Skip to content

Commit 9dc55f1

Browse files
Christoph Hellwigdjwong
Christoph Hellwig
authored and
djwong
committed
iomap: add support for sub-pagesize buffered I/O without buffer heads
After already supporting a simple implementation of buffered writes for the blocksize == PAGE_SIZE case in the last commit this adds full support even for smaller block sizes. There are three bits of per-block information in the buffer_head structure that really matter for the iomap read and write path: - uptodate status (BH_uptodate) - marked as currently under read I/O (BH_Async_Read) - marked as currently under write I/O (BH_Async_Write) Instead of having new per-block structures this now adds a per-page structure called struct iomap_page to track this information in a slightly different form: - a bitmap for the per-block uptodate status. For worst case of a 64k page size system this bitmap needs to contain 128 bits. For the typical 4k page size case it only needs 8 bits, although we still need a full unsigned long due to the way the atomic bitmap API works. - two atomic_t counters are used to track the outstanding read and write counts There is quite a bit of boilerplate code as the buffered I/O path uses various helper methods, but the actual code is very straight forward. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
1 parent ac8ee54 commit 9dc55f1

File tree

2 files changed

+290
-21
lines changed

2 files changed

+290
-21
lines changed

‎fs/iomap.c

Lines changed: 259 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,7 @@
1717
#include <linux/iomap.h>
1818
#include <linux/uaccess.h>
1919
#include <linux/gfp.h>
20+
#include <linux/migrate.h>
2021
#include <linux/mm.h>
2122
#include <linux/mm_inline.h>
2223
#include <linux/swap.h>
@@ -104,6 +105,138 @@ iomap_sector(struct iomap *iomap, loff_t pos)
104105
return (iomap->addr + pos - iomap->offset) >> SECTOR_SHIFT;
105106
}
106107

108+
static struct iomap_page *
109+
iomap_page_create(struct inode *inode, struct page *page)
110+
{
111+
struct iomap_page *iop = to_iomap_page(page);
112+
113+
if (iop || i_blocksize(inode) == PAGE_SIZE)
114+
return iop;
115+
116+
iop = kmalloc(sizeof(*iop), GFP_NOFS | __GFP_NOFAIL);
117+
atomic_set(&iop->read_count, 0);
118+
atomic_set(&iop->write_count, 0);
119+
bitmap_zero(iop->uptodate, PAGE_SIZE / SECTOR_SIZE);
120+
set_page_private(page, (unsigned long)iop);
121+
SetPagePrivate(page);
122+
return iop;
123+
}
124+
125+
static void
126+
iomap_page_release(struct page *page)
127+
{
128+
struct iomap_page *iop = to_iomap_page(page);
129+
130+
if (!iop)
131+
return;
132+
WARN_ON_ONCE(atomic_read(&iop->read_count));
133+
WARN_ON_ONCE(atomic_read(&iop->write_count));
134+
ClearPagePrivate(page);
135+
set_page_private(page, 0);
136+
kfree(iop);
137+
}
138+
139+
/*
140+
* Calculate the range inside the page that we actually need to read.
141+
*/
142+
static void
143+
iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop,
144+
loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp)
145+
{
146+
unsigned block_bits = inode->i_blkbits;
147+
unsigned block_size = (1 << block_bits);
148+
unsigned poff = *pos & (PAGE_SIZE - 1);
149+
unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
150+
unsigned first = poff >> block_bits;
151+
unsigned last = (poff + plen - 1) >> block_bits;
152+
unsigned end = (i_size_read(inode) & (PAGE_SIZE - 1)) >> block_bits;
153+
154+
/*
155+
* If the block size is smaller than the page size we need to check the
156+
* per-block uptodate status and adjust the offset and length if needed
157+
* to avoid reading in already uptodate ranges.
158+
*/
159+
if (iop) {
160+
unsigned int i;
161+
162+
/* move forward for each leading block marked uptodate */
163+
for (i = first; i <= last; i++) {
164+
if (!test_bit(i, iop->uptodate))
165+
break;
166+
*pos += block_size;
167+
poff += block_size;
168+
plen -= block_size;
169+
first++;
170+
}
171+
172+
/* truncate len if we find any trailing uptodate block(s) */
173+
for ( ; i <= last; i++) {
174+
if (test_bit(i, iop->uptodate)) {
175+
plen -= (last - i + 1) * block_size;
176+
last = i - 1;
177+
break;
178+
}
179+
}
180+
}
181+
182+
/*
183+
* If the extent spans the block that contains the i_size we need to
184+
* handle both halves separately so that we properly zero data in the
185+
* page cache for blocks that are entirely outside of i_size.
186+
*/
187+
if (first <= end && last > end)
188+
plen -= (last - end) * block_size;
189+
190+
*offp = poff;
191+
*lenp = plen;
192+
}
193+
194+
static void
195+
iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len)
196+
{
197+
struct iomap_page *iop = to_iomap_page(page);
198+
struct inode *inode = page->mapping->host;
199+
unsigned first = off >> inode->i_blkbits;
200+
unsigned last = (off + len - 1) >> inode->i_blkbits;
201+
unsigned int i;
202+
bool uptodate = true;
203+
204+
if (iop) {
205+
for (i = 0; i < PAGE_SIZE / i_blocksize(inode); i++) {
206+
if (i >= first && i <= last)
207+
set_bit(i, iop->uptodate);
208+
else if (!test_bit(i, iop->uptodate))
209+
uptodate = false;
210+
}
211+
}
212+
213+
if (uptodate && !PageError(page))
214+
SetPageUptodate(page);
215+
}
216+
217+
static void
218+
iomap_read_finish(struct iomap_page *iop, struct page *page)
219+
{
220+
if (!iop || atomic_dec_and_test(&iop->read_count))
221+
unlock_page(page);
222+
}
223+
224+
static void
225+
iomap_read_page_end_io(struct bio_vec *bvec, int error)
226+
{
227+
struct page *page = bvec->bv_page;
228+
struct iomap_page *iop = to_iomap_page(page);
229+
230+
if (unlikely(error)) {
231+
ClearPageUptodate(page);
232+
SetPageError(page);
233+
} else {
234+
iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len);
235+
}
236+
237+
iomap_read_finish(iop, page);
238+
}
239+
107240
static void
108241
iomap_read_inline_data(struct inode *inode, struct page *page,
109242
struct iomap *iomap)
@@ -132,7 +265,7 @@ iomap_read_end_io(struct bio *bio)
132265
int i;
133266

134267
bio_for_each_segment_all(bvec, bio, i)
135-
page_endio(bvec->bv_page, false, error);
268+
iomap_read_page_end_io(bvec, error);
136269
bio_put(bio);
137270
}
138271

@@ -150,9 +283,10 @@ iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
150283
{
151284
struct iomap_readpage_ctx *ctx = data;
152285
struct page *page = ctx->cur_page;
153-
unsigned poff = pos & (PAGE_SIZE - 1);
154-
unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
286+
struct iomap_page *iop = iomap_page_create(inode, page);
155287
bool is_contig = false;
288+
loff_t orig_pos = pos;
289+
unsigned poff, plen;
156290
sector_t sector;
157291

158292
if (iomap->type == IOMAP_INLINE) {
@@ -161,13 +295,14 @@ iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
161295
return PAGE_SIZE;
162296
}
163297

164-
/* we don't support blocksize < PAGE_SIZE quite yet. */
165-
WARN_ON_ONCE(pos != page_offset(page));
166-
WARN_ON_ONCE(plen != PAGE_SIZE);
298+
/* zero post-eof blocks as the page may be mapped */
299+
iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen);
300+
if (plen == 0)
301+
goto done;
167302

168303
if (iomap->type != IOMAP_MAPPED || pos >= i_size_read(inode)) {
169304
zero_user(page, poff, plen);
170-
SetPageUptodate(page);
305+
iomap_set_range_uptodate(page, poff, plen);
171306
goto done;
172307
}
173308

@@ -183,6 +318,14 @@ iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
183318
is_contig = true;
184319
}
185320

321+
/*
322+
* If we start a new segment we need to increase the read count, and we
323+
* need to do so before submitting any previous full bio to make sure
324+
* that we don't prematurely unlock the page.
325+
*/
326+
if (iop)
327+
atomic_inc(&iop->read_count);
328+
186329
if (!ctx->bio || !is_contig || bio_full(ctx->bio)) {
187330
gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
188331
int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT;
@@ -203,7 +346,13 @@ iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
203346

204347
__bio_add_page(ctx->bio, page, plen, poff);
205348
done:
206-
return plen;
349+
/*
350+
* Move the caller beyond our range so that it keeps making progress.
351+
* For that we have to include any leading non-uptodate ranges, but
352+
* we can skip trailing ones as they will be handled in the next
353+
* iteration.
354+
*/
355+
return pos - orig_pos + plen;
207356
}
208357

209358
int
@@ -214,8 +363,6 @@ iomap_readpage(struct page *page, const struct iomap_ops *ops)
214363
unsigned poff;
215364
loff_t ret;
216365

217-
WARN_ON_ONCE(page_has_buffers(page));
218-
219366
for (poff = 0; poff < PAGE_SIZE; poff += ret) {
220367
ret = iomap_apply(inode, page_offset(page) + poff,
221368
PAGE_SIZE - poff, 0, ops, &ctx,
@@ -341,6 +488,84 @@ iomap_readpages(struct address_space *mapping, struct list_head *pages,
341488
}
342489
EXPORT_SYMBOL_GPL(iomap_readpages);
343490

491+
int
492+
iomap_is_partially_uptodate(struct page *page, unsigned long from,
493+
unsigned long count)
494+
{
495+
struct iomap_page *iop = to_iomap_page(page);
496+
struct inode *inode = page->mapping->host;
497+
unsigned first = from >> inode->i_blkbits;
498+
unsigned last = (from + count - 1) >> inode->i_blkbits;
499+
unsigned i;
500+
501+
if (iop) {
502+
for (i = first; i <= last; i++)
503+
if (!test_bit(i, iop->uptodate))
504+
return 0;
505+
return 1;
506+
}
507+
508+
return 0;
509+
}
510+
EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate);
511+
512+
int
513+
iomap_releasepage(struct page *page, gfp_t gfp_mask)
514+
{
515+
/*
516+
* mm accommodates an old ext3 case where clean pages might not have had
517+
* the dirty bit cleared. Thus, it can send actual dirty pages to
518+
* ->releasepage() via shrink_active_list(), skip those here.
519+
*/
520+
if (PageDirty(page) || PageWriteback(page))
521+
return 0;
522+
iomap_page_release(page);
523+
return 1;
524+
}
525+
EXPORT_SYMBOL_GPL(iomap_releasepage);
526+
527+
void
528+
iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len)
529+
{
530+
/*
531+
* If we are invalidating the entire page, clear the dirty state from it
532+
* and release it to avoid unnecessary buildup of the LRU.
533+
*/
534+
if (offset == 0 && len == PAGE_SIZE) {
535+
WARN_ON_ONCE(PageWriteback(page));
536+
cancel_dirty_page(page);
537+
iomap_page_release(page);
538+
}
539+
}
540+
EXPORT_SYMBOL_GPL(iomap_invalidatepage);
541+
542+
#ifdef CONFIG_MIGRATION
543+
int
544+
iomap_migrate_page(struct address_space *mapping, struct page *newpage,
545+
struct page *page, enum migrate_mode mode)
546+
{
547+
int ret;
548+
549+
ret = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
550+
if (ret != MIGRATEPAGE_SUCCESS)
551+
return ret;
552+
553+
if (page_has_private(page)) {
554+
ClearPagePrivate(page);
555+
set_page_private(newpage, page_private(page));
556+
set_page_private(page, 0);
557+
SetPagePrivate(newpage);
558+
}
559+
560+
if (mode != MIGRATE_SYNC_NO_COPY)
561+
migrate_page_copy(newpage, page);
562+
else
563+
migrate_page_states(newpage, page);
564+
return MIGRATEPAGE_SUCCESS;
565+
}
566+
EXPORT_SYMBOL_GPL(iomap_migrate_page);
567+
#endif /* CONFIG_MIGRATION */
568+
344569
static void
345570
iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
346571
{
@@ -364,6 +589,7 @@ iomap_read_page_sync(struct inode *inode, loff_t block_start, struct page *page,
364589

365590
if (iomap->type != IOMAP_MAPPED || block_start >= i_size_read(inode)) {
366591
zero_user_segments(page, poff, from, to, poff + plen);
592+
iomap_set_range_uptodate(page, poff, plen);
367593
return 0;
368594
}
369595

@@ -379,21 +605,33 @@ static int
379605
__iomap_write_begin(struct inode *inode, loff_t pos, unsigned len,
380606
struct page *page, struct iomap *iomap)
381607
{
608+
struct iomap_page *iop = iomap_page_create(inode, page);
382609
loff_t block_size = i_blocksize(inode);
383610
loff_t block_start = pos & ~(block_size - 1);
384611
loff_t block_end = (pos + len + block_size - 1) & ~(block_size - 1);
385-
unsigned poff = block_start & (PAGE_SIZE - 1);
386-
unsigned plen = min_t(loff_t, PAGE_SIZE - poff, block_end - block_start);
387-
unsigned from = pos & (PAGE_SIZE - 1), to = from + len;
388-
389-
WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE);
612+
unsigned from = pos & (PAGE_SIZE - 1), to = from + len, poff, plen;
613+
int status = 0;
390614

391615
if (PageUptodate(page))
392616
return 0;
393-
if (from <= poff && to >= poff + plen)
394-
return 0;
395-
return iomap_read_page_sync(inode, block_start, page,
396-
poff, plen, from, to, iomap);
617+
618+
do {
619+
iomap_adjust_read_range(inode, iop, &block_start,
620+
block_end - block_start, &poff, &plen);
621+
if (plen == 0)
622+
break;
623+
624+
if ((from > poff && from < poff + plen) ||
625+
(to > poff && to < poff + plen)) {
626+
status = iomap_read_page_sync(inode, block_start, page,
627+
poff, plen, from, to, iomap);
628+
if (status)
629+
break;
630+
}
631+
632+
} while ((block_start += plen) < block_end);
633+
634+
return status;
397635
}
398636

399637
static int
@@ -476,7 +714,7 @@ __iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
476714
if (unlikely(copied < len && !PageUptodate(page))) {
477715
copied = 0;
478716
} else {
479-
SetPageUptodate(page);
717+
iomap_set_range_uptodate(page, pos & (PAGE_SIZE - 1), len);
480718
iomap_set_page_dirty(page);
481719
}
482720
return __generic_write_end(inode, pos, copied, page);
@@ -812,7 +1050,7 @@ iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
8121050
block_commit_write(page, 0, length);
8131051
} else {
8141052
WARN_ON_ONCE(!PageUptodate(page));
815-
WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE);
1053+
iomap_page_create(inode, page);
8161054
}
8171055

8181056
return length;

0 commit comments

Comments
 (0)