write.c 40.0 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>
L
Linus Torvalds 已提交
16 17 18 19 20

#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs_page.h>
21 22
#include <linux/backing-dev.h>

L
Linus Torvalds 已提交
23 24 25
#include <asm/uaccess.h>

#include "delegation.h"
26
#include "internal.h"
C
Chuck Lever 已提交
27
#include "iostat.h"
L
Linus Torvalds 已提交
28 29 30 31 32 33 34 35 36

#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

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

/*
 * Local function declarations
 */
37 38
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
				  struct inode *inode, int ioflags);
39
static void nfs_redirty_request(struct nfs_page *req);
40 41 42
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
43

44
static struct kmem_cache *nfs_wdata_cachep;
45
static mempool_t *nfs_wdata_mempool;
L
Linus Torvalds 已提交
46 47
static mempool_t *nfs_commit_mempool;

48
struct nfs_write_data *nfs_commitdata_alloc(void)
L
Linus Torvalds 已提交
49
{
C
Christoph Lameter 已提交
50
	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
51

L
Linus Torvalds 已提交
52 53 54
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
A
Andy Adamson 已提交
55
		p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
L
Linus Torvalds 已提交
56 57 58 59
	}
	return p;
}

60
void nfs_commit_free(struct nfs_write_data *p)
L
Linus Torvalds 已提交
61
{
62 63
	if (p && (p->pagevec != &p->page_array[0]))
		kfree(p->pagevec);
L
Linus Torvalds 已提交
64 65 66
	mempool_free(p, nfs_commit_mempool);
}

67
struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
68
{
C
Christoph Lameter 已提交
69
	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
70 71 72 73

	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
74
		p->npages = pagecount;
A
Andy Adamson 已提交
75
		p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
76 77
		if (pagecount <= ARRAY_SIZE(p->page_array))
			p->pagevec = p->page_array;
78
		else {
79 80
			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
			if (!p->pagevec) {
81 82 83 84 85 86 87 88
				mempool_free(p, nfs_wdata_mempool);
				p = NULL;
			}
		}
	}
	return p;
}

89
static void nfs_writedata_free(struct nfs_write_data *p)
90 91 92 93 94 95
{
	if (p && (p->pagevec != &p->page_array[0]))
		kfree(p->pagevec);
	mempool_free(p, nfs_wdata_mempool);
}

96
void nfs_writedata_release(void *data)
L
Linus Torvalds 已提交
97
{
98 99 100
	struct nfs_write_data *wdata = data;

	put_nfs_open_context(wdata->args.context);
L
Linus Torvalds 已提交
101 102 103
	nfs_writedata_free(wdata);
}

104 105 106 107 108 109 110
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);
}

111 112 113 114 115 116 117
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)
118
			kref_get(&req->wb_kref);
119 120 121 122 123 124
	}
	return req;
}

static struct nfs_page *nfs_page_find_request(struct page *page)
{
125
	struct inode *inode = page->mapping->host;
126 127
	struct nfs_page *req = NULL;

128
	spin_lock(&inode->i_lock);
129
	req = nfs_page_find_request_locked(page);
130
	spin_unlock(&inode->i_lock);
131 132 133
	return req;
}

L
Linus Torvalds 已提交
134 135 136 137
/* 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;
138 139
	loff_t end, i_size;
	pgoff_t end_index;
L
Linus Torvalds 已提交
140

141 142 143
	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
144
	if (i_size > 0 && page->index < end_index)
145
		goto out;
L
Linus Torvalds 已提交
146 147
	end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
	if (i_size >= end)
148
		goto out;
L
Linus Torvalds 已提交
149
	i_size_write(inode, end);
150 151 152
	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
out:
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
153 154
}

155 156 157 158 159 160 161
/* 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
162 163 164 165 166 167 168 169 170
/* 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;
171
	if (count != nfs_page_length(page))
L
Linus Torvalds 已提交
172
		return;
173
	SetPageUptodate(page);
L
Linus Torvalds 已提交
174 175 176 177 178
}

static int wb_priority(struct writeback_control *wbc)
{
	if (wbc->for_reclaim)
179
		return FLUSH_HIGHPRI | FLUSH_STABLE;
L
Linus Torvalds 已提交
180 181 182 183 184
	if (wbc->for_kupdate)
		return FLUSH_LOWPRI;
	return 0;
}

185 186 187 188 189 190 191 192 193 194
/*
 * 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))

195
static int nfs_set_page_writeback(struct page *page)
196
{
197 198 199
	int ret = test_set_page_writeback(page);

	if (!ret) {
200 201 202
		struct inode *inode = page->mapping->host;
		struct nfs_server *nfss = NFS_SERVER(inode);

203
		if (atomic_long_inc_return(&nfss->writeback) >
204 205 206
				NFS_CONGESTION_ON_THRESH)
			set_bdi_congested(&nfss->backing_dev_info, WRITE);
	}
207
	return ret;
208 209 210 211 212 213 214 215
}

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);
P
Peter Zijlstra 已提交
216
	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
217 218 219
		clear_bdi_congested(&nfss->backing_dev_info, WRITE);
}

220 221
/*
 * Find an associated nfs write request, and prepare to flush it out
T
Trond Myklebust 已提交
222
 * May return an error if the user signalled nfs_wait_on_request().
223
 */
224 225
static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
				struct page *page)
226
{
227
	struct inode *inode = page->mapping->host;
228 229 230
	struct nfs_page *req;
	int ret;

231
	spin_lock(&inode->i_lock);
232 233 234
	for(;;) {
		req = nfs_page_find_request_locked(page);
		if (req == NULL) {
235
			spin_unlock(&inode->i_lock);
T
Trond Myklebust 已提交
236
			return 0;
237
		}
238
		if (nfs_set_page_tag_locked(req))
239 240
			break;
		/* Note: If we hold the page lock, as is the case in nfs_writepage,
241
		 *	 then the call to nfs_set_page_tag_locked() will always
242 243 244
		 *	 succeed provided that someone hasn't already marked the
		 *	 request as dirty (in which case we don't care).
		 */
245
		spin_unlock(&inode->i_lock);
246 247 248 249
		ret = nfs_wait_on_request(req);
		nfs_release_request(req);
		if (ret != 0)
			return ret;
250
		spin_lock(&inode->i_lock);
251
	}
252
	if (test_bit(PG_CLEAN, &req->wb_flags)) {
253
		spin_unlock(&inode->i_lock);
254
		BUG();
255
	}
256
	if (nfs_set_page_writeback(page) != 0) {
257
		spin_unlock(&inode->i_lock);
258 259
		BUG();
	}
260
	spin_unlock(&inode->i_lock);
261 262 263 264
	if (!nfs_pageio_add_request(pgio, req)) {
		nfs_redirty_request(req);
		return pgio->pg_error;
	}
T
Trond Myklebust 已提交
265
	return 0;
266 267
}

T
Trond Myklebust 已提交
268
static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
L
Linus Torvalds 已提交
269 270 271
{
	struct inode *inode = page->mapping->host;

C
Chuck Lever 已提交
272 273 274
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);

275
	nfs_pageio_cond_complete(pgio, page->index);
T
Trond Myklebust 已提交
276 277
	return nfs_page_async_flush(pgio, page);
}
278

T
Trond Myklebust 已提交
279 280 281 282 283 284 285
/*
 * 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;
286

T
Trond Myklebust 已提交
287 288 289 290 291 292 293 294
	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;
295 296 297 298
}

int nfs_writepage(struct page *page, struct writeback_control *wbc)
{
T
Trond Myklebust 已提交
299
	int ret;
300

T
Trond Myklebust 已提交
301
	ret = nfs_writepage_locked(page, wbc);
L
Linus Torvalds 已提交
302
	unlock_page(page);
T
Trond Myklebust 已提交
303 304 305 306 307 308 309 310 311 312
	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
313 314 315 316 317
}

int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
318
	unsigned long *bitlock = &NFS_I(inode)->flags;
319
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
320 321
	int err;

322 323 324 325 326 327
	/* 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
328 329
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);

330
	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
T
Trond Myklebust 已提交
331
	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
332
	nfs_pageio_complete(&pgio);
333 334 335 336 337

	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
338
	if (err < 0)
339 340 341 342
		goto out_err;
	err = pgio.pg_error;
	if (err < 0)
		goto out_err;
343
	return 0;
344 345
out_err:
	return err;
L
Linus Torvalds 已提交
346 347 348 349 350
}

/*
 * Insert a write request into an inode
 */
351
static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
L
Linus Torvalds 已提交
352 353 354 355
{
	struct nfs_inode *nfsi = NFS_I(inode);
	int error;

356 357 358 359 360 361 362 363
	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
364
	error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
365
	BUG_ON(error);
L
Linus Torvalds 已提交
366 367 368 369 370
	if (!nfsi->npages) {
		igrab(inode);
		if (nfs_have_delegation(inode, FMODE_WRITE))
			nfsi->change_attr++;
	}
371
	SetPagePrivate(req->wb_page);
372
	set_page_private(req->wb_page, (unsigned long)req);
L
Linus Torvalds 已提交
373
	nfsi->npages++;
374
	kref_get(&req->wb_kref);
375 376
	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
				NFS_PAGE_TAG_LOCKED);
377 378 379 380
	spin_unlock(&inode->i_lock);
	radix_tree_preload_end();
out:
	return error;
L
Linus Torvalds 已提交
381 382 383
}

/*
384
 * Remove a write request from an inode
L
Linus Torvalds 已提交
385 386 387
 */
static void nfs_inode_remove_request(struct nfs_page *req)
{
388
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
389 390 391 392
	struct nfs_inode *nfsi = NFS_I(inode);

	BUG_ON (!NFS_WBACK_BUSY(req));

393
	spin_lock(&inode->i_lock);
394
	set_page_private(req->wb_page, 0);
395
	ClearPagePrivate(req->wb_page);
L
Linus Torvalds 已提交
396 397 398
	radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
	nfsi->npages--;
	if (!nfsi->npages) {
399
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
400 401
		iput(inode);
	} else
402
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
403 404 405 406
	nfs_clear_request(req);
	nfs_release_request(req);
}

407
static void
F
Fred 已提交
408
nfs_mark_request_dirty(struct nfs_page *req)
409 410 411 412
{
	__set_page_dirty_nobuffers(req->wb_page);
}

L
Linus Torvalds 已提交
413 414 415 416 417 418 419
#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)
{
420
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
421 422
	struct nfs_inode *nfsi = NFS_I(inode);

423
	spin_lock(&inode->i_lock);
424
	set_bit(PG_CLEAN, &(req)->wb_flags);
425 426 427
	radix_tree_tag_set(&nfsi->nfs_page_tree,
			req->wb_index,
			NFS_PAGE_TAG_COMMIT);
428
	spin_unlock(&inode->i_lock);
429
	inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
430
	inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
431
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
432
}
433

434 435 436 437 438 439 440 441 442 443 444 445 446
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;
}

447 448 449 450 451 452 453 454 455
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)
{
456
	if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
457 458 459 460
		nfs_mark_request_commit(req);
		return 1;
	}
	if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
F
Fred 已提交
461
		nfs_mark_request_dirty(req);
462 463 464 465 466 467 468 469 470 471
		return 1;
	}
	return 0;
}
#else
static inline void
nfs_mark_request_commit(struct nfs_page *req)
{
}

472 473 474 475 476 477
static inline int
nfs_clear_request_commit(struct nfs_page *req)
{
	return 0;
}

478 479 480 481 482 483 484 485 486 487 488
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
489 490 491 492 493
#endif

/*
 * Wait for a request to complete.
 *
494
 * Interruptible by fatal signals only.
L
Linus Torvalds 已提交
495
 */
496
static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
L
Linus Torvalds 已提交
497 498 499
{
	struct nfs_inode *nfsi = NFS_I(inode);
	struct nfs_page *req;
500
	pgoff_t idx_end, next;
L
Linus Torvalds 已提交
501 502 503 504 505 506 507 508 509
	unsigned int		res = 0;
	int			error;

	if (npages == 0)
		idx_end = ~0;
	else
		idx_end = idx_start + npages - 1;

	next = idx_start;
510
	while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) {
L
Linus Torvalds 已提交
511 512 513 514
		if (req->wb_index > idx_end)
			break;

		next = req->wb_index + 1;
515
		BUG_ON(!NFS_WBACK_BUSY(req));
L
Linus Torvalds 已提交
516

517
		kref_get(&req->wb_kref);
518
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
519 520
		error = nfs_wait_on_request(req);
		nfs_release_request(req);
521
		spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
522 523 524 525 526 527 528
		if (error < 0)
			return error;
		res++;
	}
	return res;
}

T
Trond Myklebust 已提交
529 530 531 532 533 534 535
static void nfs_cancel_commit_list(struct list_head *head)
{
	struct nfs_page *req;

	while(!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
536
		nfs_clear_request_commit(req);
T
Trond Myklebust 已提交
537
		nfs_inode_remove_request(req);
538
		nfs_unlock_request(req);
T
Trond Myklebust 已提交
539 540 541
	}
}

T
Trond Myklebust 已提交
542
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
543 544 545 546 547 548
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
549 550 551 552 553 554 555 556 557 558 559
/*
 * 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
560
nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
L
Linus Torvalds 已提交
561 562
{
	struct nfs_inode *nfsi = NFS_I(inode);
563

564 565 566 567
	if (!nfs_need_commit(nfsi))
		return 0;

	return nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
L
Linus Torvalds 已提交
568
}
569
#else
570 571 572 573 574
static inline int nfs_need_commit(struct nfs_inode *nfsi)
{
	return 0;
}

575
static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
576 577 578
{
	return 0;
}
L
Linus Torvalds 已提交
579 580 581
#endif

/*
582 583
 * 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
584
 *
585 586
 * If the attempt fails, then the existing request is flushed out
 * to disk.
L
Linus Torvalds 已提交
587
 */
588 589 590 591
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
592
{
593 594 595 596 597 598 599
	struct nfs_page *req;
	unsigned int rqend;
	unsigned int end;
	int error;

	if (!PagePrivate(page))
		return NULL;
L
Linus Torvalds 已提交
600 601

	end = offset + bytes;
602
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
603 604

	for (;;) {
605
		req = nfs_page_find_request_locked(page);
606 607 608 609 610 611 612 613 614 615
		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.
		 */
616
		if (offset > rqend
617 618 619 620
		    || end < req->wb_offset)
			goto out_flushme;

		if (nfs_set_page_tag_locked(req))
L
Linus Torvalds 已提交
621 622
			break;

623
		/* The request is locked, so wait and then retry */
624
		spin_unlock(&inode->i_lock);
625 626 627 628 629
		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
630 631
	}

632 633 634 635
	if (nfs_clear_request_commit(req))
		radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
				req->wb_index, NFS_PAGE_TAG_COMMIT);

L
Linus Torvalds 已提交
636 637 638 639 640 641 642
	/* 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;
643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
	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
669

670 671 672 673 674 675 676 677 678 679 680
	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);
	}
681
out:
T
Trond Myklebust 已提交
682
	return req;
L
Linus Torvalds 已提交
683 684
}

685 686 687 688 689 690 691 692 693 694 695 696 697 698 699
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);
	/* Update file length */
	nfs_grow_file(page, offset, count);
	nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
	nfs_clear_page_tag_locked(req);
	return 0;
}

L
Linus Torvalds 已提交
700 701
int nfs_flush_incompatible(struct file *file, struct page *page)
{
702
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
703
	struct nfs_page	*req;
T
Trond Myklebust 已提交
704
	int do_flush, status;
L
Linus Torvalds 已提交
705 706 707 708 709 710 711 712
	/*
	 * 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
713 714 715 716
	do {
		req = nfs_page_find_request(page);
		if (req == NULL)
			return 0;
717
		do_flush = req->wb_page != page || req->wb_context != ctx;
L
Linus Torvalds 已提交
718
		nfs_release_request(req);
T
Trond Myklebust 已提交
719 720 721 722 723
		if (!do_flush)
			return 0;
		status = nfs_wb_page(page->mapping->host, page);
	} while (status == 0);
	return status;
L
Linus Torvalds 已提交
724 725
}

726 727 728 729 730 731 732 733 734 735 736
/*
 * 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
737 738 739 740 741 742 743 744 745
/*
 * 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)
{
746
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
747 748 749
	struct inode	*inode = page->mapping->host;
	int		status = 0;

C
Chuck Lever 已提交
750 751
	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);

752
	dprintk("NFS:       nfs_updatepage(%s/%s %d@%lld)\n",
753 754
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name, count,
755
		(long long)(page_offset(page) + offset));
L
Linus Torvalds 已提交
756 757

	/* If we're not using byte range locks, and we know the page
758 759 760
	 * 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
761
	 */
762 763
	if (nfs_write_pageuptodate(page, inode) &&
			inode->i_flock == NULL &&
764
			!(file->f_flags & O_SYNC)) {
765
		count = max(count + offset, nfs_page_length(page));
L
Linus Torvalds 已提交
766 767 768
		offset = 0;
	}

769
	status = nfs_writepage_setup(ctx, page, offset, count);
770 771 772 773
	if (status < 0)
		nfs_set_pageerror(page);
	else
		__set_page_dirty_nobuffers(page);
L
Linus Torvalds 已提交
774

775
	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
L
Linus Torvalds 已提交
776 777 778 779 780 781 782
			status, (long long)i_size_read(inode));
	return status;
}

static void nfs_writepage_release(struct nfs_page *req)
{

T
Trond Myklebust 已提交
783
	if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) {
784 785 786 787
		nfs_end_page_writeback(req->wb_page);
		nfs_inode_remove_request(req);
	} else
		nfs_end_page_writeback(req->wb_page);
788
	nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
789 790
}

791
static int flush_task_priority(int how)
L
Linus Torvalds 已提交
792 793 794 795 796 797 798 799 800 801 802 803 804
{
	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
		case FLUSH_HIGHPRI:
			return RPC_PRIORITY_HIGH;
		case FLUSH_LOWPRI:
			return RPC_PRIORITY_LOW;
	}
	return RPC_PRIORITY_NORMAL;
}

/*
 * Set up the argument/result storage required for the RPC call.
 */
805
static int nfs_write_rpcsetup(struct nfs_page *req,
L
Linus Torvalds 已提交
806
		struct nfs_write_data *data,
807
		const struct rpc_call_ops *call_ops,
L
Linus Torvalds 已提交
808 809 810
		unsigned int count, unsigned int offset,
		int how)
{
811 812
	struct inode *inode = req->wb_context->path.dentry->d_inode;
	int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
813
	int priority = flush_task_priority(how);
814
	struct rpc_task *task;
815 816 817 818 819
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = req->wb_context->cred,
	};
820 821
	struct rpc_task_setup task_setup_data = {
		.rpc_client = NFS_CLIENT(inode),
822
		.task = &data->task,
823
		.rpc_message = &msg,
824 825
		.callback_ops = call_ops,
		.callback_data = data,
826
		.workqueue = nfsiod_workqueue,
827
		.flags = flags,
828
		.priority = priority,
829
	};
L
Linus Torvalds 已提交
830 831 832 833 834

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

	data->req = req;
835
	data->inode = inode = req->wb_context->path.dentry->d_inode;
836
	data->cred = msg.rpc_cred;
L
Linus Torvalds 已提交
837 838 839 840 841 842

	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;
843
	data->args.context = get_nfs_open_context(req->wb_context);
844 845 846
	data->args.stable  = NFS_UNSTABLE;
	if (how & FLUSH_STABLE) {
		data->args.stable = NFS_DATA_SYNC;
847
		if (!nfs_need_commit(NFS_I(inode)))
848 849
			data->args.stable = NFS_FILE_SYNC;
	}
L
Linus Torvalds 已提交
850 851 852 853

	data->res.fattr   = &data->fattr;
	data->res.count   = count;
	data->res.verf    = &data->verf;
854
	nfs_fattr_init(&data->fattr);
L
Linus Torvalds 已提交
855

856
	/* Set up the initial task struct.  */
857
	NFS_PROTO(inode)->write_setup(data, &msg);
L
Linus Torvalds 已提交
858

C
Chuck Lever 已提交
859
	dprintk("NFS: %5u initiated write call "
860
		"(req %s/%lld, %u bytes @ offset %llu)\n",
861
		data->task.tk_pid,
L
Linus Torvalds 已提交
862 863 864 865 866
		inode->i_sb->s_id,
		(long long)NFS_FILEID(inode),
		count,
		(unsigned long long)data->args.offset);

867
	task = rpc_run_task(&task_setup_data);
868 869 870 871
	if (IS_ERR(task))
		return PTR_ERR(task);
	rpc_put_task(task);
	return 0;
L
Linus Torvalds 已提交
872 873
}

F
Fred 已提交
874 875 876 877 878 879 880 881 882 883 884
/* 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)
{
	nfs_mark_request_dirty(req);
	nfs_end_page_writeback(req->wb_page);
	nfs_clear_page_tag_locked(req);
}

L
Linus Torvalds 已提交
885 886 887 888
/*
 * Generate multiple small requests to write out a single
 * contiguous dirty area on one page.
 */
889
static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
L
Linus Torvalds 已提交
890 891 892 893
{
	struct nfs_page *req = nfs_list_entry(head->next);
	struct page *page = req->wb_page;
	struct nfs_write_data *data;
894 895
	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
	unsigned int offset;
L
Linus Torvalds 已提交
896
	int requests = 0;
897
	int ret = 0;
L
Linus Torvalds 已提交
898 899 900 901
	LIST_HEAD(list);

	nfs_list_remove_request(req);

902
	nbytes = count;
903 904 905
	do {
		size_t len = min(nbytes, wsize);

906
		data = nfs_writedata_alloc(1);
L
Linus Torvalds 已提交
907 908 909 910
		if (!data)
			goto out_bad;
		list_add(&data->pages, &list);
		requests++;
911 912
		nbytes -= len;
	} while (nbytes != 0);
L
Linus Torvalds 已提交
913 914 915 916
	atomic_set(&req->wb_complete, requests);

	ClearPageError(page);
	offset = 0;
917
	nbytes = count;
L
Linus Torvalds 已提交
918
	do {
919 920
		int ret2;

L
Linus Torvalds 已提交
921 922 923 924 925
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del_init(&data->pages);

		data->pagevec[0] = page;

926 927
		if (nbytes < wsize)
			wsize = nbytes;
928
		ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
929
				   wsize, offset, how);
930 931
		if (ret == 0)
			ret = ret2;
932 933
		offset += wsize;
		nbytes -= wsize;
L
Linus Torvalds 已提交
934 935
	} while (nbytes != 0);

936
	return ret;
L
Linus Torvalds 已提交
937 938 939 940 941

out_bad:
	while (!list_empty(&list)) {
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del(&data->pages);
942
		nfs_writedata_release(data);
L
Linus Torvalds 已提交
943
	}
944
	nfs_redirty_request(req);
L
Linus Torvalds 已提交
945 946 947 948 949 950 951 952 953 954 955
	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.
 */
956
static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
L
Linus Torvalds 已提交
957 958 959 960 961
{
	struct nfs_page		*req;
	struct page		**pages;
	struct nfs_write_data	*data;

962
	data = nfs_writedata_alloc(npages);
L
Linus Torvalds 已提交
963 964 965 966 967 968 969 970 971 972 973 974 975 976
	if (!data)
		goto out_bad;

	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);

	/* Set up the argument struct */
977
	return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
L
Linus Torvalds 已提交
978 979
 out_bad:
	while (!list_empty(head)) {
980
		req = nfs_list_entry(head->next);
L
Linus Torvalds 已提交
981
		nfs_list_remove_request(req);
982
		nfs_redirty_request(req);
L
Linus Torvalds 已提交
983 984 985 986
	}
	return -ENOMEM;
}

987 988
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
				  struct inode *inode, int ioflags)
L
Linus Torvalds 已提交
989
{
990
	size_t wsize = NFS_SERVER(inode)->wsize;
L
Linus Torvalds 已提交
991

992
	if (wsize < PAGE_CACHE_SIZE)
993
		nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
994
	else
995
		nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
L
Linus Torvalds 已提交
996 997 998 999 1000
}

/*
 * Handle a write reply that flushed part of a page.
 */
1001
static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1002
{
1003
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1004

1005 1006 1007 1008 1009 1010
	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
1011

1012 1013
	nfs_writeback_done(task, data);
}
1014

1015 1016 1017 1018 1019 1020 1021 1022
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) {
1023
		nfs_set_pageerror(page);
1024 1025
		nfs_context_set_write_error(req->wb_context, status);
		dprintk(", error = %d\n", status);
1026
		goto out;
L
Linus Torvalds 已提交
1027 1028
	}

1029
	if (nfs_write_need_commit(data)) {
1030
		struct inode *inode = page->mapping->host;
1031

1032
		spin_lock(&inode->i_lock);
1033 1034 1035 1036 1037 1038 1039 1040 1041 1042
		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");
		}
1043
		spin_unlock(&inode->i_lock);
1044 1045 1046 1047
	} else
		dprintk(" OK\n");

out:
L
Linus Torvalds 已提交
1048 1049
	if (atomic_dec_and_test(&req->wb_complete))
		nfs_writepage_release(req);
1050
	nfs_writedata_release(calldata);
L
Linus Torvalds 已提交
1051 1052
}

1053 1054
static const struct rpc_call_ops nfs_write_partial_ops = {
	.rpc_call_done = nfs_writeback_done_partial,
1055
	.rpc_release = nfs_writeback_release_partial,
1056 1057
};

L
Linus Torvalds 已提交
1058 1059 1060 1061 1062 1063 1064
/*
 * 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.
 */
1065
static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1066
{
1067
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1068

1069 1070 1071 1072 1073 1074 1075
	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;
1076

L
Linus Torvalds 已提交
1077 1078
	/* Update attributes as result of writeback. */
	while (!list_empty(&data->pages)) {
1079 1080 1081
		struct nfs_page *req = nfs_list_entry(data->pages.next);
		struct page *page = req->wb_page;

L
Linus Torvalds 已提交
1082 1083
		nfs_list_remove_request(req);

1084 1085
		dprintk("NFS: %5u write (%s/%lld %d@%lld)",
			data->task.tk_pid,
1086 1087
			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
1088 1089 1090
			req->wb_bytes,
			(long long)req_offset(req));

1091
		if (status < 0) {
1092
			nfs_set_pageerror(page);
1093 1094
			nfs_context_set_write_error(req->wb_context, status);
			dprintk(", error = %d\n", status);
1095
			goto remove_request;
L
Linus Torvalds 已提交
1096 1097
		}

1098 1099 1100 1101 1102
		if (nfs_write_need_commit(data)) {
			memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
			nfs_mark_request_commit(req);
			nfs_end_page_writeback(page);
			dprintk(" marked for commit\n");
L
Linus Torvalds 已提交
1103 1104
			goto next;
		}
1105 1106 1107
		dprintk(" OK\n");
remove_request:
		nfs_end_page_writeback(page);
L
Linus Torvalds 已提交
1108 1109
		nfs_inode_remove_request(req);
	next:
1110
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1111
	}
1112
	nfs_writedata_release(calldata);
L
Linus Torvalds 已提交
1113 1114
}

1115 1116
static const struct rpc_call_ops nfs_write_full_ops = {
	.rpc_call_done = nfs_writeback_done_full,
1117
	.rpc_release = nfs_writeback_release_full,
1118 1119 1120
};


L
Linus Torvalds 已提交
1121 1122 1123
/*
 * This function is called when the WRITE call is complete.
 */
1124
int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
L
Linus Torvalds 已提交
1125 1126 1127
{
	struct nfs_writeargs	*argp = &data->args;
	struct nfs_writeres	*resp = &data->res;
1128
	int status;
L
Linus Torvalds 已提交
1129

C
Chuck Lever 已提交
1130
	dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
L
Linus Torvalds 已提交
1131 1132
		task->tk_pid, task->tk_status);

1133 1134 1135 1136 1137 1138 1139
	/*
	 * ->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.
	 */
1140 1141 1142
	status = NFS_PROTO(data->inode)->write_done(task, data);
	if (status != 0)
		return status;
C
Chuck Lever 已提交
1143 1144
	nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);

L
Linus Torvalds 已提交
1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
#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;

		if (time_before(complain, jiffies)) {
1158
			dprintk("NFS:       faulty NFS server %s:"
L
Linus Torvalds 已提交
1159
				" (committed = %d) != (stable = %d)\n",
1160
				NFS_SERVER(data->inode)->nfs_client->cl_hostname,
L
Linus Torvalds 已提交
1161 1162 1163 1164 1165 1166 1167 1168 1169
				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
1170 1171
		nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);

L
Linus Torvalds 已提交
1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186
		/* 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 */
				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;
			}
			rpc_restart_call(task);
1187
			return -EAGAIN;
L
Linus Torvalds 已提交
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
		}
		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;
	}
1198
	return 0;
L
Linus Torvalds 已提交
1199 1200 1201 1202
}


#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1203
void nfs_commitdata_release(void *data)
L
Linus Torvalds 已提交
1204
{
1205 1206 1207
	struct nfs_write_data *wdata = data;

	put_nfs_open_context(wdata->args.context);
L
Linus Torvalds 已提交
1208 1209 1210 1211 1212 1213
	nfs_commit_free(wdata);
}

/*
 * Set up the argument/result storage required for the RPC call.
 */
1214
static int nfs_commit_rpcsetup(struct list_head *head,
1215 1216
		struct nfs_write_data *data,
		int how)
L
Linus Torvalds 已提交
1217
{
1218 1219 1220
	struct nfs_page *first = nfs_list_entry(head->next);
	struct inode *inode = first->wb_context->path.dentry->d_inode;
	int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
1221
	int priority = flush_task_priority(how);
1222
	struct rpc_task *task;
1223 1224 1225 1226 1227
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = first->wb_context->cred,
	};
1228
	struct rpc_task_setup task_setup_data = {
1229
		.task = &data->task,
1230
		.rpc_client = NFS_CLIENT(inode),
1231
		.rpc_message = &msg,
1232 1233
		.callback_ops = &nfs_commit_ops,
		.callback_data = data,
1234
		.workqueue = nfsiod_workqueue,
1235
		.flags = flags,
1236
		.priority = priority,
1237
	};
L
Linus Torvalds 已提交
1238 1239 1240 1241 1242 1243 1244

	/* 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;
1245
	data->cred	  = msg.rpc_cred;
L
Linus Torvalds 已提交
1246 1247

	data->args.fh     = NFS_FH(data->inode);
1248 1249 1250
	/* Note: we always request a commit of the entire inode */
	data->args.offset = 0;
	data->args.count  = 0;
1251
	data->args.context = get_nfs_open_context(first->wb_context);
1252
	data->res.count   = 0;
L
Linus Torvalds 已提交
1253 1254
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
1255
	nfs_fattr_init(&data->fattr);
1256 1257

	/* Set up the initial task struct.  */
1258
	NFS_PROTO(inode)->commit_setup(data, &msg);
L
Linus Torvalds 已提交
1259

C
Chuck Lever 已提交
1260
	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1261

1262
	task = rpc_run_task(&task_setup_data);
1263 1264 1265 1266
	if (IS_ERR(task))
		return PTR_ERR(task);
	rpc_put_task(task);
	return 0;
L
Linus Torvalds 已提交
1267 1268 1269 1270 1271 1272
}

/*
 * Commit dirty pages
 */
static int
1273
nfs_commit_list(struct inode *inode, struct list_head *head, int how)
L
Linus Torvalds 已提交
1274 1275 1276 1277
{
	struct nfs_write_data	*data;
	struct nfs_page         *req;

1278
	data = nfs_commitdata_alloc();
L
Linus Torvalds 已提交
1279 1280 1281 1282 1283

	if (!data)
		goto out_bad;

	/* Set up the argument struct */
1284
	return nfs_commit_rpcsetup(head, data, how);
L
Linus Torvalds 已提交
1285 1286 1287 1288 1289
 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
1290
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1291 1292
		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
				BDI_RECLAIMABLE);
1293
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1294 1295 1296 1297 1298 1299 1300
	}
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
1301
static void nfs_commit_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1302
{
1303
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1304

C
Chuck Lever 已提交
1305
        dprintk("NFS: %5u nfs_commit_done (status %d)\n",
L
Linus Torvalds 已提交
1306 1307
                                task->tk_pid, task->tk_status);

1308 1309 1310
	/* Call the NFS version-specific code */
	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
		return;
1311 1312 1313 1314 1315 1316 1317
}

static void nfs_commit_release(void *calldata)
{
	struct nfs_write_data	*data = calldata;
	struct nfs_page		*req;
	int status = data->task.tk_status;
1318

L
Linus Torvalds 已提交
1319 1320 1321
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
1322
		nfs_clear_request_commit(req);
L
Linus Torvalds 已提交
1323

1324
		dprintk("NFS:       commit (%s/%lld %d@%lld)",
1325 1326
			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
1327 1328
			req->wb_bytes,
			(long long)req_offset(req));
1329 1330
		if (status < 0) {
			nfs_context_set_write_error(req->wb_context, status);
L
Linus Torvalds 已提交
1331
			nfs_inode_remove_request(req);
1332
			dprintk(", error = %d\n", status);
L
Linus Torvalds 已提交
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345
			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
1346
		nfs_mark_request_dirty(req);
L
Linus Torvalds 已提交
1347
	next:
1348
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1349
	}
1350
	nfs_commitdata_release(calldata);
L
Linus Torvalds 已提交
1351
}
1352 1353 1354 1355 1356

static const struct rpc_call_ops nfs_commit_ops = {
	.rpc_call_done = nfs_commit_done,
	.rpc_release = nfs_commit_release,
};
L
Linus Torvalds 已提交
1357

1358
int nfs_commit_inode(struct inode *inode, int how)
L
Linus Torvalds 已提交
1359 1360
{
	LIST_HEAD(head);
T
Trond Myklebust 已提交
1361
	int res;
L
Linus Torvalds 已提交
1362

1363
	spin_lock(&inode->i_lock);
1364
	res = nfs_scan_commit(inode, &head, 0, 0);
1365
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
1366
	if (res) {
T
Trond Myklebust 已提交
1367
		int error = nfs_commit_list(inode, &head, how);
1368 1369 1370
		if (error < 0)
			return error;
	}
L
Linus Torvalds 已提交
1371 1372
	return res;
}
1373 1374 1375 1376 1377
#else
static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
{
	return 0;
}
L
Linus Torvalds 已提交
1378 1379
#endif

1380
long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
L
Linus Torvalds 已提交
1381
{
1382
	struct inode *inode = mapping->host;
1383
	pgoff_t idx_start, idx_end;
1384
	unsigned int npages = 0;
1385
	LIST_HEAD(head);
1386
	int nocommit = how & FLUSH_NOCOMMIT;
T
Trond Myklebust 已提交
1387
	long pages, ret;
L
Linus Torvalds 已提交
1388

1389 1390 1391 1392 1393 1394 1395
	/* FIXME */
	if (wbc->range_cyclic)
		idx_start = 0;
	else {
		idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
		idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
		if (idx_end > idx_start) {
1396
			pgoff_t l_npages = 1 + idx_end - idx_start;
1397 1398
			npages = l_npages;
			if (sizeof(npages) != sizeof(l_npages) &&
1399
					(pgoff_t)npages != l_npages)
1400 1401 1402
				npages = 0;
		}
	}
1403
	how &= ~FLUSH_NOCOMMIT;
1404
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
1405
	do {
1406 1407
		ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
		if (ret != 0)
1408
			continue;
1409 1410
		if (nocommit)
			break;
1411
		pages = nfs_scan_commit(inode, &head, idx_start, npages);
1412
		if (pages == 0)
1413
			break;
1414
		if (how & FLUSH_INVALIDATE) {
1415
			spin_unlock(&inode->i_lock);
T
Trond Myklebust 已提交
1416
			nfs_cancel_commit_list(&head);
1417
			ret = pages;
1418
			spin_lock(&inode->i_lock);
1419 1420 1421
			continue;
		}
		pages += nfs_scan_commit(inode, &head, 0, 0);
1422
		spin_unlock(&inode->i_lock);
1423
		ret = nfs_commit_list(inode, &head, how);
1424 1425
		spin_lock(&inode->i_lock);

1426
	} while (ret >= 0);
1427
	spin_unlock(&inode->i_lock);
1428
	return ret;
L
Linus Torvalds 已提交
1429 1430
}

T
Trond Myklebust 已提交
1431
static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
1432 1433 1434
{
	int ret;

T
Trond Myklebust 已提交
1435
	ret = nfs_writepages(mapping, wbc);
1436 1437
	if (ret < 0)
		goto out;
T
Trond Myklebust 已提交
1438
	ret = nfs_sync_mapping_wait(mapping, wbc, how);
1439 1440 1441
	if (ret < 0)
		goto out;
	return 0;
1442
out:
1443
	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1444 1445 1446
	return ret;
}

T
Trond Myklebust 已提交
1447 1448 1449 1450 1451
/* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */
static int nfs_write_mapping(struct address_space *mapping, int how)
{
	struct writeback_control wbc = {
		.bdi = mapping->backing_dev_info,
1452
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1453
		.nr_to_write = LONG_MAX,
1454 1455
		.range_start = 0,
		.range_end = LLONG_MAX,
T
Trond Myklebust 已提交
1456 1457 1458 1459 1460 1461
		.for_writepages = 1,
	};

	return __nfs_write_mapping(mapping, &wbc, how);
}

1462 1463 1464 1465
/*
 * flush the inode to disk.
 */
int nfs_wb_all(struct inode *inode)
1466
{
1467 1468
	return nfs_write_mapping(inode->i_mapping, 0);
}
1469

1470 1471 1472
int nfs_wb_nocommit(struct inode *inode)
{
	return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT);
1473 1474
}

1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493
int nfs_wb_page_cancel(struct inode *inode, struct page *page)
{
	struct nfs_page *req;
	loff_t range_start = page_offset(page);
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
	struct writeback_control wbc = {
		.bdi = page->mapping->backing_dev_info,
		.sync_mode = WB_SYNC_ALL,
		.nr_to_write = LONG_MAX,
		.range_start = range_start,
		.range_end = range_end,
	};
	int ret = 0;

	BUG_ON(!PageLocked(page));
	for (;;) {
		req = nfs_page_find_request(page);
		if (req == NULL)
			goto out;
1494
		if (test_bit(PG_CLEAN, &req->wb_flags)) {
1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518
			nfs_release_request(req);
			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);
		if (ret < 0)
			goto out;
	}
	if (!PagePrivate(page))
		return 0;
	ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE);
out:
	return ret;
}

1519 1520
static int nfs_wb_page_priority(struct inode *inode, struct page *page,
				int how)
1521 1522 1523
{
	loff_t range_start = page_offset(page);
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1524 1525 1526 1527 1528 1529 1530 1531
	struct writeback_control wbc = {
		.bdi = page->mapping->backing_dev_info,
		.sync_mode = WB_SYNC_ALL,
		.nr_to_write = LONG_MAX,
		.range_start = range_start,
		.range_end = range_end,
	};
	int ret;
1532

1533 1534 1535 1536 1537 1538 1539 1540
	do {
		if (clear_page_dirty_for_io(page)) {
			ret = nfs_writepage_locked(page, &wbc);
			if (ret < 0)
				goto out_error;
		} else if (!PagePrivate(page))
			break;
		ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
1541
		if (ret < 0)
1542 1543 1544 1545
			goto out_error;
	} while (PagePrivate(page));
	return 0;
out_error:
1546
	__mark_inode_dirty(inode, I_DIRTY_PAGES);
1547
	return ret;
1548 1549 1550 1551 1552 1553 1554
}

/*
 * Write back all requests on one page - we do this before reading it.
 */
int nfs_wb_page(struct inode *inode, struct page* page)
{
1555
	return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
1556 1557
}

D
David Howells 已提交
1558
int __init nfs_init_writepagecache(void)
L
Linus Torvalds 已提交
1559 1560 1561 1562
{
	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
					     sizeof(struct nfs_write_data),
					     0, SLAB_HWCACHE_ALIGN,
1563
					     NULL);
L
Linus Torvalds 已提交
1564 1565 1566
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

1567 1568
	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
						     nfs_wdata_cachep);
L
Linus Torvalds 已提交
1569 1570 1571
	if (nfs_wdata_mempool == NULL)
		return -ENOMEM;

1572 1573
	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
						      nfs_wdata_cachep);
L
Linus Torvalds 已提交
1574 1575 1576
	if (nfs_commit_mempool == NULL)
		return -ENOMEM;

1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596
	/*
	 * 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
1597 1598 1599
	return 0;
}

1600
void nfs_destroy_writepagecache(void)
L
Linus Torvalds 已提交
1601 1602 1603
{
	mempool_destroy(nfs_commit_mempool);
	mempool_destroy(nfs_wdata_mempool);
1604
	kmem_cache_destroy(nfs_wdata_cachep);
L
Linus Torvalds 已提交
1605 1606
}