write.c 42.7 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
#include <linux/backing-dev.h>
23
#include <linux/export.h>
24

L
Linus Torvalds 已提交
25 26 27
#include <asm/uaccess.h>

#include "delegation.h"
28
#include "internal.h"
C
Chuck Lever 已提交
29
#include "iostat.h"
30
#include "nfs4_fs.h"
31
#include "fscache.h"
32
#include "pnfs.h"
L
Linus Torvalds 已提交
33

34 35
#include "nfstrace.h"

L
Linus Torvalds 已提交
36 37 38 39 40 41 42 43
#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

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

/*
 * Local function declarations
 */
44
static void nfs_redirty_request(struct nfs_page *req);
45
static const struct rpc_call_ops nfs_commit_ops;
46
static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
47
static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
48
static const struct nfs_rw_ops nfs_rw_write_ops;
L
Linus Torvalds 已提交
49

50
static struct kmem_cache *nfs_wdata_cachep;
51
static mempool_t *nfs_wdata_mempool;
52
static struct kmem_cache *nfs_cdata_cachep;
L
Linus Torvalds 已提交
53 54
static mempool_t *nfs_commit_mempool;

55
struct nfs_commit_data *nfs_commitdata_alloc(void)
L
Linus Torvalds 已提交
56
{
57
	struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
58

L
Linus Torvalds 已提交
59 60 61 62 63 64
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
	}
	return p;
}
65
EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
L
Linus Torvalds 已提交
66

67
void nfs_commit_free(struct nfs_commit_data *p)
L
Linus Torvalds 已提交
68 69 70
{
	mempool_free(p, nfs_commit_mempool);
}
71
EXPORT_SYMBOL_GPL(nfs_commit_free);
L
Linus Torvalds 已提交
72

73
static struct nfs_rw_header *nfs_writehdr_alloc(void)
74
{
75
	struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
76

77
	if (p)
78 79 80 81
		memset(p, 0, sizeof(*p));
	return p;
}

82
static void nfs_writehdr_free(struct nfs_rw_header *whdr)
83
{
84
	mempool_free(whdr, nfs_wdata_mempool);
85 86
}

87 88 89 90 91 92 93
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);
}

94 95
static struct nfs_page *
nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page)
96 97 98
{
	struct nfs_page *req = NULL;

99
	if (PagePrivate(page))
100
		req = (struct nfs_page *)page_private(page);
101 102 103 104 105 106 107 108 109 110
	else if (unlikely(PageSwapCache(page))) {
		struct nfs_page *freq, *t;

		/* Linearly search the commit list for the correct req */
		list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) {
			if (freq->wb_page == page) {
				req = freq;
				break;
			}
		}
111
	}
112 113 114 115

	if (req)
		kref_get(&req->wb_kref);

116 117 118 119 120
	return req;
}

static struct nfs_page *nfs_page_find_request(struct page *page)
{
121
	struct inode *inode = page_file_mapping(page)->host;
122 123
	struct nfs_page *req = NULL;

124
	spin_lock(&inode->i_lock);
125
	req = nfs_page_find_request_locked(NFS_I(inode), page);
126
	spin_unlock(&inode->i_lock);
127 128 129
	return req;
}

L
Linus Torvalds 已提交
130 131 132
/* 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)
{
133
	struct inode *inode = page_file_mapping(page)->host;
134 135
	loff_t end, i_size;
	pgoff_t end_index;
L
Linus Torvalds 已提交
136

137 138 139
	spin_lock(&inode->i_lock);
	i_size = i_size_read(inode);
	end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
140
	if (i_size > 0 && page_file_index(page) < end_index)
141
		goto out;
142
	end = page_file_offset(page) + ((loff_t)offset+count);
L
Linus Torvalds 已提交
143
	if (i_size >= end)
144
		goto out;
L
Linus Torvalds 已提交
145
	i_size_write(inode, end);
146 147 148
	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
out:
	spin_unlock(&inode->i_lock);
L
Linus Torvalds 已提交
149 150
}

151 152 153
/* A writeback failed: mark the page as bad, and invalidate the page cache */
static void nfs_set_pageerror(struct page *page)
{
154
	nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
155 156
}

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

static int wb_priority(struct writeback_control *wbc)
{
	if (wbc->for_reclaim)
174
		return FLUSH_HIGHPRI | FLUSH_STABLE;
175
	if (wbc->for_kupdate || wbc->for_background)
176 177
		return FLUSH_LOWPRI | FLUSH_COND_STABLE;
	return FLUSH_COND_STABLE;
L
Linus Torvalds 已提交
178 179
}

180 181 182 183 184 185 186 187 188 189
/*
 * 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))

190
static void nfs_set_page_writeback(struct page *page)
191
{
192
	struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
193 194
	int ret = test_set_page_writeback(page);

195
	WARN_ON_ONCE(ret != 0);
196

197 198 199 200
	if (atomic_long_inc_return(&nfss->writeback) >
			NFS_CONGESTION_ON_THRESH) {
		set_bdi_congested(&nfss->backing_dev_info,
					BLK_RW_ASYNC);
201 202 203
	}
}

204
static void nfs_end_page_writeback(struct nfs_page *req)
205
{
206
	struct inode *inode = page_file_mapping(req->wb_page)->host;
207 208
	struct nfs_server *nfss = NFS_SERVER(inode);

209 210 211 212
	if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
		return;

	end_page_writeback(req->wb_page);
P
Peter Zijlstra 已提交
213
	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
214
		clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
215 216
}

217
static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
218
{
219
	struct inode *inode = page_file_mapping(page)->host;
220 221 222
	struct nfs_page *req;
	int ret;

223
	spin_lock(&inode->i_lock);
224
	for (;;) {
225
		req = nfs_page_find_request_locked(NFS_I(inode), page);
226 227
		if (req == NULL)
			break;
228
		if (nfs_lock_request(req))
229 230
			break;
		/* Note: If we hold the page lock, as is the case in nfs_writepage,
231
		 *	 then the call to nfs_lock_request() will always
232 233 234
		 *	 succeed provided that someone hasn't already marked the
		 *	 request as dirty (in which case we don't care).
		 */
235
		spin_unlock(&inode->i_lock);
236 237 238 239
		if (!nonblock)
			ret = nfs_wait_on_request(req);
		else
			ret = -EAGAIN;
240 241
		nfs_release_request(req);
		if (ret != 0)
242
			return ERR_PTR(ret);
243
		spin_lock(&inode->i_lock);
244
	}
245
	spin_unlock(&inode->i_lock);
246 247 248 249 250 251 252 253
	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,
254
				struct page *page, bool nonblock)
255 256 257 258
{
	struct nfs_page *req;
	int ret = 0;

259
	req = nfs_find_and_lock_request(page, nonblock);
260 261 262 263 264 265
	if (!req)
		goto out;
	ret = PTR_ERR(req);
	if (IS_ERR(req))
		goto out;

266 267
	nfs_set_page_writeback(page);
	WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
268

269
	ret = 0;
270 271
	if (!nfs_pageio_add_request(pgio, req)) {
		nfs_redirty_request(req);
272
		ret = pgio->pg_error;
273
	}
274 275
out:
	return ret;
276 277
}

T
Trond Myklebust 已提交
278
static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
L
Linus Torvalds 已提交
279
{
280
	struct inode *inode = page_file_mapping(page)->host;
281
	int ret;
L
Linus Torvalds 已提交
282

C
Chuck Lever 已提交
283 284 285
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);

286
	nfs_pageio_cond_complete(pgio, page_file_index(page));
287
	ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
288 289 290 291 292
	if (ret == -EAGAIN) {
		redirty_page_for_writepage(wbc, page);
		ret = 0;
	}
	return ret;
T
Trond Myklebust 已提交
293
}
294

T
Trond Myklebust 已提交
295 296 297 298 299 300 301
/*
 * 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;
302

303 304
	nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
				false, &nfs_async_write_completion_ops);
T
Trond Myklebust 已提交
305 306 307 308 309 310 311
	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;
312 313 314 315
}

int nfs_writepage(struct page *page, struct writeback_control *wbc)
{
T
Trond Myklebust 已提交
316
	int ret;
317

T
Trond Myklebust 已提交
318
	ret = nfs_writepage_locked(page, wbc);
L
Linus Torvalds 已提交
319
	unlock_page(page);
T
Trond Myklebust 已提交
320 321 322 323 324 325 326 327 328 329
	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
330 331 332 333 334
}

int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
335
	unsigned long *bitlock = &NFS_I(inode)->flags;
336
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
337 338
	int err;

339 340 341 342 343 344
	/* 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
345 346
	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);

347 348
	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
				&nfs_async_write_completion_ops);
T
Trond Myklebust 已提交
349
	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
350
	nfs_pageio_complete(&pgio);
351 352 353 354 355

	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
356
	if (err < 0)
357 358 359 360
		goto out_err;
	err = pgio.pg_error;
	if (err < 0)
		goto out_err;
361
	return 0;
362 363
out_err:
	return err;
L
Linus Torvalds 已提交
364 365 366 367 368
}

/*
 * Insert a write request into an inode
 */
F
Fred Isaman 已提交
369
static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
L
Linus Torvalds 已提交
370 371
{
	struct nfs_inode *nfsi = NFS_I(inode);
372

373 374
	WARN_ON_ONCE(req->wb_this_page != req);

375
	/* Lock the request! */
376
	nfs_lock_request(req);
377 378

	spin_lock(&inode->i_lock);
379
	if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
380
		inode->i_version++;
381 382 383 384 385 386 387 388 389
	/*
	 * Swap-space should not get truncated. Hence no need to plug the race
	 * with invalidate/truncate.
	 */
	if (likely(!PageSwapCache(req->wb_page))) {
		set_bit(PG_MAPPED, &req->wb_flags);
		SetPagePrivate(req->wb_page);
		set_page_private(req->wb_page, (unsigned long)req);
	}
L
Linus Torvalds 已提交
390
	nfsi->npages++;
391
	set_bit(PG_INODE_REF, &req->wb_flags);
392
	kref_get(&req->wb_kref);
393
	spin_unlock(&inode->i_lock);
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->dentry->d_inode;
L
Linus Torvalds 已提交
402
	struct nfs_inode *nfsi = NFS_I(inode);
403
	struct nfs_page *head;
L
Linus Torvalds 已提交
404

405 406 407 408 409 410 411 412 413 414 415
	if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
		head = req->wb_head;

		spin_lock(&inode->i_lock);
		if (likely(!PageSwapCache(head->wb_page))) {
			set_page_private(head->wb_page, 0);
			ClearPagePrivate(head->wb_page);
			clear_bit(PG_MAPPED, &head->wb_flags);
		}
		nfsi->npages--;
		spin_unlock(&inode->i_lock);
416
	}
L
Linus Torvalds 已提交
417 418 419
	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);
}

B
Bryan Schumaker 已提交
426
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
427 428 429
/**
 * nfs_request_add_commit_list - add request to a commit list
 * @req: pointer to a struct nfs_page
F
Fred Isaman 已提交
430 431
 * @dst: commit list head
 * @cinfo: holds list lock and accounting info
432
 *
F
Fred Isaman 已提交
433
 * This sets the PG_CLEAN bit, updates the cinfo count of
434 435 436
 * number of outstanding requests requiring a commit as well as
 * the MM page stats.
 *
F
Fred Isaman 已提交
437
 * The caller must _not_ hold the cinfo->lock, but must be
438
 * holding the nfs_page lock.
L
Linus Torvalds 已提交
439
 */
440
void
F
Fred Isaman 已提交
441 442
nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
			    struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
443
{
444
	set_bit(PG_CLEAN, &(req)->wb_flags);
F
Fred Isaman 已提交
445 446 447 448
	spin_lock(cinfo->lock);
	nfs_list_add_request(req, dst);
	cinfo->mds->ncommit++;
	spin_unlock(cinfo->lock);
449 450
	if (!cinfo->dreq) {
		inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
451
		inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
452 453 454 455
			     BDI_RECLAIMABLE);
		__mark_inode_dirty(req->wb_context->dentry->d_inode,
				   I_DIRTY_DATASYNC);
	}
L
Linus Torvalds 已提交
456
}
457 458 459 460 461
EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);

/**
 * nfs_request_remove_commit_list - Remove request from a commit list
 * @req: pointer to a nfs_page
F
Fred Isaman 已提交
462
 * @cinfo: holds list lock and accounting info
463
 *
F
Fred Isaman 已提交
464
 * This clears the PG_CLEAN bit, and updates the cinfo's count of
465 466 467
 * number of outstanding requests requiring a commit
 * It does not update the MM page stats.
 *
F
Fred Isaman 已提交
468
 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
469 470
 */
void
F
Fred Isaman 已提交
471 472
nfs_request_remove_commit_list(struct nfs_page *req,
			       struct nfs_commit_info *cinfo)
473 474 475 476
{
	if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
		return;
	nfs_list_remove_request(req);
F
Fred Isaman 已提交
477
	cinfo->mds->ncommit--;
478 479 480
}
EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);

F
Fred Isaman 已提交
481 482 483 484 485 486
static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
				      struct inode *inode)
{
	cinfo->lock = &inode->i_lock;
	cinfo->mds = &NFS_I(inode)->commit_info;
	cinfo->ds = pnfs_get_ds_info(inode);
F
Fred Isaman 已提交
487
	cinfo->dreq = NULL;
488
	cinfo->completion_ops = &nfs_commit_completion_ops;
F
Fred Isaman 已提交
489 490 491 492 493 494
}

void nfs_init_cinfo(struct nfs_commit_info *cinfo,
		    struct inode *inode,
		    struct nfs_direct_req *dreq)
{
495 496 497 498
	if (dreq)
		nfs_init_cinfo_from_dreq(cinfo, dreq);
	else
		nfs_init_cinfo_from_inode(cinfo, inode);
F
Fred Isaman 已提交
499 500
}
EXPORT_SYMBOL_GPL(nfs_init_cinfo);
501 502 503 504

/*
 * Add a request to the inode's commit list.
 */
505
void
F
Fred Isaman 已提交
506 507
nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
			struct nfs_commit_info *cinfo)
508
{
F
Fred Isaman 已提交
509
	if (pnfs_mark_request_commit(req, lseg, cinfo))
510
		return;
F
Fred Isaman 已提交
511
	nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
512
}
513

F
Fred Isaman 已提交
514 515 516 517
static void
nfs_clear_page_commit(struct page *page)
{
	dec_zone_page_state(page, NR_UNSTABLE_NFS);
518
	dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
F
Fred Isaman 已提交
519 520
}

521
static void
522 523
nfs_clear_request_commit(struct nfs_page *req)
{
524 525
	if (test_bit(PG_CLEAN, &req->wb_flags)) {
		struct inode *inode = req->wb_context->dentry->d_inode;
F
Fred Isaman 已提交
526
		struct nfs_commit_info cinfo;
527

F
Fred Isaman 已提交
528 529 530 531 532
		nfs_init_cinfo_from_inode(&cinfo, inode);
		if (!pnfs_clear_request_commit(req, &cinfo)) {
			spin_lock(cinfo.lock);
			nfs_request_remove_commit_list(req, &cinfo);
			spin_unlock(cinfo.lock);
533
		}
F
Fred Isaman 已提交
534
		nfs_clear_page_commit(req->wb_page);
535 536 537
	}
}

538
static inline
539
int nfs_write_need_commit(struct nfs_pgio_data *data)
540
{
541
	if (data->verf.committed == NFS_DATA_SYNC)
542 543
		return data->header->lseg == NULL;
	return data->verf.committed != NFS_FILE_SYNC;
544 545 546
}

#else
547 548 549 550 551 552 553 554 555 556 557
static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
				      struct inode *inode)
{
}

void nfs_init_cinfo(struct nfs_commit_info *cinfo,
		    struct inode *inode,
		    struct nfs_direct_req *dreq)
{
}

558
void
F
Fred Isaman 已提交
559 560
nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
			struct nfs_commit_info *cinfo)
561 562 563
{
}

564
static void
565 566 567 568
nfs_clear_request_commit(struct nfs_page *req)
{
}

569
static inline
570
int nfs_write_need_commit(struct nfs_pgio_data *data)
571 572 573 574
{
	return 0;
}

575 576
#endif

577
static void nfs_write_completion(struct nfs_pgio_header *hdr)
578
{
F
Fred Isaman 已提交
579
	struct nfs_commit_info cinfo;
580
	unsigned long bytes = 0;
581
	bool do_destroy;
582 583 584

	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out;
F
Fred Isaman 已提交
585
	nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
586 587 588 589 590 591 592
	while (!list_empty(&hdr->pages)) {
		struct nfs_page *req = nfs_list_entry(hdr->pages.next);

		bytes += req->wb_bytes;
		nfs_list_remove_request(req);
		if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
		    (hdr->good_bytes < bytes)) {
593
			nfs_set_pageerror(req->wb_page);
594 595 596 597 598 599 600 601
			nfs_context_set_write_error(req->wb_context, hdr->error);
			goto remove_req;
		}
		if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
			nfs_mark_request_dirty(req);
			goto next;
		}
		if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
602
			memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
F
Fred Isaman 已提交
603
			nfs_mark_request_commit(req, hdr->lseg, &cinfo);
604 605 606 607 608
			goto next;
		}
remove_req:
		nfs_inode_remove_request(req);
next:
609
		nfs_unlock_request(req);
610
		nfs_end_page_writeback(req);
611
		do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags);
612
		nfs_release_request(req);
613 614 615
	}
out:
	hdr->release(hdr);
616
}
L
Linus Torvalds 已提交
617

B
Bryan Schumaker 已提交
618
#if  IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
619
unsigned long
F
Fred Isaman 已提交
620
nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
621
{
F
Fred Isaman 已提交
622
	return cinfo->mds->ncommit;
F
Fred Isaman 已提交
623 624
}

F
Fred Isaman 已提交
625
/* cinfo->lock held by caller */
626
int
F
Fred Isaman 已提交
627 628
nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
		     struct nfs_commit_info *cinfo, int max)
F
Fred Isaman 已提交
629 630 631 632 633
{
	struct nfs_page *req, *tmp;
	int ret = 0;

	list_for_each_entry_safe(req, tmp, src, wb_list) {
634 635
		if (!nfs_lock_request(req))
			continue;
636
		kref_get(&req->wb_kref);
F
Fred Isaman 已提交
637
		if (cond_resched_lock(cinfo->lock))
638
			list_safe_reset_next(req, tmp, wb_list);
F
Fred Isaman 已提交
639
		nfs_request_remove_commit_list(req, cinfo);
640 641
		nfs_list_add_request(req, dst);
		ret++;
642
		if ((ret == max) && !cinfo->dreq)
643
			break;
F
Fred Isaman 已提交
644 645
	}
	return ret;
646 647
}

L
Linus Torvalds 已提交
648 649 650
/*
 * nfs_scan_commit - Scan an inode for commit requests
 * @inode: NFS inode to scan
F
Fred Isaman 已提交
651 652
 * @dst: mds destination list
 * @cinfo: mds and ds lists of reqs ready to commit
L
Linus Torvalds 已提交
653 654 655 656
 *
 * Moves requests from the inode's 'commit' request list.
 * The requests are *not* checked to ensure that they form a contiguous set.
 */
657
int
F
Fred Isaman 已提交
658 659
nfs_scan_commit(struct inode *inode, struct list_head *dst,
		struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
660
{
F
Fred Isaman 已提交
661
	int ret = 0;
662

F
Fred Isaman 已提交
663 664
	spin_lock(cinfo->lock);
	if (cinfo->mds->ncommit > 0) {
665
		const int max = INT_MAX;
F
Fred Isaman 已提交
666

F
Fred Isaman 已提交
667 668 669
		ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
					   cinfo, max);
		ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
F
Fred Isaman 已提交
670
	}
F
Fred Isaman 已提交
671
	spin_unlock(cinfo->lock);
672
	return ret;
L
Linus Torvalds 已提交
673
}
F
Fred Isaman 已提交
674

675
#else
676
unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
677 678 679 680
{
	return 0;
}

681 682
int nfs_scan_commit(struct inode *inode, struct list_head *dst,
		    struct nfs_commit_info *cinfo)
683 684 685
{
	return 0;
}
L
Linus Torvalds 已提交
686 687 688
#endif

/*
689 690
 * 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
691
 *
692 693
 * If the attempt fails, then the existing request is flushed out
 * to disk.
L
Linus Torvalds 已提交
694
 */
695 696 697 698
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
699
{
700 701 702 703 704 705 706
	struct nfs_page *req;
	unsigned int rqend;
	unsigned int end;
	int error;

	if (!PagePrivate(page))
		return NULL;
L
Linus Torvalds 已提交
707 708

	end = offset + bytes;
709
	spin_lock(&inode->i_lock);
L
Linus Torvalds 已提交
710 711

	for (;;) {
712
		req = nfs_page_find_request_locked(NFS_I(inode), page);
713 714 715
		if (req == NULL)
			goto out_unlock;

716 717 718 719
		/* should be handled by nfs_flush_incompatible */
		WARN_ON_ONCE(req->wb_head != req);
		WARN_ON_ONCE(req->wb_this_page != req);

720 721 722 723 724 725 726
		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.
		 */
727
		if (offset > rqend
728 729 730
		    || end < req->wb_offset)
			goto out_flushme;

731
		if (nfs_lock_request(req))
L
Linus Torvalds 已提交
732 733
			break;

734
		/* The request is locked, so wait and then retry */
735
		spin_unlock(&inode->i_lock);
736 737 738 739 740
		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
741 742 743 744 745 746 747 748 749
	}

	/* 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;
750 751 752 753
	else
		req->wb_bytes = rqend - req->wb_offset;
out_unlock:
	spin_unlock(&inode->i_lock);
754 755
	if (req)
		nfs_clear_request_commit(req);
756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774
	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)
{
775
	struct inode *inode = page_file_mapping(page)->host;
776
	struct nfs_page	*req;
L
Linus Torvalds 已提交
777

778 779 780
	req = nfs_try_to_update_request(inode, page, offset, bytes);
	if (req != NULL)
		goto out;
781
	req = nfs_create_request(ctx, page, NULL, offset, bytes);
782 783
	if (IS_ERR(req))
		goto out;
F
Fred Isaman 已提交
784
	nfs_inode_add_request(inode, req);
785
out:
T
Trond Myklebust 已提交
786
	return req;
L
Linus Torvalds 已提交
787 788
}

789 790 791 792 793 794 795 796 797 798 799
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);
800
	nfs_mark_request_dirty(req);
801
	nfs_unlock_and_release_request(req);
802 803 804
	return 0;
}

L
Linus Torvalds 已提交
805 806
int nfs_flush_incompatible(struct file *file, struct page *page)
{
807
	struct nfs_open_context *ctx = nfs_file_open_context(file);
808
	struct nfs_lock_context *l_ctx;
L
Linus Torvalds 已提交
809
	struct nfs_page	*req;
T
Trond Myklebust 已提交
810
	int do_flush, status;
L
Linus Torvalds 已提交
811 812 813 814 815 816 817 818
	/*
	 * 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
819 820 821 822
	do {
		req = nfs_page_find_request(page);
		if (req == NULL)
			return 0;
823 824
		l_ctx = req->wb_lock_context;
		do_flush = req->wb_page != page || req->wb_context != ctx;
825 826
		/* for now, flush if more than 1 request in page_group */
		do_flush |= req->wb_this_page != req;
827
		if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
828 829 830
			do_flush |= l_ctx->lockowner.l_owner != current->files
				|| l_ctx->lockowner.l_pid != current->tgid;
		}
L
Linus Torvalds 已提交
831
		nfs_release_request(req);
T
Trond Myklebust 已提交
832 833
		if (!do_flush)
			return 0;
834
		status = nfs_wb_page(page_file_mapping(page)->host, page);
T
Trond Myklebust 已提交
835 836
	} while (status == 0);
	return status;
L
Linus Torvalds 已提交
837 838
}

839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
/*
 * Avoid buffered writes when a open context credential's key would
 * expire soon.
 *
 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
 *
 * Return 0 and set a credential flag which triggers the inode to flush
 * and performs  NFS_FILE_SYNC writes if the key will expired within
 * RPC_KEY_EXPIRE_TIMEO.
 */
int
nfs_key_timeout_notify(struct file *filp, struct inode *inode)
{
	struct nfs_open_context *ctx = nfs_file_open_context(filp);
	struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;

	return rpcauth_key_timeout_notify(auth, ctx->cred);
}

/*
 * Test if the open context credential key is marked to expire soon.
 */
bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
{
	return rpcauth_cred_key_to_expire(ctx->cred);
}

866 867 868 869 870
/*
 * 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.
 */
871
static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
872
{
873 874
	struct nfs_inode *nfsi = NFS_I(inode);

875 876
	if (nfs_have_delegated_attributes(inode))
		goto out;
877 878
	if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE))
		return false;
879
	smp_rmb();
880
	if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
881 882 883
		return false;
out:
	return PageUptodate(page) != 0;
884 885
}

886 887 888 889 890
/* If we know the page is up to date, and we're not using byte range locks (or
 * if we have the whole file locked for writing), it may be more efficient to
 * extend the write to cover the entire page in order to avoid fragmentation
 * inefficiencies.
 *
891 892
 * If the file is opened for synchronous writes then we can just skip the rest
 * of the checks.
893 894 895 896 897
 */
static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
{
	if (file->f_flags & O_DSYNC)
		return 0;
898 899
	if (!nfs_write_pageuptodate(page, inode))
		return 0;
900 901
	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
		return 1;
902
	if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
903
			inode->i_flock->fl_end == OFFSET_MAX &&
904
			inode->i_flock->fl_type != F_RDLCK))
905 906 907 908
		return 1;
	return 0;
}

L
Linus Torvalds 已提交
909 910 911 912 913 914 915 916 917
/*
 * 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)
{
918
	struct nfs_open_context *ctx = nfs_file_open_context(file);
919
	struct inode	*inode = page_file_mapping(page)->host;
L
Linus Torvalds 已提交
920 921
	int		status = 0;

C
Chuck Lever 已提交
922 923
	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);

924 925
	dprintk("NFS:       nfs_updatepage(%pD2 %d@%lld)\n",
		file, count, (long long)(page_file_offset(page) + offset));
L
Linus Torvalds 已提交
926

927
	if (nfs_can_extend_write(file, page, inode)) {
928
		count = max(count + offset, nfs_page_length(page));
L
Linus Torvalds 已提交
929 930 931
		offset = 0;
	}

932
	status = nfs_writepage_setup(ctx, page, offset, count);
933 934
	if (status < 0)
		nfs_set_pageerror(page);
935 936
	else
		__set_page_dirty_nobuffers(page);
L
Linus Torvalds 已提交
937

938
	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
L
Linus Torvalds 已提交
939 940 941 942
			status, (long long)i_size_read(inode));
	return status;
}

943
static int flush_task_priority(int how)
L
Linus Torvalds 已提交
944 945 946 947 948 949 950 951 952 953
{
	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
		case FLUSH_HIGHPRI:
			return RPC_PRIORITY_HIGH;
		case FLUSH_LOWPRI:
			return RPC_PRIORITY_LOW;
	}
	return RPC_PRIORITY_NORMAL;
}

954 955
static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg,
			       struct rpc_task_setup *task_setup_data, int how)
L
Linus Torvalds 已提交
956
{
957
	struct inode *inode = data->header->inode;
958
	int priority = flush_task_priority(how);
L
Linus Torvalds 已提交
959

960 961
	task_setup_data->priority = priority;
	NFS_PROTO(inode)->write_setup(data, msg);
962

963
	nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
964
				 &task_setup_data->rpc_client, msg, data);
965 966
}

F
Fred 已提交
967 968 969 970 971 972 973
/* 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);
974
	nfs_unlock_request(req);
975
	nfs_end_page_writeback(req);
976
	nfs_release_request(req);
F
Fred 已提交
977 978
}

979
static void nfs_async_write_error(struct list_head *head)
980 981 982 983 984 985 986 987 988 989
{
	struct nfs_page	*req;

	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_redirty_request(req);
	}
}

990 991 992 993 994
static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
	.error_cleanup = nfs_async_write_error,
	.completion = nfs_write_completion,
};

995
void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
996
			       struct inode *inode, int ioflags, bool force_mds,
997
			       const struct nfs_pgio_completion_ops *compl_ops)
L
Linus Torvalds 已提交
998
{
999
	struct nfs_server *server = NFS_SERVER(inode);
1000
	const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1001 1002 1003 1004 1005

#ifdef CONFIG_NFS_V4_1
	if (server->pnfs_curr_ld && !force_mds)
		pg_ops = server->pnfs_curr_ld->pg_write_ops;
#endif
1006 1007
	nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
			server->wsize, ioflags);
1008
}
B
Bryan Schumaker 已提交
1009
EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
L
Linus Torvalds 已提交
1010

1011 1012
void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
{
1013
	pgio->pg_ops = &nfs_pgio_rw_ops;
1014 1015
	pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
}
1016
EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1017

L
Linus Torvalds 已提交
1018

1019 1020 1021 1022 1023 1024 1025
void nfs_commit_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_commit_data *data = calldata;

	NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
}

1026
static void nfs_writeback_release_common(struct nfs_pgio_data *data)
1027
{
1028
	struct nfs_pgio_header *hdr = data->header;
1029
	int status = data->task.tk_status;
1030

1031 1032 1033 1034 1035
	if ((status >= 0) && nfs_write_need_commit(data)) {
		spin_lock(&hdr->lock);
		if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
			; /* Do nothing */
		else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
1036 1037
			memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf));
		else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf)))
1038 1039
			set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
		spin_unlock(&hdr->lock);
L
Linus Torvalds 已提交
1040 1041 1042
	}
}

1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
/*
 * Special version of should_remove_suid() that ignores capabilities.
 */
static int nfs_should_remove_suid(const struct inode *inode)
{
	umode_t mode = inode->i_mode;
	int kill = 0;

	/* suid always must be killed */
	if (unlikely(mode & S_ISUID))
		kill = ATTR_KILL_SUID;

	/*
	 * sgid without any exec bits is just a mandatory locking mark; leave
	 * it alone.  If some exec bits are set, it's a real sgid; kill it.
	 */
	if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
		kill |= ATTR_KILL_SGID;

	if (unlikely(kill && S_ISREG(mode)))
		return kill;

	return 0;
}
1067

L
Linus Torvalds 已提交
1068 1069 1070
/*
 * This function is called when the WRITE call is complete.
 */
1071 1072
static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data,
			      struct inode *inode)
L
Linus Torvalds 已提交
1073
{
1074
	int status;
L
Linus Torvalds 已提交
1075

1076 1077 1078 1079 1080 1081 1082
	/*
	 * ->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.
	 */
1083
	status = NFS_PROTO(inode)->write_done(task, data);
1084
	if (status != 0)
1085 1086
		return status;
	nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count);
C
Chuck Lever 已提交
1087

B
Bryan Schumaker 已提交
1088
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1089
	if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) {
L
Linus Torvalds 已提交
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
		/* 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;

1100
		/* Note this will print the MDS for a DS write */
L
Linus Torvalds 已提交
1101
		if (time_before(complain, jiffies)) {
1102
			dprintk("NFS:       faulty NFS server %s:"
L
Linus Torvalds 已提交
1103
				" (committed = %d) != (stable = %d)\n",
1104
				NFS_SERVER(inode)->nfs_client->cl_hostname,
1105
				data->res.verf->committed, data->args.stable);
L
Linus Torvalds 已提交
1106 1107 1108 1109
			complain = jiffies + 300 * HZ;
		}
	}
#endif
1110 1111 1112 1113

	/* Deal with the suid/sgid bit corner case */
	if (nfs_should_remove_suid(inode))
		nfs_mark_for_revalidate(inode);
1114 1115 1116 1117 1118 1119 1120 1121 1122 1123
	return 0;
}

/*
 * This function is called when the WRITE call is complete.
 */
static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data)
{
	struct nfs_pgio_args	*argp = &data->args;
	struct nfs_pgio_res	*resp = &data->res;
1124 1125

	if (resp->count < argp->count) {
L
Linus Torvalds 已提交
1126 1127
		static unsigned long    complain;

1128
		/* This a short write! */
1129
		nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE);
C
Chuck Lever 已提交
1130

L
Linus Torvalds 已提交
1131
		/* Has the server at least made some progress? */
1132 1133 1134 1135 1136 1137
		if (resp->count == 0) {
			if (time_before(complain, jiffies)) {
				printk(KERN_WARNING
				       "NFS: Server wrote zero bytes, expected %u.\n",
				       argp->count);
				complain = jiffies + 300 * HZ;
L
Linus Torvalds 已提交
1138
			}
1139 1140
			nfs_set_pgio_error(data->header, -EIO, argp->offset);
			task->tk_status = -EIO;
1141
			return;
L
Linus Torvalds 已提交
1142
		}
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
		/* Was this an NFSv2 write or an NFSv3 stable write? */
		if (resp->verf->committed != NFS_UNSTABLE) {
			/* Resend from where the server left off */
			data->mds_offset += resp->count;
			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;
L
Linus Torvalds 已提交
1155
		}
1156
		rpc_restart_call_prepare(task);
L
Linus Torvalds 已提交
1157 1158 1159 1160
	}
}


B
Bryan Schumaker 已提交
1161
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1162 1163
static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
{
1164 1165
	int ret;

1166 1167
	if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
		return 1;
1168 1169 1170 1171 1172 1173 1174
	if (!may_wait)
		return 0;
	ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
				NFS_INO_COMMIT,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
	return (ret < 0) ? ret : 1;
1175 1176
}

1177
static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1178 1179 1180 1181 1182 1183
{
	clear_bit(NFS_INO_COMMIT, &nfsi->flags);
	smp_mb__after_clear_bit();
	wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
}

1184
void nfs_commitdata_release(struct nfs_commit_data *data)
L
Linus Torvalds 已提交
1185
{
1186 1187
	put_nfs_open_context(data->context);
	nfs_commit_free(data);
L
Linus Torvalds 已提交
1188
}
1189
EXPORT_SYMBOL_GPL(nfs_commitdata_release);
L
Linus Torvalds 已提交
1190

1191
int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1192
			const struct rpc_call_ops *call_ops,
1193
			int how, int flags)
L
Linus Torvalds 已提交
1194
{
1195
	struct rpc_task *task;
1196
	int priority = flush_task_priority(how);
1197 1198 1199
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
1200
		.rpc_cred = data->cred,
1201
	};
1202
	struct rpc_task_setup task_setup_data = {
1203
		.task = &data->task,
1204
		.rpc_client = clnt,
1205
		.rpc_message = &msg,
1206
		.callback_ops = call_ops,
1207
		.callback_data = data,
1208
		.workqueue = nfsiod_workqueue,
1209
		.flags = RPC_TASK_ASYNC | flags,
1210
		.priority = priority,
1211
	};
1212 1213 1214 1215 1216
	/* Set up the initial task struct.  */
	NFS_PROTO(data->inode)->commit_setup(data, &msg);

	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);

1217 1218 1219
	nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
		NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);

1220 1221 1222 1223 1224 1225 1226 1227
	task = rpc_run_task(&task_setup_data);
	if (IS_ERR(task))
		return PTR_ERR(task);
	if (how & FLUSH_SYNC)
		rpc_wait_for_completion_task(task);
	rpc_put_task(task);
	return 0;
}
1228
EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1229 1230 1231 1232

/*
 * Set up the argument/result storage required for the RPC call.
 */
1233
void nfs_init_commit(struct nfs_commit_data *data,
1234 1235 1236
		     struct list_head *head,
		     struct pnfs_layout_segment *lseg,
		     struct nfs_commit_info *cinfo)
1237 1238
{
	struct nfs_page *first = nfs_list_entry(head->next);
1239
	struct inode *inode = first->wb_context->dentry->d_inode;
L
Linus Torvalds 已提交
1240 1241 1242 1243 1244 1245 1246

	/* 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;
1247
	data->cred	  = first->wb_context->cred;
1248
	data->lseg	  = lseg; /* reference transferred */
1249
	data->mds_ops     = &nfs_commit_ops;
1250
	data->completion_ops = cinfo->completion_ops;
F
Fred Isaman 已提交
1251
	data->dreq	  = cinfo->dreq;
L
Linus Torvalds 已提交
1252 1253

	data->args.fh     = NFS_FH(data->inode);
1254 1255 1256
	/* Note: we always request a commit of the entire inode */
	data->args.offset = 0;
	data->args.count  = 0;
1257
	data->context     = get_nfs_open_context(first->wb_context);
L
Linus Torvalds 已提交
1258 1259
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
1260
	nfs_fattr_init(&data->fattr);
L
Linus Torvalds 已提交
1261
}
1262
EXPORT_SYMBOL_GPL(nfs_init_commit);
L
Linus Torvalds 已提交
1263

1264
void nfs_retry_commit(struct list_head *page_list,
F
Fred Isaman 已提交
1265 1266
		      struct pnfs_layout_segment *lseg,
		      struct nfs_commit_info *cinfo)
1267 1268 1269 1270 1271 1272
{
	struct nfs_page *req;

	while (!list_empty(page_list)) {
		req = nfs_list_entry(page_list->next);
		nfs_list_remove_request(req);
F
Fred Isaman 已提交
1273
		nfs_mark_request_commit(req, lseg, cinfo);
1274 1275
		if (!cinfo->dreq) {
			dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1276
			dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1277 1278
				     BDI_RECLAIMABLE);
		}
1279
		nfs_unlock_and_release_request(req);
1280 1281
	}
}
1282
EXPORT_SYMBOL_GPL(nfs_retry_commit);
1283

L
Linus Torvalds 已提交
1284 1285 1286 1287
/*
 * Commit dirty pages
 */
static int
F
Fred Isaman 已提交
1288 1289
nfs_commit_list(struct inode *inode, struct list_head *head, int how,
		struct nfs_commit_info *cinfo)
L
Linus Torvalds 已提交
1290
{
1291
	struct nfs_commit_data	*data;
L
Linus Torvalds 已提交
1292

1293
	data = nfs_commitdata_alloc();
L
Linus Torvalds 已提交
1294 1295 1296 1297 1298

	if (!data)
		goto out_bad;

	/* Set up the argument struct */
1299 1300
	nfs_init_commit(data, head, NULL, cinfo);
	atomic_inc(&cinfo->mds->rpcs_out);
1301 1302
	return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
				   how, 0);
L
Linus Torvalds 已提交
1303
 out_bad:
F
Fred Isaman 已提交
1304
	nfs_retry_commit(head, NULL, cinfo);
1305
	cinfo->completion_ops->error_cleanup(NFS_I(inode));
L
Linus Torvalds 已提交
1306 1307 1308 1309 1310 1311
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
1312
static void nfs_commit_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
1313
{
1314
	struct nfs_commit_data	*data = calldata;
L
Linus Torvalds 已提交
1315

C
Chuck Lever 已提交
1316
        dprintk("NFS: %5u nfs_commit_done (status %d)\n",
L
Linus Torvalds 已提交
1317 1318
                                task->tk_pid, task->tk_status);

1319
	/* Call the NFS version-specific code */
1320
	NFS_PROTO(data->inode)->commit_done(task, data);
1321 1322
}

1323
static void nfs_commit_release_pages(struct nfs_commit_data *data)
1324
{
1325
	struct nfs_page	*req;
1326
	int status = data->task.tk_status;
1327
	struct nfs_commit_info cinfo;
1328

L
Linus Torvalds 已提交
1329 1330 1331
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
F
Fred Isaman 已提交
1332
		nfs_clear_page_commit(req->wb_page);
L
Linus Torvalds 已提交
1333

1334
		dprintk("NFS:       commit (%s/%llu %d@%lld)",
1335
			req->wb_context->dentry->d_sb->s_id,
1336
			(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
L
Linus Torvalds 已提交
1337 1338
			req->wb_bytes,
			(long long)req_offset(req));
1339 1340
		if (status < 0) {
			nfs_context_set_write_error(req->wb_context, status);
L
Linus Torvalds 已提交
1341
			nfs_inode_remove_request(req);
1342
			dprintk(", error = %d\n", status);
L
Linus Torvalds 已提交
1343 1344 1345 1346 1347
			goto next;
		}

		/* Okay, COMMIT succeeded, apparently. Check the verifier
		 * returned by the server against all stored verfs. */
1348
		if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
L
Linus Torvalds 已提交
1349 1350 1351 1352 1353 1354 1355
			/* 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
1356
		nfs_mark_request_dirty(req);
1357
		set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
L
Linus Torvalds 已提交
1358
	next:
1359
		nfs_unlock_and_release_request(req);
L
Linus Torvalds 已提交
1360
	}
1361 1362 1363
	nfs_init_cinfo(&cinfo, data->inode, data->dreq);
	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_commit_clear_lock(NFS_I(data->inode));
1364 1365 1366 1367
}

static void nfs_commit_release(void *calldata)
{
1368
	struct nfs_commit_data *data = calldata;
1369

1370
	data->completion_ops->completion(data);
1371
	nfs_commitdata_release(calldata);
L
Linus Torvalds 已提交
1372
}
1373 1374

static const struct rpc_call_ops nfs_commit_ops = {
1375
	.rpc_call_prepare = nfs_commit_prepare,
1376 1377 1378
	.rpc_call_done = nfs_commit_done,
	.rpc_release = nfs_commit_release,
};
L
Linus Torvalds 已提交
1379

1380 1381 1382 1383 1384
static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
	.completion = nfs_commit_release_pages,
	.error_cleanup = nfs_commit_clear_lock,
};

1385 1386
int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
			    int how, struct nfs_commit_info *cinfo)
1387 1388 1389
{
	int status;

F
Fred Isaman 已提交
1390
	status = pnfs_commit_list(inode, head, how, cinfo);
1391
	if (status == PNFS_NOT_ATTEMPTED)
F
Fred Isaman 已提交
1392
		status = nfs_commit_list(inode, head, how, cinfo);
1393 1394 1395
	return status;
}

1396
int nfs_commit_inode(struct inode *inode, int how)
L
Linus Torvalds 已提交
1397 1398
{
	LIST_HEAD(head);
F
Fred Isaman 已提交
1399
	struct nfs_commit_info cinfo;
1400
	int may_wait = how & FLUSH_SYNC;
1401
	int res;
L
Linus Torvalds 已提交
1402

1403 1404
	res = nfs_commit_set_lock(NFS_I(inode), may_wait);
	if (res <= 0)
1405
		goto out_mark_dirty;
F
Fred Isaman 已提交
1406 1407
	nfs_init_cinfo_from_inode(&cinfo, inode);
	res = nfs_scan_commit(inode, &head, &cinfo);
L
Linus Torvalds 已提交
1408
	if (res) {
1409 1410
		int error;

F
Fred Isaman 已提交
1411
		error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1412 1413
		if (error < 0)
			return error;
1414
		if (!may_wait)
1415
			goto out_mark_dirty;
1416 1417 1418 1419 1420 1421
		error = wait_on_bit(&NFS_I(inode)->flags,
				NFS_INO_COMMIT,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
		if (error < 0)
			return error;
1422 1423
	} else
		nfs_commit_clear_lock(NFS_I(inode));
1424 1425 1426 1427 1428 1429 1430 1431
	return res;
	/* Note: If we exit without ensuring that the commit is complete,
	 * we must mark the inode as dirty. Otherwise, future calls to
	 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
	 * that the data is on the disk.
	 */
out_mark_dirty:
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
L
Linus Torvalds 已提交
1432 1433
	return res;
}
1434 1435 1436

static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
1437 1438 1439
	struct nfs_inode *nfsi = NFS_I(inode);
	int flags = FLUSH_SYNC;
	int ret = 0;
1440

1441
	/* no commits means nothing needs to be done */
F
Fred Isaman 已提交
1442
	if (!nfsi->commit_info.ncommit)
1443 1444
		return ret;

1445 1446 1447 1448
	if (wbc->sync_mode == WB_SYNC_NONE) {
		/* Don't commit yet if this is a non-blocking flush and there
		 * are a lot of outstanding writes for this mapping.
		 */
F
Fred Isaman 已提交
1449
		if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
1450
			goto out_mark_dirty;
1451

1452
		/* don't wait for the COMMIT response */
1453
		flags = 0;
1454 1455
	}

1456 1457 1458 1459 1460 1461 1462 1463
	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;
		}
1464
		return 0;
1465 1466
	}
out_mark_dirty:
1467 1468 1469
	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
	return ret;
}
1470
#else
1471 1472 1473 1474
static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
{
	return 0;
}
L
Linus Torvalds 已提交
1475 1476
#endif

1477 1478
int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
{
1479 1480
	return nfs_commit_unstable_pages(inode, wbc);
}
B
Bryan Schumaker 已提交
1481
EXPORT_SYMBOL_GPL(nfs_write_inode);
1482

1483 1484 1485 1486
/*
 * flush the inode to disk.
 */
int nfs_wb_all(struct inode *inode)
T
Trond Myklebust 已提交
1487 1488
{
	struct writeback_control wbc = {
1489
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1490
		.nr_to_write = LONG_MAX,
1491 1492
		.range_start = 0,
		.range_end = LLONG_MAX,
T
Trond Myklebust 已提交
1493
	};
1494 1495 1496 1497 1498
	int ret;

	trace_nfs_writeback_inode_enter(inode);

	ret = sync_inode(inode, &wbc);
T
Trond Myklebust 已提交
1499

1500 1501
	trace_nfs_writeback_inode_exit(inode, ret);
	return ret;
1502
}
B
Bryan Schumaker 已提交
1503
EXPORT_SYMBOL_GPL(nfs_wb_all);
1504

1505 1506 1507 1508 1509 1510
int nfs_wb_page_cancel(struct inode *inode, struct page *page)
{
	struct nfs_page *req;
	int ret = 0;

	for (;;) {
1511
		wait_on_page_writeback(page);
1512 1513 1514
		req = nfs_page_find_request(page);
		if (req == NULL)
			break;
1515
		if (nfs_lock_request(req)) {
1516
			nfs_clear_request_commit(req);
1517 1518 1519 1520 1521 1522
			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);
1523
			nfs_unlock_and_release_request(req);
1524 1525 1526
			break;
		}
		ret = nfs_wait_on_request(req);
1527
		nfs_release_request(req);
1528
		if (ret < 0)
1529
			break;
1530 1531 1532 1533
	}
	return ret;
}

T
Trond Myklebust 已提交
1534 1535 1536 1537
/*
 * Write back all requests on one page - we do this before reading it.
 */
int nfs_wb_page(struct inode *inode, struct page *page)
1538
{
1539
	loff_t range_start = page_file_offset(page);
1540
	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1541 1542
	struct writeback_control wbc = {
		.sync_mode = WB_SYNC_ALL,
T
Trond Myklebust 已提交
1543
		.nr_to_write = 0,
1544 1545 1546 1547
		.range_start = range_start,
		.range_end = range_end,
	};
	int ret;
1548

1549 1550
	trace_nfs_writeback_page_enter(inode);

1551
	for (;;) {
1552
		wait_on_page_writeback(page);
1553 1554 1555 1556
		if (clear_page_dirty_for_io(page)) {
			ret = nfs_writepage_locked(page, &wbc);
			if (ret < 0)
				goto out_error;
1557
			continue;
T
Trond Myklebust 已提交
1558
		}
1559
		ret = 0;
1560 1561 1562
		if (!PagePrivate(page))
			break;
		ret = nfs_commit_inode(inode, FLUSH_SYNC);
1563
		if (ret < 0)
1564
			goto out_error;
T
Trond Myklebust 已提交
1565
	}
1566
out_error:
1567
	trace_nfs_writeback_page_exit(inode, ret);
1568
	return ret;
1569 1570
}

1571 1572
#ifdef CONFIG_MIGRATION
int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1573
		struct page *page, enum migrate_mode mode)
1574
{
1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
	/*
	 * If PagePrivate is set, then the page is currently associated with
	 * an in-progress read or write request. Don't try to migrate it.
	 *
	 * FIXME: we could do this in principle, but we'll need a way to ensure
	 *        that we can safely release the inode reference while holding
	 *        the page lock.
	 */
	if (PagePrivate(page))
		return -EBUSY;
1585

1586 1587
	if (!nfs_fscache_release_page(page, GFP_KERNEL))
		return -EBUSY;
1588

1589
	return migrate_page(mapping, newpage, page, mode);
1590 1591 1592
}
#endif

D
David Howells 已提交
1593
int __init nfs_init_writepagecache(void)
L
Linus Torvalds 已提交
1594 1595
{
	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1596
					     sizeof(struct nfs_rw_header),
L
Linus Torvalds 已提交
1597
					     0, SLAB_HWCACHE_ALIGN,
1598
					     NULL);
L
Linus Torvalds 已提交
1599 1600 1601
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

1602 1603
	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
						     nfs_wdata_cachep);
L
Linus Torvalds 已提交
1604
	if (nfs_wdata_mempool == NULL)
1605
		goto out_destroy_write_cache;
L
Linus Torvalds 已提交
1606

1607 1608 1609 1610 1611
	nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
					     sizeof(struct nfs_commit_data),
					     0, SLAB_HWCACHE_ALIGN,
					     NULL);
	if (nfs_cdata_cachep == NULL)
1612
		goto out_destroy_write_mempool;
1613

1614
	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1615
						      nfs_cdata_cachep);
L
Linus Torvalds 已提交
1616
	if (nfs_commit_mempool == NULL)
1617
		goto out_destroy_commit_cache;
L
Linus Torvalds 已提交
1618

1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638
	/*
	 * 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
1639
	return 0;
1640 1641 1642 1643 1644 1645 1646 1647

out_destroy_commit_cache:
	kmem_cache_destroy(nfs_cdata_cachep);
out_destroy_write_mempool:
	mempool_destroy(nfs_wdata_mempool);
out_destroy_write_cache:
	kmem_cache_destroy(nfs_wdata_cachep);
	return -ENOMEM;
L
Linus Torvalds 已提交
1648 1649
}

1650
void nfs_destroy_writepagecache(void)
L
Linus Torvalds 已提交
1651 1652
{
	mempool_destroy(nfs_commit_mempool);
1653
	kmem_cache_destroy(nfs_cdata_cachep);
L
Linus Torvalds 已提交
1654
	mempool_destroy(nfs_wdata_mempool);
1655
	kmem_cache_destroy(nfs_wdata_cachep);
L
Linus Torvalds 已提交
1656 1657
}

1658
static const struct nfs_rw_ops nfs_rw_write_ops = {
1659
	.rw_mode		= FMODE_WRITE,
1660 1661
	.rw_alloc_header	= nfs_writehdr_alloc,
	.rw_free_header		= nfs_writehdr_free,
1662
	.rw_release		= nfs_writeback_release_common,
1663 1664
	.rw_done		= nfs_writeback_done,
	.rw_result		= nfs_writeback_result,
1665
	.rw_initiate		= nfs_initiate_write,
1666
};