2 * Copyright (C) 2010 Red Hat, Inc.
3 * Copyright (c) 2016-2018 Christoph Hellwig.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 #include <linux/module.h>
15 #include <linux/compiler.h>
17 #include <linux/iomap.h>
18 #include <linux/uaccess.h>
19 #include <linux/gfp.h>
20 #include <linux/migrate.h>
22 #include <linux/mm_inline.h>
23 #include <linux/swap.h>
24 #include <linux/pagemap.h>
25 #include <linux/pagevec.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/backing-dev.h>
29 #include <linux/buffer_head.h>
30 #include <linux/task_io_accounting_ops.h>
31 #include <linux/dax.h>
32 #include <linux/sched/signal.h>
37 * Execute a iomap write on a segment of the mapping that spans a
38 * contiguous range of pages that have identical block mapping state.
40 * This avoids the need to map pages individually, do individual allocations
41 * for each page and most importantly avoid the need for filesystem specific
42 * locking per page. Instead, all the operations are amortised over the entire
43 * range of pages. It is assumed that the filesystems will lock whatever
44 * resources they require in the iomap_begin call, and release them in the
48 iomap_apply(struct inode
*inode
, loff_t pos
, loff_t length
, unsigned flags
,
49 const struct iomap_ops
*ops
, void *data
, iomap_actor_t actor
)
51 struct iomap iomap
= { 0 };
52 loff_t written
= 0, ret
;
55 * Need to map a range from start position for length bytes. This can
56 * span multiple pages - it is only guaranteed to return a range of a
57 * single type of pages (e.g. all into a hole, all mapped or all
58 * unwritten). Failure at this point has nothing to undo.
60 * If allocation is required for this range, reserve the space now so
61 * that the allocation is guaranteed to succeed later on. Once we copy
62 * the data into the page cache pages, then we cannot fail otherwise we
63 * expose transient stale data. If the reserve fails, we can safely
64 * back out at this point as there is nothing to undo.
66 ret
= ops
->iomap_begin(inode
, pos
, length
, flags
, &iomap
);
69 if (WARN_ON(iomap
.offset
> pos
))
71 if (WARN_ON(iomap
.length
== 0))
75 * Cut down the length to the one actually provided by the filesystem,
76 * as it might not be able to give us the whole size that we requested.
78 if (iomap
.offset
+ iomap
.length
< pos
+ length
)
79 length
= iomap
.offset
+ iomap
.length
- pos
;
82 * Now that we have guaranteed that the space allocation will succeed.
83 * we can do the copy-in page by page without having to worry about
84 * failures exposing transient data.
86 written
= actor(inode
, pos
, length
, data
, &iomap
);
89 * Now the data has been copied, commit the range we've copied. This
90 * should not fail unless the filesystem has had a fatal error.
93 ret
= ops
->iomap_end(inode
, pos
, length
,
94 written
> 0 ? written
: 0,
98 return written
? written
: ret
;
102 iomap_sector(struct iomap
*iomap
, loff_t pos
)
104 return (iomap
->addr
+ pos
- iomap
->offset
) >> SECTOR_SHIFT
;
107 static struct iomap_page
*
108 iomap_page_create(struct inode
*inode
, struct page
*page
)
110 struct iomap_page
*iop
= to_iomap_page(page
);
112 if (iop
|| i_blocksize(inode
) == PAGE_SIZE
)
115 iop
= kmalloc(sizeof(*iop
), GFP_NOFS
| __GFP_NOFAIL
);
116 atomic_set(&iop
->read_count
, 0);
117 atomic_set(&iop
->write_count
, 0);
118 bitmap_zero(iop
->uptodate
, PAGE_SIZE
/ SECTOR_SIZE
);
119 set_page_private(page
, (unsigned long)iop
);
120 SetPagePrivate(page
);
125 iomap_page_release(struct page
*page
)
127 struct iomap_page
*iop
= to_iomap_page(page
);
131 WARN_ON_ONCE(atomic_read(&iop
->read_count
));
132 WARN_ON_ONCE(atomic_read(&iop
->write_count
));
133 ClearPagePrivate(page
);
134 set_page_private(page
, 0);
139 * Calculate the range inside the page that we actually need to read.
142 iomap_adjust_read_range(struct inode
*inode
, struct iomap_page
*iop
,
143 loff_t
*pos
, loff_t length
, unsigned *offp
, unsigned *lenp
)
145 unsigned block_bits
= inode
->i_blkbits
;
146 unsigned block_size
= (1 << block_bits
);
147 unsigned poff
= offset_in_page(*pos
);
148 unsigned plen
= min_t(loff_t
, PAGE_SIZE
- poff
, length
);
149 unsigned first
= poff
>> block_bits
;
150 unsigned last
= (poff
+ plen
- 1) >> block_bits
;
151 unsigned end
= offset_in_page(i_size_read(inode
)) >> block_bits
;
154 * If the block size is smaller than the page size we need to check the
155 * per-block uptodate status and adjust the offset and length if needed
156 * to avoid reading in already uptodate ranges.
161 /* move forward for each leading block marked uptodate */
162 for (i
= first
; i
<= last
; i
++) {
163 if (!test_bit(i
, iop
->uptodate
))
171 /* truncate len if we find any trailing uptodate block(s) */
172 for ( ; i
<= last
; i
++) {
173 if (test_bit(i
, iop
->uptodate
)) {
174 plen
-= (last
- i
+ 1) * block_size
;
182 * If the extent spans the block that contains the i_size we need to
183 * handle both halves separately so that we properly zero data in the
184 * page cache for blocks that are entirely outside of i_size.
186 if (first
<= end
&& last
> end
)
187 plen
-= (last
- end
) * block_size
;
194 iomap_set_range_uptodate(struct page
*page
, unsigned off
, unsigned len
)
196 struct iomap_page
*iop
= to_iomap_page(page
);
197 struct inode
*inode
= page
->mapping
->host
;
198 unsigned first
= off
>> inode
->i_blkbits
;
199 unsigned last
= (off
+ len
- 1) >> inode
->i_blkbits
;
201 bool uptodate
= true;
204 for (i
= 0; i
< PAGE_SIZE
/ i_blocksize(inode
); i
++) {
205 if (i
>= first
&& i
<= last
)
206 set_bit(i
, iop
->uptodate
);
207 else if (!test_bit(i
, iop
->uptodate
))
212 if (uptodate
&& !PageError(page
))
213 SetPageUptodate(page
);
217 iomap_read_finish(struct iomap_page
*iop
, struct page
*page
)
219 if (!iop
|| atomic_dec_and_test(&iop
->read_count
))
224 iomap_read_page_end_io(struct bio_vec
*bvec
, int error
)
226 struct page
*page
= bvec
->bv_page
;
227 struct iomap_page
*iop
= to_iomap_page(page
);
229 if (unlikely(error
)) {
230 ClearPageUptodate(page
);
233 iomap_set_range_uptodate(page
, bvec
->bv_offset
, bvec
->bv_len
);
236 iomap_read_finish(iop
, page
);
240 iomap_read_inline_data(struct inode
*inode
, struct page
*page
,
243 size_t size
= i_size_read(inode
);
246 if (PageUptodate(page
))
250 BUG_ON(size
> PAGE_SIZE
- offset_in_page(iomap
->inline_data
));
252 addr
= kmap_atomic(page
);
253 memcpy(addr
, iomap
->inline_data
, size
);
254 memset(addr
+ size
, 0, PAGE_SIZE
- size
);
256 SetPageUptodate(page
);
260 iomap_read_end_io(struct bio
*bio
)
262 int error
= blk_status_to_errno(bio
->bi_status
);
263 struct bio_vec
*bvec
;
266 bio_for_each_segment_all(bvec
, bio
, i
)
267 iomap_read_page_end_io(bvec
, error
);
271 struct iomap_readpage_ctx
{
272 struct page
*cur_page
;
273 bool cur_page_in_bio
;
276 struct list_head
*pages
;
280 iomap_readpage_actor(struct inode
*inode
, loff_t pos
, loff_t length
, void *data
,
283 struct iomap_readpage_ctx
*ctx
= data
;
284 struct page
*page
= ctx
->cur_page
;
285 struct iomap_page
*iop
= iomap_page_create(inode
, page
);
286 bool is_contig
= false;
287 loff_t orig_pos
= pos
;
291 if (iomap
->type
== IOMAP_INLINE
) {
293 iomap_read_inline_data(inode
, page
, iomap
);
297 /* zero post-eof blocks as the page may be mapped */
298 iomap_adjust_read_range(inode
, iop
, &pos
, length
, &poff
, &plen
);
302 if (iomap
->type
!= IOMAP_MAPPED
|| pos
>= i_size_read(inode
)) {
303 zero_user(page
, poff
, plen
);
304 iomap_set_range_uptodate(page
, poff
, plen
);
308 ctx
->cur_page_in_bio
= true;
311 * Try to merge into a previous segment if we can.
313 sector
= iomap_sector(iomap
, pos
);
314 if (ctx
->bio
&& bio_end_sector(ctx
->bio
) == sector
) {
315 if (__bio_try_merge_page(ctx
->bio
, page
, plen
, poff
))
321 * If we start a new segment we need to increase the read count, and we
322 * need to do so before submitting any previous full bio to make sure
323 * that we don't prematurely unlock the page.
326 atomic_inc(&iop
->read_count
);
328 if (!ctx
->bio
|| !is_contig
|| bio_full(ctx
->bio
)) {
329 gfp_t gfp
= mapping_gfp_constraint(page
->mapping
, GFP_KERNEL
);
330 int nr_vecs
= (length
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
333 submit_bio(ctx
->bio
);
335 if (ctx
->is_readahead
) /* same as readahead_gfp_mask */
336 gfp
|= __GFP_NORETRY
| __GFP_NOWARN
;
337 ctx
->bio
= bio_alloc(gfp
, min(BIO_MAX_PAGES
, nr_vecs
));
338 ctx
->bio
->bi_opf
= REQ_OP_READ
;
339 if (ctx
->is_readahead
)
340 ctx
->bio
->bi_opf
|= REQ_RAHEAD
;
341 ctx
->bio
->bi_iter
.bi_sector
= sector
;
342 bio_set_dev(ctx
->bio
, iomap
->bdev
);
343 ctx
->bio
->bi_end_io
= iomap_read_end_io
;
346 __bio_add_page(ctx
->bio
, page
, plen
, poff
);
349 * Move the caller beyond our range so that it keeps making progress.
350 * For that we have to include any leading non-uptodate ranges, but
351 * we can skip trailing ones as they will be handled in the next
354 return pos
- orig_pos
+ plen
;
358 iomap_readpage(struct page
*page
, const struct iomap_ops
*ops
)
360 struct iomap_readpage_ctx ctx
= { .cur_page
= page
};
361 struct inode
*inode
= page
->mapping
->host
;
365 for (poff
= 0; poff
< PAGE_SIZE
; poff
+= ret
) {
366 ret
= iomap_apply(inode
, page_offset(page
) + poff
,
367 PAGE_SIZE
- poff
, 0, ops
, &ctx
,
368 iomap_readpage_actor
);
370 WARN_ON_ONCE(ret
== 0);
378 WARN_ON_ONCE(!ctx
.cur_page_in_bio
);
380 WARN_ON_ONCE(ctx
.cur_page_in_bio
);
385 * Just like mpage_readpages and block_read_full_page we always
386 * return 0 and just mark the page as PageError on errors. This
387 * should be cleaned up all through the stack eventually.
391 EXPORT_SYMBOL_GPL(iomap_readpage
);
394 iomap_next_page(struct inode
*inode
, struct list_head
*pages
, loff_t pos
,
395 loff_t length
, loff_t
*done
)
397 while (!list_empty(pages
)) {
398 struct page
*page
= lru_to_page(pages
);
400 if (page_offset(page
) >= (u64
)pos
+ length
)
403 list_del(&page
->lru
);
404 if (!add_to_page_cache_lru(page
, inode
->i_mapping
, page
->index
,
409 * If we already have a page in the page cache at index we are
410 * done. Upper layers don't care if it is uptodate after the
411 * readpages call itself as every page gets checked again once
422 iomap_readpages_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
423 void *data
, struct iomap
*iomap
)
425 struct iomap_readpage_ctx
*ctx
= data
;
428 for (done
= 0; done
< length
; done
+= ret
) {
429 if (ctx
->cur_page
&& offset_in_page(pos
+ done
) == 0) {
430 if (!ctx
->cur_page_in_bio
)
431 unlock_page(ctx
->cur_page
);
432 put_page(ctx
->cur_page
);
433 ctx
->cur_page
= NULL
;
435 if (!ctx
->cur_page
) {
436 ctx
->cur_page
= iomap_next_page(inode
, ctx
->pages
,
440 ctx
->cur_page_in_bio
= false;
442 ret
= iomap_readpage_actor(inode
, pos
+ done
, length
- done
,
450 iomap_readpages(struct address_space
*mapping
, struct list_head
*pages
,
451 unsigned nr_pages
, const struct iomap_ops
*ops
)
453 struct iomap_readpage_ctx ctx
= {
455 .is_readahead
= true,
457 loff_t pos
= page_offset(list_entry(pages
->prev
, struct page
, lru
));
458 loff_t last
= page_offset(list_entry(pages
->next
, struct page
, lru
));
459 loff_t length
= last
- pos
+ PAGE_SIZE
, ret
= 0;
462 ret
= iomap_apply(mapping
->host
, pos
, length
, 0, ops
,
463 &ctx
, iomap_readpages_actor
);
465 WARN_ON_ONCE(ret
== 0);
476 if (!ctx
.cur_page_in_bio
)
477 unlock_page(ctx
.cur_page
);
478 put_page(ctx
.cur_page
);
482 * Check that we didn't lose a page due to the arcance calling
485 WARN_ON_ONCE(!ret
&& !list_empty(ctx
.pages
));
488 EXPORT_SYMBOL_GPL(iomap_readpages
);
491 iomap_is_partially_uptodate(struct page
*page
, unsigned long from
,
494 struct iomap_page
*iop
= to_iomap_page(page
);
495 struct inode
*inode
= page
->mapping
->host
;
496 unsigned first
= from
>> inode
->i_blkbits
;
497 unsigned last
= (from
+ count
- 1) >> inode
->i_blkbits
;
501 for (i
= first
; i
<= last
; i
++)
502 if (!test_bit(i
, iop
->uptodate
))
509 EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate
);
512 iomap_releasepage(struct page
*page
, gfp_t gfp_mask
)
515 * mm accommodates an old ext3 case where clean pages might not have had
516 * the dirty bit cleared. Thus, it can send actual dirty pages to
517 * ->releasepage() via shrink_active_list(), skip those here.
519 if (PageDirty(page
) || PageWriteback(page
))
521 iomap_page_release(page
);
524 EXPORT_SYMBOL_GPL(iomap_releasepage
);
527 iomap_invalidatepage(struct page
*page
, unsigned int offset
, unsigned int len
)
530 * If we are invalidating the entire page, clear the dirty state from it
531 * and release it to avoid unnecessary buildup of the LRU.
533 if (offset
== 0 && len
== PAGE_SIZE
) {
534 WARN_ON_ONCE(PageWriteback(page
));
535 cancel_dirty_page(page
);
536 iomap_page_release(page
);
539 EXPORT_SYMBOL_GPL(iomap_invalidatepage
);
541 #ifdef CONFIG_MIGRATION
543 iomap_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
544 struct page
*page
, enum migrate_mode mode
)
548 ret
= migrate_page_move_mapping(mapping
, newpage
, page
, NULL
, mode
, 0);
549 if (ret
!= MIGRATEPAGE_SUCCESS
)
552 if (page_has_private(page
)) {
553 ClearPagePrivate(page
);
554 set_page_private(newpage
, page_private(page
));
555 set_page_private(page
, 0);
556 SetPagePrivate(newpage
);
559 if (mode
!= MIGRATE_SYNC_NO_COPY
)
560 migrate_page_copy(newpage
, page
);
562 migrate_page_states(newpage
, page
);
563 return MIGRATEPAGE_SUCCESS
;
565 EXPORT_SYMBOL_GPL(iomap_migrate_page
);
566 #endif /* CONFIG_MIGRATION */
569 iomap_write_failed(struct inode
*inode
, loff_t pos
, unsigned len
)
571 loff_t i_size
= i_size_read(inode
);
574 * Only truncate newly allocated pages beyoned EOF, even if the
575 * write started inside the existing inode size.
577 if (pos
+ len
> i_size
)
578 truncate_pagecache_range(inode
, max(pos
, i_size
), pos
+ len
);
582 iomap_read_page_sync(struct inode
*inode
, loff_t block_start
, struct page
*page
,
583 unsigned poff
, unsigned plen
, unsigned from
, unsigned to
,
589 if (iomap
->type
!= IOMAP_MAPPED
|| block_start
>= i_size_read(inode
)) {
590 zero_user_segments(page
, poff
, from
, to
, poff
+ plen
);
591 iomap_set_range_uptodate(page
, poff
, plen
);
595 bio_init(&bio
, &bvec
, 1);
596 bio
.bi_opf
= REQ_OP_READ
;
597 bio
.bi_iter
.bi_sector
= iomap_sector(iomap
, block_start
);
598 bio_set_dev(&bio
, iomap
->bdev
);
599 __bio_add_page(&bio
, page
, plen
, poff
);
600 return submit_bio_wait(&bio
);
604 __iomap_write_begin(struct inode
*inode
, loff_t pos
, unsigned len
,
605 struct page
*page
, struct iomap
*iomap
)
607 struct iomap_page
*iop
= iomap_page_create(inode
, page
);
608 loff_t block_size
= i_blocksize(inode
);
609 loff_t block_start
= pos
& ~(block_size
- 1);
610 loff_t block_end
= (pos
+ len
+ block_size
- 1) & ~(block_size
- 1);
611 unsigned from
= offset_in_page(pos
), to
= from
+ len
, poff
, plen
;
614 if (PageUptodate(page
))
618 iomap_adjust_read_range(inode
, iop
, &block_start
,
619 block_end
- block_start
, &poff
, &plen
);
623 if ((from
> poff
&& from
< poff
+ plen
) ||
624 (to
> poff
&& to
< poff
+ plen
)) {
625 status
= iomap_read_page_sync(inode
, block_start
, page
,
626 poff
, plen
, from
, to
, iomap
);
631 } while ((block_start
+= plen
) < block_end
);
637 iomap_write_begin(struct inode
*inode
, loff_t pos
, unsigned len
, unsigned flags
,
638 struct page
**pagep
, struct iomap
*iomap
)
640 pgoff_t index
= pos
>> PAGE_SHIFT
;
644 BUG_ON(pos
+ len
> iomap
->offset
+ iomap
->length
);
646 if (fatal_signal_pending(current
))
649 page
= grab_cache_page_write_begin(inode
->i_mapping
, index
, flags
);
653 if (iomap
->type
== IOMAP_INLINE
)
654 iomap_read_inline_data(inode
, page
, iomap
);
655 else if (iomap
->flags
& IOMAP_F_BUFFER_HEAD
)
656 status
= __block_write_begin_int(page
, pos
, len
, NULL
, iomap
);
658 status
= __iomap_write_begin(inode
, pos
, len
, page
, iomap
);
659 if (unlikely(status
)) {
664 iomap_write_failed(inode
, pos
, len
);
672 iomap_set_page_dirty(struct page
*page
)
674 struct address_space
*mapping
= page_mapping(page
);
677 if (unlikely(!mapping
))
678 return !TestSetPageDirty(page
);
681 * Lock out page->mem_cgroup migration to keep PageDirty
682 * synchronized with per-memcg dirty page counters.
684 lock_page_memcg(page
);
685 newly_dirty
= !TestSetPageDirty(page
);
687 __set_page_dirty(page
, mapping
, 0);
688 unlock_page_memcg(page
);
691 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
694 EXPORT_SYMBOL_GPL(iomap_set_page_dirty
);
697 __iomap_write_end(struct inode
*inode
, loff_t pos
, unsigned len
,
698 unsigned copied
, struct page
*page
, struct iomap
*iomap
)
700 flush_dcache_page(page
);
703 * The blocks that were entirely written will now be uptodate, so we
704 * don't have to worry about a readpage reading them and overwriting a
705 * partial write. However if we have encountered a short write and only
706 * partially written into a block, it will not be marked uptodate, so a
707 * readpage might come in and destroy our partial write.
709 * Do the simplest thing, and just treat any short write to a non
710 * uptodate page as a zero-length write, and force the caller to redo
713 if (unlikely(copied
< len
&& !PageUptodate(page
))) {
716 iomap_set_range_uptodate(page
, offset_in_page(pos
), len
);
717 iomap_set_page_dirty(page
);
719 return __generic_write_end(inode
, pos
, copied
, page
);
723 iomap_write_end_inline(struct inode
*inode
, struct page
*page
,
724 struct iomap
*iomap
, loff_t pos
, unsigned copied
)
728 WARN_ON_ONCE(!PageUptodate(page
));
729 BUG_ON(pos
+ copied
> PAGE_SIZE
- offset_in_page(iomap
->inline_data
));
731 addr
= kmap_atomic(page
);
732 memcpy(iomap
->inline_data
+ pos
, addr
+ pos
, copied
);
735 mark_inode_dirty(inode
);
736 __generic_write_end(inode
, pos
, copied
, page
);
741 iomap_write_end(struct inode
*inode
, loff_t pos
, unsigned len
,
742 unsigned copied
, struct page
*page
, struct iomap
*iomap
)
746 if (iomap
->type
== IOMAP_INLINE
) {
747 ret
= iomap_write_end_inline(inode
, page
, iomap
, pos
, copied
);
748 } else if (iomap
->flags
& IOMAP_F_BUFFER_HEAD
) {
749 ret
= generic_write_end(NULL
, inode
->i_mapping
, pos
, len
,
752 ret
= __iomap_write_end(inode
, pos
, len
, copied
, page
, iomap
);
755 if (iomap
->page_done
)
756 iomap
->page_done(inode
, pos
, copied
, page
, iomap
);
759 iomap_write_failed(inode
, pos
, len
);
764 iomap_write_actor(struct inode
*inode
, loff_t pos
, loff_t length
, void *data
,
767 struct iov_iter
*i
= data
;
770 unsigned int flags
= AOP_FLAG_NOFS
;
774 unsigned long offset
; /* Offset into pagecache page */
775 unsigned long bytes
; /* Bytes to write to page */
776 size_t copied
; /* Bytes copied from user */
778 offset
= offset_in_page(pos
);
779 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
786 * Bring in the user page that we will copy from _first_.
787 * Otherwise there's a nasty deadlock on copying from the
788 * same page as we're writing to, without it being marked
791 * Not only is this an optimisation, but it is also required
792 * to check that the address is actually valid, when atomic
793 * usercopies are used, below.
795 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
800 status
= iomap_write_begin(inode
, pos
, bytes
, flags
, &page
,
802 if (unlikely(status
))
805 if (mapping_writably_mapped(inode
->i_mapping
))
806 flush_dcache_page(page
);
808 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
810 flush_dcache_page(page
);
812 status
= iomap_write_end(inode
, pos
, bytes
, copied
, page
,
814 if (unlikely(status
< 0))
820 iov_iter_advance(i
, copied
);
821 if (unlikely(copied
== 0)) {
823 * If we were unable to copy any data at all, we must
824 * fall back to a single segment length write.
826 * If we didn't fallback here, we could livelock
827 * because not all segments in the iov can be copied at
828 * once without a pagefault.
830 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
831 iov_iter_single_seg_count(i
));
838 balance_dirty_pages_ratelimited(inode
->i_mapping
);
839 } while (iov_iter_count(i
) && length
);
841 return written
? written
: status
;
845 iomap_file_buffered_write(struct kiocb
*iocb
, struct iov_iter
*iter
,
846 const struct iomap_ops
*ops
)
848 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
849 loff_t pos
= iocb
->ki_pos
, ret
= 0, written
= 0;
851 while (iov_iter_count(iter
)) {
852 ret
= iomap_apply(inode
, pos
, iov_iter_count(iter
),
853 IOMAP_WRITE
, ops
, iter
, iomap_write_actor
);
860 return written
? written
: ret
;
862 EXPORT_SYMBOL_GPL(iomap_file_buffered_write
);
865 __iomap_read_page(struct inode
*inode
, loff_t offset
)
867 struct address_space
*mapping
= inode
->i_mapping
;
870 page
= read_mapping_page(mapping
, offset
>> PAGE_SHIFT
, NULL
);
873 if (!PageUptodate(page
)) {
875 return ERR_PTR(-EIO
);
881 iomap_dirty_actor(struct inode
*inode
, loff_t pos
, loff_t length
, void *data
,
888 struct page
*page
, *rpage
;
889 unsigned long offset
; /* Offset into pagecache page */
890 unsigned long bytes
; /* Bytes to write to page */
892 offset
= offset_in_page(pos
);
893 bytes
= min_t(loff_t
, PAGE_SIZE
- offset
, length
);
895 rpage
= __iomap_read_page(inode
, pos
);
897 return PTR_ERR(rpage
);
899 status
= iomap_write_begin(inode
, pos
, bytes
,
900 AOP_FLAG_NOFS
, &page
, iomap
);
902 if (unlikely(status
))
905 WARN_ON_ONCE(!PageUptodate(page
));
907 status
= iomap_write_end(inode
, pos
, bytes
, bytes
, page
, iomap
);
908 if (unlikely(status
<= 0)) {
909 if (WARN_ON_ONCE(status
== 0))
920 balance_dirty_pages_ratelimited(inode
->i_mapping
);
927 iomap_file_dirty(struct inode
*inode
, loff_t pos
, loff_t len
,
928 const struct iomap_ops
*ops
)
933 ret
= iomap_apply(inode
, pos
, len
, IOMAP_WRITE
, ops
, NULL
,
943 EXPORT_SYMBOL_GPL(iomap_file_dirty
);
945 static int iomap_zero(struct inode
*inode
, loff_t pos
, unsigned offset
,
946 unsigned bytes
, struct iomap
*iomap
)
951 status
= iomap_write_begin(inode
, pos
, bytes
, AOP_FLAG_NOFS
, &page
,
956 zero_user(page
, offset
, bytes
);
957 mark_page_accessed(page
);
959 return iomap_write_end(inode
, pos
, bytes
, bytes
, page
, iomap
);
962 static int iomap_dax_zero(loff_t pos
, unsigned offset
, unsigned bytes
,
965 return __dax_zero_page_range(iomap
->bdev
, iomap
->dax_dev
,
966 iomap_sector(iomap
, pos
& PAGE_MASK
), offset
, bytes
);
970 iomap_zero_range_actor(struct inode
*inode
, loff_t pos
, loff_t count
,
971 void *data
, struct iomap
*iomap
)
973 bool *did_zero
= data
;
977 /* already zeroed? we're done. */
978 if (iomap
->type
== IOMAP_HOLE
|| iomap
->type
== IOMAP_UNWRITTEN
)
982 unsigned offset
, bytes
;
984 offset
= offset_in_page(pos
);
985 bytes
= min_t(loff_t
, PAGE_SIZE
- offset
, count
);
988 status
= iomap_dax_zero(pos
, offset
, bytes
, iomap
);
990 status
= iomap_zero(inode
, pos
, offset
, bytes
, iomap
);
1005 iomap_zero_range(struct inode
*inode
, loff_t pos
, loff_t len
, bool *did_zero
,
1006 const struct iomap_ops
*ops
)
1011 ret
= iomap_apply(inode
, pos
, len
, IOMAP_ZERO
,
1012 ops
, did_zero
, iomap_zero_range_actor
);
1022 EXPORT_SYMBOL_GPL(iomap_zero_range
);
1025 iomap_truncate_page(struct inode
*inode
, loff_t pos
, bool *did_zero
,
1026 const struct iomap_ops
*ops
)
1028 unsigned int blocksize
= i_blocksize(inode
);
1029 unsigned int off
= pos
& (blocksize
- 1);
1031 /* Block boundary? Nothing to do */
1034 return iomap_zero_range(inode
, pos
, blocksize
- off
, did_zero
, ops
);
1036 EXPORT_SYMBOL_GPL(iomap_truncate_page
);
1039 iomap_page_mkwrite_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
1040 void *data
, struct iomap
*iomap
)
1042 struct page
*page
= data
;
1045 if (iomap
->flags
& IOMAP_F_BUFFER_HEAD
) {
1046 ret
= __block_write_begin_int(page
, pos
, length
, NULL
, iomap
);
1049 block_commit_write(page
, 0, length
);
1051 WARN_ON_ONCE(!PageUptodate(page
));
1052 iomap_page_create(inode
, page
);
1053 set_page_dirty(page
);
1059 vm_fault_t
iomap_page_mkwrite(struct vm_fault
*vmf
, const struct iomap_ops
*ops
)
1061 struct page
*page
= vmf
->page
;
1062 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
1063 unsigned long length
;
1064 loff_t offset
, size
;
1068 size
= i_size_read(inode
);
1069 if ((page
->mapping
!= inode
->i_mapping
) ||
1070 (page_offset(page
) > size
)) {
1071 /* We overload EFAULT to mean page got truncated */
1076 /* page is wholly or partially inside EOF */
1077 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
1078 length
= offset_in_page(size
);
1082 offset
= page_offset(page
);
1083 while (length
> 0) {
1084 ret
= iomap_apply(inode
, offset
, length
,
1085 IOMAP_WRITE
| IOMAP_FAULT
, ops
, page
,
1086 iomap_page_mkwrite_actor
);
1087 if (unlikely(ret
<= 0))
1093 wait_for_stable_page(page
);
1094 return VM_FAULT_LOCKED
;
1097 return block_page_mkwrite_return(ret
);
1099 EXPORT_SYMBOL_GPL(iomap_page_mkwrite
);
1102 struct fiemap_extent_info
*fi
;
1106 static int iomap_to_fiemap(struct fiemap_extent_info
*fi
,
1107 struct iomap
*iomap
, u32 flags
)
1109 switch (iomap
->type
) {
1113 case IOMAP_DELALLOC
:
1114 flags
|= FIEMAP_EXTENT_DELALLOC
| FIEMAP_EXTENT_UNKNOWN
;
1118 case IOMAP_UNWRITTEN
:
1119 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
1122 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
1126 if (iomap
->flags
& IOMAP_F_MERGED
)
1127 flags
|= FIEMAP_EXTENT_MERGED
;
1128 if (iomap
->flags
& IOMAP_F_SHARED
)
1129 flags
|= FIEMAP_EXTENT_SHARED
;
1131 return fiemap_fill_next_extent(fi
, iomap
->offset
,
1132 iomap
->addr
!= IOMAP_NULL_ADDR
? iomap
->addr
: 0,
1133 iomap
->length
, flags
);
1137 iomap_fiemap_actor(struct inode
*inode
, loff_t pos
, loff_t length
, void *data
,
1138 struct iomap
*iomap
)
1140 struct fiemap_ctx
*ctx
= data
;
1141 loff_t ret
= length
;
1143 if (iomap
->type
== IOMAP_HOLE
)
1146 ret
= iomap_to_fiemap(ctx
->fi
, &ctx
->prev
, 0);
1149 case 0: /* success */
1151 case 1: /* extent array full */
1158 int iomap_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fi
,
1159 loff_t start
, loff_t len
, const struct iomap_ops
*ops
)
1161 struct fiemap_ctx ctx
;
1164 memset(&ctx
, 0, sizeof(ctx
));
1166 ctx
.prev
.type
= IOMAP_HOLE
;
1168 ret
= fiemap_check_flags(fi
, FIEMAP_FLAG_SYNC
);
1172 if (fi
->fi_flags
& FIEMAP_FLAG_SYNC
) {
1173 ret
= filemap_write_and_wait(inode
->i_mapping
);
1179 ret
= iomap_apply(inode
, start
, len
, IOMAP_REPORT
, ops
, &ctx
,
1180 iomap_fiemap_actor
);
1181 /* inode with no (attribute) mapping will give ENOENT */
1193 if (ctx
.prev
.type
!= IOMAP_HOLE
) {
1194 ret
= iomap_to_fiemap(fi
, &ctx
.prev
, FIEMAP_EXTENT_LAST
);
1201 EXPORT_SYMBOL_GPL(iomap_fiemap
);
1204 * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
1205 * Returns true if found and updates @lastoff to the offset in file.
1208 page_seek_hole_data(struct inode
*inode
, struct page
*page
, loff_t
*lastoff
,
1211 const struct address_space_operations
*ops
= inode
->i_mapping
->a_ops
;
1212 unsigned int bsize
= i_blocksize(inode
), off
;
1213 bool seek_data
= whence
== SEEK_DATA
;
1214 loff_t poff
= page_offset(page
);
1216 if (WARN_ON_ONCE(*lastoff
>= poff
+ PAGE_SIZE
))
1219 if (*lastoff
< poff
) {
1221 * Last offset smaller than the start of the page means we found
1224 if (whence
== SEEK_HOLE
)
1230 * Just check the page unless we can and should check block ranges:
1232 if (bsize
== PAGE_SIZE
|| !ops
->is_partially_uptodate
)
1233 return PageUptodate(page
) == seek_data
;
1236 if (unlikely(page
->mapping
!= inode
->i_mapping
))
1237 goto out_unlock_not_found
;
1239 for (off
= 0; off
< PAGE_SIZE
; off
+= bsize
) {
1240 if (offset_in_page(*lastoff
) >= off
+ bsize
)
1242 if (ops
->is_partially_uptodate(page
, off
, bsize
) == seek_data
) {
1246 *lastoff
= poff
+ off
+ bsize
;
1249 out_unlock_not_found
:
1255 * Seek for SEEK_DATA / SEEK_HOLE in the page cache.
1257 * Within unwritten extents, the page cache determines which parts are holes
1258 * and which are data: uptodate buffer heads count as data; everything else
1261 * Returns the resulting offset on successs, and -ENOENT otherwise.
1264 page_cache_seek_hole_data(struct inode
*inode
, loff_t offset
, loff_t length
,
1267 pgoff_t index
= offset
>> PAGE_SHIFT
;
1268 pgoff_t end
= DIV_ROUND_UP(offset
+ length
, PAGE_SIZE
);
1269 loff_t lastoff
= offset
;
1270 struct pagevec pvec
;
1275 pagevec_init(&pvec
);
1278 unsigned nr_pages
, i
;
1280 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
, &index
,
1285 for (i
= 0; i
< nr_pages
; i
++) {
1286 struct page
*page
= pvec
.pages
[i
];
1288 if (page_seek_hole_data(inode
, page
, &lastoff
, whence
))
1290 lastoff
= page_offset(page
) + PAGE_SIZE
;
1292 pagevec_release(&pvec
);
1293 } while (index
< end
);
1295 /* When no page at lastoff and we are not done, we found a hole. */
1296 if (whence
!= SEEK_HOLE
)
1300 if (lastoff
< offset
+ length
)
1305 pagevec_release(&pvec
);
1311 iomap_seek_hole_actor(struct inode
*inode
, loff_t offset
, loff_t length
,
1312 void *data
, struct iomap
*iomap
)
1314 switch (iomap
->type
) {
1315 case IOMAP_UNWRITTEN
:
1316 offset
= page_cache_seek_hole_data(inode
, offset
, length
,
1322 *(loff_t
*)data
= offset
;
1330 iomap_seek_hole(struct inode
*inode
, loff_t offset
, const struct iomap_ops
*ops
)
1332 loff_t size
= i_size_read(inode
);
1333 loff_t length
= size
- offset
;
1336 /* Nothing to be found before or beyond the end of the file. */
1337 if (offset
< 0 || offset
>= size
)
1340 while (length
> 0) {
1341 ret
= iomap_apply(inode
, offset
, length
, IOMAP_REPORT
, ops
,
1342 &offset
, iomap_seek_hole_actor
);
1354 EXPORT_SYMBOL_GPL(iomap_seek_hole
);
1357 iomap_seek_data_actor(struct inode
*inode
, loff_t offset
, loff_t length
,
1358 void *data
, struct iomap
*iomap
)
1360 switch (iomap
->type
) {
1363 case IOMAP_UNWRITTEN
:
1364 offset
= page_cache_seek_hole_data(inode
, offset
, length
,
1370 *(loff_t
*)data
= offset
;
1376 iomap_seek_data(struct inode
*inode
, loff_t offset
, const struct iomap_ops
*ops
)
1378 loff_t size
= i_size_read(inode
);
1379 loff_t length
= size
- offset
;
1382 /* Nothing to be found before or beyond the end of the file. */
1383 if (offset
< 0 || offset
>= size
)
1386 while (length
> 0) {
1387 ret
= iomap_apply(inode
, offset
, length
, IOMAP_REPORT
, ops
,
1388 &offset
, iomap_seek_data_actor
);
1402 EXPORT_SYMBOL_GPL(iomap_seek_data
);
1405 * Private flags for iomap_dio, must not overlap with the public ones in
1408 #define IOMAP_DIO_WRITE_FUA (1 << 28)
1409 #define IOMAP_DIO_NEED_SYNC (1 << 29)
1410 #define IOMAP_DIO_WRITE (1 << 30)
1411 #define IOMAP_DIO_DIRTY (1 << 31)
1415 iomap_dio_end_io_t
*end_io
;
1421 bool wait_for_completion
;
1424 /* used during submission and for synchronous completion: */
1426 struct iov_iter
*iter
;
1427 struct task_struct
*waiter
;
1428 struct request_queue
*last_queue
;
1432 /* used for aio completion: */
1434 struct work_struct work
;
1439 static ssize_t
iomap_dio_complete(struct iomap_dio
*dio
)
1441 struct kiocb
*iocb
= dio
->iocb
;
1442 struct inode
*inode
= file_inode(iocb
->ki_filp
);
1443 loff_t offset
= iocb
->ki_pos
;
1447 ret
= dio
->end_io(iocb
,
1448 dio
->error
? dio
->error
: dio
->size
,
1456 /* check for short read */
1457 if (offset
+ ret
> dio
->i_size
&&
1458 !(dio
->flags
& IOMAP_DIO_WRITE
))
1459 ret
= dio
->i_size
- offset
;
1460 iocb
->ki_pos
+= ret
;
1464 * Try again to invalidate clean pages which might have been cached by
1465 * non-direct readahead, or faulted in by get_user_pages() if the source
1466 * of the write was an mmap'ed region of the file we're writing. Either
1467 * one is a pretty crazy thing to do, so we don't support it 100%. If
1468 * this invalidation fails, tough, the write still worked...
1470 * And this page cache invalidation has to be after dio->end_io(), as
1471 * some filesystems convert unwritten extents to real allocations in
1472 * end_io() when necessary, otherwise a racing buffer read would cache
1473 * zeros from unwritten extents.
1476 (dio
->flags
& IOMAP_DIO_WRITE
) && inode
->i_mapping
->nrpages
) {
1478 err
= invalidate_inode_pages2_range(inode
->i_mapping
,
1479 offset
>> PAGE_SHIFT
,
1480 (offset
+ dio
->size
- 1) >> PAGE_SHIFT
);
1482 dio_warn_stale_pagecache(iocb
->ki_filp
);
1486 * If this is a DSYNC write, make sure we push it to stable storage now
1487 * that we've written data.
1489 if (ret
> 0 && (dio
->flags
& IOMAP_DIO_NEED_SYNC
))
1490 ret
= generic_write_sync(iocb
, ret
);
1492 inode_dio_end(file_inode(iocb
->ki_filp
));
1498 static void iomap_dio_complete_work(struct work_struct
*work
)
1500 struct iomap_dio
*dio
= container_of(work
, struct iomap_dio
, aio
.work
);
1501 struct kiocb
*iocb
= dio
->iocb
;
1503 iocb
->ki_complete(iocb
, iomap_dio_complete(dio
), 0);
1507 * Set an error in the dio if none is set yet. We have to use cmpxchg
1508 * as the submission context and the completion context(s) can race to
1511 static inline void iomap_dio_set_error(struct iomap_dio
*dio
, int ret
)
1513 cmpxchg(&dio
->error
, 0, ret
);
1516 static void iomap_dio_bio_end_io(struct bio
*bio
)
1518 struct iomap_dio
*dio
= bio
->bi_private
;
1519 bool should_dirty
= (dio
->flags
& IOMAP_DIO_DIRTY
);
1522 iomap_dio_set_error(dio
, blk_status_to_errno(bio
->bi_status
));
1524 if (atomic_dec_and_test(&dio
->ref
)) {
1525 if (dio
->wait_for_completion
) {
1526 struct task_struct
*waiter
= dio
->submit
.waiter
;
1527 WRITE_ONCE(dio
->submit
.waiter
, NULL
);
1528 wake_up_process(waiter
);
1529 } else if (dio
->flags
& IOMAP_DIO_WRITE
) {
1530 struct inode
*inode
= file_inode(dio
->iocb
->ki_filp
);
1532 INIT_WORK(&dio
->aio
.work
, iomap_dio_complete_work
);
1533 queue_work(inode
->i_sb
->s_dio_done_wq
, &dio
->aio
.work
);
1535 iomap_dio_complete_work(&dio
->aio
.work
);
1540 bio_check_pages_dirty(bio
);
1542 struct bio_vec
*bvec
;
1545 bio_for_each_segment_all(bvec
, bio
, i
)
1546 put_page(bvec
->bv_page
);
1552 iomap_dio_zero(struct iomap_dio
*dio
, struct iomap
*iomap
, loff_t pos
,
1555 struct page
*page
= ZERO_PAGE(0);
1558 bio
= bio_alloc(GFP_KERNEL
, 1);
1559 bio_set_dev(bio
, iomap
->bdev
);
1560 bio
->bi_iter
.bi_sector
= iomap_sector(iomap
, pos
);
1561 bio
->bi_private
= dio
;
1562 bio
->bi_end_io
= iomap_dio_bio_end_io
;
1565 __bio_add_page(bio
, page
, len
, 0);
1566 bio_set_op_attrs(bio
, REQ_OP_WRITE
, REQ_SYNC
| REQ_IDLE
);
1568 atomic_inc(&dio
->ref
);
1569 return submit_bio(bio
);
1573 iomap_dio_bio_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
1574 struct iomap_dio
*dio
, struct iomap
*iomap
)
1576 unsigned int blkbits
= blksize_bits(bdev_logical_block_size(iomap
->bdev
));
1577 unsigned int fs_block_size
= i_blocksize(inode
), pad
;
1578 unsigned int align
= iov_iter_alignment(dio
->submit
.iter
);
1579 struct iov_iter iter
;
1581 bool need_zeroout
= false;
1582 bool use_fua
= false;
1586 if ((pos
| length
| align
) & ((1 << blkbits
) - 1))
1589 if (iomap
->type
== IOMAP_UNWRITTEN
) {
1590 dio
->flags
|= IOMAP_DIO_UNWRITTEN
;
1591 need_zeroout
= true;
1594 if (iomap
->flags
& IOMAP_F_SHARED
)
1595 dio
->flags
|= IOMAP_DIO_COW
;
1597 if (iomap
->flags
& IOMAP_F_NEW
) {
1598 need_zeroout
= true;
1601 * Use a FUA write if we need datasync semantics, this
1602 * is a pure data IO that doesn't require any metadata
1603 * updates and the underlying device supports FUA. This
1604 * allows us to avoid cache flushes on IO completion.
1606 if (!(iomap
->flags
& (IOMAP_F_SHARED
|IOMAP_F_DIRTY
)) &&
1607 (dio
->flags
& IOMAP_DIO_WRITE_FUA
) &&
1608 blk_queue_fua(bdev_get_queue(iomap
->bdev
)))
1613 * Operate on a partial iter trimmed to the extent we were called for.
1614 * We'll update the iter in the dio once we're done with this extent.
1616 iter
= *dio
->submit
.iter
;
1617 iov_iter_truncate(&iter
, length
);
1619 nr_pages
= iov_iter_npages(&iter
, BIO_MAX_PAGES
);
1624 /* zero out from the start of the block to the write offset */
1625 pad
= pos
& (fs_block_size
- 1);
1627 iomap_dio_zero(dio
, iomap
, pos
- pad
, pad
);
1633 iov_iter_revert(dio
->submit
.iter
, copied
);
1637 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
1638 bio_set_dev(bio
, iomap
->bdev
);
1639 bio
->bi_iter
.bi_sector
= iomap_sector(iomap
, pos
);
1640 bio
->bi_write_hint
= dio
->iocb
->ki_hint
;
1641 bio
->bi_ioprio
= dio
->iocb
->ki_ioprio
;
1642 bio
->bi_private
= dio
;
1643 bio
->bi_end_io
= iomap_dio_bio_end_io
;
1645 ret
= bio_iov_iter_get_pages(bio
, &iter
);
1646 if (unlikely(ret
)) {
1648 return copied
? copied
: ret
;
1651 n
= bio
->bi_iter
.bi_size
;
1652 if (dio
->flags
& IOMAP_DIO_WRITE
) {
1653 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
1655 bio
->bi_opf
|= REQ_FUA
;
1657 dio
->flags
&= ~IOMAP_DIO_WRITE_FUA
;
1658 task_io_account_write(n
);
1660 bio
->bi_opf
= REQ_OP_READ
;
1661 if (dio
->flags
& IOMAP_DIO_DIRTY
)
1662 bio_set_pages_dirty(bio
);
1665 iov_iter_advance(dio
->submit
.iter
, n
);
1671 nr_pages
= iov_iter_npages(&iter
, BIO_MAX_PAGES
);
1673 atomic_inc(&dio
->ref
);
1675 dio
->submit
.last_queue
= bdev_get_queue(iomap
->bdev
);
1676 dio
->submit
.cookie
= submit_bio(bio
);
1680 /* zero out from the end of the write to the end of the block */
1681 pad
= pos
& (fs_block_size
- 1);
1683 iomap_dio_zero(dio
, iomap
, pos
, fs_block_size
- pad
);
1689 iomap_dio_hole_actor(loff_t length
, struct iomap_dio
*dio
)
1691 length
= iov_iter_zero(length
, dio
->submit
.iter
);
1692 dio
->size
+= length
;
1697 iomap_dio_inline_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
1698 struct iomap_dio
*dio
, struct iomap
*iomap
)
1700 struct iov_iter
*iter
= dio
->submit
.iter
;
1703 BUG_ON(pos
+ length
> PAGE_SIZE
- offset_in_page(iomap
->inline_data
));
1705 if (dio
->flags
& IOMAP_DIO_WRITE
) {
1706 loff_t size
= inode
->i_size
;
1709 memset(iomap
->inline_data
+ size
, 0, pos
- size
);
1710 copied
= copy_from_iter(iomap
->inline_data
+ pos
, length
, iter
);
1712 if (pos
+ copied
> size
)
1713 i_size_write(inode
, pos
+ copied
);
1714 mark_inode_dirty(inode
);
1717 copied
= copy_to_iter(iomap
->inline_data
+ pos
, length
, iter
);
1719 dio
->size
+= copied
;
1724 iomap_dio_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
1725 void *data
, struct iomap
*iomap
)
1727 struct iomap_dio
*dio
= data
;
1729 switch (iomap
->type
) {
1731 if (WARN_ON_ONCE(dio
->flags
& IOMAP_DIO_WRITE
))
1733 return iomap_dio_hole_actor(length
, dio
);
1734 case IOMAP_UNWRITTEN
:
1735 if (!(dio
->flags
& IOMAP_DIO_WRITE
))
1736 return iomap_dio_hole_actor(length
, dio
);
1737 return iomap_dio_bio_actor(inode
, pos
, length
, dio
, iomap
);
1739 return iomap_dio_bio_actor(inode
, pos
, length
, dio
, iomap
);
1741 return iomap_dio_inline_actor(inode
, pos
, length
, dio
, iomap
);
1749 * iomap_dio_rw() always completes O_[D]SYNC writes regardless of whether the IO
1750 * is being issued as AIO or not. This allows us to optimise pure data writes
1751 * to use REQ_FUA rather than requiring generic_write_sync() to issue a
1752 * REQ_FLUSH post write. This is slightly tricky because a single request here
1753 * can be mapped into multiple disjoint IOs and only a subset of the IOs issued
1754 * may be pure data writes. In that case, we still need to do a full data sync
1758 iomap_dio_rw(struct kiocb
*iocb
, struct iov_iter
*iter
,
1759 const struct iomap_ops
*ops
, iomap_dio_end_io_t end_io
)
1761 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1762 struct inode
*inode
= file_inode(iocb
->ki_filp
);
1763 size_t count
= iov_iter_count(iter
);
1764 loff_t pos
= iocb
->ki_pos
, start
= pos
;
1765 loff_t end
= iocb
->ki_pos
+ count
- 1, ret
= 0;
1766 unsigned int flags
= IOMAP_DIRECT
;
1767 struct blk_plug plug
;
1768 struct iomap_dio
*dio
;
1770 lockdep_assert_held(&inode
->i_rwsem
);
1775 dio
= kmalloc(sizeof(*dio
), GFP_KERNEL
);
1780 atomic_set(&dio
->ref
, 1);
1782 dio
->i_size
= i_size_read(inode
);
1783 dio
->end_io
= end_io
;
1786 dio
->wait_for_completion
= is_sync_kiocb(iocb
);
1788 dio
->submit
.iter
= iter
;
1789 dio
->submit
.waiter
= current
;
1790 dio
->submit
.cookie
= BLK_QC_T_NONE
;
1791 dio
->submit
.last_queue
= NULL
;
1793 if (iov_iter_rw(iter
) == READ
) {
1794 if (pos
>= dio
->i_size
)
1797 if (iter_is_iovec(iter
) && iov_iter_rw(iter
) == READ
)
1798 dio
->flags
|= IOMAP_DIO_DIRTY
;
1800 flags
|= IOMAP_WRITE
;
1801 dio
->flags
|= IOMAP_DIO_WRITE
;
1803 /* for data sync or sync, we need sync completion processing */
1804 if (iocb
->ki_flags
& IOCB_DSYNC
)
1805 dio
->flags
|= IOMAP_DIO_NEED_SYNC
;
1808 * For datasync only writes, we optimistically try using FUA for
1809 * this IO. Any non-FUA write that occurs will clear this flag,
1810 * hence we know before completion whether a cache flush is
1813 if ((iocb
->ki_flags
& (IOCB_DSYNC
| IOCB_SYNC
)) == IOCB_DSYNC
)
1814 dio
->flags
|= IOMAP_DIO_WRITE_FUA
;
1817 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
1818 if (filemap_range_has_page(mapping
, start
, end
)) {
1822 flags
|= IOMAP_NOWAIT
;
1825 ret
= filemap_write_and_wait_range(mapping
, start
, end
);
1830 * Try to invalidate cache pages for the range we're direct
1831 * writing. If this invalidation fails, tough, the write will
1832 * still work, but racing two incompatible write paths is a
1833 * pretty crazy thing to do, so we don't support it 100%.
1835 ret
= invalidate_inode_pages2_range(mapping
,
1836 start
>> PAGE_SHIFT
, end
>> PAGE_SHIFT
);
1838 dio_warn_stale_pagecache(iocb
->ki_filp
);
1841 if (iov_iter_rw(iter
) == WRITE
&& !dio
->wait_for_completion
&&
1842 !inode
->i_sb
->s_dio_done_wq
) {
1843 ret
= sb_init_dio_done_wq(inode
->i_sb
);
1848 inode_dio_begin(inode
);
1850 blk_start_plug(&plug
);
1852 ret
= iomap_apply(inode
, pos
, count
, flags
, ops
, dio
,
1855 /* magic error code to fall back to buffered I/O */
1856 if (ret
== -ENOTBLK
) {
1857 dio
->wait_for_completion
= true;
1864 if (iov_iter_rw(iter
) == READ
&& pos
>= dio
->i_size
)
1866 } while ((count
= iov_iter_count(iter
)) > 0);
1867 blk_finish_plug(&plug
);
1870 iomap_dio_set_error(dio
, ret
);
1873 * If all the writes we issued were FUA, we don't need to flush the
1874 * cache on IO completion. Clear the sync flag for this case.
1876 if (dio
->flags
& IOMAP_DIO_WRITE_FUA
)
1877 dio
->flags
&= ~IOMAP_DIO_NEED_SYNC
;
1879 if (!atomic_dec_and_test(&dio
->ref
)) {
1880 if (!dio
->wait_for_completion
)
1881 return -EIOCBQUEUED
;
1884 set_current_state(TASK_UNINTERRUPTIBLE
);
1885 if (!READ_ONCE(dio
->submit
.waiter
))
1888 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
1889 !dio
->submit
.last_queue
||
1890 !blk_poll(dio
->submit
.last_queue
,
1891 dio
->submit
.cookie
))
1894 __set_current_state(TASK_RUNNING
);
1897 ret
= iomap_dio_complete(dio
);
1905 EXPORT_SYMBOL_GPL(iomap_dio_rw
);
1907 /* Swapfile activation */
1910 struct iomap_swapfile_info
{
1911 struct iomap iomap
; /* accumulated iomap */
1912 struct swap_info_struct
*sis
;
1913 uint64_t lowest_ppage
; /* lowest physical addr seen (pages) */
1914 uint64_t highest_ppage
; /* highest physical addr seen (pages) */
1915 unsigned long nr_pages
; /* number of pages collected */
1916 int nr_extents
; /* extent count */
1920 * Collect physical extents for this swap file. Physical extents reported to
1921 * the swap code must be trimmed to align to a page boundary. The logical
1922 * offset within the file is irrelevant since the swapfile code maps logical
1923 * page numbers of the swap device to the physical page-aligned extents.
1925 static int iomap_swapfile_add_extent(struct iomap_swapfile_info
*isi
)
1927 struct iomap
*iomap
= &isi
->iomap
;
1928 unsigned long nr_pages
;
1929 uint64_t first_ppage
;
1930 uint64_t first_ppage_reported
;
1931 uint64_t next_ppage
;
1935 * Round the start up and the end down so that the physical
1936 * extent aligns to a page boundary.
1938 first_ppage
= ALIGN(iomap
->addr
, PAGE_SIZE
) >> PAGE_SHIFT
;
1939 next_ppage
= ALIGN_DOWN(iomap
->addr
+ iomap
->length
, PAGE_SIZE
) >>
1942 /* Skip too-short physical extents. */
1943 if (first_ppage
>= next_ppage
)
1945 nr_pages
= next_ppage
- first_ppage
;
1948 * Calculate how much swap space we're adding; the first page contains
1949 * the swap header and doesn't count. The mm still wants that first
1950 * page fed to add_swap_extent, however.
1952 first_ppage_reported
= first_ppage
;
1953 if (iomap
->offset
== 0)
1954 first_ppage_reported
++;
1955 if (isi
->lowest_ppage
> first_ppage_reported
)
1956 isi
->lowest_ppage
= first_ppage_reported
;
1957 if (isi
->highest_ppage
< (next_ppage
- 1))
1958 isi
->highest_ppage
= next_ppage
- 1;
1960 /* Add extent, set up for the next call. */
1961 error
= add_swap_extent(isi
->sis
, isi
->nr_pages
, nr_pages
, first_ppage
);
1964 isi
->nr_extents
+= error
;
1965 isi
->nr_pages
+= nr_pages
;
1970 * Accumulate iomaps for this swap file. We have to accumulate iomaps because
1971 * swap only cares about contiguous page-aligned physical extents and makes no
1972 * distinction between written and unwritten extents.
1974 static loff_t
iomap_swapfile_activate_actor(struct inode
*inode
, loff_t pos
,
1975 loff_t count
, void *data
, struct iomap
*iomap
)
1977 struct iomap_swapfile_info
*isi
= data
;
1980 switch (iomap
->type
) {
1982 case IOMAP_UNWRITTEN
:
1983 /* Only real or unwritten extents. */
1986 /* No inline data. */
1987 pr_err("swapon: file is inline\n");
1990 pr_err("swapon: file has unallocated extents\n");
1994 /* No uncommitted metadata or shared blocks. */
1995 if (iomap
->flags
& IOMAP_F_DIRTY
) {
1996 pr_err("swapon: file is not committed\n");
1999 if (iomap
->flags
& IOMAP_F_SHARED
) {
2000 pr_err("swapon: file has shared extents\n");
2004 /* Only one bdev per swap file. */
2005 if (iomap
->bdev
!= isi
->sis
->bdev
) {
2006 pr_err("swapon: file is on multiple devices\n");
2010 if (isi
->iomap
.length
== 0) {
2011 /* No accumulated extent, so just store it. */
2012 memcpy(&isi
->iomap
, iomap
, sizeof(isi
->iomap
));
2013 } else if (isi
->iomap
.addr
+ isi
->iomap
.length
== iomap
->addr
) {
2014 /* Append this to the accumulated extent. */
2015 isi
->iomap
.length
+= iomap
->length
;
2017 /* Otherwise, add the retained iomap and store this one. */
2018 error
= iomap_swapfile_add_extent(isi
);
2021 memcpy(&isi
->iomap
, iomap
, sizeof(isi
->iomap
));
2027 * Iterate a swap file's iomaps to construct physical extents that can be
2028 * passed to the swapfile subsystem.
2030 int iomap_swapfile_activate(struct swap_info_struct
*sis
,
2031 struct file
*swap_file
, sector_t
*pagespan
,
2032 const struct iomap_ops
*ops
)
2034 struct iomap_swapfile_info isi
= {
2036 .lowest_ppage
= (sector_t
)-1ULL,
2038 struct address_space
*mapping
= swap_file
->f_mapping
;
2039 struct inode
*inode
= mapping
->host
;
2041 loff_t len
= ALIGN_DOWN(i_size_read(inode
), PAGE_SIZE
);
2045 * Persist all file mapping metadata so that we won't have any
2046 * IOMAP_F_DIRTY iomaps.
2048 ret
= vfs_fsync(swap_file
, 1);
2053 ret
= iomap_apply(inode
, pos
, len
, IOMAP_REPORT
,
2054 ops
, &isi
, iomap_swapfile_activate_actor
);
2062 if (isi
.iomap
.length
) {
2063 ret
= iomap_swapfile_add_extent(&isi
);
2068 *pagespan
= 1 + isi
.highest_ppage
- isi
.lowest_ppage
;
2069 sis
->max
= isi
.nr_pages
;
2070 sis
->pages
= isi
.nr_pages
- 1;
2071 sis
->highest_bit
= isi
.nr_pages
- 1;
2072 return isi
.nr_extents
;
2074 EXPORT_SYMBOL_GPL(iomap_swapfile_activate
);
2075 #endif /* CONFIG_SWAP */
2078 iomap_bmap_actor(struct inode
*inode
, loff_t pos
, loff_t length
,
2079 void *data
, struct iomap
*iomap
)
2081 sector_t
*bno
= data
, addr
;
2083 if (iomap
->type
== IOMAP_MAPPED
) {
2084 addr
= (pos
- iomap
->offset
+ iomap
->addr
) >> inode
->i_blkbits
;
2086 WARN(1, "would truncate bmap result\n");
2093 /* legacy ->bmap interface. 0 is the error return (!) */
2095 iomap_bmap(struct address_space
*mapping
, sector_t bno
,
2096 const struct iomap_ops
*ops
)
2098 struct inode
*inode
= mapping
->host
;
2099 loff_t pos
= bno
<< inode
->i_blkbits
;
2100 unsigned blocksize
= i_blocksize(inode
);
2102 if (filemap_write_and_wait(mapping
))
2106 iomap_apply(inode
, pos
, blocksize
, 0, ops
, &bno
, iomap_bmap_actor
);
2109 EXPORT_SYMBOL_GPL(iomap_bmap
);