xfs_aops.c 37.4 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 40
#include "xfs_error.h"
#include "xfs_rw.h"
#include "xfs_iomap.h"
#include <linux/mpage.h>
41
#include <linux/pagevec.h>
L
Linus Torvalds 已提交
42 43
#include <linux/writeback.h>

44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
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
66 67 68 69 70 71
#if defined(XFS_RW_TRACE)
void
xfs_page_trace(
	int		tag,
	struct inode	*inode,
	struct page	*page,
72
	unsigned long	pgoff)
L
Linus Torvalds 已提交
73 74
{
	xfs_inode_t	*ip;
75
	bhv_vnode_t	*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
76
	loff_t		isize = i_size_read(inode);
77
	loff_t		offset = page_offset(page);
L
Linus Torvalds 已提交
78 79 80 81 82
	int		delalloc = -1, unmapped = -1, unwritten = -1;

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

83
	ip = xfs_vtoi(vp);
L
Linus Torvalds 已提交
84 85 86 87 88 89 90 91
	if (!ip->i_rwtrace)
		return;

	ktrace_enter(ip->i_rwtrace,
		(void *)((unsigned long)tag),
		(void *)ip,
		(void *)inode,
		(void *)page,
92
		(void *)pgoff,
L
Linus Torvalds 已提交
93 94 95 96 97 98 99 100 101
		(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),
102
		(void *)((unsigned long)current_pid()),
L
Linus Torvalds 已提交
103 104 105
		(void *)NULL);
}
#else
106
#define xfs_page_trace(tag, inode, page, pgoff)
L
Linus Torvalds 已提交
107 108
#endif

109 110 111 112 113 114 115 116 117 118 119 120
/*
 * Schedule IO completion handling on a xfsdatad if this was
 * the final hold on this ioend.
 */
STATIC void
xfs_finish_ioend(
	xfs_ioend_t		*ioend)
{
	if (atomic_dec_and_test(&ioend->io_remaining))
		queue_work(xfsdatad_workqueue, &ioend->io_work);
}

121 122 123 124 125 126
/*
 * 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.
 */
127 128 129 130
STATIC void
xfs_destroy_ioend(
	xfs_ioend_t		*ioend)
{
131 132 133 134
	struct buffer_head	*bh, *next;

	for (bh = ioend->io_buffer_head; bh; bh = next) {
		next = bh->b_private;
135
		bh->b_end_io(bh, !ioend->io_error);
136
	}
137 138
	if (unlikely(ioend->io_error))
		vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__);
139 140 141 142
	vn_iowake(ioend->io_vnode);
	mempool_free(ioend, xfs_ioend_pool);
}

143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
/*
 * 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)
{
	xfs_inode_t		*ip;
	xfs_fsize_t		isize;
	xfs_fsize_t		bsize;

	ip = xfs_vtoi(ioend->io_vnode);

	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;
	}

	xfs_iunlock(ip, XFS_ILOCK_EXCL);
}

182
/*
183 184 185 186
 * Buffered IO write completion for delayed allocate extents.
 */
STATIC void
xfs_end_bio_delalloc(
D
David Howells 已提交
187
	struct work_struct	*work)
188
{
D
David Howells 已提交
189 190
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
191

192
	xfs_setfilesize(ioend);
193 194 195 196 197 198 199 200
	xfs_destroy_ioend(ioend);
}

/*
 * Buffered IO write completion for regular, written extents.
 */
STATIC void
xfs_end_bio_written(
D
David Howells 已提交
201
	struct work_struct	*work)
202
{
D
David Howells 已提交
203 204
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
205

206
	xfs_setfilesize(ioend);
207 208 209 210 211 212
	xfs_destroy_ioend(ioend);
}

/*
 * IO write completion for unwritten extents.
 *
213
 * Issue transactions to convert a buffer range from unwritten
214
 * to written extents.
215 216 217
 */
STATIC void
xfs_end_bio_unwritten(
D
David Howells 已提交
218
	struct work_struct	*work)
219
{
D
David Howells 已提交
220 221
	xfs_ioend_t		*ioend =
		container_of(work, xfs_ioend_t, io_work);
222
	bhv_vnode_t		*vp = ioend->io_vnode;
223 224 225
	xfs_off_t		offset = ioend->io_offset;
	size_t			size = ioend->io_size;

226
	if (likely(!ioend->io_error)) {
227
		bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL);
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
		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);

243 244 245 246 247 248 249 250 251 252 253
	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(
254 255
	struct inode		*inode,
	unsigned int		type)
256 257 258 259 260 261 262 263 264 265 266
{
	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);
267
	ioend->io_error = 0;
268 269
	ioend->io_list = NULL;
	ioend->io_type = type;
270
	ioend->io_vnode = vn_from_inode(inode);
271
	ioend->io_buffer_head = NULL;
272
	ioend->io_buffer_tail = NULL;
273 274 275 276
	atomic_inc(&ioend->io_vnode->v_iocount);
	ioend->io_offset = 0;
	ioend->io_size = 0;

277
	if (type == IOMAP_UNWRITTEN)
D
David Howells 已提交
278
		INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
279
	else if (type == IOMAP_DELAY)
D
David Howells 已提交
280
		INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
281 282
	else if (type == IOMAP_READ)
		INIT_WORK(&ioend->io_work, xfs_end_bio_read);
283
	else
D
David Howells 已提交
284
		INIT_WORK(&ioend->io_work, xfs_end_bio_written);
285 286 287 288

	return ioend;
}

L
Linus Torvalds 已提交
289 290 291 292 293 294 295 296
STATIC int
xfs_map_blocks(
	struct inode		*inode,
	loff_t			offset,
	ssize_t			count,
	xfs_iomap_t		*mapp,
	int			flags)
{
297
	bhv_vnode_t		*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
298 299
	int			error, nmaps = 1;

300
	error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps);
L
Linus Torvalds 已提交
301 302 303 304 305
	if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
		VMODIFY(vp);
	return -error;
}

306
STATIC_INLINE int
307
xfs_iomap_valid(
L
Linus Torvalds 已提交
308
	xfs_iomap_t		*iomapp,
309
	loff_t			offset)
L
Linus Torvalds 已提交
310
{
311 312
	return offset >= iomapp->iomap_offset &&
		offset < iomapp->iomap_offset + iomapp->iomap_bsize;
L
Linus Torvalds 已提交
313 314
}

315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
/*
 * BIO completion handler for buffered IO.
 */
STATIC int
xfs_end_bio(
	struct bio		*bio,
	unsigned int		bytes_done,
	int			error)
{
	xfs_ioend_t		*ioend = bio->bi_private;

	if (bio->bi_size)
		return 1;

	ASSERT(atomic_read(&bio->bi_cnt) >= 1);
330
	ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
331 332 333 334 335

	/* Toss bio and pass work off to an xfsdatad thread */
	bio->bi_private = NULL;
	bio->bi_end_io = NULL;
	bio_put(bio);
336

337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
	xfs_finish_ioend(ioend);
	return 0;
}

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)
399 400
		clear_page_dirty_for_io(page);
	set_page_writeback(page);
401 402 403 404 405 406 407 408 409 410 411 412 413
	unlock_page(page);
	if (!buffers) {
		end_page_writeback(page);
		wbc->pages_skipped++;	/* We didn't write this 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));
}

/*
414 415 416 417 418 419 420 421 422 423 424 425 426 427
 * 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
428
 * buffer_heads, and then submit them for I/O on the second pass.
429 430 431 432 433
 */
STATIC void
xfs_submit_ioend(
	xfs_ioend_t		*ioend)
{
434
	xfs_ioend_t		*head = ioend;
435 436 437 438 439
	xfs_ioend_t		*next;
	struct buffer_head	*bh;
	struct bio		*bio;
	sector_t		lastblock = 0;

440 441 442 443 444 445 446 447 448 449
	/* 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;
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
	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);
		xfs_finish_ioend(ioend);
	} 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);

		vn_iowake(ioend->io_vnode);
		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,
513
	xfs_off_t		offset,
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
	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;
}

539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
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
559 560 561
STATIC void
xfs_map_at_offset(
	struct buffer_head	*bh,
562
	loff_t			offset,
L
Linus Torvalds 已提交
563
	int			block_bits,
564
	xfs_iomap_t		*iomapp)
L
Linus Torvalds 已提交
565 566 567 568 569
{
	ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
	ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));

	lock_buffer(bh);
570
	xfs_map_buffer(bh, iomapp, offset, block_bits);
571
	bh->b_bdev = iomapp->iomap_target->bt_bdev;
L
Linus Torvalds 已提交
572 573
	set_buffer_mapped(bh);
	clear_buffer_delay(bh);
574
	clear_buffer_unwritten(bh);
L
Linus Torvalds 已提交
575 576 577
}

/*
578
 * Look for a page at index that is suitable for clustering.
L
Linus Torvalds 已提交
579 580
 */
STATIC unsigned int
581
xfs_probe_page(
582
	struct page		*page,
583 584
	unsigned int		pg_offset,
	int			mapped)
L
Linus Torvalds 已提交
585 586 587 588
{
	int			ret = 0;

	if (PageWriteback(page))
589
		return 0;
L
Linus Torvalds 已提交
590 591 592 593 594 595 596

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

			bh = head = page_buffers(page);
			do {
597 598 599
				if (!buffer_uptodate(bh))
					break;
				if (mapped != buffer_mapped(bh))
L
Linus Torvalds 已提交
600 601 602 603 604 605
					break;
				ret += bh->b_size;
				if (ret >= pg_offset)
					break;
			} while ((bh = bh->b_this_page) != head);
		} else
606
			ret = mapped ? 0 : PAGE_CACHE_SIZE;
L
Linus Torvalds 已提交
607 608 609 610 611
	}

	return ret;
}

612
STATIC size_t
613
xfs_probe_cluster(
L
Linus Torvalds 已提交
614 615 616
	struct inode		*inode,
	struct page		*startpage,
	struct buffer_head	*bh,
617 618
	struct buffer_head	*head,
	int			mapped)
L
Linus Torvalds 已提交
619
{
620
	struct pagevec		pvec;
L
Linus Torvalds 已提交
621
	pgoff_t			tindex, tlast, tloff;
622 623
	size_t			total = 0;
	int			done = 0, i;
L
Linus Torvalds 已提交
624 625 626

	/* First sum forwards in this page */
	do {
627
		if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
628
			return total;
L
Linus Torvalds 已提交
629 630 631
		total += bh->b_size;
	} while ((bh = bh->b_this_page) != head);

632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
	/* 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];
			size_t pg_offset, len = 0;

			if (tindex == tlast) {
				pg_offset =
				    i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
653 654
				if (!pg_offset) {
					done = 1;
655
					break;
656
				}
657 658 659 660
			} else
				pg_offset = PAGE_CACHE_SIZE;

			if (page->index == tindex && !TestSetPageLocked(page)) {
661
				len = xfs_probe_page(page, pg_offset, mapped);
662 663 664 665 666 667 668 669
				unlock_page(page);
			}

			if (!len) {
				done = 1;
				break;
			}

L
Linus Torvalds 已提交
670
			total += len;
671
			tindex++;
L
Linus Torvalds 已提交
672
		}
673 674 675

		pagevec_release(&pvec);
		cond_resched();
L
Linus Torvalds 已提交
676
	}
677

L
Linus Torvalds 已提交
678 679 680 681
	return total;
}

/*
682 683
 * Test if a given page is suitable for writing as part of an unwritten
 * or delayed allocate extent.
L
Linus Torvalds 已提交
684
 */
685 686 687
STATIC int
xfs_is_delayed_page(
	struct page		*page,
688
	unsigned int		type)
L
Linus Torvalds 已提交
689 690
{
	if (PageWriteback(page))
691
		return 0;
L
Linus Torvalds 已提交
692 693 694 695 696 697 698

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

		bh = head = page_buffers(page);
		do {
699 700 701 702
			if (buffer_unwritten(bh))
				acceptable = (type == IOMAP_UNWRITTEN);
			else if (buffer_delay(bh))
				acceptable = (type == IOMAP_DELAY);
703
			else if (buffer_dirty(bh) && buffer_mapped(bh))
704
				acceptable = (type == 0);
705
			else
L
Linus Torvalds 已提交
706 707 708 709
				break;
		} while ((bh = bh->b_this_page) != head);

		if (acceptable)
710
			return 1;
L
Linus Torvalds 已提交
711 712
	}

713
	return 0;
L
Linus Torvalds 已提交
714 715 716 717 718 719 720 721
}

/*
 * 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.
 */
722
STATIC int
L
Linus Torvalds 已提交
723 724 725
xfs_convert_page(
	struct inode		*inode,
	struct page		*page,
726
	loff_t			tindex,
727
	xfs_iomap_t		*mp,
728
	xfs_ioend_t		**ioendp,
L
Linus Torvalds 已提交
729 730 731 732
	struct writeback_control *wbc,
	int			startio,
	int			all_bh)
{
733
	struct buffer_head	*bh, *head;
734 735
	xfs_off_t		end_offset;
	unsigned long		p_offset;
736
	unsigned int		type;
L
Linus Torvalds 已提交
737
	int			bbits = inode->i_blkbits;
738
	int			len, page_dirty;
739
	int			count = 0, done = 0, uptodate = 1;
740
 	xfs_off_t		offset = page_offset(page);
L
Linus Torvalds 已提交
741

742 743 744 745 746 747 748 749 750 751 752
	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;

753 754
	/*
	 * page_dirty is initially a count of buffers on the page before
755
	 * EOF and is decremented as we move each into a cleanable state.
756 757 758 759 760 761 762 763 764
	 *
	 * 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.
765
	 */
766 767 768 769
	end_offset = min_t(unsigned long long,
			(xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
			i_size_read(inode));

770
	len = 1 << inode->i_blkbits;
771 772 773 774
	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;
775

L
Linus Torvalds 已提交
776 777
	bh = head = page_buffers(page);
	do {
778
		if (offset >= end_offset)
L
Linus Torvalds 已提交
779
			break;
780 781 782 783
		if (!buffer_uptodate(bh))
			uptodate = 0;
		if (!(PageUptodate(page) || buffer_uptodate(bh))) {
			done = 1;
L
Linus Torvalds 已提交
784
			continue;
785 786
		}

787 788 789 790 791 792 793
		if (buffer_unwritten(bh) || buffer_delay(bh)) {
			if (buffer_unwritten(bh))
				type = IOMAP_UNWRITTEN;
			else
				type = IOMAP_DELAY;

			if (!xfs_iomap_valid(mp, offset)) {
794
				done = 1;
795 796 797 798 799 800 801 802
				continue;
			}

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

			xfs_map_at_offset(bh, offset, bbits, mp);
			if (startio) {
803
				xfs_add_to_ioend(inode, bh, offset,
804 805 806 807 808 809 810 811 812 813 814
						type, ioendp, done);
			} else {
				set_buffer_dirty(bh);
				unlock_buffer(bh);
				mark_buffer_dirty(bh);
			}
			page_dirty--;
			count++;
		} else {
			type = 0;
			if (buffer_mapped(bh) && all_bh && startio) {
L
Linus Torvalds 已提交
815
				lock_buffer(bh);
816
				xfs_add_to_ioend(inode, bh, offset,
817 818
						type, ioendp, done);
				count++;
819
				page_dirty--;
820 821
			} else {
				done = 1;
L
Linus Torvalds 已提交
822 823
			}
		}
824
	} while (offset += len, (bh = bh->b_this_page) != head);
L
Linus Torvalds 已提交
825

826 827 828 829
	if (uptodate && bh == head)
		SetPageUptodate(page);

	if (startio) {
830 831 832 833
		if (count) {
			struct backing_dev_info *bdi;

			bdi = inode->i_mapping->backing_dev_info;
834
			wbc->nr_to_write--;
835 836 837
			if (bdi_write_congested(bdi)) {
				wbc->encountered_congestion = 1;
				done = 1;
838
			} else if (wbc->nr_to_write <= 0) {
839 840 841
				done = 1;
			}
		}
842
		xfs_start_page_writeback(page, wbc, !page_dirty, count);
L
Linus Torvalds 已提交
843
	}
844 845

	return done;
846 847 848 849
 fail_unlock_page:
	unlock_page(page);
 fail:
	return 1;
L
Linus Torvalds 已提交
850 851 852 853 854 855 856 857 858 859 860
}

/*
 * 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,
861
	xfs_ioend_t		**ioendp,
L
Linus Torvalds 已提交
862 863 864 865 866
	struct writeback_control *wbc,
	int			startio,
	int			all_bh,
	pgoff_t			tlast)
{
867 868
	struct pagevec		pvec;
	int			done = 0, i;
L
Linus Torvalds 已提交
869

870 871 872 873 874
	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
875
			break;
876 877 878 879 880 881 882 883 884 885

		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
886 887 888 889 890 891 892 893 894 895 896 897
	}
}

/*
 * 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
898
 * only valid if the page itself isn't completely uptodate.  Some layers
L
Linus Torvalds 已提交
899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915
 * 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 */
{
916
	struct buffer_head	*bh, *head;
917
	xfs_iomap_t		iomap;
918
	xfs_ioend_t		*ioend = NULL, *iohead = NULL;
L
Linus Torvalds 已提交
919 920
	loff_t			offset;
	unsigned long           p_offset = 0;
921
	unsigned int		type;
L
Linus Torvalds 已提交
922 923
	__uint64_t              end_offset;
	pgoff_t                 end_index, last_index, tlast;
924 925
	ssize_t			size, len;
	int			flags, err, iomap_valid = 0, uptodate = 1;
926 927
	int			page_dirty, count = 0;
	int			trylock = 0;
928
	int			all_bh = unmapped;
L
Linus Torvalds 已提交
929

930 931 932 933
	if (startio) {
		if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
			trylock |= BMAPI_TRYLOCK;
	}
934

L
Linus Torvalds 已提交
935 936 937 938 939 940 941
	/* 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))) {
942 943 944
			if (startio)
				unlock_page(page);
			return 0;
L
Linus Torvalds 已提交
945 946 947 948
		}
	}

	/*
949
	 * page_dirty is initially a count of buffers on the page before
950
	 * EOF and is decremented as we move each into a cleanable state.
951 952 953 954 955 956 957 958 959 960 961 962
	 *
	 * 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);
963
	len = 1 << inode->i_blkbits;
964 965 966
	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;
967 968 969
	page_dirty = p_offset / len;

	bh = head = page_buffers(page);
970
	offset = page_offset(page);
971
	flags = -1;
972
	type = IOMAP_READ;
973 974

	/* TODO: cleanup count and page_dirty */
L
Linus Torvalds 已提交
975 976 977 978 979 980

	do {
		if (offset >= end_offset)
			break;
		if (!buffer_uptodate(bh))
			uptodate = 0;
981
		if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
982 983 984 985 986
			/*
			 * 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
987
			continue;
988
		}
L
Linus Torvalds 已提交
989

990 991
		if (iomap_valid)
			iomap_valid = xfs_iomap_valid(&iomap, offset);
L
Linus Torvalds 已提交
992 993 994 995

		/*
		 * First case, map an unwritten extent and prepare for
		 * extent state conversion transaction on completion.
996
		 *
L
Linus Torvalds 已提交
997 998
		 * Second case, allocate space for a delalloc buffer.
		 * We can return EAGAIN here in the release page case.
999 1000 1001 1002 1003 1004 1005
		 *
		 * Third case, an unmapped buffer was found, and we are
		 * in a path where we need to write the whole page out.
 		 */
		if (buffer_unwritten(bh) || buffer_delay(bh) ||
		    ((buffer_uptodate(bh) || PageUptodate(page)) &&
		     !buffer_mapped(bh) && (unmapped || startio))) {
1006 1007 1008 1009 1010 1011
		     	/*
			 * Make sure we don't use a read-only iomap
			 */
		     	if (flags == BMAPI_READ)
				iomap_valid = 0;

1012 1013
			if (buffer_unwritten(bh)) {
				type = IOMAP_UNWRITTEN;
1014
				flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1015
			} else if (buffer_delay(bh)) {
1016
				type = IOMAP_DELAY;
1017
				flags = BMAPI_ALLOCATE | trylock;
1018
			} else {
1019
				type = IOMAP_NEW;
1020
				flags = BMAPI_WRITE | BMAPI_MMAP;
1021 1022
			}

1023
			if (!iomap_valid) {
1024 1025 1026
				if (type == IOMAP_NEW) {
					size = xfs_probe_cluster(inode,
							page, bh, head, 0);
1027 1028 1029 1030 1031 1032
				} else {
					size = len;
				}

				err = xfs_map_blocks(inode, offset, size,
						&iomap, flags);
1033
				if (err)
L
Linus Torvalds 已提交
1034
					goto error;
1035
				iomap_valid = xfs_iomap_valid(&iomap, offset);
L
Linus Torvalds 已提交
1036
			}
1037 1038 1039
			if (iomap_valid) {
				xfs_map_at_offset(bh, offset,
						inode->i_blkbits, &iomap);
L
Linus Torvalds 已提交
1040
				if (startio) {
1041
					xfs_add_to_ioend(inode, bh, offset,
1042 1043
							type, &ioend,
							!iomap_valid);
L
Linus Torvalds 已提交
1044 1045 1046 1047 1048 1049
				} else {
					set_buffer_dirty(bh);
					unlock_buffer(bh);
					mark_buffer_dirty(bh);
				}
				page_dirty--;
1050
				count++;
L
Linus Torvalds 已提交
1051
			}
1052
		} else if (buffer_uptodate(bh) && startio) {
1053 1054 1055 1056 1057
			/*
			 * 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.
			 */
1058
			if (!iomap_valid || type != IOMAP_READ) {
1059 1060 1061 1062 1063 1064 1065 1066 1067
				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);
			}
1068

1069
			type = IOMAP_READ;
1070 1071
			if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
				ASSERT(buffer_mapped(bh));
1072 1073
				if (iomap_valid)
					all_bh = 1;
1074
				xfs_add_to_ioend(inode, bh, offset, type,
1075 1076 1077
						&ioend, !iomap_valid);
				page_dirty--;
				count++;
1078
			} else {
1079
				iomap_valid = 0;
L
Linus Torvalds 已提交
1080
			}
1081 1082 1083
		} else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
			   (unmapped || startio)) {
			iomap_valid = 0;
L
Linus Torvalds 已提交
1084
		}
1085 1086 1087 1088 1089

		if (!iohead)
			iohead = ioend;

	} while (offset += len, ((bh = bh->b_this_page) != head));
L
Linus Torvalds 已提交
1090 1091 1092 1093

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

1094 1095
	if (startio)
		xfs_start_page_writeback(page, wbc, 1, count);
L
Linus Torvalds 已提交
1096

1097 1098
	if (ioend && iomap_valid) {
		offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
L
Linus Torvalds 已提交
1099
					PAGE_CACHE_SHIFT;
1100
		tlast = min_t(pgoff_t, offset, last_index);
1101
		xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1102
					wbc, startio, all_bh, tlast);
L
Linus Torvalds 已提交
1103 1104
	}

1105 1106 1107
	if (iohead)
		xfs_submit_ioend(iohead);

L
Linus Torvalds 已提交
1108 1109 1110
	return page_dirty;

error:
1111 1112
	if (iohead)
		xfs_cancel_ioend(iohead);
L
Linus Torvalds 已提交
1113 1114 1115 1116 1117 1118 1119

	/*
	 * 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) {
1120
		if (!unmapped)
L
Linus Torvalds 已提交
1121 1122 1123 1124 1125 1126
			block_invalidatepage(page, 0);
		ClearPageUptodate(page);
	}
	return err;
}

1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
/*
 * 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
1148
xfs_vm_writepage(
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
	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.
	 */
1183
	if (current_test_flags(PF_FSTRANS) && need_trans)
1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
		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;
}

1214 1215 1216 1217 1218
STATIC int
xfs_vm_writepages(
	struct address_space	*mapping,
	struct writeback_control *wbc)
{
1219
	struct bhv_vnode	*vp = vn_from_inode(mapping->host);
1220 1221 1222 1223 1224 1225

	if (VN_TRUNC(vp))
		VUNTRUNCATE(vp);
	return generic_writepages(mapping, wbc);
}

1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
/*
 * 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
1246
xfs_vm_releasepage(
1247 1248 1249 1250 1251 1252 1253 1254 1255 1256
	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,
	};

1257
	xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1258

1259 1260 1261
	if (!page_has_buffers(page))
		return 0;

1262 1263 1264 1265 1266 1267 1268 1269 1270 1271
	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.
	 */
1272
	if (current_test_flags(PF_FSTRANS))
1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
		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
1290
STATIC int
1291
__xfs_get_blocks(
L
Linus Torvalds 已提交
1292 1293 1294 1295 1296 1297 1298
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create,
	int			direct,
	bmapi_flags_t		flags)
{
1299
	bhv_vnode_t		*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
1300
	xfs_iomap_t		iomap;
1301 1302
	xfs_off_t		offset;
	ssize_t			size;
1303
	int			niomap = 1;
L
Linus Torvalds 已提交
1304 1305
	int			error;

1306
	offset = (xfs_off_t)iblock << inode->i_blkbits;
1307 1308
	ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
	size = bh_result->b_size;
1309 1310
	error = bhv_vop_bmap(vp, offset, size,
			     create ? flags : BMAPI_READ, &iomap, &niomap);
L
Linus Torvalds 已提交
1311 1312
	if (error)
		return -error;
1313
	if (niomap == 0)
L
Linus Torvalds 已提交
1314 1315 1316
		return 0;

	if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1317 1318
		/*
		 * For unwritten extents do not report a disk address on
L
Linus Torvalds 已提交
1319 1320 1321
		 * the read case (treat as if we're reading into a hole).
		 */
		if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1322 1323
			xfs_map_buffer(bh_result, &iomap, offset,
				       inode->i_blkbits);
L
Linus Torvalds 已提交
1324 1325 1326 1327 1328 1329 1330 1331
		}
		if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
			if (direct)
				bh_result->b_private = inode;
			set_buffer_unwritten(bh_result);
		}
	}

1332 1333 1334 1335
	/*
	 * If this is a realtime file, data may be on a different device.
	 * to that pointed to from the buffer_head b_bdev currently.
	 */
1336
	bh_result->b_bdev = iomap.iomap_target->bt_bdev;
L
Linus Torvalds 已提交
1337

1338
	/*
1339 1340 1341 1342 1343 1344 1345
	 * 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
1346 1347 1348
	 */
	if (create &&
	    ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1349 1350
	     (offset >= i_size_read(inode)) ||
	     (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
L
Linus Torvalds 已提交
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361
		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);
		}
	}

1362
	if (direct || size > (1 << inode->i_blkbits)) {
1363 1364
		ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
		offset = min_t(xfs_off_t,
1365 1366
				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
1367 1368 1369 1370 1371 1372
	}

	return 0;
}

int
1373
xfs_get_blocks(
L
Linus Torvalds 已提交
1374 1375 1376 1377 1378
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create)
{
1379
	return __xfs_get_blocks(inode, iblock,
1380
				bh_result, create, 0, BMAPI_WRITE);
L
Linus Torvalds 已提交
1381 1382 1383
}

STATIC int
1384
xfs_get_blocks_direct(
L
Linus Torvalds 已提交
1385 1386 1387 1388 1389
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create)
{
1390
	return __xfs_get_blocks(inode, iblock,
1391
				bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
L
Linus Torvalds 已提交
1392 1393
}

1394
STATIC void
1395
xfs_end_io_direct(
1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
	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
1406
	 * to happen from process context but aio+dio I/O completion
1407
	 * happens from irq context so we need to defer it to a workqueue.
1408
	 * This is not necessary for synchronous direct I/O, but we do
1409 1410 1411 1412 1413 1414
	 * it anyway to keep the code uniform and simpler.
	 *
	 * 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.
	 */
1415 1416 1417 1418 1419
	ioend->io_offset = offset;
	ioend->io_size = size;
	if (ioend->io_type == IOMAP_READ) {
		xfs_finish_ioend(ioend);
	} else if (private && size > 0) {
1420 1421
		xfs_finish_ioend(ioend);
	} else {
1422 1423 1424 1425 1426 1427 1428 1429
		/*
		 * 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);
		xfs_finish_ioend(ioend);
1430 1431 1432
	}

	/*
1433
	 * blockdev_direct_IO can return an error even after the I/O
1434 1435 1436 1437 1438 1439
	 * completion handler was called.  Thus we need to protect
	 * against double-freeing.
	 */
	iocb->private = NULL;
}

L
Linus Torvalds 已提交
1440
STATIC ssize_t
1441
xfs_vm_direct_IO(
L
Linus Torvalds 已提交
1442 1443 1444 1445 1446 1447 1448 1449
	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;
1450
	bhv_vnode_t	*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
1451 1452 1453
	xfs_iomap_t	iomap;
	int		maps = 1;
	int		error;
1454
	ssize_t		ret;
L
Linus Torvalds 已提交
1455

1456
	error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps);
L
Linus Torvalds 已提交
1457 1458 1459
	if (error)
		return -error;

1460
	if (rw == WRITE) {
1461
		iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1462 1463 1464 1465 1466 1467
		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 {
1468
		iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1469 1470 1471 1472 1473 1474
		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);
	}
1475

1476
	if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1477 1478
		xfs_destroy_ioend(iocb->private);
	return ret;
L
Linus Torvalds 已提交
1479 1480
}

1481
STATIC int
1482
xfs_vm_prepare_write(
1483 1484 1485 1486 1487
	struct file		*file,
	struct page		*page,
	unsigned int		from,
	unsigned int		to)
{
1488
	return block_prepare_write(page, from, to, xfs_get_blocks);
1489
}
L
Linus Torvalds 已提交
1490 1491

STATIC sector_t
1492
xfs_vm_bmap(
L
Linus Torvalds 已提交
1493 1494 1495 1496
	struct address_space	*mapping,
	sector_t		block)
{
	struct inode		*inode = (struct inode *)mapping->host;
1497
	bhv_vnode_t		*vp = vn_from_inode(inode);
L
Linus Torvalds 已提交
1498

1499
	vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1500 1501 1502
	bhv_vop_rwlock(vp, VRWLOCK_READ);
	bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
	bhv_vop_rwunlock(vp, VRWLOCK_READ);
1503
	return generic_block_bmap(mapping, block, xfs_get_blocks);
L
Linus Torvalds 已提交
1504 1505 1506
}

STATIC int
1507
xfs_vm_readpage(
L
Linus Torvalds 已提交
1508 1509 1510
	struct file		*unused,
	struct page		*page)
{
1511
	return mpage_readpage(page, xfs_get_blocks);
L
Linus Torvalds 已提交
1512 1513 1514
}

STATIC int
1515
xfs_vm_readpages(
L
Linus Torvalds 已提交
1516 1517 1518 1519 1520
	struct file		*unused,
	struct address_space	*mapping,
	struct list_head	*pages,
	unsigned		nr_pages)
{
1521
	return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
L
Linus Torvalds 已提交
1522 1523
}

1524
STATIC void
1525
xfs_vm_invalidatepage(
1526 1527 1528 1529 1530
	struct page		*page,
	unsigned long		offset)
{
	xfs_page_trace(XFS_INVALIDPAGE_ENTER,
			page->mapping->host, page, offset);
1531
	block_invalidatepage(page, offset);
1532 1533
}

1534
const struct address_space_operations xfs_address_space_operations = {
1535 1536 1537
	.readpage		= xfs_vm_readpage,
	.readpages		= xfs_vm_readpages,
	.writepage		= xfs_vm_writepage,
1538
	.writepages		= xfs_vm_writepages,
L
Linus Torvalds 已提交
1539
	.sync_page		= block_sync_page,
1540 1541
	.releasepage		= xfs_vm_releasepage,
	.invalidatepage		= xfs_vm_invalidatepage,
1542
	.prepare_write		= xfs_vm_prepare_write,
L
Linus Torvalds 已提交
1543
	.commit_write		= generic_commit_write,
1544 1545
	.bmap			= xfs_vm_bmap,
	.direct_IO		= xfs_vm_direct_IO,
1546
	.migratepage		= buffer_migrate_page,
L
Linus Torvalds 已提交
1547
};