes performance issues with concurrent
464 : * O_SYNC writers to a block device.
465 : */
466 0 : error = blkdev_issue_flush(bdev);
467 0 : if (error == -EOPNOTSUPP)
468 0 : error = 0;
469 :
470 : return error;
471 : }
472 :
473 0 : static int blkdev_open(struct inode *inode, struct file *filp)
474 : {
475 : struct block_device *bdev;
476 :
477 : /*
478 : * Preserve backwards compatibility and allow large file access
479 : * even if userspace doesn't ask for it explicitly. Some mkfs
480 : * binary needs it. We might want to drop this workaround
481 : * during an unstable branch.
482 : */
483 0 : filp->f_flags |= O_LARGEFILE;
484 0 : filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
485 :
486 0 : if (filp->f_flags & O_NDELAY)
487 0 : filp->f_mode |= FMODE_NDELAY;
488 0 : if (filp->f_flags & O_EXCL)
489 0 : filp->f_mode |= FMODE_EXCL;
490 0 : if ((filp->f_flags & O_ACCMODE) == 3)
491 0 : filp->f_mode |= FMODE_WRITE_IOCTL;
492 :
493 0 : bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
494 0 : if (IS_ERR(bdev))
495 0 : return PTR_ERR(bdev);
496 :
497 0 : filp->private_data = bdev;
498 0 : filp->f_mapping = bdev->bd_inode->i_mapping;
499 0 : filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
500 0 : return 0;
501 : }
502 :
503 0 : static int blkdev_close(struct inode *inode, struct file *filp)
504 : {
505 0 : struct block_device *bdev = filp->private_data;
506 :
507 0 : blkdev_put(bdev, filp->f_mode);
508 0 : return 0;
509 : }
510 :
511 : /*
512 : * Write data to the block device. Only intended for the block device itself
513 : * and the raw driver which basically is a fake block device.
514 : *
515 : * Does not take i_mutex for the write and thus is not for general purpose
516 : * use.
517 : */
518 0 : static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
519 : {
520 0 : struct block_device *bdev = iocb->ki_filp->private_data;
521 0 : struct inode *bd_inode = bdev->bd_inode;
522 0 : loff_t size = bdev_nr_bytes(bdev);
523 : struct blk_plug plug;
524 0 : size_t shorted = 0;
525 : ssize_t ret;
526 :
527 0 : if (bdev_read_only(bdev))
528 : return -EPERM;
529 :
530 0 : if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
531 : return -ETXTBSY;
532 :
533 0 : if (!iov_iter_count(from))
534 : return 0;
535 :
536 0 : if (iocb->ki_pos >= size)
537 : return -ENOSPC;
538 :
539 0 : if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
540 : return -EOPNOTSUPP;
541 :
542 0 : size -= iocb->ki_pos;
543 0 : if (iov_iter_count(from) > size) {
544 0 : shorted = iov_iter_count(from) - size;
545 0 : iov_iter_truncate(from, size);
546 : }
547 :
548 0 : blk_start_plug(&plug);
549 0 : ret = __generic_file_write_iter(iocb, from);
550 0 : if (ret > 0)
551 0 : ret = generic_write_sync(iocb, ret);
552 0 : iov_iter_reexpand(from, iov_iter_count(from) + shorted);
553 0 : blk_finish_plug(&plug);
554 0 : return ret;
555 : }
556 :
557 0 : static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
558 : {
559 0 : struct block_device *bdev = iocb->ki_filp->private_data;
560 0 : loff_t size = bdev_nr_bytes(bdev);
561 0 : loff_t pos = iocb->ki_pos;
562 0 : size_t shorted = 0;
563 0 : ssize_t ret = 0;
564 : size_t count;
565 :
566 0 : if (unlikely(pos + iov_iter_count(to) > size)) {
567 0 : if (pos >= size)
568 : return 0;
569 0 : size -= pos;
570 0 : shorted = iov_iter_count(to) - size;
571 0 : iov_iter_truncate(to, size);
572 : }
573 :
574 0 : count = iov_iter_count(to);
575 0 : if (!count)
576 : goto reexpand; /* skip atime */
577 :
578 0 : if (iocb->ki_flags & IOCB_DIRECT) {
579 0 : struct address_space *mapping = iocb->ki_filp->f_mapping;
580 :
581 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
582 0 : if (filemap_range_needs_writeback(mapping, pos,
583 0 : pos + count - 1)) {
584 : ret = -EAGAIN;
585 : goto reexpand;
586 : }
587 : } else {
588 0 : ret = filemap_write_and_wait_range(mapping, pos,
589 0 : pos + count - 1);
590 0 : if (ret < 0)
591 : goto reexpand;
592 : }
593 :
594 0 : file_accessed(iocb->ki_filp);
595 :
596 0 : ret = blkdev_direct_IO(iocb, to);
597 0 : if (ret >= 0) {
598 0 : iocb->ki_pos += ret;
599 0 : count -= ret;
600 : }
601 0 : iov_iter_revert(to, count - iov_iter_count(to));
602 0 : if (ret < 0 || !count)
603 : goto reexpand;
604 : }
605 :
606 0 : ret = filemap_read(iocb, to, ret);
607 :
608 : reexpand:
609 0 : if (unlikely(shorted))
610 0 : iov_iter_reexpand(to, iov_iter_count(to) + shorted);
611 : return ret;
612 : }
613 :
614 : #define BLKDEV_FALLOC_FL_SUPPORTED \
615 : (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
616 : FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
617 :
618 0 : static long blkdev_fallocate(struct file *file, int mode, loff_t start,
619 : loff_t len)
620 : {
621 0 : struct inode *inode = bdev_file_inode(file);
622 0 : struct block_device *bdev = I_BDEV(inode);
623 0 : loff_t end = start + len - 1;
624 : loff_t isize;
625 : int error;
626 :
627 : /* Fail if we don't recognize the flags. */
628 0 : if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
629 : return -EOPNOTSUPP;
630 :
631 : /* Don't go off the end of the device. */
632 0 : isize = bdev_nr_bytes(bdev);
633 0 : if (start >= isize)
634 : return -EINVAL;
635 0 : if (end >= isize) {
636 0 : if (mode & FALLOC_FL_KEEP_SIZE) {
637 0 : len = isize - start;
638 0 : end = start + len - 1;
639 : } else
640 : return -EINVAL;
641 : }
642 :
643 : /*
644 : * Don't allow IO that isn't aligned to logical block size.
645 : */
646 0 : if ((start | len) & (bdev_logical_block_size(bdev) - 1))
647 : return -EINVAL;
648 :
649 0 : filemap_invalidate_lock(inode->i_mapping);
650 :
651 : /* Invalidate the page cache, including dirty pages. */
652 0 : error = truncate_bdev_range(bdev, file->f_mode, start, end);
653 0 : if (error)
654 : goto fail;
655 :
656 0 : switch (mode) {
657 : case FALLOC_FL_ZERO_RANGE:
658 : case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
659 0 : error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
660 0 : len >> SECTOR_SHIFT, GFP_KERNEL,
661 : BLKDEV_ZERO_NOUNMAP);
662 0 : break;
663 : case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
664 0 : error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
665 0 : len >> SECTOR_SHIFT, GFP_KERNEL,
666 : BLKDEV_ZERO_NOFALLBACK);
667 0 : break;
668 : case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
669 0 : error = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
670 0 : len >> SECTOR_SHIFT, GFP_KERNEL);
671 0 : break;
672 : default:
673 : error = -EOPNOTSUPP;
674 : }
675 :
676 : fail:
677 0 : filemap_invalidate_unlock(inode->i_mapping);
678 0 : return error;
679 : }
680 :
681 : const struct file_operations def_blk_fops = {
682 : .open = blkdev_open,
683 : .release = blkdev_close,
684 : .llseek = blkdev_llseek,
685 : .read_iter = blkdev_read_iter,
686 : .write_iter = blkdev_write_iter,
687 : .iopoll = iocb_bio_iopoll,
688 : .mmap = generic_file_mmap,
689 : .fsync = blkdev_fsync,
690 : .unlocked_ioctl = blkdev_ioctl,
691 : #ifdef CONFIG_COMPAT
692 : .compat_ioctl = compat_blkdev_ioctl,
693 : #endif
694 : .splice_read = generic_file_splice_read,
695 : .splice_write = iter_file_splice_write,
696 : .fallocate = blkdev_fallocate,
697 : };
698 :
699 1 : static __init int blkdev_init(void)
700 : {
701 1 : return bioset_init(&blkdev_dio_pool, 4,
702 : offsetof(struct blkdev_dio, bio),
703 : BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
704 : }
705 : module_init(blkdev_init);
|
