direct.c 27.5 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 69 70

/*
 * This represents a set of asynchronous requests that we're waiting on
 */
struct nfs_direct_req {
	struct kref		kref;		/* release manager */
71 72

	/* I/O parameters */
73
	struct nfs_open_context	*ctx;		/* file open context info */
74
	struct nfs_lock_context *l_ctx;		/* Lock context info */
75
	struct kiocb *		iocb;		/* controlling i/o request */
76
	struct inode *		inode;		/* target file of i/o */
77 78

	/* completion state */
79
	atomic_t		io_count;	/* i/os we're waiting for */
80 81
	spinlock_t		lock;		/* protect completion state */
	ssize_t			count,		/* bytes actually processed */
P
Peng Tao 已提交
82
				bytes_left,	/* bytes left to be sent */
L
Linus Torvalds 已提交
83
				error;		/* any reported error */
84
	struct completion	completion;	/* wait for i/o completion */
85 86

	/* commit state */
87 88 89
	struct nfs_mds_commit_info mds_cinfo;	/* Storage for cinfo */
	struct pnfs_ds_commit_info ds_cinfo;	/* Storage for cinfo */
	struct work_struct	work;
90 91 92 93
	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
94 95
};

96 97
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops;
static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops;
98
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
99
static void nfs_direct_write_schedule_work(struct work_struct *work);
100 101 102 103 104 105 106 107 108 109 110

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);
}

111 112 113 114
/*
 * 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
115
 * @commit_idx - commit bucket index for the DS
116 117 118 119 120 121
 *
 * 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,
122
		       int commit_idx)
123 124 125 126 127 128
{
	struct nfs_writeverf *verfp = &dreq->verf;

#ifdef CONFIG_NFS_V4_1
	if (ds_clp) {
		/* pNFS is in use, use the DS verf */
129 130
		if (commit_idx >= 0 && commit_idx < dreq->ds_cinfo.nbuckets)
			verfp = &dreq->ds_cinfo.buckets[commit_idx].direct_verf;
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
		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;

151
	verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
	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;

171
	verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
	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);
	WARN_ON_ONCE(verfp->committed < 0);
	return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf));
}

L
Linus Torvalds 已提交
198
/**
199 200 201 202 203 204 205 206
 * nfs_direct_IO - NFS address space operation for direct I/O
 * @rw: direction (read or write)
 * @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
207 208 209
 * 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
210
 */
A
Al Viro 已提交
211
ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t pos)
212
{
M
Mel Gorman 已提交
213
#ifndef CONFIG_NFS_SWAP
214
	dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n",
A
Al Viro 已提交
215
			iocb->ki_filp, (long long) pos, iter->nr_segs);
216 217

	return -EINVAL;
M
Mel Gorman 已提交
218 219 220
#else
	VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE);

M
Martin K. Petersen 已提交
221 222 223
	if (rw == READ)
		return nfs_file_direct_read(iocb, iter, pos);
	return nfs_file_direct_write(iocb, iter, pos);
M
Mel Gorman 已提交
224
#endif /* CONFIG_NFS_SWAP */
225 226
}

227
static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
228
{
229
	unsigned int i;
230 231
	for (i = 0; i < npages; i++)
		page_cache_release(pages[i]);
232 233
}

234 235 236 237 238 239 240 241 242 243
void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo,
			      struct nfs_direct_req *dreq)
{
	cinfo->lock = &dreq->lock;
	cinfo->mds = &dreq->mds_cinfo;
	cinfo->ds = &dreq->ds_cinfo;
	cinfo->dreq = dreq;
	cinfo->completion_ops = &nfs_direct_commit_completion_ops;
}

244
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
L
Linus Torvalds 已提交
245
{
246 247
	struct nfs_direct_req *dreq;

248
	dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL);
249 250 251 252
	if (!dreq)
		return NULL;

	kref_init(&dreq->kref);
253
	kref_get(&dreq->kref);
254
	init_completion(&dreq->completion);
255
	INIT_LIST_HEAD(&dreq->mds_cinfo.list);
256
	dreq->verf.committed = NFS_INVALID_STABLE_HOW;	/* not set yet */
257
	INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
258
	spin_lock_init(&dreq->lock);
259 260

	return dreq;
L
Linus Torvalds 已提交
261 262
}

263
static void nfs_direct_req_free(struct kref *kref)
L
Linus Torvalds 已提交
264 265
{
	struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
266

P
Peng Tao 已提交
267
	nfs_free_pnfs_ds_cinfo(&dreq->ds_cinfo);
268 269
	if (dreq->l_ctx != NULL)
		nfs_put_lock_context(dreq->l_ctx);
270 271
	if (dreq->ctx != NULL)
		put_nfs_open_context(dreq->ctx);
L
Linus Torvalds 已提交
272 273 274
	kmem_cache_free(nfs_direct_cachep, dreq);
}

275 276 277 278 279
static void nfs_direct_req_release(struct nfs_direct_req *dreq)
{
	kref_put(&dreq->kref, nfs_direct_req_free);
}

280 281 282 283 284 285
ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq)
{
	return dreq->bytes_left;
}
EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left);

286 287 288 289 290
/*
 * Collects and returns the final error value/byte-count.
 */
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
291
	ssize_t result = -EIOCBQUEUED;
292 293 294 295 296

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

297
	result = wait_for_completion_killable(&dreq->completion);
298 299

	if (!result)
300
		result = dreq->error;
301
	if (!result)
302
		result = dreq->count;
303 304 305 306 307

out:
	return (ssize_t) result;
}

308
/*
309 310
 * 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.
311
 */
312
static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write)
313
{
314 315
	struct inode *inode = dreq->inode;

316
	if (dreq->iocb && write) {
317
		loff_t pos = dreq->iocb->ki_pos + dreq->count;
318 319 320 321 322 323 324

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

325
	if (write)
326
		nfs_zap_mapping(inode, inode->i_mapping);
327 328

	inode_dio_done(inode);
329 330

	if (dreq->iocb) {
331
		long res = (long) dreq->error;
332
		if (!res)
333
			res = (long) dreq->count;
334
		aio_complete(dreq->iocb, res, 0);
335
	}
336

337
	complete_all(&dreq->completion);
338

339
	nfs_direct_req_release(dreq);
340 341
}

T
Trond Myklebust 已提交
342
static void nfs_direct_readpage_release(struct nfs_page *req)
L
Linus Torvalds 已提交
343
{
344
	dprintk("NFS: direct read done (%s/%llu %d@%lld)\n",
345
		req->wb_context->dentry->d_inode->i_sb->s_id,
346
		(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
347 348 349
		req->wb_bytes,
		(long long)req_offset(req));
	nfs_release_request(req);
350 351
}

352
static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
353
{
354 355
	unsigned long bytes = 0;
	struct nfs_direct_req *dreq = hdr->dreq;
356

357 358
	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out_put;
359 360

	spin_lock(&dreq->lock);
361 362
	if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0))
		dreq->error = hdr->error;
363 364 365 366 367 368 369 370
	else {
		/*
		 * FIXME: right now this only accounts for bytes written
		 *        to the first mirror
		 */
		if (hdr->pgio_mirror_idx == 0)
			dreq->count += hdr->good_bytes;
	}
371 372
	spin_unlock(&dreq->lock);

373 374 375 376
	while (!list_empty(&hdr->pages)) {
		struct nfs_page *req = nfs_list_entry(hdr->pages.next);
		struct page *page = req->wb_page;

377 378
		if (!PageCompound(page) && bytes < hdr->good_bytes)
			set_page_dirty(page);
379 380 381
		bytes += req->wb_bytes;
		nfs_list_remove_request(req);
		nfs_direct_readpage_release(req);
382
	}
383
out_put:
384
	if (put_dreq(dreq))
385
		nfs_direct_complete(dreq, false);
386
	hdr->release(hdr);
L
Linus Torvalds 已提交
387 388
}

389
static void nfs_read_sync_pgio_error(struct list_head *head)
390
{
391
	struct nfs_page *req;
392

393 394 395 396 397
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_release_request(req);
	}
398 399
}

400 401 402 403 404 405
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 = {
406
	.error_cleanup = nfs_read_sync_pgio_error,
407 408 409 410
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_read_completion,
};

411
/*
412 413 414 415 416
 * 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
417 418
 */

A
Al Viro 已提交
419 420 421 422 423 424 425 426 427
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);
428

429
	nfs_pageio_init_read(&desc, dreq->inode, false,
A
Al Viro 已提交
430 431 432 433
			     &nfs_direct_read_completion_ops);
	get_dreq(dreq);
	desc.pg_dreq = dreq;
	atomic_inc(&inode->i_dio_count);
M
Mel Gorman 已提交
434

A
Al Viro 已提交
435 436 437 438 439
	while (iov_iter_count(iter)) {
		struct page **pagevec;
		size_t bytes;
		size_t pgbase;
		unsigned npages, i;
440

A
Al Viro 已提交
441 442 443 444 445 446 447 448
		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;
449 450
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
451
			unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
452
			/* XXX do we need to do the eof zeroing found in async_filler? */
453
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
454 455 456 457 458 459 460
						 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
461 462
			if (!nfs_pageio_add_request(&desc, req)) {
				result = desc.pg_error;
463 464 465 466 467
				nfs_release_request(req);
				break;
			}
			pgbase = 0;
			bytes -= req_len;
A
Al Viro 已提交
468
			requested_bytes += req_len;
469
			pos += req_len;
P
Peng Tao 已提交
470
			dreq->bytes_left -= req_len;
471
		}
472
		nfs_direct_release_pages(pagevec, npages);
A
Al Viro 已提交
473
		kvfree(pagevec);
474 475 476 477
		if (result < 0)
			break;
	}

478 479
	nfs_pageio_complete(&desc);

480 481 482 483 484
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
485
		inode_dio_done(inode);
486 487 488 489
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

490
	if (put_dreq(dreq))
491
		nfs_direct_complete(dreq, false);
492
	return 0;
493 494
}

495 496 497
/**
 * nfs_file_direct_read - file direct read operation for NFS files
 * @iocb: target I/O control block
498
 * @iter: vector of user buffers into which to read data
499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514
 * @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.
 */
515
ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter,
M
Martin K. Petersen 已提交
516
				loff_t pos)
L
Linus Torvalds 已提交
517
{
518 519 520
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
	struct inode *inode = mapping->host;
L
Linus Torvalds 已提交
521
	struct nfs_direct_req *dreq;
522
	struct nfs_lock_context *l_ctx;
523
	ssize_t result = -EINVAL;
524
	size_t count = iov_iter_count(iter);
525 526 527 528 529 530 531 532 533
	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;

534
	mutex_lock(&inode->i_mutex);
535 536
	result = nfs_sync_mapping(mapping);
	if (result)
537
		goto out_unlock;
L
Linus Torvalds 已提交
538

539 540 541
	task_io_account_read(count);

	result = -ENOMEM;
542
	dreq = nfs_direct_req_alloc();
543
	if (dreq == NULL)
544
		goto out_unlock;
L
Linus Torvalds 已提交
545

C
Chuck Lever 已提交
546
	dreq->inode = inode;
547
	dreq->bytes_left = count;
548
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
549 550 551
	l_ctx = nfs_get_lock_context(dreq->ctx);
	if (IS_ERR(l_ctx)) {
		result = PTR_ERR(l_ctx);
552
		goto out_release;
553 554
	}
	dreq->l_ctx = l_ctx;
555 556
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
557

558
	NFS_I(inode)->read_io += count;
A
Al Viro 已提交
559
	result = nfs_direct_read_schedule_iovec(dreq, iter, pos);
560 561 562

	mutex_unlock(&inode->i_mutex);

563
	if (!result) {
564
		result = nfs_direct_wait(dreq);
565 566 567
		if (result > 0)
			iocb->ki_pos = pos + result;
	}
568 569 570 571

	nfs_direct_req_release(dreq);
	return result;

572
out_release:
573
	nfs_direct_req_release(dreq);
574 575
out_unlock:
	mutex_unlock(&inode->i_mutex);
576
out:
L
Linus Torvalds 已提交
577 578 579
	return result;
}

580 581 582 583 584 585 586 587 588 589 590 591 592 593
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);
}

594 595
static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
{
596 597 598 599 600 601 602
	struct nfs_pageio_descriptor desc;
	struct nfs_page *req, *tmp;
	LIST_HEAD(reqs);
	struct nfs_commit_info cinfo;
	LIST_HEAD(failed);

	nfs_init_cinfo_from_dreq(&cinfo, dreq);
603
	nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo);
L
Linus Torvalds 已提交
604

605
	dreq->count = 0;
606 607
	get_dreq(dreq);

608
	nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false,
609 610
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
611

612 613
	list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
		if (!nfs_pageio_add_request(&desc, req)) {
614
			nfs_list_remove_request(req);
615 616 617 618 619 620
			nfs_list_add_request(req, &failed);
			spin_lock(cinfo.lock);
			dreq->flags = 0;
			dreq->error = -EIO;
			spin_unlock(cinfo.lock);
		}
621
		nfs_release_request(req);
622 623
	}
	nfs_pageio_complete(&desc);
624

625 626 627
	while (!list_empty(&failed)) {
		req = nfs_list_entry(failed.next);
		nfs_list_remove_request(req);
628
		nfs_unlock_and_release_request(req);
629
	}
630

631 632
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
633 634
}

635
static void nfs_direct_commit_complete(struct nfs_commit_data *data)
636
{
637
	struct nfs_direct_req *dreq = data->dreq;
638 639
	struct nfs_commit_info cinfo;
	struct nfs_page *req;
640 641
	int status = data->task.tk_status;

642
	nfs_init_cinfo_from_dreq(&cinfo, dreq);
643
	if (status < 0) {
644
		dprintk("NFS: %5u commit failed with error %d.\n",
645
			data->task.tk_pid, status);
646
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
647
	} else if (nfs_direct_cmp_commit_data_verf(dreq, data)) {
648
		dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
649
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
L
Linus Torvalds 已提交
650 651
	}

652
	dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
653 654 655 656 657
	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 */
658
			nfs_mark_request_commit(req, NULL, &cinfo, 0);
659 660
		} else
			nfs_release_request(req);
661
		nfs_unlock_and_release_request(req);
662 663 664 665
	}

	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_direct_write_complete(dreq, data->inode);
L
Linus Torvalds 已提交
666 667
}

668 669 670 671 672 673 674 675
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,
676 677 678
};

static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
679
{
680 681 682 683 684 685 686 687 688
	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);
689
}
L
Linus Torvalds 已提交
690

691
static void nfs_direct_write_schedule_work(struct work_struct *work)
692
{
693
	struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work);
694
	int flags = dreq->flags;
L
Linus Torvalds 已提交
695

696 697 698 699
	dreq->flags = 0;
	switch (flags) {
		case NFS_ODIRECT_DO_COMMIT:
			nfs_direct_commit_schedule(dreq);
L
Linus Torvalds 已提交
700
			break;
701 702 703 704
		case NFS_ODIRECT_RESCHED_WRITES:
			nfs_direct_write_reschedule(dreq);
			break;
		default:
705
			nfs_direct_complete(dreq, true);
706 707
	}
}
L
Linus Torvalds 已提交
708

709
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
710
{
711
	schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */
712
}
713 714 715 716 717

static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
{
	struct nfs_direct_req *dreq = hdr->dreq;
	struct nfs_commit_info cinfo;
718
	bool request_commit = false;
719 720 721 722 723 724 725 726 727 728 729 730 731
	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;
	}
732
	if (dreq->error == 0) {
733 734 735 736 737 738
		/*
		 * FIXME: right now this only accounts for bytes written
		 *        to the first mirror
		 */
		if (hdr->pgio_mirror_idx == 0)
			dreq->count += hdr->good_bytes;
739
		if (nfs_write_need_commit(hdr)) {
740
			if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES)
741
				request_commit = true;
742
			else if (dreq->flags == 0) {
743
				nfs_direct_set_hdr_verf(dreq, hdr);
744
				request_commit = true;
745 746
				dreq->flags = NFS_ODIRECT_DO_COMMIT;
			} else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) {
747 748
				request_commit = true;
				if (nfs_direct_set_or_cmp_hdr_verf(dreq, hdr))
749 750
					dreq->flags =
						NFS_ODIRECT_RESCHED_WRITES;
751 752 753 754 755 756
			}
		}
	}
	spin_unlock(&dreq->lock);

	while (!list_empty(&hdr->pages)) {
757

758 759
		req = nfs_list_entry(hdr->pages.next);
		nfs_list_remove_request(req);
760
		if (request_commit) {
761
			kref_get(&req->wb_kref);
762 763
			nfs_mark_request_commit(req, hdr->lseg, &cinfo,
				hdr->ds_commit_idx);
764
		}
765
		nfs_unlock_and_release_request(req);
766 767 768 769 770 771 772 773
	}

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

774 775 776 777 778 779 780
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);
781
		nfs_unlock_and_release_request(req);
782 783 784
	}
}

785
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
786
	.error_cleanup = nfs_write_sync_pgio_error,
787 788 789 790
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_write_completion,
};

A
Al Viro 已提交
791 792 793 794 795 796 797 798 799 800 801 802

/*
 * 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.
 */
803
static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
804
					       struct iov_iter *iter,
A
Al Viro 已提交
805
					       loff_t pos)
806
{
807
	struct nfs_pageio_descriptor desc;
808
	struct inode *inode = dreq->inode;
809 810
	ssize_t result = 0;
	size_t requested_bytes = 0;
A
Al Viro 已提交
811
	size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE);
812

813
	nfs_pageio_init_write(&desc, inode, FLUSH_COND_STABLE, false,
814 815
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
816
	get_dreq(dreq);
817
	atomic_inc(&inode->i_dio_count);
818

A
Al Viro 已提交
819 820 821 822 823 824 825 826 827
	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);
828 829
		if (result < 0)
			break;
A
Al Viro 已提交
830 831 832 833 834 835 836 837

		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);

838
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
A
Al Viro 已提交
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860
						 pgbase, req_len);
			if (IS_ERR(req)) {
				result = PTR_ERR(req);
				break;
			}
			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)
861 862
			break;
	}
863
	nfs_pageio_complete(&desc);
864

865 866 867 868 869
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
870
		inode_dio_done(inode);
871 872 873 874
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

875 876
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
877
	return 0;
878 879
}

L
Linus Torvalds 已提交
880 881 882
/**
 * nfs_file_direct_write - file direct write operation for NFS files
 * @iocb: target I/O control block
883
 * @iter: vector of user buffers from which to write data
884
 * @pos: byte offset in file where writing starts
L
Linus Torvalds 已提交
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
 *
 * 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.
 */
901
ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter,
M
Martin K. Petersen 已提交
902
				loff_t pos)
L
Linus Torvalds 已提交
903
{
904
	ssize_t result = -EINVAL;
L
Linus Torvalds 已提交
905 906
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
907 908 909
	struct inode *inode = mapping->host;
	struct nfs_direct_req *dreq;
	struct nfs_lock_context *l_ctx;
910
	loff_t end;
911
	size_t count = iov_iter_count(iter);
912 913
	end = (pos + count - 1) >> PAGE_CACHE_SHIFT;

914 915
	nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);

916 917
	dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n",
		file, count, (long long) pos);
918

919 920
	result = generic_write_checks(file, &pos, &count, 0);
	if (result)
L
Linus Torvalds 已提交
921
		goto out;
922

923
	result = -EINVAL;
924
	if ((ssize_t) count < 0)
L
Linus Torvalds 已提交
925
		goto out;
926
	result = 0;
L
Linus Torvalds 已提交
927 928
	if (!count)
		goto out;
929

930 931
	mutex_lock(&inode->i_mutex);

932 933
	result = nfs_sync_mapping(mapping);
	if (result)
934 935 936 937 938 939 940 941
		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
942

943 944
	task_io_account_write(count);

945 946 947
	result = -ENOMEM;
	dreq = nfs_direct_req_alloc();
	if (!dreq)
948
		goto out_unlock;
949

950 951 952 953 954 955 956 957 958 959 960 961
	dreq->inode = inode;
	dreq->bytes_left = count;
	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
962
	result = nfs_direct_write_schedule_iovec(dreq, iter, pos);
963 964 965 966 967 968 969 970

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

	mutex_unlock(&inode->i_mutex);

971 972 973 974 975 976 977 978 979 980 981
	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);
		}
982
	}
983 984 985
	nfs_direct_req_release(dreq);
	return result;

986 987
out_release:
	nfs_direct_req_release(dreq);
988 989
out_unlock:
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
990
out:
991
	return result;
L
Linus Torvalds 已提交
992 993
}

994 995 996 997
/**
 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 *
 */
D
David Howells 已提交
998
int __init nfs_init_directcache(void)
L
Linus Torvalds 已提交
999 1000 1001
{
	nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
						sizeof(struct nfs_direct_req),
1002 1003
						0, (SLAB_RECLAIM_ACCOUNT|
							SLAB_MEM_SPREAD),
1004
						NULL);
L
Linus Torvalds 已提交
1005 1006 1007 1008 1009 1010
	if (nfs_direct_cachep == NULL)
		return -ENOMEM;

	return 0;
}

1011
/**
D
David Howells 已提交
1012
 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1013 1014
 *
 */
1015
void nfs_destroy_directcache(void)
L
Linus Torvalds 已提交
1016
{
1017
	kmem_cache_destroy(nfs_direct_cachep);
L
Linus Torvalds 已提交
1018
}