1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2010 Red Hat, Inc.
4 * Copyright (C) 2016-2019 Christoph Hellwig.
6 #include <linux/module.h>
7 #include <linux/compiler.h>
9 #include <linux/iomap.h>
10 #include <linux/pagemap.h>
11 #include <linux/uio.h>
12 #include <linux/buffer_head.h>
13 #include <linux/dax.h>
14 #include <linux/writeback.h>
15 #include <linux/list_sort.h>
16 #include <linux/swap.h>
17 #include <linux/bio.h>
18 #include <linux/sched/signal.h>
19 #include <linux/migrate.h>
22 #include "../internal.h"
25 * Structure allocated for each page or THP when block size < page size
26 * to track sub-page uptodate status and I/O completions.
31 spinlock_t uptodate_lock;
32 unsigned long uptodate[];
35 static inline struct iomap_page *to_iomap_page(struct page *page)
38 * per-block data is stored in the head page. Callers should
39 * not be dealing with tail pages (and if they are, they can
40 * call thp_head() first.
42 VM_BUG_ON_PGFLAGS(PageTail(page), page);
44 if (page_has_private(page))
45 return (struct iomap_page *)page_private(page);
49 static struct bio_set iomap_ioend_bioset;
51 static struct iomap_page *
52 iomap_page_create(struct inode *inode, struct page *page)
54 struct iomap_page *iop = to_iomap_page(page);
55 unsigned int nr_blocks = i_blocks_per_page(inode, page);
57 if (iop || nr_blocks <= 1)
60 iop = kzalloc(struct_size(iop, uptodate, BITS_TO_LONGS(nr_blocks)),
61 GFP_NOFS | __GFP_NOFAIL);
62 spin_lock_init(&iop->uptodate_lock);
63 attach_page_private(page, iop);
68 iomap_page_release(struct page *page)
70 struct iomap_page *iop = detach_page_private(page);
71 unsigned int nr_blocks = i_blocks_per_page(page->mapping->host, page);
75 WARN_ON_ONCE(atomic_read(&iop->read_count));
76 WARN_ON_ONCE(atomic_read(&iop->write_count));
77 WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) !=
83 * Calculate the range inside the page that we actually need to read.
86 iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop,
87 loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp)
89 loff_t orig_pos = *pos;
90 loff_t isize = i_size_read(inode);
91 unsigned block_bits = inode->i_blkbits;
92 unsigned block_size = (1 << block_bits);
93 unsigned poff = offset_in_page(*pos);
94 unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
95 unsigned first = poff >> block_bits;
96 unsigned last = (poff + plen - 1) >> block_bits;
99 * If the block size is smaller than the page size we need to check the
100 * per-block uptodate status and adjust the offset and length if needed
101 * to avoid reading in already uptodate ranges.
106 /* move forward for each leading block marked uptodate */
107 for (i = first; i <= last; i++) {
108 if (!test_bit(i, iop->uptodate))
116 /* truncate len if we find any trailing uptodate block(s) */
117 for ( ; i <= last; i++) {
118 if (test_bit(i, iop->uptodate)) {
119 plen -= (last - i + 1) * block_size;
127 * If the extent spans the block that contains the i_size we need to
128 * handle both halves separately so that we properly zero data in the
129 * page cache for blocks that are entirely outside of i_size.
131 if (orig_pos <= isize && orig_pos + length > isize) {
132 unsigned end = offset_in_page(isize - 1) >> block_bits;
134 if (first <= end && last > end)
135 plen -= (last - end) * block_size;
143 iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len)
145 struct iomap_page *iop = to_iomap_page(page);
146 struct inode *inode = page->mapping->host;
147 unsigned first = off >> inode->i_blkbits;
148 unsigned last = (off + len - 1) >> inode->i_blkbits;
151 spin_lock_irqsave(&iop->uptodate_lock, flags);
152 bitmap_set(iop->uptodate, first, last - first + 1);
153 if (bitmap_full(iop->uptodate, i_blocks_per_page(inode, page)))
154 SetPageUptodate(page);
155 spin_unlock_irqrestore(&iop->uptodate_lock, flags);
159 iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len)
164 if (page_has_private(page))
165 iomap_iop_set_range_uptodate(page, off, len);
167 SetPageUptodate(page);
171 iomap_read_finish(struct iomap_page *iop, struct page *page)
173 if (!iop || atomic_dec_and_test(&iop->read_count))
178 iomap_read_page_end_io(struct bio_vec *bvec, int error)
180 struct page *page = bvec->bv_page;
181 struct iomap_page *iop = to_iomap_page(page);
183 if (unlikely(error)) {
184 ClearPageUptodate(page);
187 iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len);
190 iomap_read_finish(iop, page);
194 iomap_read_end_io(struct bio *bio)
196 int error = blk_status_to_errno(bio->bi_status);
197 struct bio_vec *bvec;
198 struct bvec_iter_all iter_all;
200 bio_for_each_segment_all(bvec, bio, iter_all)
201 iomap_read_page_end_io(bvec, error);
205 struct iomap_readpage_ctx {
206 struct page *cur_page;
207 bool cur_page_in_bio;
209 struct readahead_control *rac;
213 iomap_read_inline_data(struct inode *inode, struct page *page,
216 size_t size = i_size_read(inode);
219 if (PageUptodate(page))
223 BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data));
225 addr = kmap_atomic(page);
226 memcpy(addr, iomap->inline_data, size);
227 memset(addr + size, 0, PAGE_SIZE - size);
229 SetPageUptodate(page);
232 static inline bool iomap_block_needs_zeroing(struct inode *inode,
233 struct iomap *iomap, loff_t pos)
235 return iomap->type != IOMAP_MAPPED ||
236 (iomap->flags & IOMAP_F_NEW) ||
237 pos >= i_size_read(inode);
241 iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
242 struct iomap *iomap, struct iomap *srcmap)
244 struct iomap_readpage_ctx *ctx = data;
245 struct page *page = ctx->cur_page;
246 struct iomap_page *iop = iomap_page_create(inode, page);
247 bool same_page = false, is_contig = false;
248 loff_t orig_pos = pos;
252 if (iomap->type == IOMAP_INLINE) {
254 iomap_read_inline_data(inode, page, iomap);
258 /* zero post-eof blocks as the page may be mapped */
259 iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen);
263 if (iomap_block_needs_zeroing(inode, iomap, pos)) {
264 zero_user(page, poff, plen);
265 iomap_set_range_uptodate(page, poff, plen);
269 ctx->cur_page_in_bio = true;
272 * Try to merge into a previous segment if we can.
274 sector = iomap_sector(iomap, pos);
275 if (ctx->bio && bio_end_sector(ctx->bio) == sector)
279 __bio_try_merge_page(ctx->bio, page, plen, poff, &same_page)) {
280 if (!same_page && iop)
281 atomic_inc(&iop->read_count);
286 * If we start a new segment we need to increase the read count, and we
287 * need to do so before submitting any previous full bio to make sure
288 * that we don't prematurely unlock the page.
291 atomic_inc(&iop->read_count);
293 if (!ctx->bio || !is_contig || bio_full(ctx->bio, plen)) {
294 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
295 gfp_t orig_gfp = gfp;
296 int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT;
299 submit_bio(ctx->bio);
301 if (ctx->rac) /* same as readahead_gfp_mask */
302 gfp |= __GFP_NORETRY | __GFP_NOWARN;
303 ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs));
305 * If the bio_alloc fails, try it again for a single page to
306 * avoid having to deal with partial page reads. This emulates
307 * what do_mpage_readpage does.
310 ctx->bio = bio_alloc(orig_gfp, 1);
311 ctx->bio->bi_opf = REQ_OP_READ;
313 ctx->bio->bi_opf |= REQ_RAHEAD;
314 ctx->bio->bi_iter.bi_sector = sector;
315 bio_set_dev(ctx->bio, iomap->bdev);
316 ctx->bio->bi_end_io = iomap_read_end_io;
319 bio_add_page(ctx->bio, page, plen, poff);
322 * Move the caller beyond our range so that it keeps making progress.
323 * For that we have to include any leading non-uptodate ranges, but
324 * we can skip trailing ones as they will be handled in the next
327 return pos - orig_pos + plen;
331 iomap_readpage(struct page *page, const struct iomap_ops *ops)
333 struct iomap_readpage_ctx ctx = { .cur_page = page };
334 struct inode *inode = page->mapping->host;
338 trace_iomap_readpage(page->mapping->host, 1);
340 for (poff = 0; poff < PAGE_SIZE; poff += ret) {
341 ret = iomap_apply(inode, page_offset(page) + poff,
342 PAGE_SIZE - poff, 0, ops, &ctx,
343 iomap_readpage_actor);
345 WARN_ON_ONCE(ret == 0);
353 WARN_ON_ONCE(!ctx.cur_page_in_bio);
355 WARN_ON_ONCE(ctx.cur_page_in_bio);
360 * Just like mpage_readahead and block_read_full_page we always
361 * return 0 and just mark the page as PageError on errors. This
362 * should be cleaned up all through the stack eventually.
366 EXPORT_SYMBOL_GPL(iomap_readpage);
369 iomap_readahead_actor(struct inode *inode, loff_t pos, loff_t length,
370 void *data, struct iomap *iomap, struct iomap *srcmap)
372 struct iomap_readpage_ctx *ctx = data;
375 for (done = 0; done < length; done += ret) {
376 if (ctx->cur_page && offset_in_page(pos + done) == 0) {
377 if (!ctx->cur_page_in_bio)
378 unlock_page(ctx->cur_page);
379 put_page(ctx->cur_page);
380 ctx->cur_page = NULL;
382 if (!ctx->cur_page) {
383 ctx->cur_page = readahead_page(ctx->rac);
384 ctx->cur_page_in_bio = false;
386 ret = iomap_readpage_actor(inode, pos + done, length - done,
394 * iomap_readahead - Attempt to read pages from a file.
395 * @rac: Describes the pages to be read.
396 * @ops: The operations vector for the filesystem.
398 * This function is for filesystems to call to implement their readahead
399 * address_space operation.
401 * Context: The @ops callbacks may submit I/O (eg to read the addresses of
402 * blocks from disc), and may wait for it. The caller may be trying to
403 * access a different page, and so sleeping excessively should be avoided.
404 * It may allocate memory, but should avoid costly allocations. This
405 * function is called with memalloc_nofs set, so allocations will not cause
406 * the filesystem to be reentered.
408 void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops)
410 struct inode *inode = rac->mapping->host;
411 loff_t pos = readahead_pos(rac);
412 loff_t length = readahead_length(rac);
413 struct iomap_readpage_ctx ctx = {
417 trace_iomap_readahead(inode, readahead_count(rac));
420 loff_t ret = iomap_apply(inode, pos, length, 0, ops,
421 &ctx, iomap_readahead_actor);
423 WARN_ON_ONCE(ret == 0);
433 if (!ctx.cur_page_in_bio)
434 unlock_page(ctx.cur_page);
435 put_page(ctx.cur_page);
438 EXPORT_SYMBOL_GPL(iomap_readahead);
441 * iomap_is_partially_uptodate checks whether blocks within a page are
444 * Returns true if all blocks which correspond to a file portion
445 * we want to read within the page are uptodate.
448 iomap_is_partially_uptodate(struct page *page, unsigned long from,
451 struct iomap_page *iop = to_iomap_page(page);
452 struct inode *inode = page->mapping->host;
453 unsigned len, first, last;
456 /* Limit range to one page */
457 len = min_t(unsigned, PAGE_SIZE - from, count);
459 /* First and last blocks in range within page */
460 first = from >> inode->i_blkbits;
461 last = (from + len - 1) >> inode->i_blkbits;
464 for (i = first; i <= last; i++)
465 if (!test_bit(i, iop->uptodate))
472 EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate);
475 iomap_releasepage(struct page *page, gfp_t gfp_mask)
477 trace_iomap_releasepage(page->mapping->host, page_offset(page),
481 * mm accommodates an old ext3 case where clean pages might not have had
482 * the dirty bit cleared. Thus, it can send actual dirty pages to
483 * ->releasepage() via shrink_active_list(), skip those here.
485 if (PageDirty(page) || PageWriteback(page))
487 iomap_page_release(page);
490 EXPORT_SYMBOL_GPL(iomap_releasepage);
493 iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len)
495 trace_iomap_invalidatepage(page->mapping->host, offset, len);
498 * If we are invalidating the entire page, clear the dirty state from it
499 * and release it to avoid unnecessary buildup of the LRU.
501 if (offset == 0 && len == PAGE_SIZE) {
502 WARN_ON_ONCE(PageWriteback(page));
503 cancel_dirty_page(page);
504 iomap_page_release(page);
507 EXPORT_SYMBOL_GPL(iomap_invalidatepage);
509 #ifdef CONFIG_MIGRATION
511 iomap_migrate_page(struct address_space *mapping, struct page *newpage,
512 struct page *page, enum migrate_mode mode)
516 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
517 if (ret != MIGRATEPAGE_SUCCESS)
520 if (page_has_private(page))
521 attach_page_private(newpage, detach_page_private(page));
523 if (mode != MIGRATE_SYNC_NO_COPY)
524 migrate_page_copy(newpage, page);
526 migrate_page_states(newpage, page);
527 return MIGRATEPAGE_SUCCESS;
529 EXPORT_SYMBOL_GPL(iomap_migrate_page);
530 #endif /* CONFIG_MIGRATION */
533 IOMAP_WRITE_F_UNSHARE = (1 << 0),
537 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
539 loff_t i_size = i_size_read(inode);
542 * Only truncate newly allocated pages beyoned EOF, even if the
543 * write started inside the existing inode size.
545 if (pos + len > i_size)
546 truncate_pagecache_range(inode, max(pos, i_size), pos + len);
550 iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff,
551 unsigned plen, struct iomap *iomap)
556 bio_init(&bio, &bvec, 1);
557 bio.bi_opf = REQ_OP_READ;
558 bio.bi_iter.bi_sector = iomap_sector(iomap, block_start);
559 bio_set_dev(&bio, iomap->bdev);
560 __bio_add_page(&bio, page, plen, poff);
561 return submit_bio_wait(&bio);
565 __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags,
566 struct page *page, struct iomap *srcmap)
568 struct iomap_page *iop = iomap_page_create(inode, page);
569 loff_t block_size = i_blocksize(inode);
570 loff_t block_start = round_down(pos, block_size);
571 loff_t block_end = round_up(pos + len, block_size);
572 unsigned from = offset_in_page(pos), to = from + len, poff, plen;
574 if (PageUptodate(page))
576 ClearPageError(page);
579 iomap_adjust_read_range(inode, iop, &block_start,
580 block_end - block_start, &poff, &plen);
584 if (!(flags & IOMAP_WRITE_F_UNSHARE) &&
585 (from <= poff || from >= poff + plen) &&
586 (to <= poff || to >= poff + plen))
589 if (iomap_block_needs_zeroing(inode, srcmap, block_start)) {
590 if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE))
592 zero_user_segments(page, poff, from, to, poff + plen);
594 int status = iomap_read_page_sync(block_start, page,
599 iomap_set_range_uptodate(page, poff, plen);
600 } while ((block_start += plen) < block_end);
606 iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
607 struct page **pagep, struct iomap *iomap, struct iomap *srcmap)
609 const struct iomap_page_ops *page_ops = iomap->page_ops;
613 BUG_ON(pos + len > iomap->offset + iomap->length);
615 BUG_ON(pos + len > srcmap->offset + srcmap->length);
617 if (fatal_signal_pending(current))
620 if (page_ops && page_ops->page_prepare) {
621 status = page_ops->page_prepare(inode, pos, len, iomap);
626 page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT,
633 if (srcmap->type == IOMAP_INLINE)
634 iomap_read_inline_data(inode, page, srcmap);
635 else if (iomap->flags & IOMAP_F_BUFFER_HEAD)
636 status = __block_write_begin_int(page, pos, len, NULL, srcmap);
638 status = __iomap_write_begin(inode, pos, len, flags, page,
641 if (unlikely(status))
650 iomap_write_failed(inode, pos, len);
653 if (page_ops && page_ops->page_done)
654 page_ops->page_done(inode, pos, 0, NULL, iomap);
659 iomap_set_page_dirty(struct page *page)
661 struct address_space *mapping = page_mapping(page);
664 if (unlikely(!mapping))
665 return !TestSetPageDirty(page);
668 * Lock out page->mem_cgroup migration to keep PageDirty
669 * synchronized with per-memcg dirty page counters.
671 lock_page_memcg(page);
672 newly_dirty = !TestSetPageDirty(page);
674 __set_page_dirty(page, mapping, 0);
675 unlock_page_memcg(page);
678 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
681 EXPORT_SYMBOL_GPL(iomap_set_page_dirty);
684 __iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
685 unsigned copied, struct page *page)
687 flush_dcache_page(page);
690 * The blocks that were entirely written will now be uptodate, so we
691 * don't have to worry about a readpage reading them and overwriting a
692 * partial write. However if we have encountered a short write and only
693 * partially written into a block, it will not be marked uptodate, so a
694 * readpage might come in and destroy our partial write.
696 * Do the simplest thing, and just treat any short write to a non
697 * uptodate page as a zero-length write, and force the caller to redo
700 if (unlikely(copied < len && !PageUptodate(page)))
702 iomap_set_range_uptodate(page, offset_in_page(pos), len);
703 iomap_set_page_dirty(page);
708 iomap_write_end_inline(struct inode *inode, struct page *page,
709 struct iomap *iomap, loff_t pos, unsigned copied)
713 WARN_ON_ONCE(!PageUptodate(page));
714 BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data));
716 flush_dcache_page(page);
717 addr = kmap_atomic(page);
718 memcpy(iomap->inline_data + pos, addr + pos, copied);
721 mark_inode_dirty(inode);
726 iomap_write_end(struct inode *inode, loff_t pos, unsigned len, unsigned copied,
727 struct page *page, struct iomap *iomap, struct iomap *srcmap)
729 const struct iomap_page_ops *page_ops = iomap->page_ops;
730 loff_t old_size = inode->i_size;
733 if (srcmap->type == IOMAP_INLINE) {
734 ret = iomap_write_end_inline(inode, page, iomap, pos, copied);
735 } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) {
736 ret = block_write_end(NULL, inode->i_mapping, pos, len, copied,
739 ret = __iomap_write_end(inode, pos, len, copied, page);
743 * Update the in-memory inode size after copying the data into the page
744 * cache. It's up to the file system to write the updated size to disk,
745 * preferably after I/O completion so that no stale data is exposed.
747 if (pos + ret > old_size) {
748 i_size_write(inode, pos + ret);
749 iomap->flags |= IOMAP_F_SIZE_CHANGED;
754 pagecache_isize_extended(inode, old_size, pos);
755 if (page_ops && page_ops->page_done)
756 page_ops->page_done(inode, pos, ret, page, iomap);
760 iomap_write_failed(inode, pos, len);
765 iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
766 struct iomap *iomap, struct iomap *srcmap)
768 struct iov_iter *i = data;
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, 0, &page, iomap,
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 status = iomap_write_end(inode, pos, bytes, copied, page, iomap,
812 if (unlikely(status < 0))
818 iov_iter_advance(i, copied);
819 if (unlikely(copied == 0)) {
821 * If we were unable to copy any data at all, we must
822 * fall back to a single segment length write.
824 * If we didn't fallback here, we could livelock
825 * because not all segments in the iov can be copied at
826 * once without a pagefault.
828 bytes = min_t(unsigned long, PAGE_SIZE - offset,
829 iov_iter_single_seg_count(i));
836 balance_dirty_pages_ratelimited(inode->i_mapping);
837 } while (iov_iter_count(i) && length);
839 return written ? written : status;
843 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
844 const struct iomap_ops *ops)
846 struct inode *inode = iocb->ki_filp->f_mapping->host;
847 loff_t pos = iocb->ki_pos, ret = 0, written = 0;
849 while (iov_iter_count(iter)) {
850 ret = iomap_apply(inode, pos, iov_iter_count(iter),
851 IOMAP_WRITE, ops, iter, iomap_write_actor);
858 return written ? written : ret;
860 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
863 iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
864 struct iomap *iomap, struct iomap *srcmap)
869 /* don't bother with blocks that are not shared to start with */
870 if (!(iomap->flags & IOMAP_F_SHARED))
872 /* don't bother with holes or unwritten extents */
873 if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
877 unsigned long offset = offset_in_page(pos);
878 unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length);
881 status = iomap_write_begin(inode, pos, bytes,
882 IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap);
883 if (unlikely(status))
886 status = iomap_write_end(inode, pos, bytes, bytes, page, iomap,
888 if (unlikely(status <= 0)) {
889 if (WARN_ON_ONCE(status == 0))
900 balance_dirty_pages_ratelimited(inode->i_mapping);
907 iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len,
908 const struct iomap_ops *ops)
913 ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
914 iomap_unshare_actor);
923 EXPORT_SYMBOL_GPL(iomap_file_unshare);
925 static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
926 unsigned bytes, struct iomap *iomap, struct iomap *srcmap)
931 status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap);
935 zero_user(page, offset, bytes);
936 mark_page_accessed(page);
938 return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap);
942 iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
943 void *data, struct iomap *iomap, struct iomap *srcmap)
945 bool *did_zero = data;
949 /* already zeroed? we're done. */
950 if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
954 unsigned offset, bytes;
956 offset = offset_in_page(pos);
957 bytes = min_t(loff_t, PAGE_SIZE - offset, count);
960 status = dax_iomap_zero(pos, offset, bytes, iomap);
962 status = iomap_zero(inode, pos, offset, bytes, iomap,
978 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
979 const struct iomap_ops *ops)
984 ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
985 ops, did_zero, iomap_zero_range_actor);
995 EXPORT_SYMBOL_GPL(iomap_zero_range);
998 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
999 const struct iomap_ops *ops)
1001 unsigned int blocksize = i_blocksize(inode);
1002 unsigned int off = pos & (blocksize - 1);
1004 /* Block boundary? Nothing to do */
1007 return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
1009 EXPORT_SYMBOL_GPL(iomap_truncate_page);
1012 iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
1013 void *data, struct iomap *iomap, struct iomap *srcmap)
1015 struct page *page = data;
1018 if (iomap->flags & IOMAP_F_BUFFER_HEAD) {
1019 ret = __block_write_begin_int(page, pos, length, NULL, iomap);
1022 block_commit_write(page, 0, length);
1024 WARN_ON_ONCE(!PageUptodate(page));
1025 iomap_page_create(inode, page);
1026 set_page_dirty(page);
1032 vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops)
1034 struct page *page = vmf->page;
1035 struct inode *inode = file_inode(vmf->vma->vm_file);
1036 unsigned long length;
1041 ret = page_mkwrite_check_truncate(page, inode);
1046 offset = page_offset(page);
1047 while (length > 0) {
1048 ret = iomap_apply(inode, offset, length,
1049 IOMAP_WRITE | IOMAP_FAULT, ops, page,
1050 iomap_page_mkwrite_actor);
1051 if (unlikely(ret <= 0))
1057 wait_for_stable_page(page);
1058 return VM_FAULT_LOCKED;
1061 return block_page_mkwrite_return(ret);
1063 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
1066 iomap_finish_page_writeback(struct inode *inode, struct page *page,
1069 struct iomap_page *iop = to_iomap_page(page);
1073 mapping_set_error(inode->i_mapping, -EIO);
1076 WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop);
1077 WARN_ON_ONCE(iop && atomic_read(&iop->write_count) <= 0);
1079 if (!iop || atomic_dec_and_test(&iop->write_count))
1080 end_page_writeback(page);
1084 * We're now finished for good with this ioend structure. Update the page
1085 * state, release holds on bios, and finally free up memory. Do not use the
1089 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
1091 struct inode *inode = ioend->io_inode;
1092 struct bio *bio = &ioend->io_inline_bio;
1093 struct bio *last = ioend->io_bio, *next;
1094 u64 start = bio->bi_iter.bi_sector;
1095 loff_t offset = ioend->io_offset;
1096 bool quiet = bio_flagged(bio, BIO_QUIET);
1098 for (bio = &ioend->io_inline_bio; bio; bio = next) {
1100 struct bvec_iter_all iter_all;
1103 * For the last bio, bi_private points to the ioend, so we
1104 * need to explicitly end the iteration here.
1109 next = bio->bi_private;
1111 /* walk each page on bio, ending page IO on them */
1112 bio_for_each_segment_all(bv, bio, iter_all)
1113 iomap_finish_page_writeback(inode, bv->bv_page, error);
1116 /* The ioend has been freed by bio_put() */
1118 if (unlikely(error && !quiet)) {
1119 printk_ratelimited(KERN_ERR
1120 "%s: writeback error on inode %lu, offset %lld, sector %llu",
1121 inode->i_sb->s_id, inode->i_ino, offset, start);
1126 iomap_finish_ioends(struct iomap_ioend *ioend, int error)
1128 struct list_head tmp;
1130 list_replace_init(&ioend->io_list, &tmp);
1131 iomap_finish_ioend(ioend, error);
1133 while (!list_empty(&tmp)) {
1134 ioend = list_first_entry(&tmp, struct iomap_ioend, io_list);
1135 list_del_init(&ioend->io_list);
1136 iomap_finish_ioend(ioend, error);
1139 EXPORT_SYMBOL_GPL(iomap_finish_ioends);
1142 * We can merge two adjacent ioends if they have the same set of work to do.
1145 iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next)
1147 if (ioend->io_bio->bi_status != next->io_bio->bi_status)
1149 if ((ioend->io_flags & IOMAP_F_SHARED) ^
1150 (next->io_flags & IOMAP_F_SHARED))
1152 if ((ioend->io_type == IOMAP_UNWRITTEN) ^
1153 (next->io_type == IOMAP_UNWRITTEN))
1155 if (ioend->io_offset + ioend->io_size != next->io_offset)
1161 iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends,
1162 void (*merge_private)(struct iomap_ioend *ioend,
1163 struct iomap_ioend *next))
1165 struct iomap_ioend *next;
1167 INIT_LIST_HEAD(&ioend->io_list);
1169 while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend,
1171 if (!iomap_ioend_can_merge(ioend, next))
1173 list_move_tail(&next->io_list, &ioend->io_list);
1174 ioend->io_size += next->io_size;
1175 if (next->io_private && merge_private)
1176 merge_private(ioend, next);
1179 EXPORT_SYMBOL_GPL(iomap_ioend_try_merge);
1182 iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b)
1184 struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list);
1185 struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list);
1187 if (ia->io_offset < ib->io_offset)
1189 if (ia->io_offset > ib->io_offset)
1195 iomap_sort_ioends(struct list_head *ioend_list)
1197 list_sort(NULL, ioend_list, iomap_ioend_compare);
1199 EXPORT_SYMBOL_GPL(iomap_sort_ioends);
1201 static void iomap_writepage_end_bio(struct bio *bio)
1203 struct iomap_ioend *ioend = bio->bi_private;
1205 iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status));
1209 * Submit the final bio for an ioend.
1211 * If @error is non-zero, it means that we have a situation where some part of
1212 * the submission process has failed after we have marked paged for writeback
1213 * and unlocked them. In this situation, we need to fail the bio instead of
1214 * submitting it. This typically only happens on a filesystem shutdown.
1217 iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend,
1220 ioend->io_bio->bi_private = ioend;
1221 ioend->io_bio->bi_end_io = iomap_writepage_end_bio;
1223 if (wpc->ops->prepare_ioend)
1224 error = wpc->ops->prepare_ioend(ioend, error);
1227 * If we are failing the IO now, just mark the ioend with an
1228 * error and finish it. This will run IO completion immediately
1229 * as there is only one reference to the ioend at this point in
1232 ioend->io_bio->bi_status = errno_to_blk_status(error);
1233 bio_endio(ioend->io_bio);
1237 submit_bio(ioend->io_bio);
1241 static struct iomap_ioend *
1242 iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc,
1243 loff_t offset, sector_t sector, struct writeback_control *wbc)
1245 struct iomap_ioend *ioend;
1248 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &iomap_ioend_bioset);
1249 bio_set_dev(bio, wpc->iomap.bdev);
1250 bio->bi_iter.bi_sector = sector;
1251 bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
1252 bio->bi_write_hint = inode->i_write_hint;
1253 wbc_init_bio(wbc, bio);
1255 ioend = container_of(bio, struct iomap_ioend, io_inline_bio);
1256 INIT_LIST_HEAD(&ioend->io_list);
1257 ioend->io_type = wpc->iomap.type;
1258 ioend->io_flags = wpc->iomap.flags;
1259 ioend->io_inode = inode;
1261 ioend->io_offset = offset;
1262 ioend->io_private = NULL;
1263 ioend->io_bio = bio;
1268 * Allocate a new bio, and chain the old bio to the new one.
1270 * Note that we have to do perform the chaining in this unintuitive order
1271 * so that the bi_private linkage is set up in the right direction for the
1272 * traversal in iomap_finish_ioend().
1275 iomap_chain_bio(struct bio *prev)
1279 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
1280 bio_copy_dev(new, prev);/* also copies over blkcg information */
1281 new->bi_iter.bi_sector = bio_end_sector(prev);
1282 new->bi_opf = prev->bi_opf;
1283 new->bi_write_hint = prev->bi_write_hint;
1285 bio_chain(prev, new);
1286 bio_get(prev); /* for iomap_finish_ioend */
1292 iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset,
1295 if ((wpc->iomap.flags & IOMAP_F_SHARED) !=
1296 (wpc->ioend->io_flags & IOMAP_F_SHARED))
1298 if (wpc->iomap.type != wpc->ioend->io_type)
1300 if (offset != wpc->ioend->io_offset + wpc->ioend->io_size)
1302 if (sector != bio_end_sector(wpc->ioend->io_bio))
1308 * Test to see if we have an existing ioend structure that we could append to
1309 * first, otherwise finish off the current ioend and start another.
1312 iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page,
1313 struct iomap_page *iop, struct iomap_writepage_ctx *wpc,
1314 struct writeback_control *wbc, struct list_head *iolist)
1316 sector_t sector = iomap_sector(&wpc->iomap, offset);
1317 unsigned len = i_blocksize(inode);
1318 unsigned poff = offset & (PAGE_SIZE - 1);
1319 bool merged, same_page = false;
1321 if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) {
1323 list_add(&wpc->ioend->io_list, iolist);
1324 wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc);
1327 merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff,
1329 if (iop && !same_page)
1330 atomic_inc(&iop->write_count);
1333 if (bio_full(wpc->ioend->io_bio, len)) {
1334 wpc->ioend->io_bio =
1335 iomap_chain_bio(wpc->ioend->io_bio);
1337 bio_add_page(wpc->ioend->io_bio, page, len, poff);
1340 wpc->ioend->io_size += len;
1341 wbc_account_cgroup_owner(wbc, page, len);
1345 * We implement an immediate ioend submission policy here to avoid needing to
1346 * chain multiple ioends and hence nest mempool allocations which can violate
1347 * forward progress guarantees we need to provide. The current ioend we are
1348 * adding blocks to is cached on the writepage context, and if the new block
1349 * does not append to the cached ioend it will create a new ioend and cache that
1352 * If a new ioend is created and cached, the old ioend is returned and queued
1353 * locally for submission once the entire page is processed or an error has been
1354 * detected. While ioends are submitted immediately after they are completed,
1355 * batching optimisations are provided by higher level block plugging.
1357 * At the end of a writeback pass, there will be a cached ioend remaining on the
1358 * writepage context that the caller will need to submit.
1361 iomap_writepage_map(struct iomap_writepage_ctx *wpc,
1362 struct writeback_control *wbc, struct inode *inode,
1363 struct page *page, u64 end_offset)
1365 struct iomap_page *iop = to_iomap_page(page);
1366 struct iomap_ioend *ioend, *next;
1367 unsigned len = i_blocksize(inode);
1368 u64 file_offset; /* file offset of page */
1369 int error = 0, count = 0, i;
1370 LIST_HEAD(submit_list);
1372 WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop);
1373 WARN_ON_ONCE(iop && atomic_read(&iop->write_count) != 0);
1376 * Walk through the page to find areas to write back. If we run off the
1377 * end of the current map or find the current map invalid, grab a new
1380 for (i = 0, file_offset = page_offset(page);
1381 i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
1382 i++, file_offset += len) {
1383 if (iop && !test_bit(i, iop->uptodate))
1386 error = wpc->ops->map_blocks(wpc, inode, file_offset);
1389 if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE))
1391 if (wpc->iomap.type == IOMAP_HOLE)
1393 iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
1398 WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list));
1399 WARN_ON_ONCE(!PageLocked(page));
1400 WARN_ON_ONCE(PageWriteback(page));
1403 * We cannot cancel the ioend directly here on error. We may have
1404 * already set other pages under writeback and hence we have to run I/O
1405 * completion to mark the error state of the pages under writeback
1408 if (unlikely(error)) {
1411 * If the current page hasn't been added to ioend, it
1412 * won't be affected by I/O completions and we must
1413 * discard and unlock it right here.
1415 if (wpc->ops->discard_page)
1416 wpc->ops->discard_page(page);
1417 ClearPageUptodate(page);
1423 * If the page was not fully cleaned, we need to ensure that the
1424 * higher layers come back to it correctly. That means we need
1425 * to keep the page dirty, and for WB_SYNC_ALL writeback we need
1426 * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed
1427 * so another attempt to write this page in this writeback sweep
1430 set_page_writeback_keepwrite(page);
1432 clear_page_dirty_for_io(page);
1433 set_page_writeback(page);
1439 * Preserve the original error if there was one, otherwise catch
1440 * submission errors here and propagate into subsequent ioend
1443 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
1446 list_del_init(&ioend->io_list);
1447 error2 = iomap_submit_ioend(wpc, ioend, error);
1448 if (error2 && !error)
1453 * We can end up here with no error and nothing to write only if we race
1454 * with a partial page truncate on a sub-page block sized filesystem.
1457 end_page_writeback(page);
1459 mapping_set_error(page->mapping, error);
1464 * Write out a dirty page.
1466 * For delalloc space on the page we need to allocate space and flush it.
1467 * For unwritten space on the page we need to start the conversion to
1468 * regular allocated space.
1471 iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data)
1473 struct iomap_writepage_ctx *wpc = data;
1474 struct inode *inode = page->mapping->host;
1479 trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE);
1482 * Refuse to write the page out if we are called from reclaim context.
1484 * This avoids stack overflows when called from deeply used stacks in
1485 * random callers for direct reclaim or memcg reclaim. We explicitly
1486 * allow reclaim from kswapd as the stack usage there is relatively low.
1488 * This should never happen except in the case of a VM regression so
1491 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1496 * Given that we do not allow direct reclaim to call us, we should
1497 * never be called in a recursive filesystem reclaim context.
1499 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
1503 * Is this page beyond the end of the file?
1505 * The page index is less than the end_index, adjust the end_offset
1506 * to the highest offset that this page should represent.
1507 * -----------------------------------------------------
1508 * | file mapping | <EOF> |
1509 * -----------------------------------------------------
1510 * | Page ... | Page N-2 | Page N-1 | Page N | |
1511 * ^--------------------------------^----------|--------
1512 * | desired writeback range | see else |
1513 * ---------------------------------^------------------|
1515 offset = i_size_read(inode);
1516 end_index = offset >> PAGE_SHIFT;
1517 if (page->index < end_index)
1518 end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT;
1521 * Check whether the page to write out is beyond or straddles
1523 * -------------------------------------------------------
1524 * | file mapping | <EOF> |
1525 * -------------------------------------------------------
1526 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1527 * ^--------------------------------^-----------|---------
1529 * ---------------------------------^-----------|--------|
1531 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1534 * Skip the page if it is fully outside i_size, e.g. due to a
1535 * truncate operation that is in progress. We must redirty the
1536 * page so that reclaim stops reclaiming it. Otherwise
1537 * iomap_vm_releasepage() is called on it and gets confused.
1539 * Note that the end_index is unsigned long, it would overflow
1540 * if the given offset is greater than 16TB on 32-bit system
1541 * and if we do check the page is fully outside i_size or not
1542 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1543 * will be evaluated to 0. Hence this page will be redirtied
1544 * and be written out repeatedly which would result in an
1545 * infinite loop, the user program that perform this operation
1546 * will hang. Instead, we can verify this situation by checking
1547 * if the page to write is totally beyond the i_size or if it's
1548 * offset is just equal to the EOF.
1550 if (page->index > end_index ||
1551 (page->index == end_index && offset_into_page == 0))
1555 * The page straddles i_size. It must be zeroed out on each
1556 * and every writepage invocation because it may be mmapped.
1557 * "A file is mapped in multiples of the page size. For a file
1558 * that is not a multiple of the page size, the remaining
1559 * memory is zeroed when mapped, and writes to that region are
1560 * not written out to the file."
1562 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1564 /* Adjust the end_offset to the end of file */
1565 end_offset = offset;
1568 return iomap_writepage_map(wpc, wbc, inode, page, end_offset);
1571 redirty_page_for_writepage(wbc, page);
1577 iomap_writepage(struct page *page, struct writeback_control *wbc,
1578 struct iomap_writepage_ctx *wpc,
1579 const struct iomap_writeback_ops *ops)
1584 ret = iomap_do_writepage(page, wbc, wpc);
1587 return iomap_submit_ioend(wpc, wpc->ioend, ret);
1589 EXPORT_SYMBOL_GPL(iomap_writepage);
1592 iomap_writepages(struct address_space *mapping, struct writeback_control *wbc,
1593 struct iomap_writepage_ctx *wpc,
1594 const struct iomap_writeback_ops *ops)
1599 ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc);
1602 return iomap_submit_ioend(wpc, wpc->ioend, ret);
1604 EXPORT_SYMBOL_GPL(iomap_writepages);
1606 static int __init iomap_init(void)
1608 return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1609 offsetof(struct iomap_ioend, io_inline_bio),
1612 fs_initcall(iomap_init);