write.c 41.6 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3
/*
 * linux/fs/nfs/write.c
 *
4
 * Write file data over NFS.
L
Linus Torvalds 已提交
5 6 7 8 9 10 11 12 13 14
 *
 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
 */

#include <linux/types.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/file.h>
#include <linux/writeback.h>
15
#include <linux/swap.h>
16
#include <linux/migrate.h>
L
Linus Torvalds 已提交
17 18 19 20 21

#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs_page.h>
22 23
#include <linux/backing-dev.h>

L
Linus Torvalds 已提交
24 25 26
#include <asm/uaccess.h>

#include "delegation.h"
27
#include "internal.h"
C
Chuck Lever 已提交
28
#include "iostat.h"
29
#include "nfs4_fs.h"
30
#include "fscache.h"
31
#include "pnfs.h"
L
Linus Torvalds 已提交
32 33 34 35 36 37 38 39 40

#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

#define MIN_POOL_WRITE		(32)
#define MIN_POOL_COMMIT		(4)

/*
 * Local function declarations
 */
41 42
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
				  struct inode *inode, int ioflags);
43
static void nfs_redirty_request(struct nfs_page *req);
44 45 46
static const struct rpc_call_ops nfs_write_partial_ops;
static const struct rpc_call_ops nfs_write_full_ops;
static const struct rpc_call_ops nfs_commit_ops;
L
Linus Torvalds 已提交
47

48
static struct kmem_cache *nfs_wdata_cachep;
49
static mempool_t *nfs_wdata_mempool;
L
Linus Torvalds 已提交
50 51
static mempool_t *nfs_commit_mempool;

52
struct nfs_write_data *nfs_commitdata_alloc(void)
L
Linus Torvalds 已提交
53
{
C
Christoph Lameter 已提交
54
	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
55

L
Linus Torvalds 已提交
56 57 58 59 60 61 62
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
	}
	return p;
}

63
void nfs_commit_free(struct nfs_write_data *p)
L
Linus Torvalds 已提交
64
{
65 66
	if (p && (p->pagevec != &p->page_array[0]))
		kfree(p->pagevec);
L
Linus Torvalds 已提交
67 68 69
	mempool_free(p, nfs_commit_mempool);
}

70
struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
71
{
C
Christoph Lameter 已提交
72
	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
73 74 75 76

	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
77
		p->npages = pagecount;
78 79
		if (pagecount <= ARRAY_SIZE(p->page_array))
			p->pagevec = p->page_array;
80
		else {
81 82
			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
			if (!p->pagevec) {
83 84 85 86 87 88 89 90
				mempool_free(p, nfs_wdata_mempool);
				p = NULL;
			}
		}
	}
	return p;
}

T
Trond Myklebust 已提交
91
void nfs_writedata_free(struct nfs_write_data *p)
92 93 94 95 96 97
{
	if (p && (p->pagevec != &p->page_array[0]))
		kfree(p->pagevec);
	mempool_free(p, nfs_wdata_mempool);
}

T
Trond Myklebust 已提交
98
static void nfs_writedata_release(struct nfs_write_data *wdata)
L
Linus Torvalds 已提交
99
{
100
	put_lseg(wdata->lseg);
101
	put_nfs_open_context(wdata->args.context);
L
Linus Torvalds 已提交
102 103 104
	nfs_writedata_free(wdata);
}

105 106 107 108 109 110 111
static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
{
	ctx->error = error;
	smp_wmb();
	set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
}

112 113 114 115 116 117 118
static struct nfs_page *nfs_page_find_request_locked(struct page *page)
{
	struct nfs_page *req = NULL;

	if (PagePrivate(page)) {
		req = (struct nfs_page *)page_private(page);
		if (req != NULL)
119
			kref_get(&req->wb_kref);
120 121 122 123 124 125
	}
	return req;
}

static struct nfs_page *nfs_page_find_request(struct page *page)
{
126
	struct inode *inode = page->mapping->host;
127 128
	struct nfs_page *req = NULL;

129
	spin_lock(&inode->i_lock);
130
	req = nfs_page_find_request_locked(page);
131
	spin_unlock(&inode->i_lock);
132 133 134
	return req;
}

L
Linus Torvalds 已提交
135 136 137 138
/* Adjust the file length if we're writing beyond the end */
static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
{
	struct inode *inode = page->mapping->host;
139 140
	loff_t end, i_size;
	pgoff_t end_index;
L
Linus Torvalds 已提交
141

142 143 144
	spin_lock(&inode->i_lock);
	i_size = i_size_read(inode);
	end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
L
Linus Torvalds 已提交
145
	if (i_size > 0 && page->index < end_index)
146
		goto out;
L
Linus Torvalds 已提交
147 148
	end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
	if (i_size >= end)
149
		goto out;
L
Linus Torvalds 已提交
150
	i_size_write(inode, end);
151 152 153
	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
out:
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
154 155
}

156 157 158 159 160 161 162
/* A writeback failed: mark the page as bad, and invalidate the page cache */
static void nfs_set_pageerror(struct page *page)
{
	SetPageError(page);
	nfs_zap_mapping(page->mapping->host, page->mapping);
}

L
Linus Torvalds 已提交
163 164 165 166 167 168 169 170 171
/* We can set the PG_uptodate flag if we see that a write request
 * covers the full page.
 */
static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
{
	if (PageUptodate(page))
		return;
	if (base != 0)
		return;
172
	if (count != nfs_page_length(page))
L
Linus Torvalds 已提交
173
		return;
174
	SetPageUptodate(page);
L
Linus Torvalds 已提交
175 176 177 178 179
}

static int wb_priority(struct writeback_control *wbc)
{
	if (wbc->for_reclaim)
180
		return FLUSH_HIGHPRI | FLUSH_STABLE;
181
	if (wbc->for_kupdate || wbc->for_background)
L
Linus Torvalds 已提交
182 183 184 185
		return FLUSH_LOWPRI;
	return 0;
}

186 187 188 189 190 191 192 193 194 195
/*
 * NFS congestion control
 */

int nfs_congestion_kb;

#define NFS_CONGESTION_ON_THRESH 	(nfs_congestion_kb >> (PAGE_SHIFT-10))
#define NFS_CONGESTION_OFF_THRESH	\
	(NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))

196
static int nfs_set_page_writeback(struct page *page)
197
{
198 199 200
	int ret = test_set_page_writeback(page);

	if (!ret) {
201 202 203
		struct inode *inode = page->mapping->host;
		struct nfs_server *nfss = NFS_SERVER(inode);

204
		page_cache_get(page);
205
		if (atomic_long_inc_return(&nfss->writeback) >
206 207 208 209
				NFS_CONGESTION_ON_THRESH) {
			set_bdi_congested(&nfss->backing_dev_info,
						BLK_RW_ASYNC);
		}
210
	}
211
	return ret;
212 213 214 215 216 217 218 219
}

static void nfs_end_page_writeback(struct page *page)
{
	struct inode *inode = page->mapping->host;
	struct nfs_server *nfss = NFS_SERVER(inode);

	end_page_writeback(page);
220
	page_cache_release(page);
P
Peter Zijlstra 已提交
221
	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
222
		clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
223 224
}

225
static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
226
{
227
	struct inode *inode = page->mapping->host;
228 229 230
	struct nfs_page *req;
	int ret;

231
	spin_lock(&inode->i_lock);
232
	for (;;) {
233
		req = nfs_page_find_request_locked(page);
234 235
		if (req == NULL)
			break;
236
		if (nfs_set_page_tag_locked(req))
237 238
			break;
		/* Note: If we hold the page lock, as is the case in nfs_writepage,
239
		 *	 then the call to nfs_set_page_tag_locked() will always
240 241 242
		 *	 succeed provided that someone hasn't already marked the
		 *	 request as dirty (in which case we don't care).
		 */
243
		spin_unlock(&inode->i_lock);
244 245 246 247
		if (!nonblock)
			ret = nfs_wait_on_request(req);
		else
			ret = -EAGAIN;
248 249
		nfs_release_request(req);
		if (ret != 0)
250
			return ERR_PTR(ret);
251
		spin_lock(&inode->i_lock);
252
	}
253
	spin_unlock(&inode->i_lock);
254 255 256 257 258 259 260 261
	return req;
}

/*
 * Find an associated nfs write request, and prepare to flush it out
 * May return an error if the user signalled nfs_wait_on_request().
 */
static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
262
				struct page *page, bool nonblock)
263 264 265 266
{
	struct nfs_page *req;
	int ret = 0;

267
	req = nfs_find_and_lock_request(page, nonblock);
268 269 270 271 272 273 274 275 276 277
	if (!req)
		goto out;
	ret = PTR_ERR(req);
	if (IS_ERR(req))
		goto out;

	ret = nfs_set_page_writeback(page);
	BUG_ON(ret != 0);
	BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));

278 279
	if (!nfs_pageio_add_request(pgio, req)) {
		nfs_redirty_request(req);
280
		ret = pgio->pg_error;
281
	}
282 283
out:
	return ret;
284 285
}

T
Trond Myklebust 已提交
286
static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
L
Linus Torvalds 已提交
287 288
{
	struct inode *inode = page->mapping->host;
289
	int ret;
L
Linus Torvalds 已提交
290

C
Chuck Lever 已提交
291 292 293
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);

294
	nfs_pageio_cond_complete(pgio, page->index);
295
	ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
296 297 298 299 300
	if (ret == -EAGAIN) {
		redirty_page_for_writepage(wbc, page);
		ret = 0;
	}
	return ret;
T
Trond Myklebust 已提交
301
}
302

T
Trond Myklebust 已提交
303 304 305 306 307 308 309
/*
 * Write an mmapped page to the server.
 */
static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
{
	struct nfs_pageio_descriptor pgio;
	int err;
310

T
Trond Myklebust 已提交
311 312 313 314 315 316 317 318
	nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
	err = nfs_do_writepage(page, wbc, &pgio);
	nfs_pageio_complete(&pgio);
	if (err < 0)
		return err;
	if (pgio.pg_error < 0)
		return pgio.pg_error;
	return 0;
319 320 321 322
}

int nfs_writepage(struct page *page, struct writeback_control *wbc)
{
T
Trond Myklebust 已提交
323
	int ret;
324

T
Trond Myklebust 已提交
325
	ret = nfs_writepage_locked(page, wbc);
L
Linus Torvalds 已提交
326
	unlock_page(page);
T
Trond Myklebust 已提交
327 328 329 330 331 332 333 334 335 336
	return ret;
}

static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
{
	int ret;

	ret = nfs_do_writepage(page, wbc, data);
	unlock_page(page);
	return ret;
L
Linus Torvalds 已提交
337 338 339 340 341
}

int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
342
	unsigned long *bitlock = &NFS_I(inode)->flags;
343
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
344 345
	int err;

346 347 348 349 350 351
	/* Stop dirtying of new pages while we sync */
	err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
			nfs_wait_bit_killable, TASK_KILLABLE);
	if (err)
		goto out_err;

C
Chuck Lever 已提交
352 353
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);

354
	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
T
Trond Myklebust 已提交
355
	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
356
	nfs_pageio_complete(&pgio);
357 358 359 360 361

	clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
	smp_mb__after_clear_bit();
	wake_up_bit(bitlock, NFS_INO_FLUSHING);

T
Trond Myklebust 已提交
362
	if (err < 0)
363 364 365 366
		goto out_err;
	err = pgio.pg_error;
	if (err < 0)
		goto out_err;
367
	return 0;
368 369
out_err:
	return err;
L
Linus Torvalds 已提交
370 371 372 373 374
}

/*
 * Insert a write request into an inode
 */
375
static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
L
Linus Torvalds 已提交
376 377 378 379
{
	struct nfs_inode *nfsi = NFS_I(inode);
	int error;

380 381 382 383 384 385 386 387
	error = radix_tree_preload(GFP_NOFS);
	if (error != 0)
		goto out;

	/* Lock the request! */
	nfs_lock_request_dontget(req);

	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
388
	error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
389
	BUG_ON(error);
L
Linus Torvalds 已提交
390 391 392 393 394
	if (!nfsi->npages) {
		igrab(inode);
		if (nfs_have_delegation(inode, FMODE_WRITE))
			nfsi->change_attr++;
	}
395
	set_bit(PG_MAPPED, &req->wb_flags);
396
	SetPagePrivate(req->wb_page);
397
	set_page_private(req->wb_page, (unsigned long)req);
L
Linus Torvalds 已提交
398
	nfsi->npages++;
399
	kref_get(&req->wb_kref);
400 401
	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
				NFS_PAGE_TAG_LOCKED);
402 403 404 405
	spin_unlock(&inode->i_lock);
	radix_tree_preload_end();
out:
	return error;
L
Linus Torvalds 已提交
406 407 408
}

/*
409
 * Remove a write request from an inode
L
Linus Torvalds 已提交
410 411 412
 */
static void nfs_inode_remove_request(struct nfs_page *req)
{
413
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
414 415 416 417
	struct nfs_inode *nfsi = NFS_I(inode);

	BUG_ON (!NFS_WBACK_BUSY(req));

418
	spin_lock(&inode->i_lock);
419
	set_page_private(req->wb_page, 0);
420
	ClearPagePrivate(req->wb_page);
421
	clear_bit(PG_MAPPED, &req->wb_flags);
L
Linus Torvalds 已提交
422 423 424
	radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
	nfsi->npages--;
	if (!nfsi->npages) {
425
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
426 427
		iput(inode);
	} else
428
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
429 430 431
	nfs_release_request(req);
}

432
static void
F
Fred 已提交
433
nfs_mark_request_dirty(struct nfs_page *req)
434 435
{
	__set_page_dirty_nobuffers(req->wb_page);
436
	__mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC);
437 438
}

L
Linus Torvalds 已提交
439 440 441 442 443 444 445
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
/*
 * Add a request to the inode's commit list.
 */
static void
nfs_mark_request_commit(struct nfs_page *req)
{
446
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
447 448
	struct nfs_inode *nfsi = NFS_I(inode);

449
	spin_lock(&inode->i_lock);
450
	set_bit(PG_CLEAN, &(req)->wb_flags);
451 452 453
	radix_tree_tag_set(&nfsi->nfs_page_tree,
			req->wb_index,
			NFS_PAGE_TAG_COMMIT);
454
	nfsi->ncommit++;
455
	spin_unlock(&inode->i_lock);
456
	inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
457
	inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
458
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
459
}
460

461 462 463 464 465 466 467 468 469 470 471 472 473
static int
nfs_clear_request_commit(struct nfs_page *req)
{
	struct page *page = req->wb_page;

	if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
		dec_zone_page_state(page, NR_UNSTABLE_NFS);
		dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
		return 1;
	}
	return 0;
}

474 475 476 477 478 479 480 481 482
static inline
int nfs_write_need_commit(struct nfs_write_data *data)
{
	return data->verf.committed != NFS_FILE_SYNC;
}

static inline
int nfs_reschedule_unstable_write(struct nfs_page *req)
{
483
	if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
484 485 486 487
		nfs_mark_request_commit(req);
		return 1;
	}
	if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
F
Fred 已提交
488
		nfs_mark_request_dirty(req);
489 490 491 492 493 494 495 496 497 498
		return 1;
	}
	return 0;
}
#else
static inline void
nfs_mark_request_commit(struct nfs_page *req)
{
}

499 500 501 502 503 504
static inline int
nfs_clear_request_commit(struct nfs_page *req)
{
	return 0;
}

505 506 507 508 509 510 511 512 513 514 515
static inline
int nfs_write_need_commit(struct nfs_write_data *data)
{
	return 0;
}

static inline
int nfs_reschedule_unstable_write(struct nfs_page *req)
{
	return 0;
}
L
Linus Torvalds 已提交
516 517
#endif

T
Trond Myklebust 已提交
518
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
519 520 521 522 523 524
static int
nfs_need_commit(struct nfs_inode *nfsi)
{
	return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
}

L
Linus Torvalds 已提交
525 526 527 528 529 530 531 532 533 534 535
/*
 * nfs_scan_commit - Scan an inode for commit requests
 * @inode: NFS inode to scan
 * @dst: destination list
 * @idx_start: lower bound of page->index to scan.
 * @npages: idx_start + npages sets the upper bound to scan.
 *
 * Moves requests from the inode's 'commit' request list.
 * The requests are *not* checked to ensure that they form a contiguous set.
 */
static int
536
nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
L
Linus Torvalds 已提交
537 538
{
	struct nfs_inode *nfsi = NFS_I(inode);
539
	int ret;
540

541 542 543
	if (!nfs_need_commit(nfsi))
		return 0;

544 545 546
	ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
	if (ret > 0)
		nfsi->ncommit -= ret;
547 548
	if (nfs_need_commit(NFS_I(inode)))
		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
549
	return ret;
L
Linus Torvalds 已提交
550
}
551
#else
552 553 554 555 556
static inline int nfs_need_commit(struct nfs_inode *nfsi)
{
	return 0;
}

557
static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
558 559 560
{
	return 0;
}
L
Linus Torvalds 已提交
561 562 563
#endif

/*
564 565
 * Search for an existing write request, and attempt to update
 * it to reflect a new dirty region on a given page.
L
Linus Torvalds 已提交
566
 *
567 568
 * If the attempt fails, then the existing request is flushed out
 * to disk.
L
Linus Torvalds 已提交
569
 */
570 571 572 573
static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
		struct page *page,
		unsigned int offset,
		unsigned int bytes)
L
Linus Torvalds 已提交
574
{
575 576 577 578 579 580 581
	struct nfs_page *req;
	unsigned int rqend;
	unsigned int end;
	int error;

	if (!PagePrivate(page))
		return NULL;
L
Linus Torvalds 已提交
582 583

	end = offset + bytes;
584
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
585 586

	for (;;) {
587
		req = nfs_page_find_request_locked(page);
588 589 590 591 592 593 594 595 596 597
		if (req == NULL)
			goto out_unlock;

		rqend = req->wb_offset + req->wb_bytes;
		/*
		 * Tell the caller to flush out the request if
		 * the offsets are non-contiguous.
		 * Note: nfs_flush_incompatible() will already
		 * have flushed out requests having wrong owners.
		 */
598
		if (offset > rqend
599 600 601 602
		    || end < req->wb_offset)
			goto out_flushme;

		if (nfs_set_page_tag_locked(req))
L
Linus Torvalds 已提交
603 604
			break;

605
		/* The request is locked, so wait and then retry */
606
		spin_unlock(&inode->i_lock);
607 608 609 610 611
		error = nfs_wait_on_request(req);
		nfs_release_request(req);
		if (error != 0)
			goto out_err;
		spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
612 613
	}

614 615 616 617
	if (nfs_clear_request_commit(req) &&
			radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
				req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL)
		NFS_I(inode)->ncommit--;
618

L
Linus Torvalds 已提交
619 620 621 622 623 624 625
	/* Okay, the request matches. Update the region */
	if (offset < req->wb_offset) {
		req->wb_offset = offset;
		req->wb_pgbase = offset;
	}
	if (end > rqend)
		req->wb_bytes = end - req->wb_offset;
626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651
	else
		req->wb_bytes = rqend - req->wb_offset;
out_unlock:
	spin_unlock(&inode->i_lock);
	return req;
out_flushme:
	spin_unlock(&inode->i_lock);
	nfs_release_request(req);
	error = nfs_wb_page(inode, page);
out_err:
	return ERR_PTR(error);
}

/*
 * Try to update an existing write request, or create one if there is none.
 *
 * Note: Should always be called with the Page Lock held to prevent races
 * if we have to add a new request. Also assumes that the caller has
 * already called nfs_flush_incompatible() if necessary.
 */
static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
		struct page *page, unsigned int offset, unsigned int bytes)
{
	struct inode *inode = page->mapping->host;
	struct nfs_page	*req;
	int error;
L
Linus Torvalds 已提交
652

653 654 655 656 657 658 659 660 661 662 663
	req = nfs_try_to_update_request(inode, page, offset, bytes);
	if (req != NULL)
		goto out;
	req = nfs_create_request(ctx, inode, page, offset, bytes);
	if (IS_ERR(req))
		goto out;
	error = nfs_inode_add_request(inode, req);
	if (error != 0) {
		nfs_release_request(req);
		req = ERR_PTR(error);
	}
664
out:
T
Trond Myklebust 已提交
665
	return req;
L
Linus Torvalds 已提交
666 667
}

668 669 670 671 672 673 674 675
static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
		unsigned int offset, unsigned int count)
{
	struct nfs_page	*req;

	req = nfs_setup_write_request(ctx, page, offset, count);
	if (IS_ERR(req))
		return PTR_ERR(req);
676
	nfs_mark_request_dirty(req);
677 678 679
	/* Update file length */
	nfs_grow_file(page, offset, count);
	nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
680
	nfs_mark_request_dirty(req);
681 682 683 684
	nfs_clear_page_tag_locked(req);
	return 0;
}

L
Linus Torvalds 已提交
685 686
int nfs_flush_incompatible(struct file *file, struct page *page)
{
687
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
688
	struct nfs_page	*req;
T
Trond Myklebust 已提交
689
	int do_flush, status;
L
Linus Torvalds 已提交
690 691 692 693 694 695 696 697
	/*
	 * Look for a request corresponding to this page. If there
	 * is one, and it belongs to another file, we flush it out
	 * before we try to copy anything into the page. Do this
	 * due to the lack of an ACCESS-type call in NFSv2.
	 * Also do the same if we find a request from an existing
	 * dropped page.
	 */
T
Trond Myklebust 已提交
698 699 700 701
	do {
		req = nfs_page_find_request(page);
		if (req == NULL)
			return 0;
702 703 704
		do_flush = req->wb_page != page || req->wb_context != ctx ||
			req->wb_lock_context->lockowner != current->files ||
			req->wb_lock_context->pid != current->tgid;
L
Linus Torvalds 已提交
705
		nfs_release_request(req);
T
Trond Myklebust 已提交
706 707 708 709 710
		if (!do_flush)
			return 0;
		status = nfs_wb_page(page->mapping->host, page);
	} while (status == 0);
	return status;
L
Linus Torvalds 已提交
711 712
}

713 714 715 716 717 718 719 720 721 722 723
/*
 * If the page cache is marked as unsafe or invalid, then we can't rely on
 * the PageUptodate() flag. In this case, we will need to turn off
 * write optimisations that depend on the page contents being correct.
 */
static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
{
	return PageUptodate(page) &&
		!(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
}

L
Linus Torvalds 已提交
724 725 726 727 728 729 730 731 732
/*
 * Update and possibly write a cached page of an NFS file.
 *
 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
 * things with a page scheduled for an RPC call (e.g. invalidate it).
 */
int nfs_updatepage(struct file *file, struct page *page,
		unsigned int offset, unsigned int count)
{
733
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
734 735 736
	struct inode	*inode = page->mapping->host;
	int		status = 0;

C
Chuck Lever 已提交
737 738
	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);

739
	dprintk("NFS:       nfs_updatepage(%s/%s %d@%lld)\n",
740 741
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name, count,
742
		(long long)(page_offset(page) + offset));
L
Linus Torvalds 已提交
743 744

	/* If we're not using byte range locks, and we know the page
745 746 747
	 * is up to date, it may be more efficient to extend the write
	 * to cover the entire page in order to avoid fragmentation
	 * inefficiencies.
L
Linus Torvalds 已提交
748
	 */
749 750
	if (nfs_write_pageuptodate(page, inode) &&
			inode->i_flock == NULL &&
751
			!(file->f_flags & O_DSYNC)) {
752
		count = max(count + offset, nfs_page_length(page));
L
Linus Torvalds 已提交
753 754 755
		offset = 0;
	}

756
	status = nfs_writepage_setup(ctx, page, offset, count);
757 758
	if (status < 0)
		nfs_set_pageerror(page);
L
Linus Torvalds 已提交
759

760
	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
L
Linus Torvalds 已提交
761 762 763 764 765 766
			status, (long long)i_size_read(inode));
	return status;
}

static void nfs_writepage_release(struct nfs_page *req)
{
767
	struct page *page = req->wb_page;
L
Linus Torvalds 已提交
768

769
	if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req))
770
		nfs_inode_remove_request(req);
771
	nfs_clear_page_tag_locked(req);
772
	nfs_end_page_writeback(page);
L
Linus Torvalds 已提交
773 774
}

775
static int flush_task_priority(int how)
L
Linus Torvalds 已提交
776 777 778 779 780 781 782 783 784 785
{
	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
		case FLUSH_HIGHPRI:
			return RPC_PRIORITY_HIGH;
		case FLUSH_LOWPRI:
			return RPC_PRIORITY_LOW;
	}
	return RPC_PRIORITY_NORMAL;
}

786
int nfs_initiate_write(struct nfs_write_data *data,
787 788 789
		       struct rpc_clnt *clnt,
		       const struct rpc_call_ops *call_ops,
		       int how)
L
Linus Torvalds 已提交
790
{
791
	struct inode *inode = data->inode;
792
	int priority = flush_task_priority(how);
793
	struct rpc_task *task;
794 795 796
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
797
		.rpc_cred = data->cred,
798
	};
799
	struct rpc_task_setup task_setup_data = {
800
		.rpc_client = clnt,
801
		.task = &data->task,
802
		.rpc_message = &msg,
803 804
		.callback_ops = call_ops,
		.callback_data = data,
805
		.workqueue = nfsiod_workqueue,
806
		.flags = RPC_TASK_ASYNC,
807
		.priority = priority,
808
	};
809
	int ret = 0;
L
Linus Torvalds 已提交
810

811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835
	/* Set up the initial task struct.  */
	NFS_PROTO(inode)->write_setup(data, &msg);

	dprintk("NFS: %5u initiated write call "
		"(req %s/%lld, %u bytes @ offset %llu)\n",
		data->task.tk_pid,
		inode->i_sb->s_id,
		(long long)NFS_FILEID(inode),
		data->args.count,
		(unsigned long long)data->args.offset);

	task = rpc_run_task(&task_setup_data);
	if (IS_ERR(task)) {
		ret = PTR_ERR(task);
		goto out;
	}
	if (how & FLUSH_SYNC) {
		ret = rpc_wait_for_completion_task(task);
		if (ret == 0)
			ret = task->tk_status;
	}
	rpc_put_task(task);
out:
	return ret;
}
836
EXPORT_SYMBOL_GPL(nfs_initiate_write);
837 838 839 840 841 842 843 844

/*
 * Set up the argument/result storage required for the RPC call.
 */
static int nfs_write_rpcsetup(struct nfs_page *req,
		struct nfs_write_data *data,
		const struct rpc_call_ops *call_ops,
		unsigned int count, unsigned int offset,
845
		struct pnfs_layout_segment *lseg,
846 847 848 849
		int how)
{
	struct inode *inode = req->wb_context->path.dentry->d_inode;

L
Linus Torvalds 已提交
850 851 852 853
	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

	data->req = req;
854
	data->inode = inode = req->wb_context->path.dentry->d_inode;
855
	data->cred = req->wb_context->cred;
856
	data->lseg = get_lseg(lseg);
L
Linus Torvalds 已提交
857 858 859 860 861 862

	data->args.fh     = NFS_FH(inode);
	data->args.offset = req_offset(req) + offset;
	data->args.pgbase = req->wb_pgbase + offset;
	data->args.pages  = data->pagevec;
	data->args.count  = count;
863
	data->args.context = get_nfs_open_context(req->wb_context);
864
	data->args.lock_context = req->wb_lock_context;
865 866 867
	data->args.stable  = NFS_UNSTABLE;
	if (how & FLUSH_STABLE) {
		data->args.stable = NFS_DATA_SYNC;
868
		if (!nfs_need_commit(NFS_I(inode)))
869 870
			data->args.stable = NFS_FILE_SYNC;
	}
L
Linus Torvalds 已提交
871 872 873 874

	data->res.fattr   = &data->fattr;
	data->res.count   = count;
	data->res.verf    = &data->verf;
875
	nfs_fattr_init(&data->fattr);
L
Linus Torvalds 已提交
876

877 878 879 880
	if (data->lseg &&
	    (pnfs_try_to_write_data(data, call_ops, how) == PNFS_ATTEMPTED))
		return 0;

881
	return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
L
Linus Torvalds 已提交
882 883
}

F
Fred 已提交
884 885 886 887 888 889
/* If a nfs_flush_* function fails, it should remove reqs from @head and
 * call this on each, which will prepare them to be retried on next
 * writeback using standard nfs.
 */
static void nfs_redirty_request(struct nfs_page *req)
{
890 891
	struct page *page = req->wb_page;

F
Fred 已提交
892 893
	nfs_mark_request_dirty(req);
	nfs_clear_page_tag_locked(req);
894
	nfs_end_page_writeback(page);
F
Fred 已提交
895 896
}

L
Linus Torvalds 已提交
897 898 899 900
/*
 * Generate multiple small requests to write out a single
 * contiguous dirty area on one page.
 */
901
static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how, struct pnfs_layout_segment *lseg)
L
Linus Torvalds 已提交
902 903 904 905
{
	struct nfs_page *req = nfs_list_entry(head->next);
	struct page *page = req->wb_page;
	struct nfs_write_data *data;
906 907
	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
	unsigned int offset;
L
Linus Torvalds 已提交
908
	int requests = 0;
909
	int ret = 0;
L
Linus Torvalds 已提交
910 911 912 913
	LIST_HEAD(list);

	nfs_list_remove_request(req);

914
	nbytes = count;
915 916 917
	do {
		size_t len = min(nbytes, wsize);

918
		data = nfs_writedata_alloc(1);
L
Linus Torvalds 已提交
919 920 921 922
		if (!data)
			goto out_bad;
		list_add(&data->pages, &list);
		requests++;
923 924
		nbytes -= len;
	} while (nbytes != 0);
L
Linus Torvalds 已提交
925 926
	atomic_set(&req->wb_complete, requests);

927 928
	BUG_ON(lseg);
	lseg = pnfs_update_layout(inode, req->wb_context, IOMODE_RW);
L
Linus Torvalds 已提交
929 930
	ClearPageError(page);
	offset = 0;
931
	nbytes = count;
L
Linus Torvalds 已提交
932
	do {
933 934
		int ret2;

L
Linus Torvalds 已提交
935 936 937 938 939
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del_init(&data->pages);

		data->pagevec[0] = page;

940 941
		if (nbytes < wsize)
			wsize = nbytes;
942
		ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
943
					  wsize, offset, lseg, how);
944 945
		if (ret == 0)
			ret = ret2;
946 947
		offset += wsize;
		nbytes -= wsize;
L
Linus Torvalds 已提交
948 949
	} while (nbytes != 0);

950
	put_lseg(lseg);
951
	return ret;
L
Linus Torvalds 已提交
952 953 954 955 956

out_bad:
	while (!list_empty(&list)) {
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del(&data->pages);
957
		nfs_writedata_free(data);
L
Linus Torvalds 已提交
958
	}
959
	nfs_redirty_request(req);
L
Linus Torvalds 已提交
960 961 962 963 964 965 966 967 968 969 970
	return -ENOMEM;
}

/*
 * Create an RPC task for the given write request and kick it.
 * The page must have been locked by the caller.
 *
 * It may happen that the page we're passed is not marked dirty.
 * This is the case if nfs_updatepage detects a conflicting request
 * that has been written but not committed.
 */
971
static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how, struct pnfs_layout_segment *lseg)
L
Linus Torvalds 已提交
972 973 974 975
{
	struct nfs_page		*req;
	struct page		**pages;
	struct nfs_write_data	*data;
976
	int ret;
L
Linus Torvalds 已提交
977

978
	data = nfs_writedata_alloc(npages);
979 980 981 982 983 984 985 986 987
	if (!data) {
		while (!list_empty(head)) {
			req = nfs_list_entry(head->next);
			nfs_list_remove_request(req);
			nfs_redirty_request(req);
		}
		ret = -ENOMEM;
		goto out;
	}
L
Linus Torvalds 已提交
988 989 990 991 992 993 994 995 996
	pages = data->pagevec;
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_list_add_request(req, &data->pages);
		ClearPageError(req->wb_page);
		*pages++ = req->wb_page;
	}
	req = nfs_list_entry(data->pages.next);
997 998
	if ((!lseg) && list_is_singular(&data->pages))
		lseg = pnfs_update_layout(inode, req->wb_context, IOMODE_RW);
L
Linus Torvalds 已提交
999 1000

	/* Set up the argument struct */
1001 1002 1003 1004
	ret = nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, lseg, how);
out:
	put_lseg(lseg); /* Cleans any gotten in ->pg_test */
	return ret;
L
Linus Torvalds 已提交
1005 1006
}

1007 1008
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
				  struct inode *inode, int ioflags)
L
Linus Torvalds 已提交
1009
{
1010
	size_t wsize = NFS_SERVER(inode)->wsize;
L
Linus Torvalds 已提交
1011

1012
	pnfs_pageio_init_write(pgio, inode);
1013

1014
	if (wsize < PAGE_CACHE_SIZE)
1015
		nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
1016
	else
1017
		nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
L
Linus Torvalds 已提交
1018 1019 1020 1021 1022
}

/*
 * Handle a write reply that flushed part of a page.
 */
1023
static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1024
{
1025
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1026

1027 1028 1029 1030 1031 1032
	dprintk("NFS: %5u write(%s/%lld %d@%lld)",
		task->tk_pid,
		data->req->wb_context->path.dentry->d_inode->i_sb->s_id,
		(long long)
		  NFS_FILEID(data->req->wb_context->path.dentry->d_inode),
		data->req->wb_bytes, (long long)req_offset(data->req));
L
Linus Torvalds 已提交
1033

1034 1035
	nfs_writeback_done(task, data);
}
1036

1037 1038 1039 1040 1041 1042 1043 1044
static void nfs_writeback_release_partial(void *calldata)
{
	struct nfs_write_data	*data = calldata;
	struct nfs_page		*req = data->req;
	struct page		*page = req->wb_page;
	int status = data->task.tk_status;

	if (status < 0) {
1045
		nfs_set_pageerror(page);
1046 1047
		nfs_context_set_write_error(req->wb_context, status);
		dprintk(", error = %d\n", status);
1048
		goto out;
L
Linus Torvalds 已提交
1049 1050
	}

1051
	if (nfs_write_need_commit(data)) {
1052
		struct inode *inode = page->mapping->host;
1053

1054
		spin_lock(&inode->i_lock);
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
		if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
			/* Do nothing we need to resend the writes */
		} else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
			memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
			dprintk(" defer commit\n");
		} else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
			set_bit(PG_NEED_RESCHED, &req->wb_flags);
			clear_bit(PG_NEED_COMMIT, &req->wb_flags);
			dprintk(" server reboot detected\n");
		}
1065
		spin_unlock(&inode->i_lock);
1066 1067 1068 1069
	} else
		dprintk(" OK\n");

out:
L
Linus Torvalds 已提交
1070 1071
	if (atomic_dec_and_test(&req->wb_complete))
		nfs_writepage_release(req);
1072
	nfs_writedata_release(calldata);
L
Linus Torvalds 已提交
1073 1074
}

1075 1076 1077 1078 1079
#if defined(CONFIG_NFS_V4_1)
void nfs_write_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_write_data *data = calldata;

1080 1081
	if (nfs4_setup_sequence(NFS_SERVER(data->inode),
				&data->args.seq_args,
1082 1083 1084 1085 1086 1087
				&data->res.seq_res, 1, task))
		return;
	rpc_call_start(task);
}
#endif /* CONFIG_NFS_V4_1 */

1088
static const struct rpc_call_ops nfs_write_partial_ops = {
1089 1090 1091
#if defined(CONFIG_NFS_V4_1)
	.rpc_call_prepare = nfs_write_prepare,
#endif /* CONFIG_NFS_V4_1 */
1092
	.rpc_call_done = nfs_writeback_done_partial,
1093
	.rpc_release = nfs_writeback_release_partial,
1094 1095
};

L
Linus Torvalds 已提交
1096 1097 1098 1099 1100 1101 1102
/*
 * Handle a write reply that flushes a whole page.
 *
 * FIXME: There is an inherent race with invalidate_inode_pages and
 *	  writebacks since the page->count is kept > 1 for as long
 *	  as the page has a write request pending.
 */
1103
static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1104
{
1105
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1106

1107 1108 1109 1110 1111 1112 1113
	nfs_writeback_done(task, data);
}

static void nfs_writeback_release_full(void *calldata)
{
	struct nfs_write_data	*data = calldata;
	int status = data->task.tk_status;
1114

L
Linus Torvalds 已提交
1115 1116
	/* Update attributes as result of writeback. */
	while (!list_empty(&data->pages)) {
1117 1118 1119
		struct nfs_page *req = nfs_list_entry(data->pages.next);
		struct page *page = req->wb_page;

L
Linus Torvalds 已提交
1120 1121
		nfs_list_remove_request(req);

1122 1123
		dprintk("NFS: %5u write (%s/%lld %d@%lld)",
			data->task.tk_pid,
1124 1125
			req->wb_context->path.dentry->d_inode->i_sb->s_id,
			(long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
L
Linus Torvalds 已提交
1126 1127 1128
			req->wb_bytes,
			(long long)req_offset(req));

1129
		if (status < 0) {
1130
			nfs_set_pageerror(page);
1131 1132
			nfs_context_set_write_error(req->wb_context, status);
			dprintk(", error = %d\n", status);
1133
			goto remove_request;
L
Linus Torvalds 已提交
1134 1135
		}

1136 1137 1138 1139
		if (nfs_write_need_commit(data)) {
			memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
			nfs_mark_request_commit(req);
			dprintk(" marked for commit\n");
L
Linus Torvalds 已提交
1140 1141
			goto next;
		}
1142 1143
		dprintk(" OK\n");
remove_request:
L
Linus Torvalds 已提交
1144 1145
		nfs_inode_remove_request(req);
	next:
1146
		nfs_clear_page_tag_locked(req);
1147
		nfs_end_page_writeback(page);
L
Linus Torvalds 已提交
1148
	}
1149
	nfs_writedata_release(calldata);
L
Linus Torvalds 已提交
1150 1151
}

1152
static const struct rpc_call_ops nfs_write_full_ops = {
1153 1154 1155
#if defined(CONFIG_NFS_V4_1)
	.rpc_call_prepare = nfs_write_prepare,
#endif /* CONFIG_NFS_V4_1 */
1156
	.rpc_call_done = nfs_writeback_done_full,
1157
	.rpc_release = nfs_writeback_release_full,
1158 1159 1160
};


L
Linus Torvalds 已提交
1161 1162 1163
/*
 * This function is called when the WRITE call is complete.
 */
1164
void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
L
Linus Torvalds 已提交
1165 1166 1167
{
	struct nfs_writeargs	*argp = &data->args;
	struct nfs_writeres	*resp = &data->res;
1168
	struct nfs_server	*server = NFS_SERVER(data->inode);
1169
	int status;
L
Linus Torvalds 已提交
1170

C
Chuck Lever 已提交
1171
	dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
L
Linus Torvalds 已提交
1172 1173
		task->tk_pid, task->tk_status);

1174 1175 1176 1177 1178 1179 1180
	/*
	 * ->write_done will attempt to use post-op attributes to detect
	 * conflicting writes by other clients.  A strict interpretation
	 * of close-to-open would allow us to continue caching even if
	 * another writer had changed the file, but some applications
	 * depend on tighter cache coherency when writing.
	 */
1181 1182
	status = NFS_PROTO(data->inode)->write_done(task, data);
	if (status != 0)
1183
		return;
C
Chuck Lever 已提交
1184 1185
	nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);

L
Linus Torvalds 已提交
1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
	if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
		/* We tried a write call, but the server did not
		 * commit data to stable storage even though we
		 * requested it.
		 * Note: There is a known bug in Tru64 < 5.0 in which
		 *	 the server reports NFS_DATA_SYNC, but performs
		 *	 NFS_FILE_SYNC. We therefore implement this checking
		 *	 as a dprintk() in order to avoid filling syslog.
		 */
		static unsigned long    complain;

1198
		/* Note this will print the MDS for a DS write */
L
Linus Torvalds 已提交
1199
		if (time_before(complain, jiffies)) {
1200
			dprintk("NFS:       faulty NFS server %s:"
L
Linus Torvalds 已提交
1201
				" (committed = %d) != (stable = %d)\n",
1202
				server->nfs_client->cl_hostname,
L
Linus Torvalds 已提交
1203 1204 1205 1206 1207 1208 1209 1210 1211
				resp->verf->committed, argp->stable);
			complain = jiffies + 300 * HZ;
		}
	}
#endif
	/* Is this a short write? */
	if (task->tk_status >= 0 && resp->count < argp->count) {
		static unsigned long    complain;

C
Chuck Lever 已提交
1212 1213
		nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);

L
Linus Torvalds 已提交
1214 1215 1216 1217 1218
		/* Has the server at least made some progress? */
		if (resp->count != 0) {
			/* Was this an NFSv2 write or an NFSv3 stable write? */
			if (resp->verf->committed != NFS_UNSTABLE) {
				/* Resend from where the server left off */
1219
				data->mds_offset += resp->count;
L
Linus Torvalds 已提交
1220 1221 1222 1223 1224 1225 1226 1227 1228
				argp->offset += resp->count;
				argp->pgbase += resp->count;
				argp->count -= resp->count;
			} else {
				/* Resend as a stable write in order to avoid
				 * headaches in the case of a server crash.
				 */
				argp->stable = NFS_FILE_SYNC;
			}
1229
			nfs_restart_rpc(task, server->nfs_client);
1230
			return;
L
Linus Torvalds 已提交
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240
		}
		if (time_before(complain, jiffies)) {
			printk(KERN_WARNING
			       "NFS: Server wrote zero bytes, expected %u.\n",
					argp->count);
			complain = jiffies + 300 * HZ;
		}
		/* Can't do anything about it except throw an error. */
		task->tk_status = -EIO;
	}
1241
	return;
L
Linus Torvalds 已提交
1242 1243 1244 1245
}


#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264
static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
{
	if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
		return 1;
	if (may_wait && !out_of_line_wait_on_bit_lock(&nfsi->flags,
				NFS_INO_COMMIT, nfs_wait_bit_killable,
				TASK_KILLABLE))
		return 1;
	return 0;
}

static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
{
	clear_bit(NFS_INO_COMMIT, &nfsi->flags);
	smp_mb__after_clear_bit();
	wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
}


1265
static void nfs_commitdata_release(void *data)
L
Linus Torvalds 已提交
1266
{
1267 1268 1269
	struct nfs_write_data *wdata = data;

	put_nfs_open_context(wdata->args.context);
L
Linus Torvalds 已提交
1270 1271 1272 1273 1274 1275
	nfs_commit_free(wdata);
}

/*
 * Set up the argument/result storage required for the RPC call.
 */
1276
static int nfs_commit_rpcsetup(struct list_head *head,
1277 1278
		struct nfs_write_data *data,
		int how)
L
Linus Torvalds 已提交
1279
{
1280 1281
	struct nfs_page *first = nfs_list_entry(head->next);
	struct inode *inode = first->wb_context->path.dentry->d_inode;
1282
	int priority = flush_task_priority(how);
1283
	struct rpc_task *task;
1284 1285 1286 1287 1288
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = first->wb_context->cred,
	};
1289
	struct rpc_task_setup task_setup_data = {
1290
		.task = &data->task,
1291
		.rpc_client = NFS_CLIENT(inode),
1292
		.rpc_message = &msg,
1293 1294
		.callback_ops = &nfs_commit_ops,
		.callback_data = data,
1295
		.workqueue = nfsiod_workqueue,
1296
		.flags = RPC_TASK_ASYNC,
1297
		.priority = priority,
1298
	};
L
Linus Torvalds 已提交
1299 1300 1301 1302 1303 1304 1305

	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

	list_splice_init(head, &data->pages);

	data->inode	  = inode;
1306
	data->cred	  = msg.rpc_cred;
L
Linus Torvalds 已提交
1307 1308

	data->args.fh     = NFS_FH(data->inode);
1309 1310 1311
	/* Note: we always request a commit of the entire inode */
	data->args.offset = 0;
	data->args.count  = 0;
1312
	data->args.context = get_nfs_open_context(first->wb_context);
1313
	data->res.count   = 0;
L
Linus Torvalds 已提交
1314 1315
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
1316
	nfs_fattr_init(&data->fattr);
1317 1318

	/* Set up the initial task struct.  */
1319
	NFS_PROTO(inode)->commit_setup(data, &msg);
L
Linus Torvalds 已提交
1320

C
Chuck Lever 已提交
1321
	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1322

1323
	task = rpc_run_task(&task_setup_data);
1324 1325
	if (IS_ERR(task))
		return PTR_ERR(task);
1326 1327
	if (how & FLUSH_SYNC)
		rpc_wait_for_completion_task(task);
1328 1329
	rpc_put_task(task);
	return 0;
L
Linus Torvalds 已提交
1330 1331 1332 1333 1334 1335
}

/*
 * Commit dirty pages
 */
static int
1336
nfs_commit_list(struct inode *inode, struct list_head *head, int how)
L
Linus Torvalds 已提交
1337 1338 1339 1340
{
	struct nfs_write_data	*data;
	struct nfs_page         *req;

1341
	data = nfs_commitdata_alloc();
L
Linus Torvalds 已提交
1342 1343 1344 1345 1346

	if (!data)
		goto out_bad;

	/* Set up the argument struct */
1347
	return nfs_commit_rpcsetup(head, data, how);
L
Linus Torvalds 已提交
1348 1349 1350 1351 1352
 out_bad:
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_mark_request_commit(req);
T
Trond Myklebust 已提交
1353
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1354 1355
		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
				BDI_RECLAIMABLE);
1356
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1357
	}
1358
	nfs_commit_clear_lock(NFS_I(inode));
L
Linus Torvalds 已提交
1359 1360 1361 1362 1363 1364
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
1365
static void nfs_commit_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1366
{
1367
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1368

C
Chuck Lever 已提交
1369
        dprintk("NFS: %5u nfs_commit_done (status %d)\n",
L
Linus Torvalds 已提交
1370 1371
                                task->tk_pid, task->tk_status);

1372 1373 1374
	/* Call the NFS version-specific code */
	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
		return;
1375 1376 1377 1378 1379 1380 1381
}

static void nfs_commit_release(void *calldata)
{
	struct nfs_write_data	*data = calldata;
	struct nfs_page		*req;
	int status = data->task.tk_status;
1382

L
Linus Torvalds 已提交
1383 1384 1385
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
1386
		nfs_clear_request_commit(req);
L
Linus Torvalds 已提交
1387

1388
		dprintk("NFS:       commit (%s/%lld %d@%lld)",
1389 1390
			req->wb_context->path.dentry->d_inode->i_sb->s_id,
			(long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
L
Linus Torvalds 已提交
1391 1392
			req->wb_bytes,
			(long long)req_offset(req));
1393 1394
		if (status < 0) {
			nfs_context_set_write_error(req->wb_context, status);
L
Linus Torvalds 已提交
1395
			nfs_inode_remove_request(req);
1396
			dprintk(", error = %d\n", status);
L
Linus Torvalds 已提交
1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
			goto next;
		}

		/* Okay, COMMIT succeeded, apparently. Check the verifier
		 * returned by the server against all stored verfs. */
		if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
			/* We have a match */
			nfs_inode_remove_request(req);
			dprintk(" OK\n");
			goto next;
		}
		/* We have a mismatch. Write the page again */
		dprintk(" mismatch\n");
F
Fred 已提交
1410
		nfs_mark_request_dirty(req);
L
Linus Torvalds 已提交
1411
	next:
1412
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1413
	}
1414
	nfs_commit_clear_lock(NFS_I(data->inode));
1415
	nfs_commitdata_release(calldata);
L
Linus Torvalds 已提交
1416
}
1417 1418

static const struct rpc_call_ops nfs_commit_ops = {
1419 1420 1421
#if defined(CONFIG_NFS_V4_1)
	.rpc_call_prepare = nfs_write_prepare,
#endif /* CONFIG_NFS_V4_1 */
1422 1423 1424
	.rpc_call_done = nfs_commit_done,
	.rpc_release = nfs_commit_release,
};
L
Linus Torvalds 已提交
1425

1426
int nfs_commit_inode(struct inode *inode, int how)
L
Linus Torvalds 已提交
1427 1428
{
	LIST_HEAD(head);
1429 1430
	int may_wait = how & FLUSH_SYNC;
	int res = 0;
L
Linus Torvalds 已提交
1431

1432
	if (!nfs_commit_set_lock(NFS_I(inode), may_wait))
1433
		goto out_mark_dirty;
1434
	spin_lock(&inode->i_lock);
1435
	res = nfs_scan_commit(inode, &head, 0, 0);
1436
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
1437
	if (res) {
T
Trond Myklebust 已提交
1438
		int error = nfs_commit_list(inode, &head, how);
1439 1440
		if (error < 0)
			return error;
1441 1442 1443 1444
		if (may_wait)
			wait_on_bit(&NFS_I(inode)->flags, NFS_INO_COMMIT,
					nfs_wait_bit_killable,
					TASK_KILLABLE);
1445 1446
		else
			goto out_mark_dirty;
1447 1448
	} else
		nfs_commit_clear_lock(NFS_I(inode));
1449 1450 1451 1452 1453 1454 1455 1456
	return res;
	/* Note: If we exit without ensuring that the commit is complete,
	 * we must mark the inode as dirty. Otherwise, future calls to
	 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
	 * that the data is on the disk.
	 */
out_mark_dirty:
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
1457 1458
	return res;
}
1459 1460 1461

static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
1462 1463 1464
	struct nfs_inode *nfsi = NFS_I(inode);
	int flags = FLUSH_SYNC;
	int ret = 0;
1465

1466 1467 1468 1469 1470 1471
	if (wbc->sync_mode == WB_SYNC_NONE) {
		/* Don't commit yet if this is a non-blocking flush and there
		 * are a lot of outstanding writes for this mapping.
		 */
		if (nfsi->ncommit <= (nfsi->npages >> 1))
			goto out_mark_dirty;
1472

1473
		/* don't wait for the COMMIT response */
1474
		flags = 0;
1475 1476
	}

1477 1478 1479 1480 1481 1482 1483 1484
	ret = nfs_commit_inode(inode, flags);
	if (ret >= 0) {
		if (wbc->sync_mode == WB_SYNC_NONE) {
			if (ret < wbc->nr_to_write)
				wbc->nr_to_write -= ret;
			else
				wbc->nr_to_write = 0;
		}
1485
		return 0;
1486 1487
	}
out_mark_dirty:
1488 1489 1490
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
	return ret;
}
1491
#else
1492 1493 1494 1495
static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
	return 0;
}
L
Linus Torvalds 已提交
1496 1497
#endif

1498 1499 1500 1501 1502
int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
{
	return nfs_commit_unstable_pages(inode, wbc);
}

1503 1504 1505 1506
/*
 * flush the inode to disk.
 */
int nfs_wb_all(struct inode *inode)
T
Trond Myklebust 已提交
1507 1508
{
	struct writeback_control wbc = {
1509
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1510
		.nr_to_write = LONG_MAX,
1511 1512
		.range_start = 0,
		.range_end = LLONG_MAX,
T
Trond Myklebust 已提交
1513 1514
	};

1515
	return sync_inode(inode, &wbc);
1516 1517
}

1518 1519 1520 1521 1522 1523 1524
int nfs_wb_page_cancel(struct inode *inode, struct page *page)
{
	struct nfs_page *req;
	int ret = 0;

	BUG_ON(!PageLocked(page));
	for (;;) {
1525
		wait_on_page_writeback(page);
1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539
		req = nfs_page_find_request(page);
		if (req == NULL)
			break;
		if (nfs_lock_request_dontget(req)) {
			nfs_inode_remove_request(req);
			/*
			 * In case nfs_inode_remove_request has marked the
			 * page as being dirty
			 */
			cancel_dirty_page(page, PAGE_CACHE_SIZE);
			nfs_unlock_request(req);
			break;
		}
		ret = nfs_wait_on_request(req);
1540
		nfs_release_request(req);
1541
		if (ret < 0)
1542
			break;
1543 1544 1545 1546
	}
	return ret;
}

T
Trond Myklebust 已提交
1547 1548 1549 1550
/*
 * Write back all requests on one page - we do this before reading it.
 */
int nfs_wb_page(struct inode *inode, struct page *page)
1551 1552 1553
{
	loff_t range_start = page_offset(page);
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1554 1555
	struct writeback_control wbc = {
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1556
		.nr_to_write = 0,
1557 1558 1559 1560
		.range_start = range_start,
		.range_end = range_end,
	};
	int ret;
1561

1562
	for (;;) {
1563
		wait_on_page_writeback(page);
1564 1565 1566 1567
		if (clear_page_dirty_for_io(page)) {
			ret = nfs_writepage_locked(page, &wbc);
			if (ret < 0)
				goto out_error;
1568
			continue;
T
Trond Myklebust 已提交
1569
		}
1570 1571 1572
		if (!PagePrivate(page))
			break;
		ret = nfs_commit_inode(inode, FLUSH_SYNC);
1573
		if (ret < 0)
1574
			goto out_error;
T
Trond Myklebust 已提交
1575
	}
1576 1577
	return 0;
out_error:
1578
	return ret;
1579 1580
}

1581 1582 1583 1584 1585 1586 1587
#ifdef CONFIG_MIGRATION
int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
		struct page *page)
{
	struct nfs_page *req;
	int ret;

1588
	nfs_fscache_release_page(page, GFP_KERNEL);
1589

1590
	req = nfs_find_and_lock_request(page, false);
1591 1592 1593 1594 1595 1596 1597 1598 1599 1600
	ret = PTR_ERR(req);
	if (IS_ERR(req))
		goto out;

	ret = migrate_page(mapping, newpage, page);
	if (!req)
		goto out;
	if (ret)
		goto out_unlock;
	page_cache_get(newpage);
T
Trond Myklebust 已提交
1601
	spin_lock(&mapping->host->i_lock);
1602 1603
	req->wb_page = newpage;
	SetPagePrivate(newpage);
T
Trond Myklebust 已提交
1604
	set_page_private(newpage, (unsigned long)req);
1605 1606
	ClearPagePrivate(page);
	set_page_private(page, 0);
T
Trond Myklebust 已提交
1607
	spin_unlock(&mapping->host->i_lock);
1608 1609 1610 1611 1612 1613 1614 1615
	page_cache_release(page);
out_unlock:
	nfs_clear_page_tag_locked(req);
out:
	return ret;
}
#endif

D
David Howells 已提交
1616
int __init nfs_init_writepagecache(void)
L
Linus Torvalds 已提交
1617 1618 1619 1620
{
	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
					     sizeof(struct nfs_write_data),
					     0, SLAB_HWCACHE_ALIGN,
1621
					     NULL);
L
Linus Torvalds 已提交
1622 1623 1624
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

1625 1626
	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
						     nfs_wdata_cachep);
L
Linus Torvalds 已提交
1627 1628 1629
	if (nfs_wdata_mempool == NULL)
		return -ENOMEM;

1630 1631
	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
						      nfs_wdata_cachep);
L
Linus Torvalds 已提交
1632 1633 1634
	if (nfs_commit_mempool == NULL)
		return -ENOMEM;

1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654
	/*
	 * NFS congestion size, scale with available memory.
	 *
	 *  64MB:    8192k
	 * 128MB:   11585k
	 * 256MB:   16384k
	 * 512MB:   23170k
	 *   1GB:   32768k
	 *   2GB:   46340k
	 *   4GB:   65536k
	 *   8GB:   92681k
	 *  16GB:  131072k
	 *
	 * This allows larger machines to have larger/more transfers.
	 * Limit the default to 256M
	 */
	nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
	if (nfs_congestion_kb > 256*1024)
		nfs_congestion_kb = 256*1024;

L
Linus Torvalds 已提交
1655 1656 1657
	return 0;
}

1658
void nfs_destroy_writepagecache(void)
L
Linus Torvalds 已提交
1659 1660 1661
{
	mempool_destroy(nfs_commit_mempool);
	mempool_destroy(nfs_wdata_mempool);
1662
	kmem_cache_destroy(nfs_wdata_cachep);
L
Linus Torvalds 已提交
1663 1664
}