xfs_aops.c 38.6 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2 3
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
L
Linus Torvalds 已提交
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
L
Linus Torvalds 已提交
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
L
Linus Torvalds 已提交
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
L
Linus Torvalds 已提交
17 18
 */
#include "xfs.h"
19
#include "xfs_bit.h"
L
Linus Torvalds 已提交
20
#include "xfs_log.h"
21
#include "xfs_inum.h"
L
Linus Torvalds 已提交
22
#include "xfs_sb.h"
23
#include "xfs_ag.h"
L
Linus Torvalds 已提交
24 25 26 27 28 29 30 31
#include "xfs_dir2.h"
#include "xfs_trans.h"
#include "xfs_dmapi.h"
#include "xfs_mount.h"
#include "xfs_bmap_btree.h"
#include "xfs_alloc_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_dir2_sf.h"
32
#include "xfs_attr_sf.h"
L
Linus Torvalds 已提交
33 34
#include "xfs_dinode.h"
#include "xfs_inode.h"
35 36
#include "xfs_alloc.h"
#include "xfs_btree.h"
L
Linus Torvalds 已提交
37 38 39
#include "xfs_error.h"
#include "xfs_rw.h"
#include "xfs_iomap.h"
40
#include "xfs_vnodeops.h"
L
Linus Torvalds 已提交
41
#include <linux/mpage.h>
42
#include <linux/pagevec.h>
L
Linus Torvalds 已提交
43 44
#include <linux/writeback.h>

45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
STATIC void
xfs_count_page_state(
	struct page		*page,
	int			*delalloc,
	int			*unmapped,
	int			*unwritten)
{
	struct buffer_head	*bh, *head;

	*delalloc = *unmapped = *unwritten = 0;

	bh = head = page_buffers(page);
	do {
		if (buffer_uptodate(bh) && !buffer_mapped(bh))
			(*unmapped) = 1;
		else if (buffer_unwritten(bh))
			(*unwritten) = 1;
		else if (buffer_delay(bh))
			(*delalloc) = 1;
	} while ((bh = bh->b_this_page) != head);
}

L
Linus Torvalds 已提交
67 68 69 70 71 72
#if defined(XFS_RW_TRACE)
void
xfs_page_trace(
	int		tag,
	struct inode	*inode,
	struct page	*page,
73
	unsigned long	pgoff)
L
Linus Torvalds 已提交
74 75
{
	xfs_inode_t	*ip;
76
	bhv_vnode_t	*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
77
	loff_t		isize = i_size_read(inode);
78
	loff_t		offset = page_offset(page);
L
Linus Torvalds 已提交
79 80 81 82 83
	int		delalloc = -1, unmapped = -1, unwritten = -1;

	if (page_has_buffers(page))
		xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);

84
	ip = xfs_vtoi(vp);
L
Linus Torvalds 已提交
85 86 87 88 89 90 91 92
	if (!ip->i_rwtrace)
		return;

	ktrace_enter(ip->i_rwtrace,
		(void *)((unsigned long)tag),
		(void *)ip,
		(void *)inode,
		(void *)page,
93
		(void *)pgoff,
L
Linus Torvalds 已提交
94 95 96 97 98 99 100 101 102
		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
		(void *)((unsigned long)((isize >> 32) & 0xffffffff)),
		(void *)((unsigned long)(isize & 0xffffffff)),
		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
		(void *)((unsigned long)(offset & 0xffffffff)),
		(void *)((unsigned long)delalloc),
		(void *)((unsigned long)unmapped),
		(void *)((unsigned long)unwritten),
103
		(void *)((unsigned long)current_pid()),
L
Linus Torvalds 已提交
104 105 106
		(void *)NULL);
}
#else
107
#define xfs_page_trace(tag, inode, page, pgoff)
L
Linus Torvalds 已提交
108 109
#endif

110 111
/*
 * Schedule IO completion handling on a xfsdatad if this was
112 113
 * the final hold on this ioend. If we are asked to wait,
 * flush the workqueue.
114 115 116
 */
STATIC void
xfs_finish_ioend(
117 118
	xfs_ioend_t	*ioend,
	int		wait)
119
{
120
	if (atomic_dec_and_test(&ioend->io_remaining)) {
121
		queue_work(xfsdatad_workqueue, &ioend->io_work);
122 123 124
		if (wait)
			flush_workqueue(xfsdatad_workqueue);
	}
125 126
}

127 128 129 130 131 132
/*
 * We're now finished for good with this ioend structure.
 * Update the page state via the associated buffer_heads,
 * release holds on the inode and bio, and finally free
 * up memory.  Do not use the ioend after this.
 */
133 134 135 136
STATIC void
xfs_destroy_ioend(
	xfs_ioend_t		*ioend)
{
137 138 139 140
	struct buffer_head	*bh, *next;

	for (bh = ioend->io_buffer_head; bh; bh = next) {
		next = bh->b_private;
141
		bh->b_end_io(bh, !ioend->io_error);
142
	}
143 144 145 146 147
	if (unlikely(ioend->io_error)) {
		vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
				__FILE__,__LINE__);
	}
	vn_iowake(XFS_I(ioend->io_inode));
148 149 150
	mempool_free(ioend, xfs_ioend_pool);
}

151 152 153 154 155 156 157 158 159 160 161
/*
 * Update on-disk file size now that data has been written to disk.
 * The current in-memory file size is i_size.  If a write is beyond
 * eof io_new_size will be the intended file size until i_size is
 * updated.  If this write does not extend all the way to the valid
 * file size then restrict this update to the end of the write.
 */
STATIC void
xfs_setfilesize(
	xfs_ioend_t		*ioend)
{
162
	xfs_inode_t		*ip = XFS_I(ioend->io_inode);
163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
	xfs_fsize_t		isize;
	xfs_fsize_t		bsize;

	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
	ASSERT(ioend->io_type != IOMAP_READ);

	if (unlikely(ioend->io_error))
		return;

	bsize = ioend->io_offset + ioend->io_size;

	xfs_ilock(ip, XFS_ILOCK_EXCL);

	isize = MAX(ip->i_size, ip->i_iocore.io_new_size);
	isize = MIN(isize, bsize);

	if (ip->i_d.di_size < isize) {
		ip->i_d.di_size = isize;
		ip->i_update_core = 1;
		ip->i_update_size = 1;
183
		mark_inode_dirty_sync(ioend->io_inode);
184 185 186 187 188
	}

	xfs_iunlock(ip, XFS_ILOCK_EXCL);
}

189
/*
190 191 192 193
 * Buffered IO write completion for delayed allocate extents.
 */
STATIC void
xfs_end_bio_delalloc(
D
David Howells 已提交
194
	struct work_struct	*work)
195
{
D
David Howells 已提交
196 197
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
198

199
	xfs_setfilesize(ioend);
200 201 202 203 204 205 206 207
	xfs_destroy_ioend(ioend);
}

/*
 * Buffered IO write completion for regular, written extents.
 */
STATIC void
xfs_end_bio_written(
D
David Howells 已提交
208
	struct work_struct	*work)
209
{
D
David Howells 已提交
210 211
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
212

213
	xfs_setfilesize(ioend);
214 215 216 217 218 219
	xfs_destroy_ioend(ioend);
}

/*
 * IO write completion for unwritten extents.
 *
220
 * Issue transactions to convert a buffer range from unwritten
221
 * to written extents.
222 223 224
 */
STATIC void
xfs_end_bio_unwritten(
D
David Howells 已提交
225
	struct work_struct	*work)
226
{
D
David Howells 已提交
227 228
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
229 230 231
	xfs_off_t		offset = ioend->io_offset;
	size_t			size = ioend->io_size;

232
	if (likely(!ioend->io_error)) {
233
		xfs_bmap(XFS_I(ioend->io_inode), offset, size,
234
				BMAPI_UNWRITTEN, NULL, NULL);
235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
		xfs_setfilesize(ioend);
	}
	xfs_destroy_ioend(ioend);
}

/*
 * IO read completion for regular, written extents.
 */
STATIC void
xfs_end_bio_read(
	struct work_struct	*work)
{
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);

250 251 252 253 254 255 256 257 258 259 260
	xfs_destroy_ioend(ioend);
}

/*
 * Allocate and initialise an IO completion structure.
 * We need to track unwritten extent write completion here initially.
 * We'll need to extend this for updating the ondisk inode size later
 * (vs. incore size).
 */
STATIC xfs_ioend_t *
xfs_alloc_ioend(
261 262
	struct inode		*inode,
	unsigned int		type)
263 264 265 266 267 268 269 270 271 272 273
{
	xfs_ioend_t		*ioend;

	ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);

	/*
	 * Set the count to 1 initially, which will prevent an I/O
	 * completion callback from happening before we have started
	 * all the I/O from calling the completion routine too early.
	 */
	atomic_set(&ioend->io_remaining, 1);
274
	ioend->io_error = 0;
275 276
	ioend->io_list = NULL;
	ioend->io_type = type;
277
	ioend->io_inode = inode;
278
	ioend->io_buffer_head = NULL;
279
	ioend->io_buffer_tail = NULL;
280
	atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
281 282 283
	ioend->io_offset = 0;
	ioend->io_size = 0;

284
	if (type == IOMAP_UNWRITTEN)
D
David Howells 已提交
285
		INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
286
	else if (type == IOMAP_DELAY)
D
David Howells 已提交
287
		INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
288 289
	else if (type == IOMAP_READ)
		INIT_WORK(&ioend->io_work, xfs_end_bio_read);
290
	else
D
David Howells 已提交
291
		INIT_WORK(&ioend->io_work, xfs_end_bio_written);
292 293 294 295

	return ioend;
}

L
Linus Torvalds 已提交
296 297 298 299 300 301 302 303
STATIC int
xfs_map_blocks(
	struct inode		*inode,
	loff_t			offset,
	ssize_t			count,
	xfs_iomap_t		*mapp,
	int			flags)
{
304
	xfs_inode_t		*ip = XFS_I(inode);
L
Linus Torvalds 已提交
305 306
	int			error, nmaps = 1;

307
	error = xfs_bmap(ip, offset, count,
308
				flags, mapp, &nmaps);
L
Linus Torvalds 已提交
309
	if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
310
		xfs_iflags_set(ip, XFS_IMODIFIED);
L
Linus Torvalds 已提交
311 312 313
	return -error;
}

314
STATIC_INLINE int
315
xfs_iomap_valid(
L
Linus Torvalds 已提交
316
	xfs_iomap_t		*iomapp,
317
	loff_t			offset)
L
Linus Torvalds 已提交
318
{
319 320
	return offset >= iomapp->iomap_offset &&
		offset < iomapp->iomap_offset + iomapp->iomap_bsize;
L
Linus Torvalds 已提交
321 322
}

323 324 325
/*
 * BIO completion handler for buffered IO.
 */
A
Al Viro 已提交
326
STATIC void
327 328 329 330 331 332 333
xfs_end_bio(
	struct bio		*bio,
	int			error)
{
	xfs_ioend_t		*ioend = bio->bi_private;

	ASSERT(atomic_read(&bio->bi_cnt) >= 1);
334
	ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
335 336 337 338 339

	/* Toss bio and pass work off to an xfsdatad thread */
	bio->bi_private = NULL;
	bio->bi_end_io = NULL;
	bio_put(bio);
340

341
	xfs_finish_ioend(ioend, 0);
342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401
}

STATIC void
xfs_submit_ioend_bio(
	xfs_ioend_t	*ioend,
	struct bio	*bio)
{
	atomic_inc(&ioend->io_remaining);

	bio->bi_private = ioend;
	bio->bi_end_io = xfs_end_bio;

	submit_bio(WRITE, bio);
	ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
	bio_put(bio);
}

STATIC struct bio *
xfs_alloc_ioend_bio(
	struct buffer_head	*bh)
{
	struct bio		*bio;
	int			nvecs = bio_get_nr_vecs(bh->b_bdev);

	do {
		bio = bio_alloc(GFP_NOIO, nvecs);
		nvecs >>= 1;
	} while (!bio);

	ASSERT(bio->bi_private == NULL);
	bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
	bio->bi_bdev = bh->b_bdev;
	bio_get(bio);
	return bio;
}

STATIC void
xfs_start_buffer_writeback(
	struct buffer_head	*bh)
{
	ASSERT(buffer_mapped(bh));
	ASSERT(buffer_locked(bh));
	ASSERT(!buffer_delay(bh));
	ASSERT(!buffer_unwritten(bh));

	mark_buffer_async_write(bh);
	set_buffer_uptodate(bh);
	clear_buffer_dirty(bh);
}

STATIC void
xfs_start_page_writeback(
	struct page		*page,
	struct writeback_control *wbc,
	int			clear_dirty,
	int			buffers)
{
	ASSERT(PageLocked(page));
	ASSERT(!PageWriteback(page));
	if (clear_dirty)
402 403
		clear_page_dirty_for_io(page);
	set_page_writeback(page);
404
	unlock_page(page);
405 406
	/* If no buffers on the page are to be written, finish it here */
	if (!buffers)
407 408 409 410 411 412 413 414 415
		end_page_writeback(page);
}

static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
{
	return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
}

/*
416 417 418 419 420 421 422 423 424 425 426 427 428 429
 * Submit all of the bios for all of the ioends we have saved up, covering the
 * initial writepage page and also any probed pages.
 *
 * Because we may have multiple ioends spanning a page, we need to start
 * writeback on all the buffers before we submit them for I/O. If we mark the
 * buffers as we got, then we can end up with a page that only has buffers
 * marked async write and I/O complete on can occur before we mark the other
 * buffers async write.
 *
 * The end result of this is that we trip a bug in end_page_writeback() because
 * we call it twice for the one page as the code in end_buffer_async_write()
 * assumes that all buffers on the page are started at the same time.
 *
 * The fix is two passes across the ioend list - one to start writeback on the
430
 * buffer_heads, and then submit them for I/O on the second pass.
431 432 433 434 435
 */
STATIC void
xfs_submit_ioend(
	xfs_ioend_t		*ioend)
{
436
	xfs_ioend_t		*head = ioend;
437 438 439 440 441
	xfs_ioend_t		*next;
	struct buffer_head	*bh;
	struct bio		*bio;
	sector_t		lastblock = 0;

442 443 444 445 446 447 448 449 450 451
	/* Pass 1 - start writeback */
	do {
		next = ioend->io_list;
		for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
			xfs_start_buffer_writeback(bh);
		}
	} while ((ioend = next) != NULL);

	/* Pass 2 - submit I/O */
	ioend = head;
452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
	do {
		next = ioend->io_list;
		bio = NULL;

		for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {

			if (!bio) {
 retry:
				bio = xfs_alloc_ioend_bio(bh);
			} else if (bh->b_blocknr != lastblock + 1) {
				xfs_submit_ioend_bio(ioend, bio);
				goto retry;
			}

			if (bio_add_buffer(bio, bh) != bh->b_size) {
				xfs_submit_ioend_bio(ioend, bio);
				goto retry;
			}

			lastblock = bh->b_blocknr;
		}
		if (bio)
			xfs_submit_ioend_bio(ioend, bio);
475
		xfs_finish_ioend(ioend, 0);
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
	} while ((ioend = next) != NULL);
}

/*
 * Cancel submission of all buffer_heads so far in this endio.
 * Toss the endio too.  Only ever called for the initial page
 * in a writepage request, so only ever one page.
 */
STATIC void
xfs_cancel_ioend(
	xfs_ioend_t		*ioend)
{
	xfs_ioend_t		*next;
	struct buffer_head	*bh, *next_bh;

	do {
		next = ioend->io_list;
		bh = ioend->io_buffer_head;
		do {
			next_bh = bh->b_private;
			clear_buffer_async_write(bh);
			unlock_buffer(bh);
		} while ((bh = next_bh) != NULL);

500
		vn_iowake(XFS_I(ioend->io_inode));
501 502 503 504 505 506 507 508 509 510 511 512 513 514
		mempool_free(ioend, xfs_ioend_pool);
	} while ((ioend = next) != NULL);
}

/*
 * Test to see if we've been building up a completion structure for
 * earlier buffers -- if so, we try to append to this ioend if we
 * can, otherwise we finish off any current ioend and start another.
 * Return true if we've finished the given ioend.
 */
STATIC void
xfs_add_to_ioend(
	struct inode		*inode,
	struct buffer_head	*bh,
515
	xfs_off_t		offset,
516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540
	unsigned int		type,
	xfs_ioend_t		**result,
	int			need_ioend)
{
	xfs_ioend_t		*ioend = *result;

	if (!ioend || need_ioend || type != ioend->io_type) {
		xfs_ioend_t	*previous = *result;

		ioend = xfs_alloc_ioend(inode, type);
		ioend->io_offset = offset;
		ioend->io_buffer_head = bh;
		ioend->io_buffer_tail = bh;
		if (previous)
			previous->io_list = ioend;
		*result = ioend;
	} else {
		ioend->io_buffer_tail->b_private = bh;
		ioend->io_buffer_tail = bh;
	}

	bh->b_private = NULL;
	ioend->io_size += bh->b_size;
}

541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
STATIC void
xfs_map_buffer(
	struct buffer_head	*bh,
	xfs_iomap_t		*mp,
	xfs_off_t		offset,
	uint			block_bits)
{
	sector_t		bn;

	ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);

	bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
	      ((offset - mp->iomap_offset) >> block_bits);

	ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));

	bh->b_blocknr = bn;
	set_buffer_mapped(bh);
}

L
Linus Torvalds 已提交
561 562 563
STATIC void
xfs_map_at_offset(
	struct buffer_head	*bh,
564
	loff_t			offset,
L
Linus Torvalds 已提交
565
	int			block_bits,
566
	xfs_iomap_t		*iomapp)
L
Linus Torvalds 已提交
567 568 569 570 571
{
	ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
	ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));

	lock_buffer(bh);
572
	xfs_map_buffer(bh, iomapp, offset, block_bits);
573
	bh->b_bdev = iomapp->iomap_target->bt_bdev;
L
Linus Torvalds 已提交
574 575
	set_buffer_mapped(bh);
	clear_buffer_delay(bh);
576
	clear_buffer_unwritten(bh);
L
Linus Torvalds 已提交
577 578 579
}

/*
580
 * Look for a page at index that is suitable for clustering.
L
Linus Torvalds 已提交
581 582
 */
STATIC unsigned int
583
xfs_probe_page(
584
	struct page		*page,
585 586
	unsigned int		pg_offset,
	int			mapped)
L
Linus Torvalds 已提交
587 588 589 590
{
	int			ret = 0;

	if (PageWriteback(page))
591
		return 0;
L
Linus Torvalds 已提交
592 593 594 595 596 597 598

	if (page->mapping && PageDirty(page)) {
		if (page_has_buffers(page)) {
			struct buffer_head	*bh, *head;

			bh = head = page_buffers(page);
			do {
599 600 601
				if (!buffer_uptodate(bh))
					break;
				if (mapped != buffer_mapped(bh))
L
Linus Torvalds 已提交
602 603 604 605 606 607
					break;
				ret += bh->b_size;
				if (ret >= pg_offset)
					break;
			} while ((bh = bh->b_this_page) != head);
		} else
608
			ret = mapped ? 0 : PAGE_CACHE_SIZE;
L
Linus Torvalds 已提交
609 610 611 612 613
	}

	return ret;
}

614
STATIC size_t
615
xfs_probe_cluster(
L
Linus Torvalds 已提交
616 617 618
	struct inode		*inode,
	struct page		*startpage,
	struct buffer_head	*bh,
619 620
	struct buffer_head	*head,
	int			mapped)
L
Linus Torvalds 已提交
621
{
622
	struct pagevec		pvec;
L
Linus Torvalds 已提交
623
	pgoff_t			tindex, tlast, tloff;
624 625
	size_t			total = 0;
	int			done = 0, i;
L
Linus Torvalds 已提交
626 627 628

	/* First sum forwards in this page */
	do {
629
		if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
630
			return total;
L
Linus Torvalds 已提交
631 632 633
		total += bh->b_size;
	} while ((bh = bh->b_this_page) != head);

634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649
	/* if we reached the end of the page, sum forwards in following pages */
	tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
	tindex = startpage->index + 1;

	/* Prune this back to avoid pathological behavior */
	tloff = min(tlast, startpage->index + 64);

	pagevec_init(&pvec, 0);
	while (!done && tindex <= tloff) {
		unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);

		if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
			break;

		for (i = 0; i < pagevec_count(&pvec); i++) {
			struct page *page = pvec.pages[i];
650
			size_t pg_offset, pg_len = 0;
651 652 653 654

			if (tindex == tlast) {
				pg_offset =
				    i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
655 656
				if (!pg_offset) {
					done = 1;
657
					break;
658
				}
659 660 661 662
			} else
				pg_offset = PAGE_CACHE_SIZE;

			if (page->index == tindex && !TestSetPageLocked(page)) {
663
				pg_len = xfs_probe_page(page, pg_offset, mapped);
664 665 666
				unlock_page(page);
			}

667
			if (!pg_len) {
668 669 670 671
				done = 1;
				break;
			}

672
			total += pg_len;
673
			tindex++;
L
Linus Torvalds 已提交
674
		}
675 676 677

		pagevec_release(&pvec);
		cond_resched();
L
Linus Torvalds 已提交
678
	}
679

L
Linus Torvalds 已提交
680 681 682 683
	return total;
}

/*
684 685
 * Test if a given page is suitable for writing as part of an unwritten
 * or delayed allocate extent.
L
Linus Torvalds 已提交
686
 */
687 688 689
STATIC int
xfs_is_delayed_page(
	struct page		*page,
690
	unsigned int		type)
L
Linus Torvalds 已提交
691 692
{
	if (PageWriteback(page))
693
		return 0;
L
Linus Torvalds 已提交
694 695 696 697 698 699 700

	if (page->mapping && page_has_buffers(page)) {
		struct buffer_head	*bh, *head;
		int			acceptable = 0;

		bh = head = page_buffers(page);
		do {
701 702 703 704
			if (buffer_unwritten(bh))
				acceptable = (type == IOMAP_UNWRITTEN);
			else if (buffer_delay(bh))
				acceptable = (type == IOMAP_DELAY);
705
			else if (buffer_dirty(bh) && buffer_mapped(bh))
706
				acceptable = (type == IOMAP_NEW);
707
			else
L
Linus Torvalds 已提交
708 709 710 711
				break;
		} while ((bh = bh->b_this_page) != head);

		if (acceptable)
712
			return 1;
L
Linus Torvalds 已提交
713 714
	}

715
	return 0;
L
Linus Torvalds 已提交
716 717 718 719 720 721 722 723
}

/*
 * Allocate & map buffers for page given the extent map. Write it out.
 * except for the original page of a writepage, this is called on
 * delalloc/unwritten pages only, for the original page it is possible
 * that the page has no mapping at all.
 */
724
STATIC int
L
Linus Torvalds 已提交
725 726 727
xfs_convert_page(
	struct inode		*inode,
	struct page		*page,
728
	loff_t			tindex,
729
	xfs_iomap_t		*mp,
730
	xfs_ioend_t		**ioendp,
L
Linus Torvalds 已提交
731 732 733 734
	struct writeback_control *wbc,
	int			startio,
	int			all_bh)
{
735
	struct buffer_head	*bh, *head;
736 737
	xfs_off_t		end_offset;
	unsigned long		p_offset;
738
	unsigned int		type;
L
Linus Torvalds 已提交
739
	int			bbits = inode->i_blkbits;
740
	int			len, page_dirty;
741
	int			count = 0, done = 0, uptodate = 1;
742
 	xfs_off_t		offset = page_offset(page);
L
Linus Torvalds 已提交
743

744 745 746 747 748 749 750 751 752 753 754
	if (page->index != tindex)
		goto fail;
	if (TestSetPageLocked(page))
		goto fail;
	if (PageWriteback(page))
		goto fail_unlock_page;
	if (page->mapping != inode->i_mapping)
		goto fail_unlock_page;
	if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
		goto fail_unlock_page;

755 756
	/*
	 * page_dirty is initially a count of buffers on the page before
757
	 * EOF and is decremented as we move each into a cleanable state.
758 759 760 761 762 763 764 765 766
	 *
	 * Derivation:
	 *
	 * End offset is the highest offset that this page should represent.
	 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
	 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
	 * hence give us the correct page_dirty count. On any other page,
	 * it will be zero and in that case we need page_dirty to be the
	 * count of buffers on the page.
767
	 */
768 769 770 771
	end_offset = min_t(unsigned long long,
			(xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
			i_size_read(inode));

772
	len = 1 << inode->i_blkbits;
773 774 775 776
	p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
					PAGE_CACHE_SIZE);
	p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
	page_dirty = p_offset / len;
777

L
Linus Torvalds 已提交
778 779
	bh = head = page_buffers(page);
	do {
780
		if (offset >= end_offset)
L
Linus Torvalds 已提交
781
			break;
782 783 784 785
		if (!buffer_uptodate(bh))
			uptodate = 0;
		if (!(PageUptodate(page) || buffer_uptodate(bh))) {
			done = 1;
L
Linus Torvalds 已提交
786
			continue;
787 788
		}

789 790 791 792 793 794 795
		if (buffer_unwritten(bh) || buffer_delay(bh)) {
			if (buffer_unwritten(bh))
				type = IOMAP_UNWRITTEN;
			else
				type = IOMAP_DELAY;

			if (!xfs_iomap_valid(mp, offset)) {
796
				done = 1;
797 798 799 800 801 802 803 804
				continue;
			}

			ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
			ASSERT(!(mp->iomap_flags & IOMAP_DELAY));

			xfs_map_at_offset(bh, offset, bbits, mp);
			if (startio) {
805
				xfs_add_to_ioend(inode, bh, offset,
806 807 808 809 810 811 812 813 814
						type, ioendp, done);
			} else {
				set_buffer_dirty(bh);
				unlock_buffer(bh);
				mark_buffer_dirty(bh);
			}
			page_dirty--;
			count++;
		} else {
815
			type = IOMAP_NEW;
816
			if (buffer_mapped(bh) && all_bh && startio) {
L
Linus Torvalds 已提交
817
				lock_buffer(bh);
818
				xfs_add_to_ioend(inode, bh, offset,
819 820
						type, ioendp, done);
				count++;
821
				page_dirty--;
822 823
			} else {
				done = 1;
L
Linus Torvalds 已提交
824 825
			}
		}
826
	} while (offset += len, (bh = bh->b_this_page) != head);
L
Linus Torvalds 已提交
827

828 829 830 831
	if (uptodate && bh == head)
		SetPageUptodate(page);

	if (startio) {
832 833 834 835
		if (count) {
			struct backing_dev_info *bdi;

			bdi = inode->i_mapping->backing_dev_info;
836
			wbc->nr_to_write--;
837 838 839
			if (bdi_write_congested(bdi)) {
				wbc->encountered_congestion = 1;
				done = 1;
840
			} else if (wbc->nr_to_write <= 0) {
841 842 843
				done = 1;
			}
		}
844
		xfs_start_page_writeback(page, wbc, !page_dirty, count);
L
Linus Torvalds 已提交
845
	}
846 847

	return done;
848 849 850 851
 fail_unlock_page:
	unlock_page(page);
 fail:
	return 1;
L
Linus Torvalds 已提交
852 853 854 855 856 857 858 859 860 861 862
}

/*
 * Convert & write out a cluster of pages in the same extent as defined
 * by mp and following the start page.
 */
STATIC void
xfs_cluster_write(
	struct inode		*inode,
	pgoff_t			tindex,
	xfs_iomap_t		*iomapp,
863
	xfs_ioend_t		**ioendp,
L
Linus Torvalds 已提交
864 865 866 867 868
	struct writeback_control *wbc,
	int			startio,
	int			all_bh,
	pgoff_t			tlast)
{
869 870
	struct pagevec		pvec;
	int			done = 0, i;
L
Linus Torvalds 已提交
871

872 873 874 875 876
	pagevec_init(&pvec, 0);
	while (!done && tindex <= tlast) {
		unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);

		if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
L
Linus Torvalds 已提交
877
			break;
878 879 880 881 882 883 884 885 886 887

		for (i = 0; i < pagevec_count(&pvec); i++) {
			done = xfs_convert_page(inode, pvec.pages[i], tindex++,
					iomapp, ioendp, wbc, startio, all_bh);
			if (done)
				break;
		}

		pagevec_release(&pvec);
		cond_resched();
L
Linus Torvalds 已提交
888 889 890 891 892 893 894 895 896 897 898 899
	}
}

/*
 * Calling this without startio set means we are being asked to make a dirty
 * page ready for freeing it's buffers.  When called with startio set then
 * we are coming from writepage.
 *
 * When called with startio set it is important that we write the WHOLE
 * page if possible.
 * The bh->b_state's cannot know if any of the blocks or which block for
 * that matter are dirty due to mmap writes, and therefore bh uptodate is
900
 * only valid if the page itself isn't completely uptodate.  Some layers
L
Linus Torvalds 已提交
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
 * may clear the page dirty flag prior to calling write page, under the
 * assumption the entire page will be written out; by not writing out the
 * whole page the page can be reused before all valid dirty data is
 * written out.  Note: in the case of a page that has been dirty'd by
 * mapwrite and but partially setup by block_prepare_write the
 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
 * valid state, thus the whole page must be written out thing.
 */

STATIC int
xfs_page_state_convert(
	struct inode	*inode,
	struct page	*page,
	struct writeback_control *wbc,
	int		startio,
	int		unmapped) /* also implies page uptodate */
{
918
	struct buffer_head	*bh, *head;
919
	xfs_iomap_t		iomap;
920
	xfs_ioend_t		*ioend = NULL, *iohead = NULL;
L
Linus Torvalds 已提交
921 922
	loff_t			offset;
	unsigned long           p_offset = 0;
923
	unsigned int		type;
L
Linus Torvalds 已提交
924 925
	__uint64_t              end_offset;
	pgoff_t                 end_index, last_index, tlast;
926 927
	ssize_t			size, len;
	int			flags, err, iomap_valid = 0, uptodate = 1;
928 929
	int			page_dirty, count = 0;
	int			trylock = 0;
930
	int			all_bh = unmapped;
L
Linus Torvalds 已提交
931

932 933 934 935
	if (startio) {
		if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
			trylock |= BMAPI_TRYLOCK;
	}
936

L
Linus Torvalds 已提交
937 938 939 940 941 942 943
	/* Is this page beyond the end of the file? */
	offset = i_size_read(inode);
	end_index = offset >> PAGE_CACHE_SHIFT;
	last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
	if (page->index >= end_index) {
		if ((page->index >= end_index + 1) ||
		    !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
944 945 946
			if (startio)
				unlock_page(page);
			return 0;
L
Linus Torvalds 已提交
947 948 949 950
		}
	}

	/*
951
	 * page_dirty is initially a count of buffers on the page before
952
	 * EOF and is decremented as we move each into a cleanable state.
953 954 955 956 957 958 959 960 961 962 963 964
	 *
	 * Derivation:
	 *
	 * End offset is the highest offset that this page should represent.
	 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
	 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
	 * hence give us the correct page_dirty count. On any other page,
	 * it will be zero and in that case we need page_dirty to be the
	 * count of buffers on the page.
 	 */
	end_offset = min_t(unsigned long long,
			(xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
965
	len = 1 << inode->i_blkbits;
966 967 968
	p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
					PAGE_CACHE_SIZE);
	p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
969 970 971
	page_dirty = p_offset / len;

	bh = head = page_buffers(page);
972
	offset = page_offset(page);
973 974
	flags = BMAPI_READ;
	type = IOMAP_NEW;
975 976

	/* TODO: cleanup count and page_dirty */
L
Linus Torvalds 已提交
977 978 979 980 981 982

	do {
		if (offset >= end_offset)
			break;
		if (!buffer_uptodate(bh))
			uptodate = 0;
983
		if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
984 985 986 987 988
			/*
			 * the iomap is actually still valid, but the ioend
			 * isn't.  shouldn't happen too often.
			 */
			iomap_valid = 0;
L
Linus Torvalds 已提交
989
			continue;
990
		}
L
Linus Torvalds 已提交
991

992 993
		if (iomap_valid)
			iomap_valid = xfs_iomap_valid(&iomap, offset);
L
Linus Torvalds 已提交
994 995 996 997

		/*
		 * First case, map an unwritten extent and prepare for
		 * extent state conversion transaction on completion.
998
		 *
L
Linus Torvalds 已提交
999 1000
		 * Second case, allocate space for a delalloc buffer.
		 * We can return EAGAIN here in the release page case.
1001 1002 1003
		 *
		 * Third case, an unmapped buffer was found, and we are
		 * in a path where we need to write the whole page out.
1004
		 */
1005 1006 1007
		if (buffer_unwritten(bh) || buffer_delay(bh) ||
		    ((buffer_uptodate(bh) || PageUptodate(page)) &&
		     !buffer_mapped(bh) && (unmapped || startio))) {
1008 1009
			int new_ioend = 0;

1010
			/*
1011 1012
			 * Make sure we don't use a read-only iomap
			 */
1013
			if (flags == BMAPI_READ)
1014 1015
				iomap_valid = 0;

1016 1017
			if (buffer_unwritten(bh)) {
				type = IOMAP_UNWRITTEN;
1018
				flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1019
			} else if (buffer_delay(bh)) {
1020
				type = IOMAP_DELAY;
1021
				flags = BMAPI_ALLOCATE | trylock;
1022
			} else {
1023
				type = IOMAP_NEW;
1024
				flags = BMAPI_WRITE | BMAPI_MMAP;
1025 1026
			}

1027
			if (!iomap_valid) {
1028 1029 1030 1031 1032 1033 1034 1035 1036
				/*
				 * if we didn't have a valid mapping then we
				 * need to ensure that we put the new mapping
				 * in a new ioend structure. This needs to be
				 * done to ensure that the ioends correctly
				 * reflect the block mappings at io completion
				 * for unwritten extent conversion.
				 */
				new_ioend = 1;
1037 1038 1039
				if (type == IOMAP_NEW) {
					size = xfs_probe_cluster(inode,
							page, bh, head, 0);
1040 1041 1042 1043 1044 1045
				} else {
					size = len;
				}

				err = xfs_map_blocks(inode, offset, size,
						&iomap, flags);
1046
				if (err)
L
Linus Torvalds 已提交
1047
					goto error;
1048
				iomap_valid = xfs_iomap_valid(&iomap, offset);
L
Linus Torvalds 已提交
1049
			}
1050 1051 1052
			if (iomap_valid) {
				xfs_map_at_offset(bh, offset,
						inode->i_blkbits, &iomap);
L
Linus Torvalds 已提交
1053
				if (startio) {
1054
					xfs_add_to_ioend(inode, bh, offset,
1055
							type, &ioend,
1056
							new_ioend);
L
Linus Torvalds 已提交
1057 1058 1059 1060 1061 1062
				} else {
					set_buffer_dirty(bh);
					unlock_buffer(bh);
					mark_buffer_dirty(bh);
				}
				page_dirty--;
1063
				count++;
L
Linus Torvalds 已提交
1064
			}
1065
		} else if (buffer_uptodate(bh) && startio) {
1066 1067 1068 1069 1070
			/*
			 * we got here because the buffer is already mapped.
			 * That means it must already have extents allocated
			 * underneath it. Map the extent by reading it.
			 */
1071
			if (!iomap_valid || flags != BMAPI_READ) {
1072 1073 1074 1075 1076 1077 1078 1079 1080
				flags = BMAPI_READ;
				size = xfs_probe_cluster(inode, page, bh,
								head, 1);
				err = xfs_map_blocks(inode, offset, size,
						&iomap, flags);
				if (err)
					goto error;
				iomap_valid = xfs_iomap_valid(&iomap, offset);
			}
1081

1082 1083 1084 1085 1086 1087 1088 1089 1090
			/*
			 * We set the type to IOMAP_NEW in case we are doing a
			 * small write at EOF that is extending the file but
			 * without needing an allocation. We need to update the
			 * file size on I/O completion in this case so it is
			 * the same case as having just allocated a new extent
			 * that we are writing into for the first time.
			 */
			type = IOMAP_NEW;
1091 1092
			if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
				ASSERT(buffer_mapped(bh));
1093 1094
				if (iomap_valid)
					all_bh = 1;
1095
				xfs_add_to_ioend(inode, bh, offset, type,
1096 1097 1098
						&ioend, !iomap_valid);
				page_dirty--;
				count++;
1099
			} else {
1100
				iomap_valid = 0;
L
Linus Torvalds 已提交
1101
			}
1102 1103 1104
		} else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
			   (unmapped || startio)) {
			iomap_valid = 0;
L
Linus Torvalds 已提交
1105
		}
1106 1107 1108 1109 1110

		if (!iohead)
			iohead = ioend;

	} while (offset += len, ((bh = bh->b_this_page) != head));
L
Linus Torvalds 已提交
1111 1112 1113 1114

	if (uptodate && bh == head)
		SetPageUptodate(page);

1115 1116
	if (startio)
		xfs_start_page_writeback(page, wbc, 1, count);
L
Linus Torvalds 已提交
1117

1118 1119
	if (ioend && iomap_valid) {
		offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
L
Linus Torvalds 已提交
1120
					PAGE_CACHE_SHIFT;
1121
		tlast = min_t(pgoff_t, offset, last_index);
1122
		xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1123
					wbc, startio, all_bh, tlast);
L
Linus Torvalds 已提交
1124 1125
	}

1126 1127 1128
	if (iohead)
		xfs_submit_ioend(iohead);

L
Linus Torvalds 已提交
1129 1130 1131
	return page_dirty;

error:
1132 1133
	if (iohead)
		xfs_cancel_ioend(iohead);
L
Linus Torvalds 已提交
1134 1135 1136 1137 1138 1139 1140

	/*
	 * If it's delalloc and we have nowhere to put it,
	 * throw it away, unless the lower layers told
	 * us to try again.
	 */
	if (err != -EAGAIN) {
1141
		if (!unmapped)
L
Linus Torvalds 已提交
1142 1143 1144 1145 1146 1147
			block_invalidatepage(page, 0);
		ClearPageUptodate(page);
	}
	return err;
}

1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168
/*
 * writepage: Called from one of two places:
 *
 * 1. we are flushing a delalloc buffer head.
 *
 * 2. we are writing out a dirty page. Typically the page dirty
 *    state is cleared before we get here. In this case is it
 *    conceivable we have no buffer heads.
 *
 * For delalloc space on the page we need to allocate space and
 * flush it. For unmapped buffer heads on the page we should
 * allocate space if the page is uptodate. For any other dirty
 * buffer heads on the page we should flush them.
 *
 * If we detect that a transaction would be required to flush
 * the page, we have to check the process flags first, if we
 * are already in a transaction or disk I/O during allocations
 * is off, we need to fail the writepage and redirty the page.
 */

STATIC int
1169
xfs_vm_writepage(
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203
	struct page		*page,
	struct writeback_control *wbc)
{
	int			error;
	int			need_trans;
	int			delalloc, unmapped, unwritten;
	struct inode		*inode = page->mapping->host;

	xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);

	/*
	 * We need a transaction if:
	 *  1. There are delalloc buffers on the page
	 *  2. The page is uptodate and we have unmapped buffers
	 *  3. The page is uptodate and we have no buffers
	 *  4. There are unwritten buffers on the page
	 */

	if (!page_has_buffers(page)) {
		unmapped = 1;
		need_trans = 1;
	} else {
		xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
		if (!PageUptodate(page))
			unmapped = 0;
		need_trans = delalloc + unmapped + unwritten;
	}

	/*
	 * If we need a transaction and the process flags say
	 * we are already in a transaction, or no IO is allowed
	 * then mark the page dirty again and leave the page
	 * as is.
	 */
1204
	if (current_test_flags(PF_FSTRANS) && need_trans)
1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
		goto out_fail;

	/*
	 * Delay hooking up buffer heads until we have
	 * made our go/no-go decision.
	 */
	if (!page_has_buffers(page))
		create_empty_buffers(page, 1 << inode->i_blkbits, 0);

	/*
	 * Convert delayed allocate, unwritten or unmapped space
	 * to real space and flush out to disk.
	 */
	error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
	if (error == -EAGAIN)
		goto out_fail;
	if (unlikely(error < 0))
		goto out_unlock;

	return 0;

out_fail:
	redirty_page_for_writepage(wbc, page);
	unlock_page(page);
	return 0;
out_unlock:
	unlock_page(page);
	return error;
}

1235 1236 1237 1238 1239
STATIC int
xfs_vm_writepages(
	struct address_space	*mapping,
	struct writeback_control *wbc)
{
1240
	xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1241 1242 1243
	return generic_writepages(mapping, wbc);
}

1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263
/*
 * Called to move a page into cleanable state - and from there
 * to be released. Possibly the page is already clean. We always
 * have buffer heads in this call.
 *
 * Returns 0 if the page is ok to release, 1 otherwise.
 *
 * Possible scenarios are:
 *
 * 1. We are being called to release a page which has been written
 *    to via regular I/O. buffer heads will be dirty and possibly
 *    delalloc. If no delalloc buffer heads in this case then we
 *    can just return zero.
 *
 * 2. We are called to release a page which has been written via
 *    mmap, all we need to do is ensure there is no delalloc
 *    state in the buffer heads, if not we can let the caller
 *    free them and we should come back later via writepage.
 */
STATIC int
1264
xfs_vm_releasepage(
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
	struct page		*page,
	gfp_t			gfp_mask)
{
	struct inode		*inode = page->mapping->host;
	int			dirty, delalloc, unmapped, unwritten;
	struct writeback_control wbc = {
		.sync_mode = WB_SYNC_ALL,
		.nr_to_write = 1,
	};

1275
	xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1276

1277 1278 1279
	if (!page_has_buffers(page))
		return 0;

1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
	xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
	if (!delalloc && !unwritten)
		goto free_buffers;

	if (!(gfp_mask & __GFP_FS))
		return 0;

	/* If we are already inside a transaction or the thread cannot
	 * do I/O, we cannot release this page.
	 */
1290
	if (current_test_flags(PF_FSTRANS))
1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
		return 0;

	/*
	 * Convert delalloc space to real space, do not flush the
	 * data out to disk, that will be done by the caller.
	 * Never need to allocate space here - we will always
	 * come back to writepage in that case.
	 */
	dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
	if (dirty == 0 && !unwritten)
		goto free_buffers;
	return 0;

free_buffers:
	return try_to_free_buffers(page);
}

L
Linus Torvalds 已提交
1308
STATIC int
1309
__xfs_get_blocks(
L
Linus Torvalds 已提交
1310 1311 1312 1313 1314 1315 1316 1317
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create,
	int			direct,
	bmapi_flags_t		flags)
{
	xfs_iomap_t		iomap;
1318 1319
	xfs_off_t		offset;
	ssize_t			size;
1320
	int			niomap = 1;
L
Linus Torvalds 已提交
1321 1322
	int			error;

1323
	offset = (xfs_off_t)iblock << inode->i_blkbits;
1324 1325
	ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
	size = bh_result->b_size;
1326
	error = xfs_bmap(XFS_I(inode), offset, size,
1327
			     create ? flags : BMAPI_READ, &iomap, &niomap);
L
Linus Torvalds 已提交
1328 1329
	if (error)
		return -error;
1330
	if (niomap == 0)
L
Linus Torvalds 已提交
1331 1332 1333
		return 0;

	if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1334 1335
		/*
		 * For unwritten extents do not report a disk address on
L
Linus Torvalds 已提交
1336 1337 1338
		 * the read case (treat as if we're reading into a hole).
		 */
		if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1339 1340
			xfs_map_buffer(bh_result, &iomap, offset,
				       inode->i_blkbits);
L
Linus Torvalds 已提交
1341 1342 1343 1344 1345 1346 1347 1348
		}
		if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
			if (direct)
				bh_result->b_private = inode;
			set_buffer_unwritten(bh_result);
		}
	}

1349 1350 1351 1352
	/*
	 * If this is a realtime file, data may be on a different device.
	 * to that pointed to from the buffer_head b_bdev currently.
	 */
1353
	bh_result->b_bdev = iomap.iomap_target->bt_bdev;
L
Linus Torvalds 已提交
1354

1355
	/*
1356 1357 1358 1359 1360 1361 1362
	 * If we previously allocated a block out beyond eof and we are now
	 * coming back to use it then we will need to flag it as new even if it
	 * has a disk address.
	 *
	 * With sub-block writes into unwritten extents we also need to mark
	 * the buffer as new so that the unwritten parts of the buffer gets
	 * correctly zeroed.
L
Linus Torvalds 已提交
1363 1364 1365
	 */
	if (create &&
	    ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1366 1367
	     (offset >= i_size_read(inode)) ||
	     (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
L
Linus Torvalds 已提交
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
		set_buffer_new(bh_result);

	if (iomap.iomap_flags & IOMAP_DELAY) {
		BUG_ON(direct);
		if (create) {
			set_buffer_uptodate(bh_result);
			set_buffer_mapped(bh_result);
			set_buffer_delay(bh_result);
		}
	}

1379
	if (direct || size > (1 << inode->i_blkbits)) {
1380 1381
		ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
		offset = min_t(xfs_off_t,
1382 1383
				iomap.iomap_bsize - iomap.iomap_delta, size);
		bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
L
Linus Torvalds 已提交
1384 1385 1386 1387 1388 1389
	}

	return 0;
}

int
1390
xfs_get_blocks(
L
Linus Torvalds 已提交
1391 1392 1393 1394 1395
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create)
{
1396
	return __xfs_get_blocks(inode, iblock,
1397
				bh_result, create, 0, BMAPI_WRITE);
L
Linus Torvalds 已提交
1398 1399 1400
}

STATIC int
1401
xfs_get_blocks_direct(
L
Linus Torvalds 已提交
1402 1403 1404 1405 1406
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create)
{
1407
	return __xfs_get_blocks(inode, iblock,
1408
				bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
L
Linus Torvalds 已提交
1409 1410
}

1411
STATIC void
1412
xfs_end_io_direct(
1413 1414 1415 1416 1417 1418 1419 1420 1421 1422
	struct kiocb	*iocb,
	loff_t		offset,
	ssize_t		size,
	void		*private)
{
	xfs_ioend_t	*ioend = iocb->private;

	/*
	 * Non-NULL private data means we need to issue a transaction to
	 * convert a range from unwritten to written extents.  This needs
1423
	 * to happen from process context but aio+dio I/O completion
1424
	 * happens from irq context so we need to defer it to a workqueue.
1425
	 * This is not necessary for synchronous direct I/O, but we do
1426 1427
	 * it anyway to keep the code uniform and simpler.
	 *
1428 1429 1430 1431 1432 1433 1434
	 * Well, if only it were that simple. Because synchronous direct I/O
	 * requires extent conversion to occur *before* we return to userspace,
	 * we have to wait for extent conversion to complete. Look at the
	 * iocb that has been passed to us to determine if this is AIO or
	 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
	 * workqueue and wait for it to complete.
	 *
1435 1436 1437 1438
	 * The core direct I/O code might be changed to always call the
	 * completion handler in the future, in which case all this can
	 * go away.
	 */
1439 1440 1441
	ioend->io_offset = offset;
	ioend->io_size = size;
	if (ioend->io_type == IOMAP_READ) {
1442
		xfs_finish_ioend(ioend, 0);
1443
	} else if (private && size > 0) {
1444
		xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1445
	} else {
1446 1447 1448 1449 1450 1451 1452
		/*
		 * A direct I/O write ioend starts it's life in unwritten
		 * state in case they map an unwritten extent.  This write
		 * didn't map an unwritten extent so switch it's completion
		 * handler.
		 */
		INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1453
		xfs_finish_ioend(ioend, 0);
1454 1455 1456
	}

	/*
1457
	 * blockdev_direct_IO can return an error even after the I/O
1458 1459 1460 1461 1462 1463
	 * completion handler was called.  Thus we need to protect
	 * against double-freeing.
	 */
	iocb->private = NULL;
}

L
Linus Torvalds 已提交
1464
STATIC ssize_t
1465
xfs_vm_direct_IO(
L
Linus Torvalds 已提交
1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476
	int			rw,
	struct kiocb		*iocb,
	const struct iovec	*iov,
	loff_t			offset,
	unsigned long		nr_segs)
{
	struct file	*file = iocb->ki_filp;
	struct inode	*inode = file->f_mapping->host;
	xfs_iomap_t	iomap;
	int		maps = 1;
	int		error;
1477
	ssize_t		ret;
L
Linus Torvalds 已提交
1478

1479 1480
	error = xfs_bmap(XFS_I(inode), offset, 0,
				BMAPI_DEVICE, &iomap, &maps);
L
Linus Torvalds 已提交
1481 1482 1483
	if (error)
		return -error;

1484
	if (rw == WRITE) {
1485
		iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1486 1487 1488 1489 1490 1491
		ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
			iomap.iomap_target->bt_bdev,
			iov, offset, nr_segs,
			xfs_get_blocks_direct,
			xfs_end_io_direct);
	} else {
1492
		iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1493 1494 1495 1496 1497 1498
		ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
			iomap.iomap_target->bt_bdev,
			iov, offset, nr_segs,
			xfs_get_blocks_direct,
			xfs_end_io_direct);
	}
1499

1500
	if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1501 1502
		xfs_destroy_ioend(iocb->private);
	return ret;
L
Linus Torvalds 已提交
1503 1504
}

1505
STATIC int
N
Nick Piggin 已提交
1506
xfs_vm_write_begin(
1507
	struct file		*file,
N
Nick Piggin 已提交
1508 1509 1510 1511 1512 1513
	struct address_space	*mapping,
	loff_t			pos,
	unsigned		len,
	unsigned		flags,
	struct page		**pagep,
	void			**fsdata)
1514
{
N
Nick Piggin 已提交
1515 1516 1517
	*pagep = NULL;
	return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
								xfs_get_blocks);
1518
}
L
Linus Torvalds 已提交
1519 1520

STATIC sector_t
1521
xfs_vm_bmap(
L
Linus Torvalds 已提交
1522 1523 1524 1525
	struct address_space	*mapping,
	sector_t		block)
{
	struct inode		*inode = (struct inode *)mapping->host;
1526
	struct xfs_inode	*ip = XFS_I(inode);
L
Linus Torvalds 已提交
1527

1528
	xfs_itrace_entry(XFS_I(inode));
1529 1530 1531
	xfs_rwlock(ip, VRWLOCK_READ);
	xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
	xfs_rwunlock(ip, VRWLOCK_READ);
1532
	return generic_block_bmap(mapping, block, xfs_get_blocks);
L
Linus Torvalds 已提交
1533 1534 1535
}

STATIC int
1536
xfs_vm_readpage(
L
Linus Torvalds 已提交
1537 1538 1539
	struct file		*unused,
	struct page		*page)
{
1540
	return mpage_readpage(page, xfs_get_blocks);
L
Linus Torvalds 已提交
1541 1542 1543
}

STATIC int
1544
xfs_vm_readpages(
L
Linus Torvalds 已提交
1545 1546 1547 1548 1549
	struct file		*unused,
	struct address_space	*mapping,
	struct list_head	*pages,
	unsigned		nr_pages)
{
1550
	return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
L
Linus Torvalds 已提交
1551 1552
}

1553
STATIC void
1554
xfs_vm_invalidatepage(
1555 1556 1557 1558 1559
	struct page		*page,
	unsigned long		offset)
{
	xfs_page_trace(XFS_INVALIDPAGE_ENTER,
			page->mapping->host, page, offset);
1560
	block_invalidatepage(page, offset);
1561 1562
}

1563
const struct address_space_operations xfs_address_space_operations = {
1564 1565 1566
	.readpage		= xfs_vm_readpage,
	.readpages		= xfs_vm_readpages,
	.writepage		= xfs_vm_writepage,
1567
	.writepages		= xfs_vm_writepages,
L
Linus Torvalds 已提交
1568
	.sync_page		= block_sync_page,
1569 1570
	.releasepage		= xfs_vm_releasepage,
	.invalidatepage		= xfs_vm_invalidatepage,
N
Nick Piggin 已提交
1571 1572
	.write_begin		= xfs_vm_write_begin,
	.write_end		= generic_write_end,
1573 1574
	.bmap			= xfs_vm_bmap,
	.direct_IO		= xfs_vm_direct_IO,
1575
	.migratepage		= buffer_migrate_page,
L
Linus Torvalds 已提交
1576
};