write.c 42.8 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"
L
Linus Torvalds 已提交
31 32 33 34 35 36 37 38 39

#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

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

/*
 * Local function declarations
 */
40 41
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
				  struct inode *inode, int ioflags);
42
static void nfs_redirty_request(struct nfs_page *req);
43 44 45
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
46

47
static struct kmem_cache *nfs_wdata_cachep;
48
static mempool_t *nfs_wdata_mempool;
L
Linus Torvalds 已提交
49 50
static mempool_t *nfs_commit_mempool;

51
struct nfs_write_data *nfs_commitdata_alloc(void)
L
Linus Torvalds 已提交
52
{
C
Christoph Lameter 已提交
53
	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
54

L
Linus Torvalds 已提交
55 56 57
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
A
Andy Adamson 已提交
58
		p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
L
Linus Torvalds 已提交
59 60 61 62
	}
	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;
A
Andy Adamson 已提交
78
		p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
79 80
		if (pagecount <= ARRAY_SIZE(p->page_array))
			p->pagevec = p->page_array;
81
		else {
82 83
			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
			if (!p->pagevec) {
84 85 86 87 88 89 90 91
				mempool_free(p, nfs_wdata_mempool);
				p = NULL;
			}
		}
	}
	return p;
}

T
Trond Myklebust 已提交
92
void nfs_writedata_free(struct nfs_write_data *p)
93 94 95 96 97 98
{
	if (p && (p->pagevec != &p->page_array[0]))
		kfree(p->pagevec);
	mempool_free(p, nfs_wdata_mempool);
}

T
Trond Myklebust 已提交
99
static void nfs_writedata_release(struct nfs_write_data *wdata)
L
Linus Torvalds 已提交
100
{
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
		if (atomic_long_inc_return(&nfss->writeback) >
205 206 207 208
				NFS_CONGESTION_ON_THRESH) {
			set_bdi_congested(&nfss->backing_dev_info,
						BLK_RW_ASYNC);
		}
209
	}
210
	return ret;
211 212 213 214 215 216 217 218
}

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
219
	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
220
		clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
221 222
}

223
static struct nfs_page *nfs_find_and_lock_request(struct page *page)
224
{
225
	struct inode *inode = page->mapping->host;
226 227 228
	struct nfs_page *req;
	int ret;

229
	spin_lock(&inode->i_lock);
230
	for (;;) {
231
		req = nfs_page_find_request_locked(page);
232 233
		if (req == NULL)
			break;
234
		if (nfs_set_page_tag_locked(req))
235 236
			break;
		/* Note: If we hold the page lock, as is the case in nfs_writepage,
237
		 *	 then the call to nfs_set_page_tag_locked() will always
238 239 240
		 *	 succeed provided that someone hasn't already marked the
		 *	 request as dirty (in which case we don't care).
		 */
241
		spin_unlock(&inode->i_lock);
242 243 244
		ret = nfs_wait_on_request(req);
		nfs_release_request(req);
		if (ret != 0)
245
			return ERR_PTR(ret);
246
		spin_lock(&inode->i_lock);
247
	}
248
	spin_unlock(&inode->i_lock);
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
	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,
				struct page *page)
{
	struct nfs_page *req;
	int ret = 0;

	req = nfs_find_and_lock_request(page);
	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));

273 274
	if (!nfs_pageio_add_request(pgio, req)) {
		nfs_redirty_request(req);
275
		ret = pgio->pg_error;
276
	}
277 278
out:
	return ret;
279 280
}

T
Trond Myklebust 已提交
281
static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
L
Linus Torvalds 已提交
282 283 284
{
	struct inode *inode = page->mapping->host;

C
Chuck Lever 已提交
285 286 287
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);

288
	nfs_pageio_cond_complete(pgio, page->index);
T
Trond Myklebust 已提交
289 290
	return nfs_page_async_flush(pgio, page);
}
291

T
Trond Myklebust 已提交
292 293 294 295 296 297 298
/*
 * 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;
299

T
Trond Myklebust 已提交
300 301 302 303 304 305 306 307
	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;
308 309 310 311
}

int nfs_writepage(struct page *page, struct writeback_control *wbc)
{
T
Trond Myklebust 已提交
312
	int ret;
313

T
Trond Myklebust 已提交
314
	ret = nfs_writepage_locked(page, wbc);
L
Linus Torvalds 已提交
315
	unlock_page(page);
T
Trond Myklebust 已提交
316 317 318 319 320 321 322 323 324 325
	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
326 327 328 329 330
}

int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
331
	unsigned long *bitlock = &NFS_I(inode)->flags;
332
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
333 334
	int err;

335 336 337 338 339 340
	/* 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
341 342
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);

343
	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
T
Trond Myklebust 已提交
344
	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
345
	nfs_pageio_complete(&pgio);
346 347 348 349 350

	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
351
	if (err < 0)
352 353 354 355
		goto out_err;
	err = pgio.pg_error;
	if (err < 0)
		goto out_err;
356
	return 0;
357 358
out_err:
	return err;
L
Linus Torvalds 已提交
359 360 361 362 363
}

/*
 * Insert a write request into an inode
 */
364
static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
L
Linus Torvalds 已提交
365 366 367 368
{
	struct nfs_inode *nfsi = NFS_I(inode);
	int error;

369 370 371 372 373 374 375 376
	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
377
	error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
378
	BUG_ON(error);
L
Linus Torvalds 已提交
379 380 381 382 383
	if (!nfsi->npages) {
		igrab(inode);
		if (nfs_have_delegation(inode, FMODE_WRITE))
			nfsi->change_attr++;
	}
384
	SetPagePrivate(req->wb_page);
385
	set_page_private(req->wb_page, (unsigned long)req);
L
Linus Torvalds 已提交
386
	nfsi->npages++;
387
	kref_get(&req->wb_kref);
388 389
	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
				NFS_PAGE_TAG_LOCKED);
390 391 392 393
	spin_unlock(&inode->i_lock);
	radix_tree_preload_end();
out:
	return error;
L
Linus Torvalds 已提交
394 395 396
}

/*
397
 * Remove a write request from an inode
L
Linus Torvalds 已提交
398 399 400
 */
static void nfs_inode_remove_request(struct nfs_page *req)
{
401
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
402 403 404 405
	struct nfs_inode *nfsi = NFS_I(inode);

	BUG_ON (!NFS_WBACK_BUSY(req));

406
	spin_lock(&inode->i_lock);
407
	set_page_private(req->wb_page, 0);
408
	ClearPagePrivate(req->wb_page);
L
Linus Torvalds 已提交
409 410 411
	radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
	nfsi->npages--;
	if (!nfsi->npages) {
412
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
413 414
		iput(inode);
	} else
415
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
416 417 418 419
	nfs_clear_request(req);
	nfs_release_request(req);
}

420
static void
F
Fred 已提交
421
nfs_mark_request_dirty(struct nfs_page *req)
422 423 424 425
{
	__set_page_dirty_nobuffers(req->wb_page);
}

L
Linus Torvalds 已提交
426 427 428 429 430 431 432
#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)
{
433
	struct inode *inode = req->wb_context->path.dentry->d_inode;
L
Linus Torvalds 已提交
434 435
	struct nfs_inode *nfsi = NFS_I(inode);

436
	spin_lock(&inode->i_lock);
437
	set_bit(PG_CLEAN, &(req)->wb_flags);
438 439 440
	radix_tree_tag_set(&nfsi->nfs_page_tree,
			req->wb_index,
			NFS_PAGE_TAG_COMMIT);
441
	nfsi->ncommit++;
442
	spin_unlock(&inode->i_lock);
443
	inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
444
	inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
445
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
446
}
447

448 449 450 451 452 453 454 455 456 457 458 459 460
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;
}

461 462 463 464 465 466 467 468 469
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)
{
470
	if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
471 472 473 474
		nfs_mark_request_commit(req);
		return 1;
	}
	if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
F
Fred 已提交
475
		nfs_mark_request_dirty(req);
476 477 478 479 480 481 482 483 484 485
		return 1;
	}
	return 0;
}
#else
static inline void
nfs_mark_request_commit(struct nfs_page *req)
{
}

486 487 488 489 490 491
static inline int
nfs_clear_request_commit(struct nfs_page *req)
{
	return 0;
}

492 493 494 495 496 497 498 499 500 501 502
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
503 504 505 506 507
#endif

/*
 * Wait for a request to complete.
 *
508
 * Interruptible by fatal signals only.
L
Linus Torvalds 已提交
509
 */
510
static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
L
Linus Torvalds 已提交
511 512 513
{
	struct nfs_inode *nfsi = NFS_I(inode);
	struct nfs_page *req;
514
	pgoff_t idx_end, next;
L
Linus Torvalds 已提交
515 516 517 518 519 520 521 522 523
	unsigned int		res = 0;
	int			error;

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

	next = idx_start;
524
	while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) {
L
Linus Torvalds 已提交
525 526 527 528
		if (req->wb_index > idx_end)
			break;

		next = req->wb_index + 1;
529
		BUG_ON(!NFS_WBACK_BUSY(req));
L
Linus Torvalds 已提交
530

531
		kref_get(&req->wb_kref);
532
		spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
533 534
		error = nfs_wait_on_request(req);
		nfs_release_request(req);
535
		spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
536 537 538 539 540 541 542
		if (error < 0)
			return error;
		res++;
	}
	return res;
}

T
Trond Myklebust 已提交
543 544 545 546 547 548 549
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);
550
		nfs_clear_request_commit(req);
T
Trond Myklebust 已提交
551
		nfs_inode_remove_request(req);
552
		nfs_unlock_request(req);
T
Trond Myklebust 已提交
553 554 555
	}
}

T
Trond Myklebust 已提交
556
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
557 558 559 560 561 562
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
563 564 565 566 567 568 569 570 571 572 573
/*
 * 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
574
nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
L
Linus Torvalds 已提交
575 576
{
	struct nfs_inode *nfsi = NFS_I(inode);
577
	int ret;
578

579 580 581
	if (!nfs_need_commit(nfsi))
		return 0;

582 583 584
	ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
	if (ret > 0)
		nfsi->ncommit -= ret;
585 586
	if (nfs_need_commit(NFS_I(inode)))
		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
587
	return ret;
L
Linus Torvalds 已提交
588
}
589
#else
590 591 592 593 594
static inline int nfs_need_commit(struct nfs_inode *nfsi)
{
	return 0;
}

595
static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
596 597 598
{
	return 0;
}
L
Linus Torvalds 已提交
599 600 601
#endif

/*
602 603
 * 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
604
 *
605 606
 * If the attempt fails, then the existing request is flushed out
 * to disk.
L
Linus Torvalds 已提交
607
 */
608 609 610 611
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
612
{
613 614 615 616 617 618 619
	struct nfs_page *req;
	unsigned int rqend;
	unsigned int end;
	int error;

	if (!PagePrivate(page))
		return NULL;
L
Linus Torvalds 已提交
620 621

	end = offset + bytes;
622
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
623 624

	for (;;) {
625
		req = nfs_page_find_request_locked(page);
626 627 628 629 630 631 632 633 634 635
		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.
		 */
636
		if (offset > rqend
637 638 639 640
		    || end < req->wb_offset)
			goto out_flushme;

		if (nfs_set_page_tag_locked(req))
L
Linus Torvalds 已提交
641 642
			break;

643
		/* The request is locked, so wait and then retry */
644
		spin_unlock(&inode->i_lock);
645 646 647 648 649
		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
650 651
	}

652 653 654 655
	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--;
656

L
Linus Torvalds 已提交
657 658 659 660 661 662 663
	/* 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;
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
	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
690

691 692 693 694 695 696 697 698 699 700 701
	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);
	}
702
out:
T
Trond Myklebust 已提交
703
	return req;
L
Linus Torvalds 已提交
704 705
}

706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
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
721 722
int nfs_flush_incompatible(struct file *file, struct page *page)
{
723
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
724
	struct nfs_page	*req;
T
Trond Myklebust 已提交
725
	int do_flush, status;
L
Linus Torvalds 已提交
726 727 728 729 730 731 732 733
	/*
	 * 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
734 735 736 737
	do {
		req = nfs_page_find_request(page);
		if (req == NULL)
			return 0;
738
		do_flush = req->wb_page != page || req->wb_context != ctx;
L
Linus Torvalds 已提交
739
		nfs_release_request(req);
T
Trond Myklebust 已提交
740 741 742 743 744
		if (!do_flush)
			return 0;
		status = nfs_wb_page(page->mapping->host, page);
	} while (status == 0);
	return status;
L
Linus Torvalds 已提交
745 746
}

747 748 749 750 751 752 753 754 755 756 757
/*
 * 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
758 759 760 761 762 763 764 765 766
/*
 * 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)
{
767
	struct nfs_open_context *ctx = nfs_file_open_context(file);
L
Linus Torvalds 已提交
768 769 770
	struct inode	*inode = page->mapping->host;
	int		status = 0;

C
Chuck Lever 已提交
771 772
	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);

773
	dprintk("NFS:       nfs_updatepage(%s/%s %d@%lld)\n",
774 775
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name, count,
776
		(long long)(page_offset(page) + offset));
L
Linus Torvalds 已提交
777 778

	/* If we're not using byte range locks, and we know the page
779 780 781
	 * 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
782
	 */
783 784
	if (nfs_write_pageuptodate(page, inode) &&
			inode->i_flock == NULL &&
785
			!(file->f_flags & O_DSYNC)) {
786
		count = max(count + offset, nfs_page_length(page));
L
Linus Torvalds 已提交
787 788 789
		offset = 0;
	}

790
	status = nfs_writepage_setup(ctx, page, offset, count);
791 792 793 794
	if (status < 0)
		nfs_set_pageerror(page);
	else
		__set_page_dirty_nobuffers(page);
L
Linus Torvalds 已提交
795

796
	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
L
Linus Torvalds 已提交
797 798 799 800 801 802 803
			status, (long long)i_size_read(inode));
	return status;
}

static void nfs_writepage_release(struct nfs_page *req)
{

T
Trond Myklebust 已提交
804
	if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) {
805 806 807 808
		nfs_end_page_writeback(req->wb_page);
		nfs_inode_remove_request(req);
	} else
		nfs_end_page_writeback(req->wb_page);
809
	nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
810 811
}

812
static int flush_task_priority(int how)
L
Linus Torvalds 已提交
813 814 815 816 817 818 819 820 821 822 823 824 825
{
	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.
 */
826
static int nfs_write_rpcsetup(struct nfs_page *req,
L
Linus Torvalds 已提交
827
		struct nfs_write_data *data,
828
		const struct rpc_call_ops *call_ops,
L
Linus Torvalds 已提交
829 830 831
		unsigned int count, unsigned int offset,
		int how)
{
832 833
	struct inode *inode = req->wb_context->path.dentry->d_inode;
	int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
834
	int priority = flush_task_priority(how);
835
	struct rpc_task *task;
836 837 838 839 840
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = req->wb_context->cred,
	};
841 842
	struct rpc_task_setup task_setup_data = {
		.rpc_client = NFS_CLIENT(inode),
843
		.task = &data->task,
844
		.rpc_message = &msg,
845 846
		.callback_ops = call_ops,
		.callback_data = data,
847
		.workqueue = nfsiod_workqueue,
848
		.flags = flags,
849
		.priority = priority,
850
	};
L
Linus Torvalds 已提交
851 852 853 854 855

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

	data->req = req;
856
	data->inode = inode = req->wb_context->path.dentry->d_inode;
857
	data->cred = msg.rpc_cred;
L
Linus Torvalds 已提交
858 859 860 861 862 863

	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;
864
	data->args.context = get_nfs_open_context(req->wb_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
	/* Set up the initial task struct.  */
878
	NFS_PROTO(inode)->write_setup(data, &msg);
L
Linus Torvalds 已提交
879

C
Chuck Lever 已提交
880
	dprintk("NFS: %5u initiated write call "
881
		"(req %s/%lld, %u bytes @ offset %llu)\n",
882
		data->task.tk_pid,
L
Linus Torvalds 已提交
883 884 885 886 887
		inode->i_sb->s_id,
		(long long)NFS_FILEID(inode),
		count,
		(unsigned long long)data->args.offset);

888
	task = rpc_run_task(&task_setup_data);
889 890 891 892
	if (IS_ERR(task))
		return PTR_ERR(task);
	rpc_put_task(task);
	return 0;
L
Linus Torvalds 已提交
893 894
}

F
Fred 已提交
895 896 897 898 899 900 901 902 903 904 905
/* 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
906 907 908 909
/*
 * Generate multiple small requests to write out a single
 * contiguous dirty area on one page.
 */
910
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
911 912 913 914
{
	struct nfs_page *req = nfs_list_entry(head->next);
	struct page *page = req->wb_page;
	struct nfs_write_data *data;
915 916
	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
	unsigned int offset;
L
Linus Torvalds 已提交
917
	int requests = 0;
918
	int ret = 0;
L
Linus Torvalds 已提交
919 920 921 922
	LIST_HEAD(list);

	nfs_list_remove_request(req);

923
	nbytes = count;
924 925 926
	do {
		size_t len = min(nbytes, wsize);

927
		data = nfs_writedata_alloc(1);
L
Linus Torvalds 已提交
928 929 930 931
		if (!data)
			goto out_bad;
		list_add(&data->pages, &list);
		requests++;
932 933
		nbytes -= len;
	} while (nbytes != 0);
L
Linus Torvalds 已提交
934 935 936 937
	atomic_set(&req->wb_complete, requests);

	ClearPageError(page);
	offset = 0;
938
	nbytes = count;
L
Linus Torvalds 已提交
939
	do {
940 941
		int ret2;

L
Linus Torvalds 已提交
942 943 944 945 946
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del_init(&data->pages);

		data->pagevec[0] = page;

947 948
		if (nbytes < wsize)
			wsize = nbytes;
949
		ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
950
				   wsize, offset, how);
951 952
		if (ret == 0)
			ret = ret2;
953 954
		offset += wsize;
		nbytes -= wsize;
L
Linus Torvalds 已提交
955 956
	} while (nbytes != 0);

957
	return ret;
L
Linus Torvalds 已提交
958 959 960 961 962

out_bad:
	while (!list_empty(&list)) {
		data = list_entry(list.next, struct nfs_write_data, pages);
		list_del(&data->pages);
963
		nfs_writedata_release(data);
L
Linus Torvalds 已提交
964
	}
965
	nfs_redirty_request(req);
L
Linus Torvalds 已提交
966 967 968 969 970 971 972 973 974 975 976
	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.
 */
977
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
978 979 980 981 982
{
	struct nfs_page		*req;
	struct page		**pages;
	struct nfs_write_data	*data;

983
	data = nfs_writedata_alloc(npages);
L
Linus Torvalds 已提交
984 985 986 987 988 989 990 991 992 993 994 995 996 997
	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 */
998
	return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
L
Linus Torvalds 已提交
999 1000
 out_bad:
	while (!list_empty(head)) {
1001
		req = nfs_list_entry(head->next);
L
Linus Torvalds 已提交
1002
		nfs_list_remove_request(req);
1003
		nfs_redirty_request(req);
L
Linus Torvalds 已提交
1004 1005 1006 1007
	}
	return -ENOMEM;
}

1008 1009
static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
				  struct inode *inode, int ioflags)
L
Linus Torvalds 已提交
1010
{
1011
	size_t wsize = NFS_SERVER(inode)->wsize;
L
Linus Torvalds 已提交
1012

1013
	if (wsize < PAGE_CACHE_SIZE)
1014
		nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
1015
	else
1016
		nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
L
Linus Torvalds 已提交
1017 1018 1019 1020 1021
}

/*
 * Handle a write reply that flushed part of a page.
 */
1022
static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1023
{
1024
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1025

1026 1027 1028 1029 1030 1031
	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
1032

1033 1034
	nfs_writeback_done(task, data);
}
1035

1036 1037 1038 1039 1040 1041 1042 1043
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) {
1044
		nfs_set_pageerror(page);
1045 1046
		nfs_context_set_write_error(req->wb_context, status);
		dprintk(", error = %d\n", status);
1047
		goto out;
L
Linus Torvalds 已提交
1048 1049
	}

1050
	if (nfs_write_need_commit(data)) {
1051
		struct inode *inode = page->mapping->host;
1052

1053
		spin_lock(&inode->i_lock);
1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
		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");
		}
1064
		spin_unlock(&inode->i_lock);
1065 1066 1067 1068
	} else
		dprintk(" OK\n");

out:
L
Linus Torvalds 已提交
1069 1070
	if (atomic_dec_and_test(&req->wb_complete))
		nfs_writepage_release(req);
1071
	nfs_writedata_release(calldata);
L
Linus Torvalds 已提交
1072 1073
}

1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
#if defined(CONFIG_NFS_V4_1)
void nfs_write_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_write_data *data = calldata;
	struct nfs_client *clp = (NFS_SERVER(data->inode))->nfs_client;

	if (nfs4_setup_sequence(clp, &data->args.seq_args,
				&data->res.seq_res, 1, task))
		return;
	rpc_call_start(task);
}
#endif /* CONFIG_NFS_V4_1 */

1087
static const struct rpc_call_ops nfs_write_partial_ops = {
1088 1089 1090
#if defined(CONFIG_NFS_V4_1)
	.rpc_call_prepare = nfs_write_prepare,
#endif /* CONFIG_NFS_V4_1 */
1091
	.rpc_call_done = nfs_writeback_done_partial,
1092
	.rpc_release = nfs_writeback_release_partial,
1093 1094
};

L
Linus Torvalds 已提交
1095 1096 1097 1098 1099 1100 1101
/*
 * 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.
 */
1102
static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1103
{
1104
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1105

1106 1107 1108 1109 1110 1111 1112
	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;
1113

L
Linus Torvalds 已提交
1114 1115
	/* Update attributes as result of writeback. */
	while (!list_empty(&data->pages)) {
1116 1117 1118
		struct nfs_page *req = nfs_list_entry(data->pages.next);
		struct page *page = req->wb_page;

L
Linus Torvalds 已提交
1119 1120
		nfs_list_remove_request(req);

1121 1122
		dprintk("NFS: %5u write (%s/%lld %d@%lld)",
			data->task.tk_pid,
1123 1124
			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
1125 1126 1127
			req->wb_bytes,
			(long long)req_offset(req));

1128
		if (status < 0) {
1129
			nfs_set_pageerror(page);
1130 1131
			nfs_context_set_write_error(req->wb_context, status);
			dprintk(", error = %d\n", status);
1132
			goto remove_request;
L
Linus Torvalds 已提交
1133 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);
			nfs_end_page_writeback(page);
			dprintk(" marked for commit\n");
L
Linus Torvalds 已提交
1140 1141
			goto next;
		}
1142 1143 1144
		dprintk(" OK\n");
remove_request:
		nfs_end_page_writeback(page);
L
Linus Torvalds 已提交
1145 1146
		nfs_inode_remove_request(req);
	next:
1147
		nfs_clear_page_tag_locked(req);
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
int 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 1183
	status = NFS_PROTO(data->inode)->write_done(task, data);
	if (status != 0)
		return status;
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 1198
#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)) {
1199
			dprintk("NFS:       faulty NFS server %s:"
L
Linus Torvalds 已提交
1200
				" (committed = %d) != (stable = %d)\n",
1201
				server->nfs_client->cl_hostname,
L
Linus Torvalds 已提交
1202 1203 1204 1205 1206 1207 1208 1209 1210
				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
1211 1212
		nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);

L
Linus Torvalds 已提交
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
		/* 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;
			}
1227
			nfs_restart_rpc(task, server->nfs_client);
1228
			return -EAGAIN;
L
Linus Torvalds 已提交
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
		}
		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;
	}
1239
	return 0;
L
Linus Torvalds 已提交
1240 1241 1242 1243
}


#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1244
static void nfs_commitdata_release(void *data)
L
Linus Torvalds 已提交
1245
{
1246 1247 1248
	struct nfs_write_data *wdata = data;

	put_nfs_open_context(wdata->args.context);
L
Linus Torvalds 已提交
1249 1250 1251 1252 1253 1254
	nfs_commit_free(wdata);
}

/*
 * Set up the argument/result storage required for the RPC call.
 */
1255
static int nfs_commit_rpcsetup(struct list_head *head,
1256 1257
		struct nfs_write_data *data,
		int how)
L
Linus Torvalds 已提交
1258
{
1259 1260 1261
	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;
1262
	int priority = flush_task_priority(how);
1263
	struct rpc_task *task;
1264 1265 1266 1267 1268
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = first->wb_context->cred,
	};
1269
	struct rpc_task_setup task_setup_data = {
1270
		.task = &data->task,
1271
		.rpc_client = NFS_CLIENT(inode),
1272
		.rpc_message = &msg,
1273 1274
		.callback_ops = &nfs_commit_ops,
		.callback_data = data,
1275
		.workqueue = nfsiod_workqueue,
1276
		.flags = flags,
1277
		.priority = priority,
1278
	};
L
Linus Torvalds 已提交
1279 1280 1281 1282 1283 1284 1285

	/* 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;
1286
	data->cred	  = msg.rpc_cred;
L
Linus Torvalds 已提交
1287 1288

	data->args.fh     = NFS_FH(data->inode);
1289 1290 1291
	/* Note: we always request a commit of the entire inode */
	data->args.offset = 0;
	data->args.count  = 0;
1292
	data->args.context = get_nfs_open_context(first->wb_context);
1293
	data->res.count   = 0;
L
Linus Torvalds 已提交
1294 1295
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
1296
	nfs_fattr_init(&data->fattr);
1297 1298

	/* Set up the initial task struct.  */
1299
	NFS_PROTO(inode)->commit_setup(data, &msg);
L
Linus Torvalds 已提交
1300

C
Chuck Lever 已提交
1301
	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1302

1303
	task = rpc_run_task(&task_setup_data);
1304 1305 1306 1307
	if (IS_ERR(task))
		return PTR_ERR(task);
	rpc_put_task(task);
	return 0;
L
Linus Torvalds 已提交
1308 1309 1310 1311 1312 1313
}

/*
 * Commit dirty pages
 */
static int
1314
nfs_commit_list(struct inode *inode, struct list_head *head, int how)
L
Linus Torvalds 已提交
1315 1316 1317 1318
{
	struct nfs_write_data	*data;
	struct nfs_page         *req;

1319
	data = nfs_commitdata_alloc();
L
Linus Torvalds 已提交
1320 1321 1322 1323 1324

	if (!data)
		goto out_bad;

	/* Set up the argument struct */
1325
	return nfs_commit_rpcsetup(head, data, how);
L
Linus Torvalds 已提交
1326 1327 1328 1329 1330
 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
1331
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1332 1333
		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
				BDI_RECLAIMABLE);
1334
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1335 1336 1337 1338 1339 1340 1341
	}
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
1342
static void nfs_commit_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1343
{
1344
	struct nfs_write_data	*data = calldata;
L
Linus Torvalds 已提交
1345

C
Chuck Lever 已提交
1346
        dprintk("NFS: %5u nfs_commit_done (status %d)\n",
L
Linus Torvalds 已提交
1347 1348
                                task->tk_pid, task->tk_status);

1349 1350 1351
	/* Call the NFS version-specific code */
	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
		return;
1352 1353 1354 1355 1356 1357 1358
}

static void nfs_commit_release(void *calldata)
{
	struct nfs_write_data	*data = calldata;
	struct nfs_page		*req;
	int status = data->task.tk_status;
1359

L
Linus Torvalds 已提交
1360 1361 1362
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
1363
		nfs_clear_request_commit(req);
L
Linus Torvalds 已提交
1364

1365
		dprintk("NFS:       commit (%s/%lld %d@%lld)",
1366 1367
			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
1368 1369
			req->wb_bytes,
			(long long)req_offset(req));
1370 1371
		if (status < 0) {
			nfs_context_set_write_error(req->wb_context, status);
L
Linus Torvalds 已提交
1372
			nfs_inode_remove_request(req);
1373
			dprintk(", error = %d\n", status);
L
Linus Torvalds 已提交
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
			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
1387
		nfs_mark_request_dirty(req);
L
Linus Torvalds 已提交
1388
	next:
1389
		nfs_clear_page_tag_locked(req);
L
Linus Torvalds 已提交
1390
	}
1391
	nfs_commitdata_release(calldata);
L
Linus Torvalds 已提交
1392
}
1393 1394

static const struct rpc_call_ops nfs_commit_ops = {
1395 1396 1397
#if defined(CONFIG_NFS_V4_1)
	.rpc_call_prepare = nfs_write_prepare,
#endif /* CONFIG_NFS_V4_1 */
1398 1399 1400
	.rpc_call_done = nfs_commit_done,
	.rpc_release = nfs_commit_release,
};
L
Linus Torvalds 已提交
1401

1402
static int nfs_commit_inode(struct inode *inode, int how)
L
Linus Torvalds 已提交
1403 1404
{
	LIST_HEAD(head);
T
Trond Myklebust 已提交
1405
	int res;
L
Linus Torvalds 已提交
1406

1407
	spin_lock(&inode->i_lock);
1408
	res = nfs_scan_commit(inode, &head, 0, 0);
1409
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
1410
	if (res) {
T
Trond Myklebust 已提交
1411
		int error = nfs_commit_list(inode, &head, how);
1412 1413 1414
		if (error < 0)
			return error;
	}
L
Linus Torvalds 已提交
1415 1416
	return res;
}
1417 1418 1419

static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
1420 1421 1422
	struct nfs_inode *nfsi = NFS_I(inode);
	int flags = FLUSH_SYNC;
	int ret = 0;
1423

1424 1425 1426 1427 1428 1429 1430
	/* Don't commit yet if this is a non-blocking flush and there are
	 * lots of outstanding writes for this mapping.
	 */
	if (wbc->sync_mode == WB_SYNC_NONE &&
	    nfsi->ncommit <= (nfsi->npages >> 1))
		goto out_mark_dirty;

1431
	if (wbc->nonblocking || wbc->for_background)
1432 1433 1434 1435 1436 1437 1438 1439 1440
		flags = 0;
	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;
		}
1441
		return 0;
1442 1443
	}
out_mark_dirty:
1444 1445 1446
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
	return ret;
}
1447 1448 1449 1450 1451
#else
static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
{
	return 0;
}
1452 1453 1454 1455 1456

static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
	return 0;
}
L
Linus Torvalds 已提交
1457 1458
#endif

1459 1460 1461 1462 1463
int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
{
	return nfs_commit_unstable_pages(inode, wbc);
}

1464
long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
L
Linus Torvalds 已提交
1465
{
1466
	struct inode *inode = mapping->host;
1467
	pgoff_t idx_start, idx_end;
1468
	unsigned int npages = 0;
1469
	LIST_HEAD(head);
1470
	int nocommit = how & FLUSH_NOCOMMIT;
T
Trond Myklebust 已提交
1471
	long pages, ret;
L
Linus Torvalds 已提交
1472

1473 1474 1475 1476 1477 1478 1479
	/* 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) {
1480
			pgoff_t l_npages = 1 + idx_end - idx_start;
1481 1482
			npages = l_npages;
			if (sizeof(npages) != sizeof(l_npages) &&
1483
					(pgoff_t)npages != l_npages)
1484 1485 1486
				npages = 0;
		}
	}
1487
	how &= ~FLUSH_NOCOMMIT;
1488
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
1489
	do {
1490 1491
		ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
		if (ret != 0)
1492
			continue;
1493 1494
		if (nocommit)
			break;
1495
		pages = nfs_scan_commit(inode, &head, idx_start, npages);
1496
		if (pages == 0)
1497
			break;
1498
		if (how & FLUSH_INVALIDATE) {
1499
			spin_unlock(&inode->i_lock);
T
Trond Myklebust 已提交
1500
			nfs_cancel_commit_list(&head);
1501
			ret = pages;
1502
			spin_lock(&inode->i_lock);
1503 1504 1505
			continue;
		}
		pages += nfs_scan_commit(inode, &head, 0, 0);
1506
		spin_unlock(&inode->i_lock);
1507
		ret = nfs_commit_list(inode, &head, how);
1508 1509
		spin_lock(&inode->i_lock);

1510
	} while (ret >= 0);
1511
	spin_unlock(&inode->i_lock);
1512
	return ret;
L
Linus Torvalds 已提交
1513 1514
}

T
Trond Myklebust 已提交
1515
static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
1516 1517 1518
{
	int ret;

T
Trond Myklebust 已提交
1519
	ret = nfs_writepages(mapping, wbc);
1520 1521
	if (ret < 0)
		goto out;
T
Trond Myklebust 已提交
1522
	ret = nfs_sync_mapping_wait(mapping, wbc, how);
1523 1524 1525
	if (ret < 0)
		goto out;
	return 0;
1526
out:
1527
	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1528 1529 1530
	return ret;
}

T
Trond Myklebust 已提交
1531 1532 1533 1534 1535
/* 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,
1536
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1537
		.nr_to_write = LONG_MAX,
1538 1539
		.range_start = 0,
		.range_end = LLONG_MAX,
T
Trond Myklebust 已提交
1540 1541 1542 1543 1544
	};

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

1545 1546 1547 1548
/*
 * flush the inode to disk.
 */
int nfs_wb_all(struct inode *inode)
1549
{
1550 1551
	return nfs_write_mapping(inode->i_mapping, 0);
}
1552

1553 1554 1555
int nfs_wb_nocommit(struct inode *inode)
{
	return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT);
1556 1557
}

1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576
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;
1577
		if (test_bit(PG_CLEAN, &req->wb_flags)) {
1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
			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);
1592
		nfs_release_request(req);
1593 1594 1595 1596 1597 1598 1599 1600 1601 1602
		if (ret < 0)
			goto out;
	}
	if (!PagePrivate(page))
		return 0;
	ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE);
out:
	return ret;
}

1603 1604
static int nfs_wb_page_priority(struct inode *inode, struct page *page,
				int how)
1605 1606 1607
{
	loff_t range_start = page_offset(page);
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1608 1609 1610 1611 1612 1613 1614 1615
	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;
1616

1617 1618 1619 1620 1621 1622 1623 1624
	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);
1625
		if (ret < 0)
1626 1627 1628 1629
			goto out_error;
	} while (PagePrivate(page));
	return 0;
out_error:
1630
	__mark_inode_dirty(inode, I_DIRTY_PAGES);
1631
	return ret;
1632 1633 1634 1635 1636 1637 1638
}

/*
 * Write back all requests on one page - we do this before reading it.
 */
int nfs_wb_page(struct inode *inode, struct page* page)
{
1639
	return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
1640 1641
}

1642 1643 1644 1645 1646 1647 1648
#ifdef CONFIG_MIGRATION
int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
		struct page *page)
{
	struct nfs_page *req;
	int ret;

1649
	nfs_fscache_release_page(page, GFP_KERNEL);
1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661

	req = nfs_find_and_lock_request(page);
	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
1662
	spin_lock(&mapping->host->i_lock);
1663 1664
	req->wb_page = newpage;
	SetPagePrivate(newpage);
T
Trond Myklebust 已提交
1665
	set_page_private(newpage, (unsigned long)req);
1666 1667
	ClearPagePrivate(page);
	set_page_private(page, 0);
T
Trond Myklebust 已提交
1668
	spin_unlock(&mapping->host->i_lock);
1669 1670 1671 1672 1673 1674 1675 1676
	page_cache_release(page);
out_unlock:
	nfs_clear_page_tag_locked(req);
out:
	return ret;
}
#endif

D
David Howells 已提交
1677
int __init nfs_init_writepagecache(void)
L
Linus Torvalds 已提交
1678 1679 1680 1681
{
	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
					     sizeof(struct nfs_write_data),
					     0, SLAB_HWCACHE_ALIGN,
1682
					     NULL);
L
Linus Torvalds 已提交
1683 1684 1685
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

1686 1687
	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
						     nfs_wdata_cachep);
L
Linus Torvalds 已提交
1688 1689 1690
	if (nfs_wdata_mempool == NULL)
		return -ENOMEM;

1691 1692
	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
						      nfs_wdata_cachep);
L
Linus Torvalds 已提交
1693 1694 1695
	if (nfs_commit_mempool == NULL)
		return -ENOMEM;

1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715
	/*
	 * 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
1716 1717 1718
	return 0;
}

1719
void nfs_destroy_writepagecache(void)
L
Linus Torvalds 已提交
1720 1721 1722
{
	mempool_destroy(nfs_commit_mempool);
	mempool_destroy(nfs_wdata_mempool);
1723
	kmem_cache_destroy(nfs_wdata_cachep);
L
Linus Torvalds 已提交
1724 1725
}