direct.c 28.8 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9
/*
 * linux/fs/nfs/direct.c
 *
 * Copyright (C) 2003 by Chuck Lever <cel@netapp.com>
 *
 * High-performance uncached I/O for the Linux NFS client
 *
 * There are important applications whose performance or correctness
 * depends on uncached access to file data.  Database clusters
10
 * (multiple copies of the same instance running on separate hosts)
L
Linus Torvalds 已提交
11
 * implement their own cache coherency protocol that subsumes file
12 13 14
 * system cache protocols.  Applications that process datasets
 * considerably larger than the client's memory do not always benefit
 * from a local cache.  A streaming video server, for instance, has no
L
Linus Torvalds 已提交
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 * need to cache the contents of a file.
 *
 * When an application requests uncached I/O, all read and write requests
 * are made directly to the server; data stored or fetched via these
 * requests is not cached in the Linux page cache.  The client does not
 * correct unaligned requests from applications.  All requested bytes are
 * held on permanent storage before a direct write system call returns to
 * an application.
 *
 * Solaris implements an uncached I/O facility called directio() that
 * is used for backups and sequential I/O to very large files.  Solaris
 * also supports uncaching whole NFS partitions with "-o forcedirectio,"
 * an undocumented mount option.
 *
 * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with
 * help from Andrew Morton.
 *
 * 18 Dec 2001	Initial implementation for 2.4  --cel
 * 08 Jul 2002	Version for 2.4.19, with bug fixes --trondmy
 * 08 Jun 2003	Port to 2.5 APIs  --cel
 * 31 Mar 2004	Handle direct I/O without VFS support  --cel
 * 15 Sep 2004	Parallel async reads  --cel
37
 * 04 May 2005	support O_DIRECT with aio  --cel
L
Linus Torvalds 已提交
38 39 40 41 42 43 44 45 46
 *
 */

#include <linux/errno.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/file.h>
#include <linux/pagemap.h>
#include <linux/kref.h>
47
#include <linux/slab.h>
48
#include <linux/task_io_accounting_ops.h>
49
#include <linux/module.h>
L
Linus Torvalds 已提交
50 51 52 53 54 55

#include <linux/nfs_fs.h>
#include <linux/nfs_page.h>
#include <linux/sunrpc/clnt.h>

#include <asm/uaccess.h>
A
Arun Sharma 已提交
56
#include <linux/atomic.h>
L
Linus Torvalds 已提交
57

58
#include "internal.h"
C
Chuck Lever 已提交
59
#include "iostat.h"
60
#include "pnfs.h"
C
Chuck Lever 已提交
61

L
Linus Torvalds 已提交
62 63
#define NFSDBG_FACILITY		NFSDBG_VFS

64
static struct kmem_cache *nfs_direct_cachep;
L
Linus Torvalds 已提交
65 66 67 68

/*
 * This represents a set of asynchronous requests that we're waiting on
 */
69 70 71 72
struct nfs_direct_mirror {
	ssize_t count;
};

L
Linus Torvalds 已提交
73 74
struct nfs_direct_req {
	struct kref		kref;		/* release manager */
75 76

	/* I/O parameters */
77
	struct nfs_open_context	*ctx;		/* file open context info */
78
	struct nfs_lock_context *l_ctx;		/* Lock context info */
79
	struct kiocb *		iocb;		/* controlling i/o request */
80
	struct inode *		inode;		/* target file of i/o */
81 82

	/* completion state */
83
	atomic_t		io_count;	/* i/os we're waiting for */
84
	spinlock_t		lock;		/* protect completion state */
85 86 87 88

	struct nfs_direct_mirror mirrors[NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX];
	int			mirror_count;

89
	ssize_t			count,		/* bytes actually processed */
P
Peng Tao 已提交
90
				bytes_left,	/* bytes left to be sent */
91
				io_start,	/* start of IO */
L
Linus Torvalds 已提交
92
				error;		/* any reported error */
93
	struct completion	completion;	/* wait for i/o completion */
94 95

	/* commit state */
96 97 98
	struct nfs_mds_commit_info mds_cinfo;	/* Storage for cinfo */
	struct pnfs_ds_commit_info ds_cinfo;	/* Storage for cinfo */
	struct work_struct	work;
99 100 101 102
	int			flags;
#define NFS_ODIRECT_DO_COMMIT		(1)	/* an unstable reply was received */
#define NFS_ODIRECT_RESCHED_WRITES	(2)	/* write verification failed */
	struct nfs_writeverf	verf;		/* unstable write verifier */
L
Linus Torvalds 已提交
103 104
};

105 106
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops;
static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops;
107
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
108
static void nfs_direct_write_schedule_work(struct work_struct *work);
109 110 111 112 113 114 115 116 117 118 119

static inline void get_dreq(struct nfs_direct_req *dreq)
{
	atomic_inc(&dreq->io_count);
}

static inline int put_dreq(struct nfs_direct_req *dreq)
{
	return atomic_dec_and_test(&dreq->io_count);
}

120 121 122 123 124 125
void nfs_direct_set_resched_writes(struct nfs_direct_req *dreq)
{
	dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
}
EXPORT_SYMBOL_GPL(nfs_direct_set_resched_writes);

126 127 128 129 130 131 132 133
static void
nfs_direct_good_bytes(struct nfs_direct_req *dreq, struct nfs_pgio_header *hdr)
{
	int i;
	ssize_t count;

	WARN_ON_ONCE(hdr->pgio_mirror_idx >= dreq->mirror_count);

134 135 136 137 138
	count = dreq->mirrors[hdr->pgio_mirror_idx].count;
	if (count + dreq->io_start < hdr->io_start + hdr->good_bytes) {
		count = hdr->io_start + hdr->good_bytes - dreq->io_start;
		dreq->mirrors[hdr->pgio_mirror_idx].count = count;
	}
139 140 141 142 143 144 145 146 147 148 149

	/* update the dreq->count by finding the minimum agreed count from all
	 * mirrors */
	count = dreq->mirrors[0].count;

	for (i = 1; i < dreq->mirror_count; i++)
		count = min(count, dreq->mirrors[i].count);

	dreq->count = count;
}

150 151 152 153
/*
 * nfs_direct_select_verf - select the right verifier
 * @dreq - direct request possibly spanning multiple servers
 * @ds_clp - nfs_client of data server or NULL if MDS / non-pnfs
154
 * @commit_idx - commit bucket index for the DS
155 156 157 158 159 160
 *
 * returns the correct verifier to use given the role of the server
 */
static struct nfs_writeverf *
nfs_direct_select_verf(struct nfs_direct_req *dreq,
		       struct nfs_client *ds_clp,
161
		       int commit_idx)
162 163 164 165 166 167
{
	struct nfs_writeverf *verfp = &dreq->verf;

#ifdef CONFIG_NFS_V4_1
	if (ds_clp) {
		/* pNFS is in use, use the DS verf */
168 169
		if (commit_idx >= 0 && commit_idx < dreq->ds_cinfo.nbuckets)
			verfp = &dreq->ds_cinfo.buckets[commit_idx].direct_verf;
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
		else
			WARN_ON_ONCE(1);
	}
#endif
	return verfp;
}


/*
 * nfs_direct_set_hdr_verf - set the write/commit verifier
 * @dreq - direct request possibly spanning multiple servers
 * @hdr - pageio header to validate against previously seen verfs
 *
 * Set the server's (MDS or DS) "seen" verifier
 */
static void nfs_direct_set_hdr_verf(struct nfs_direct_req *dreq,
				    struct nfs_pgio_header *hdr)
{
	struct nfs_writeverf *verfp;

190
	verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
	WARN_ON_ONCE(verfp->committed >= 0);
	memcpy(verfp, &hdr->verf, sizeof(struct nfs_writeverf));
	WARN_ON_ONCE(verfp->committed < 0);
}

/*
 * nfs_direct_cmp_hdr_verf - compare verifier for pgio header
 * @dreq - direct request possibly spanning multiple servers
 * @hdr - pageio header to validate against previously seen verf
 *
 * set the server's "seen" verf if not initialized.
 * returns result of comparison between @hdr->verf and the "seen"
 * verf of the server used by @hdr (DS or MDS)
 */
static int nfs_direct_set_or_cmp_hdr_verf(struct nfs_direct_req *dreq,
					  struct nfs_pgio_header *hdr)
{
	struct nfs_writeverf *verfp;

210
	verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
	if (verfp->committed < 0) {
		nfs_direct_set_hdr_verf(dreq, hdr);
		return 0;
	}
	return memcmp(verfp, &hdr->verf, sizeof(struct nfs_writeverf));
}

/*
 * nfs_direct_cmp_commit_data_verf - compare verifier for commit data
 * @dreq - direct request possibly spanning multiple servers
 * @data - commit data to validate against previously seen verf
 *
 * returns result of comparison between @data->verf and the verf of
 * the server used by @data (DS or MDS)
 */
static int nfs_direct_cmp_commit_data_verf(struct nfs_direct_req *dreq,
					   struct nfs_commit_data *data)
{
	struct nfs_writeverf *verfp;

	verfp = nfs_direct_select_verf(dreq, data->ds_clp,
					 data->ds_commit_index);
233 234 235 236 237

	/* verifier not set so always fail */
	if (verfp->committed < 0)
		return 1;

238 239 240
	return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf));
}

L
Linus Torvalds 已提交
241
/**
242 243 244 245 246 247 248
 * nfs_direct_IO - NFS address space operation for direct I/O
 * @iocb: target I/O control block
 * @iov: array of vectors that define I/O buffer
 * @pos: offset in file to begin the operation
 * @nr_segs: size of iovec array
 *
 * The presence of this routine in the address space ops vector means
M
Mel Gorman 已提交
249 250 251
 * the NFS client supports direct I/O. However, for most direct IO, we
 * shunt off direct read and write requests before the VFS gets them,
 * so this method is only ever called for swap.
L
Linus Torvalds 已提交
252
 */
253
ssize_t nfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t pos)
254
{
255 256 257 258 259 260
	struct inode *inode = iocb->ki_filp->f_mapping->host;

	/* we only support swap file calling nfs_direct_IO */
	if (!IS_SWAPFILE(inode))
		return 0;

M
Mel Gorman 已提交
261
#ifndef CONFIG_NFS_SWAP
262
	dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n",
A
Al Viro 已提交
263
			iocb->ki_filp, (long long) pos, iter->nr_segs);
264 265

	return -EINVAL;
M
Mel Gorman 已提交
266
#else
C
Christoph Hellwig 已提交
267
	VM_BUG_ON(iov_iter_count(iter) != PAGE_SIZE);
M
Mel Gorman 已提交
268

269
	if (iov_iter_rw(iter) == READ)
M
Martin K. Petersen 已提交
270
		return nfs_file_direct_read(iocb, iter, pos);
271
	return nfs_file_direct_write(iocb, iter);
M
Mel Gorman 已提交
272
#endif /* CONFIG_NFS_SWAP */
273 274
}

275
static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
276
{
277
	unsigned int i;
278 279
	for (i = 0; i < npages; i++)
		page_cache_release(pages[i]);
280 281
}

282 283 284
void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo,
			      struct nfs_direct_req *dreq)
{
285
	cinfo->lock = &dreq->inode->i_lock;
286 287 288 289 290 291
	cinfo->mds = &dreq->mds_cinfo;
	cinfo->ds = &dreq->ds_cinfo;
	cinfo->dreq = dreq;
	cinfo->completion_ops = &nfs_direct_commit_completion_ops;
}

292 293 294 295 296 297 298 299 300 301 302 303
static inline void nfs_direct_setup_mirroring(struct nfs_direct_req *dreq,
					     struct nfs_pageio_descriptor *pgio,
					     struct nfs_page *req)
{
	int mirror_count = 1;

	if (pgio->pg_ops->pg_get_mirror_count)
		mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);

	dreq->mirror_count = mirror_count;
}

304
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
L
Linus Torvalds 已提交
305
{
306 307
	struct nfs_direct_req *dreq;

308
	dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL);
309 310 311 312
	if (!dreq)
		return NULL;

	kref_init(&dreq->kref);
313
	kref_get(&dreq->kref);
314
	init_completion(&dreq->completion);
315
	INIT_LIST_HEAD(&dreq->mds_cinfo.list);
316
	dreq->verf.committed = NFS_INVALID_STABLE_HOW;	/* not set yet */
317
	INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
318
	dreq->mirror_count = 1;
319
	spin_lock_init(&dreq->lock);
320 321

	return dreq;
L
Linus Torvalds 已提交
322 323
}

324
static void nfs_direct_req_free(struct kref *kref)
L
Linus Torvalds 已提交
325 326
{
	struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
327

P
Peng Tao 已提交
328
	nfs_free_pnfs_ds_cinfo(&dreq->ds_cinfo);
329 330
	if (dreq->l_ctx != NULL)
		nfs_put_lock_context(dreq->l_ctx);
331 332
	if (dreq->ctx != NULL)
		put_nfs_open_context(dreq->ctx);
L
Linus Torvalds 已提交
333 334 335
	kmem_cache_free(nfs_direct_cachep, dreq);
}

336 337 338 339 340
static void nfs_direct_req_release(struct nfs_direct_req *dreq)
{
	kref_put(&dreq->kref, nfs_direct_req_free);
}

341 342 343 344 345 346
ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq)
{
	return dreq->bytes_left;
}
EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left);

347 348 349 350 351
/*
 * Collects and returns the final error value/byte-count.
 */
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
352
	ssize_t result = -EIOCBQUEUED;
353 354 355 356 357

	/* Async requests don't wait here */
	if (dreq->iocb)
		goto out;

358
	result = wait_for_completion_killable(&dreq->completion);
359 360

	if (!result)
361
		result = dreq->error;
362
	if (!result)
363
		result = dreq->count;
364 365 366 367 368

out:
	return (ssize_t) result;
}

369
/*
370 371
 * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
 * the iocb is still valid here if this is a synchronous request.
372
 */
373
static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write)
374
{
375 376
	struct inode *inode = dreq->inode;

377
	if (dreq->iocb && write) {
378
		loff_t pos = dreq->iocb->ki_pos + dreq->count;
379 380 381 382 383 384 385

		spin_lock(&inode->i_lock);
		if (i_size_read(inode) < pos)
			i_size_write(inode, pos);
		spin_unlock(&inode->i_lock);
	}

386
	if (write)
387
		nfs_zap_mapping(inode, inode->i_mapping);
388 389

	inode_dio_done(inode);
390 391

	if (dreq->iocb) {
392
		long res = (long) dreq->error;
393
		if (!res)
394
			res = (long) dreq->count;
395
		dreq->iocb->ki_complete(dreq->iocb, res, 0);
396
	}
397

398
	complete_all(&dreq->completion);
399

400
	nfs_direct_req_release(dreq);
401 402
}

T
Trond Myklebust 已提交
403
static void nfs_direct_readpage_release(struct nfs_page *req)
L
Linus Torvalds 已提交
404
{
405
	dprintk("NFS: direct read done (%s/%llu %d@%lld)\n",
406
		req->wb_context->dentry->d_inode->i_sb->s_id,
407
		(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
408 409 410
		req->wb_bytes,
		(long long)req_offset(req));
	nfs_release_request(req);
411 412
}

413
static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
414
{
415 416
	unsigned long bytes = 0;
	struct nfs_direct_req *dreq = hdr->dreq;
417

418 419
	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out_put;
420 421

	spin_lock(&dreq->lock);
422 423 424
	if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0))
		dreq->error = hdr->error;
	else
425 426
		nfs_direct_good_bytes(dreq, hdr);

427 428
	spin_unlock(&dreq->lock);

429 430 431 432
	while (!list_empty(&hdr->pages)) {
		struct nfs_page *req = nfs_list_entry(hdr->pages.next);
		struct page *page = req->wb_page;

433 434
		if (!PageCompound(page) && bytes < hdr->good_bytes)
			set_page_dirty(page);
435 436 437
		bytes += req->wb_bytes;
		nfs_list_remove_request(req);
		nfs_direct_readpage_release(req);
438
	}
439
out_put:
440
	if (put_dreq(dreq))
441
		nfs_direct_complete(dreq, false);
442
	hdr->release(hdr);
L
Linus Torvalds 已提交
443 444
}

445
static void nfs_read_sync_pgio_error(struct list_head *head)
446
{
447
	struct nfs_page *req;
448

449 450 451 452 453
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_release_request(req);
	}
454 455
}

456 457 458 459 460 461
static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr)
{
	get_dreq(hdr->dreq);
}

static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = {
462
	.error_cleanup = nfs_read_sync_pgio_error,
463 464 465 466
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_read_completion,
};

467
/*
468 469 470 471 472
 * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 * bail and stop sending more reads.  Read length accounting is
 * handled automatically by nfs_direct_read_result().  Otherwise, if
 * no requests have been sent, just return an error.
L
Linus Torvalds 已提交
473 474
 */

A
Al Viro 已提交
475 476 477 478 479 480 481 482 483
static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
					      struct iov_iter *iter,
					      loff_t pos)
{
	struct nfs_pageio_descriptor desc;
	struct inode *inode = dreq->inode;
	ssize_t result = -EINVAL;
	size_t requested_bytes = 0;
	size_t rsize = max_t(size_t, NFS_SERVER(inode)->rsize, PAGE_SIZE);
484

485
	nfs_pageio_init_read(&desc, dreq->inode, false,
A
Al Viro 已提交
486 487 488 489
			     &nfs_direct_read_completion_ops);
	get_dreq(dreq);
	desc.pg_dreq = dreq;
	atomic_inc(&inode->i_dio_count);
M
Mel Gorman 已提交
490

A
Al Viro 已提交
491 492 493 494 495
	while (iov_iter_count(iter)) {
		struct page **pagevec;
		size_t bytes;
		size_t pgbase;
		unsigned npages, i;
496

A
Al Viro 已提交
497 498 499 500 501 502 503 504
		result = iov_iter_get_pages_alloc(iter, &pagevec, 
						  rsize, &pgbase);
		if (result < 0)
			break;
	
		bytes = result;
		iov_iter_advance(iter, bytes);
		npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE;
505 506
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
507
			unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
508
			/* XXX do we need to do the eof zeroing found in async_filler? */
509
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
510 511 512 513 514 515 516
						 pgbase, req_len);
			if (IS_ERR(req)) {
				result = PTR_ERR(req);
				break;
			}
			req->wb_index = pos >> PAGE_SHIFT;
			req->wb_offset = pos & ~PAGE_MASK;
A
Al Viro 已提交
517 518
			if (!nfs_pageio_add_request(&desc, req)) {
				result = desc.pg_error;
519 520 521 522 523
				nfs_release_request(req);
				break;
			}
			pgbase = 0;
			bytes -= req_len;
A
Al Viro 已提交
524
			requested_bytes += req_len;
525
			pos += req_len;
P
Peng Tao 已提交
526
			dreq->bytes_left -= req_len;
527
		}
528
		nfs_direct_release_pages(pagevec, npages);
A
Al Viro 已提交
529
		kvfree(pagevec);
530 531 532 533
		if (result < 0)
			break;
	}

534 535
	nfs_pageio_complete(&desc);

536 537 538 539 540
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
541
		inode_dio_done(inode);
542 543 544 545
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

546
	if (put_dreq(dreq))
547
		nfs_direct_complete(dreq, false);
548
	return 0;
549 550
}

551 552 553
/**
 * nfs_file_direct_read - file direct read operation for NFS files
 * @iocb: target I/O control block
554
 * @iter: vector of user buffers into which to read data
555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570
 * @pos: byte offset in file where reading starts
 *
 * We use this function for direct reads instead of calling
 * generic_file_aio_read() in order to avoid gfar's check to see if
 * the request starts before the end of the file.  For that check
 * to work, we must generate a GETATTR before each direct read, and
 * even then there is a window between the GETATTR and the subsequent
 * READ where the file size could change.  Our preference is simply
 * to do all reads the application wants, and the server will take
 * care of managing the end of file boundary.
 *
 * This function also eliminates unnecessarily updating the file's
 * atime locally, as the NFS server sets the file's atime, and this
 * client must read the updated atime from the server back into its
 * cache.
 */
571
ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter,
M
Martin K. Petersen 已提交
572
				loff_t pos)
L
Linus Torvalds 已提交
573
{
574 575 576
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
	struct inode *inode = mapping->host;
L
Linus Torvalds 已提交
577
	struct nfs_direct_req *dreq;
578
	struct nfs_lock_context *l_ctx;
579
	ssize_t result = -EINVAL;
580
	size_t count = iov_iter_count(iter);
581 582 583 584 585 586 587 588 589
	nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);

	dfprintk(FILE, "NFS: direct read(%pD2, %zd@%Ld)\n",
		file, count, (long long) pos);

	result = 0;
	if (!count)
		goto out;

590
	mutex_lock(&inode->i_mutex);
591 592
	result = nfs_sync_mapping(mapping);
	if (result)
593
		goto out_unlock;
L
Linus Torvalds 已提交
594

595 596 597
	task_io_account_read(count);

	result = -ENOMEM;
598
	dreq = nfs_direct_req_alloc();
599
	if (dreq == NULL)
600
		goto out_unlock;
L
Linus Torvalds 已提交
601

C
Chuck Lever 已提交
602
	dreq->inode = inode;
603
	dreq->bytes_left = count;
604
	dreq->io_start = pos;
605
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
606 607 608
	l_ctx = nfs_get_lock_context(dreq->ctx);
	if (IS_ERR(l_ctx)) {
		result = PTR_ERR(l_ctx);
609
		goto out_release;
610 611
	}
	dreq->l_ctx = l_ctx;
612 613
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
614

615
	NFS_I(inode)->read_io += count;
A
Al Viro 已提交
616
	result = nfs_direct_read_schedule_iovec(dreq, iter, pos);
617 618 619

	mutex_unlock(&inode->i_mutex);

620
	if (!result) {
621
		result = nfs_direct_wait(dreq);
622 623 624
		if (result > 0)
			iocb->ki_pos = pos + result;
	}
625 626 627 628

	nfs_direct_req_release(dreq);
	return result;

629
out_release:
630
	nfs_direct_req_release(dreq);
631 632
out_unlock:
	mutex_unlock(&inode->i_mutex);
633
out:
L
Linus Torvalds 已提交
634 635 636
	return result;
}

637 638 639 640 641 642 643 644 645 646 647 648 649 650
static void
nfs_direct_write_scan_commit_list(struct inode *inode,
				  struct list_head *list,
				  struct nfs_commit_info *cinfo)
{
	spin_lock(cinfo->lock);
#ifdef CONFIG_NFS_V4_1
	if (cinfo->ds != NULL && cinfo->ds->nwritten != 0)
		NFS_SERVER(inode)->pnfs_curr_ld->recover_commit_reqs(list, cinfo);
#endif
	nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0);
	spin_unlock(cinfo->lock);
}

651 652
static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
{
653 654 655 656 657
	struct nfs_pageio_descriptor desc;
	struct nfs_page *req, *tmp;
	LIST_HEAD(reqs);
	struct nfs_commit_info cinfo;
	LIST_HEAD(failed);
658
	int i;
659 660

	nfs_init_cinfo_from_dreq(&cinfo, dreq);
661
	nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo);
L
Linus Torvalds 已提交
662

663
	dreq->count = 0;
664 665
	for (i = 0; i < dreq->mirror_count; i++)
		dreq->mirrors[i].count = 0;
666 667
	get_dreq(dreq);

668
	nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false,
669 670
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
671

672 673 674
	req = nfs_list_entry(reqs.next);
	nfs_direct_setup_mirroring(dreq, &desc, req);

675 676
	list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
		if (!nfs_pageio_add_request(&desc, req)) {
677
			nfs_list_remove_request(req);
678 679 680 681 682 683
			nfs_list_add_request(req, &failed);
			spin_lock(cinfo.lock);
			dreq->flags = 0;
			dreq->error = -EIO;
			spin_unlock(cinfo.lock);
		}
684
		nfs_release_request(req);
685 686
	}
	nfs_pageio_complete(&desc);
687

688 689 690
	while (!list_empty(&failed)) {
		req = nfs_list_entry(failed.next);
		nfs_list_remove_request(req);
691
		nfs_unlock_and_release_request(req);
692
	}
693

694 695
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
696 697
}

698
static void nfs_direct_commit_complete(struct nfs_commit_data *data)
699
{
700
	struct nfs_direct_req *dreq = data->dreq;
701 702
	struct nfs_commit_info cinfo;
	struct nfs_page *req;
703 704
	int status = data->task.tk_status;

705
	nfs_init_cinfo_from_dreq(&cinfo, dreq);
706
	if (status < 0) {
707
		dprintk("NFS: %5u commit failed with error %d.\n",
708
			data->task.tk_pid, status);
709
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
710
	} else if (nfs_direct_cmp_commit_data_verf(dreq, data)) {
711
		dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
712
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
L
Linus Torvalds 已提交
713 714
	}

715
	dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
716 717 718 719 720
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
		if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) {
			/* Note the rewrite will go through mds */
721
			nfs_mark_request_commit(req, NULL, &cinfo, 0);
722 723
		} else
			nfs_release_request(req);
724
		nfs_unlock_and_release_request(req);
725 726 727 728
	}

	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_direct_write_complete(dreq, data->inode);
L
Linus Torvalds 已提交
729 730
}

731 732 733 734 735 736 737 738
static void nfs_direct_error_cleanup(struct nfs_inode *nfsi)
{
	/* There is no lock to clear */
}

static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = {
	.completion = nfs_direct_commit_complete,
	.error_cleanup = nfs_direct_error_cleanup,
739 740 741
};

static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
742
{
743 744 745 746 747 748 749 750 751
	int res;
	struct nfs_commit_info cinfo;
	LIST_HEAD(mds_list);

	nfs_init_cinfo_from_dreq(&cinfo, dreq);
	nfs_scan_commit(dreq->inode, &mds_list, &cinfo);
	res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo);
	if (res < 0) /* res == -ENOMEM */
		nfs_direct_write_reschedule(dreq);
752
}
L
Linus Torvalds 已提交
753

754
static void nfs_direct_write_schedule_work(struct work_struct *work)
755
{
756
	struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work);
757
	int flags = dreq->flags;
L
Linus Torvalds 已提交
758

759 760 761 762
	dreq->flags = 0;
	switch (flags) {
		case NFS_ODIRECT_DO_COMMIT:
			nfs_direct_commit_schedule(dreq);
L
Linus Torvalds 已提交
763
			break;
764 765 766 767
		case NFS_ODIRECT_RESCHED_WRITES:
			nfs_direct_write_reschedule(dreq);
			break;
		default:
768
			nfs_direct_complete(dreq, true);
769 770
	}
}
L
Linus Torvalds 已提交
771

772
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
773
{
774
	schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */
775
}
776 777 778 779 780

static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
{
	struct nfs_direct_req *dreq = hdr->dreq;
	struct nfs_commit_info cinfo;
781
	bool request_commit = false;
782 783 784 785 786 787 788 789 790 791 792 793 794
	struct nfs_page *req = nfs_list_entry(hdr->pages.next);

	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out_put;

	nfs_init_cinfo_from_dreq(&cinfo, dreq);

	spin_lock(&dreq->lock);

	if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
		dreq->flags = 0;
		dreq->error = hdr->error;
	}
795
	if (dreq->error == 0) {
796
		nfs_direct_good_bytes(dreq, hdr);
797
		if (nfs_write_need_commit(hdr)) {
798
			if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES)
799
				request_commit = true;
800
			else if (dreq->flags == 0) {
801
				nfs_direct_set_hdr_verf(dreq, hdr);
802
				request_commit = true;
803 804
				dreq->flags = NFS_ODIRECT_DO_COMMIT;
			} else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) {
805 806
				request_commit = true;
				if (nfs_direct_set_or_cmp_hdr_verf(dreq, hdr))
807 808
					dreq->flags =
						NFS_ODIRECT_RESCHED_WRITES;
809 810 811 812 813 814
			}
		}
	}
	spin_unlock(&dreq->lock);

	while (!list_empty(&hdr->pages)) {
815

816 817
		req = nfs_list_entry(hdr->pages.next);
		nfs_list_remove_request(req);
818
		if (request_commit) {
819
			kref_get(&req->wb_kref);
820 821
			nfs_mark_request_commit(req, hdr->lseg, &cinfo,
				hdr->ds_commit_idx);
822
		}
823
		nfs_unlock_and_release_request(req);
824 825 826 827 828 829 830 831
	}

out_put:
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, hdr->inode);
	hdr->release(hdr);
}

832 833 834 835 836 837 838
static void nfs_write_sync_pgio_error(struct list_head *head)
{
	struct nfs_page *req;

	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
839
		nfs_unlock_and_release_request(req);
840 841 842
	}
}

843
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
844
	.error_cleanup = nfs_write_sync_pgio_error,
845 846 847 848
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_write_completion,
};

A
Al Viro 已提交
849 850 851 852 853 854 855 856 857 858 859 860

/*
 * NB: Return the value of the first error return code.  Subsequent
 *     errors after the first one are ignored.
 */
/*
 * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 * bail and stop sending more writes.  Write length accounting is
 * handled automatically by nfs_direct_write_result().  Otherwise, if
 * no requests have been sent, just return an error.
 */
861
static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
862
					       struct iov_iter *iter,
A
Al Viro 已提交
863
					       loff_t pos)
864
{
865
	struct nfs_pageio_descriptor desc;
866
	struct inode *inode = dreq->inode;
867 868
	ssize_t result = 0;
	size_t requested_bytes = 0;
A
Al Viro 已提交
869
	size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE);
870

871
	nfs_pageio_init_write(&desc, inode, FLUSH_COND_STABLE, false,
872 873
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
874
	get_dreq(dreq);
875
	atomic_inc(&inode->i_dio_count);
876

A
Al Viro 已提交
877 878 879 880 881 882 883 884 885
	NFS_I(inode)->write_io += iov_iter_count(iter);
	while (iov_iter_count(iter)) {
		struct page **pagevec;
		size_t bytes;
		size_t pgbase;
		unsigned npages, i;

		result = iov_iter_get_pages_alloc(iter, &pagevec, 
						  wsize, &pgbase);
886 887
		if (result < 0)
			break;
A
Al Viro 已提交
888 889 890 891 892 893 894 895

		bytes = result;
		iov_iter_advance(iter, bytes);
		npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE;
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
			unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);

896
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
A
Al Viro 已提交
897 898 899 900 901
						 pgbase, req_len);
			if (IS_ERR(req)) {
				result = PTR_ERR(req);
				break;
			}
902 903 904

			nfs_direct_setup_mirroring(dreq, &desc, req);

A
Al Viro 已提交
905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921
			nfs_lock_request(req);
			req->wb_index = pos >> PAGE_SHIFT;
			req->wb_offset = pos & ~PAGE_MASK;
			if (!nfs_pageio_add_request(&desc, req)) {
				result = desc.pg_error;
				nfs_unlock_and_release_request(req);
				break;
			}
			pgbase = 0;
			bytes -= req_len;
			requested_bytes += req_len;
			pos += req_len;
			dreq->bytes_left -= req_len;
		}
		nfs_direct_release_pages(pagevec, npages);
		kvfree(pagevec);
		if (result < 0)
922 923
			break;
	}
924
	nfs_pageio_complete(&desc);
925

926 927 928 929 930
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
931
		inode_dio_done(inode);
932 933 934 935
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

936 937
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
938
	return 0;
939 940
}

L
Linus Torvalds 已提交
941 942 943
/**
 * nfs_file_direct_write - file direct write operation for NFS files
 * @iocb: target I/O control block
944
 * @iter: vector of user buffers from which to write data
945
 * @pos: byte offset in file where writing starts
L
Linus Torvalds 已提交
946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
 *
 * We use this function for direct writes instead of calling
 * generic_file_aio_write() in order to avoid taking the inode
 * semaphore and updating the i_size.  The NFS server will set
 * the new i_size and this client must read the updated size
 * back into its cache.  We let the server do generic write
 * parameter checking and report problems.
 *
 * We eliminate local atime updates, see direct read above.
 *
 * We avoid unnecessary page cache invalidations for normal cached
 * readers of this file.
 *
 * Note that O_APPEND is not supported for NFS direct writes, as there
 * is no atomic O_APPEND write facility in the NFS protocol.
 */
962
ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter)
L
Linus Torvalds 已提交
963
{
964
	ssize_t result = -EINVAL;
L
Linus Torvalds 已提交
965 966
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
967 968 969
	struct inode *inode = mapping->host;
	struct nfs_direct_req *dreq;
	struct nfs_lock_context *l_ctx;
970
	loff_t pos, end;
971

972
	dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n",
973
		file, iov_iter_count(iter), (long long) iocb->ki_pos);
974

975 976 977 978 979
	nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES,
		      iov_iter_count(iter));

	pos = iocb->ki_pos;
	end = (pos + iov_iter_count(iter) - 1) >> PAGE_CACHE_SHIFT;
980

981 982
	mutex_lock(&inode->i_mutex);

983 984
	result = nfs_sync_mapping(mapping);
	if (result)
985 986 987 988 989 990 991 992
		goto out_unlock;

	if (mapping->nrpages) {
		result = invalidate_inode_pages2_range(mapping,
					pos >> PAGE_CACHE_SHIFT, end);
		if (result)
			goto out_unlock;
	}
L
Linus Torvalds 已提交
993

994
	task_io_account_write(iov_iter_count(iter));
995

996 997 998
	result = -ENOMEM;
	dreq = nfs_direct_req_alloc();
	if (!dreq)
999
		goto out_unlock;
1000

1001
	dreq->inode = inode;
1002
	dreq->bytes_left = iov_iter_count(iter);
1003
	dreq->io_start = pos;
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
	l_ctx = nfs_get_lock_context(dreq->ctx);
	if (IS_ERR(l_ctx)) {
		result = PTR_ERR(l_ctx);
		goto out_release;
	}
	dreq->l_ctx = l_ctx;
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;

A
Al Viro 已提交
1014
	result = nfs_direct_write_schedule_iovec(dreq, iter, pos);
1015 1016 1017 1018 1019 1020 1021 1022

	if (mapping->nrpages) {
		invalidate_inode_pages2_range(mapping,
					      pos >> PAGE_CACHE_SHIFT, end);
	}

	mutex_unlock(&inode->i_mutex);

1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
	if (!result) {
		result = nfs_direct_wait(dreq);
		if (result > 0) {
			struct inode *inode = mapping->host;

			iocb->ki_pos = pos + result;
			spin_lock(&inode->i_lock);
			if (i_size_read(inode) < iocb->ki_pos)
				i_size_write(inode, iocb->ki_pos);
			spin_unlock(&inode->i_lock);
		}
1034
	}
1035 1036 1037
	nfs_direct_req_release(dreq);
	return result;

1038 1039
out_release:
	nfs_direct_req_release(dreq);
1040 1041
out_unlock:
	mutex_unlock(&inode->i_mutex);
1042
	return result;
L
Linus Torvalds 已提交
1043 1044
}

1045 1046 1047 1048
/**
 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 *
 */
D
David Howells 已提交
1049
int __init nfs_init_directcache(void)
L
Linus Torvalds 已提交
1050 1051 1052
{
	nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
						sizeof(struct nfs_direct_req),
1053 1054
						0, (SLAB_RECLAIM_ACCOUNT|
							SLAB_MEM_SPREAD),
1055
						NULL);
L
Linus Torvalds 已提交
1056 1057 1058 1059 1060 1061
	if (nfs_direct_cachep == NULL)
		return -ENOMEM;

	return 0;
}

1062
/**
D
David Howells 已提交
1063
 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1064 1065
 *
 */
1066
void nfs_destroy_directcache(void)
L
Linus Torvalds 已提交
1067
{
1068
	kmem_cache_destroy(nfs_direct_cachep);
L
Linus Torvalds 已提交
1069
}