direct.c 27.7 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);
}

111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
/*
 * 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
 * @ds_idx - index of data server in data server list, only valid if ds_clp set
 *
 * 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,
		       int ds_idx)
{
	struct nfs_writeverf *verfp = &dreq->verf;

#ifdef CONFIG_NFS_V4_1
	if (ds_clp) {
		/* pNFS is in use, use the DS verf */
		if (ds_idx >= 0 && ds_idx < dreq->ds_cinfo.nbuckets)
			verfp = &dreq->ds_cinfo.buckets[ds_idx].direct_verf;
		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;

151 152
	verfp = nfs_direct_select_verf(dreq, hdr->data.ds_clp,
				      hdr->data.ds_idx);
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
	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;

172 173
	verfp = nfs_direct_select_verf(dreq, hdr->data.ds_clp,
					 hdr->data.ds_idx);
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
	if (verfp->committed < 0) {
		nfs_direct_set_hdr_verf(dreq, hdr);
		return 0;
	}
	return memcmp(verfp, &hdr->verf, sizeof(struct nfs_writeverf));
}

#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
/*
 * 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));
}
#endif

L
Linus Torvalds 已提交
202
/**
203 204 205 206 207 208 209 210
 * 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
211 212 213
 * 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
214
 */
A
Al Viro 已提交
215
ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t pos)
216
{
M
Mel Gorman 已提交
217
#ifndef CONFIG_NFS_SWAP
218
	dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n",
A
Al Viro 已提交
219
			iocb->ki_filp, (long long) pos, iter->nr_segs);
220 221

	return -EINVAL;
M
Mel Gorman 已提交
222 223 224 225
#else
	VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE);

	if (rw == READ || rw == KERNEL_READ)
226
		return nfs_file_direct_read(iocb, iter, pos,
M
Mel Gorman 已提交
227
				rw == READ ? true : false);
228
	return nfs_file_direct_write(iocb, iter, pos,
M
Mel Gorman 已提交
229 230
				rw == WRITE ? true : false);
#endif /* CONFIG_NFS_SWAP */
231 232
}

233
static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
234
{
235
	unsigned int i;
236 237
	for (i = 0; i < npages; i++)
		page_cache_release(pages[i]);
238 239
}

240 241 242 243 244 245 246 247 248 249
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;
}

250
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
L
Linus Torvalds 已提交
251
{
252 253
	struct nfs_direct_req *dreq;

254
	dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL);
255 256 257 258
	if (!dreq)
		return NULL;

	kref_init(&dreq->kref);
259
	kref_get(&dreq->kref);
260
	init_completion(&dreq->completion);
261
	INIT_LIST_HEAD(&dreq->mds_cinfo.list);
262
	dreq->verf.committed = NFS_INVALID_STABLE_HOW;	/* not set yet */
263
	INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
264
	spin_lock_init(&dreq->lock);
265 266

	return dreq;
L
Linus Torvalds 已提交
267 268
}

269
static void nfs_direct_req_free(struct kref *kref)
L
Linus Torvalds 已提交
270 271
{
	struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
272

273 274
	if (dreq->l_ctx != NULL)
		nfs_put_lock_context(dreq->l_ctx);
275 276
	if (dreq->ctx != NULL)
		put_nfs_open_context(dreq->ctx);
L
Linus Torvalds 已提交
277 278 279
	kmem_cache_free(nfs_direct_cachep, dreq);
}

280 281 282 283 284
static void nfs_direct_req_release(struct nfs_direct_req *dreq)
{
	kref_put(&dreq->kref, nfs_direct_req_free);
}

285 286 287 288 289 290
ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq)
{
	return dreq->bytes_left;
}
EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left);

291 292 293 294 295
/*
 * Collects and returns the final error value/byte-count.
 */
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
296
	ssize_t result = -EIOCBQUEUED;
297 298 299 300 301

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

302
	result = wait_for_completion_killable(&dreq->completion);
303 304

	if (!result)
305
		result = dreq->error;
306
	if (!result)
307
		result = dreq->count;
308 309 310 311 312

out:
	return (ssize_t) result;
}

313
/*
314 315
 * 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.
316
 */
317
static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write)
318
{
319 320
	struct inode *inode = dreq->inode;

321
	if (dreq->iocb && write) {
322
		loff_t pos = dreq->iocb->ki_pos + dreq->count;
323 324 325 326 327 328 329

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

330
	if (write)
331
		nfs_zap_mapping(inode, inode->i_mapping);
332 333

	inode_dio_done(inode);
334 335

	if (dreq->iocb) {
336
		long res = (long) dreq->error;
337
		if (!res)
338
			res = (long) dreq->count;
339
		aio_complete(dreq->iocb, res, 0);
340
	}
341

342
	complete_all(&dreq->completion);
343

344
	nfs_direct_req_release(dreq);
345 346
}

T
Trond Myklebust 已提交
347
static void nfs_direct_readpage_release(struct nfs_page *req)
L
Linus Torvalds 已提交
348
{
349
	dprintk("NFS: direct read done (%s/%llu %d@%lld)\n",
350
		req->wb_context->dentry->d_inode->i_sb->s_id,
351
		(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
352 353 354
		req->wb_bytes,
		(long long)req_offset(req));
	nfs_release_request(req);
355 356
}

357
static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
358
{
359 360
	unsigned long bytes = 0;
	struct nfs_direct_req *dreq = hdr->dreq;
361

362 363
	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out_put;
364 365

	spin_lock(&dreq->lock);
366 367 368 369 370 371
	if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0))
		dreq->error = hdr->error;
	else
		dreq->count += hdr->good_bytes;
	spin_unlock(&dreq->lock);

372 373 374 375
	while (!list_empty(&hdr->pages)) {
		struct nfs_page *req = nfs_list_entry(hdr->pages.next);
		struct page *page = req->wb_page;

376 377
		if (!PageCompound(page) && bytes < hdr->good_bytes)
			set_page_dirty(page);
378 379 380
		bytes += req->wb_bytes;
		nfs_list_remove_request(req);
		nfs_direct_readpage_release(req);
381
	}
382
out_put:
383
	if (put_dreq(dreq))
384
		nfs_direct_complete(dreq, false);
385
	hdr->release(hdr);
L
Linus Torvalds 已提交
386 387
}

388
static void nfs_read_sync_pgio_error(struct list_head *head)
389
{
390
	struct nfs_page *req;
391

392 393 394 395 396
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_release_request(req);
	}
397 398
}

399 400 401 402 403 404
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 = {
405
	.error_cleanup = nfs_read_sync_pgio_error,
406 407 408 409
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_read_completion,
};

410
/*
411 412 413 414 415
 * 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
416 417
 */

A
Al Viro 已提交
418 419 420 421 422 423 424 425 426
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);
427

428
	nfs_pageio_init_read(&desc, dreq->inode, false,
A
Al Viro 已提交
429 430 431 432
			     &nfs_direct_read_completion_ops);
	get_dreq(dreq);
	desc.pg_dreq = dreq;
	atomic_inc(&inode->i_dio_count);
M
Mel Gorman 已提交
433

A
Al Viro 已提交
434 435 436 437 438
	while (iov_iter_count(iter)) {
		struct page **pagevec;
		size_t bytes;
		size_t pgbase;
		unsigned npages, i;
439

A
Al Viro 已提交
440 441 442 443 444 445 446 447
		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;
448 449
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
450
			unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
451
			/* XXX do we need to do the eof zeroing found in async_filler? */
452
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
453 454 455 456 457 458 459
						 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
460 461
			if (!nfs_pageio_add_request(&desc, req)) {
				result = desc.pg_error;
462 463 464 465 466
				nfs_release_request(req);
				break;
			}
			pgbase = 0;
			bytes -= req_len;
A
Al Viro 已提交
467
			requested_bytes += req_len;
468
			pos += req_len;
P
Peng Tao 已提交
469
			dreq->bytes_left -= req_len;
470
		}
471
		nfs_direct_release_pages(pagevec, npages);
A
Al Viro 已提交
472
		kvfree(pagevec);
473 474 475 476
		if (result < 0)
			break;
	}

477 478
	nfs_pageio_complete(&desc);

479 480 481 482 483
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
484
		inode_dio_done(inode);
485 486 487 488
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

489
	if (put_dreq(dreq))
490
		nfs_direct_complete(dreq, false);
491
	return 0;
492 493
}

494 495 496
/**
 * nfs_file_direct_read - file direct read operation for NFS files
 * @iocb: target I/O control block
497
 * @iter: vector of user buffers into which to read data
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513
 * @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.
 */
514 515
ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter,
				loff_t pos, bool uio)
L
Linus Torvalds 已提交
516
{
517 518 519
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
	struct inode *inode = mapping->host;
L
Linus Torvalds 已提交
520
	struct nfs_direct_req *dreq;
521
	struct nfs_lock_context *l_ctx;
522
	ssize_t result = -EINVAL;
523
	size_t count = iov_iter_count(iter);
524 525 526 527 528 529 530 531 532
	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;

533
	mutex_lock(&inode->i_mutex);
534 535
	result = nfs_sync_mapping(mapping);
	if (result)
536
		goto out_unlock;
L
Linus Torvalds 已提交
537

538 539 540
	task_io_account_read(count);

	result = -ENOMEM;
541
	dreq = nfs_direct_req_alloc();
542
	if (dreq == NULL)
543
		goto out_unlock;
L
Linus Torvalds 已提交
544

C
Chuck Lever 已提交
545
	dreq->inode = inode;
546
	dreq->bytes_left = count;
547
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
548 549 550
	l_ctx = nfs_get_lock_context(dreq->ctx);
	if (IS_ERR(l_ctx)) {
		result = PTR_ERR(l_ctx);
551
		goto out_release;
552 553
	}
	dreq->l_ctx = l_ctx;
554 555
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
556

557
	NFS_I(inode)->read_io += count;
A
Al Viro 已提交
558
	result = nfs_direct_read_schedule_iovec(dreq, iter, pos);
559 560 561

	mutex_unlock(&inode->i_mutex);

562
	if (!result) {
563
		result = nfs_direct_wait(dreq);
564 565 566
		if (result > 0)
			iocb->ki_pos = pos + result;
	}
567 568 569 570

	nfs_direct_req_release(dreq);
	return result;

571
out_release:
572
	nfs_direct_req_release(dreq);
573 574
out_unlock:
	mutex_unlock(&inode->i_mutex);
575
out:
L
Linus Torvalds 已提交
576 577 578
	return result;
}

B
Bryan Schumaker 已提交
579
#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
580 581
static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
{
582 583 584 585 586 587 588 589 590 591 592
	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);
	pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo);
	spin_lock(cinfo.lock);
	nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0);
	spin_unlock(cinfo.lock);
L
Linus Torvalds 已提交
593

594
	dreq->count = 0;
595 596
	get_dreq(dreq);

597
	nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false,
598 599
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
600

601 602
	list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
		if (!nfs_pageio_add_request(&desc, req)) {
603
			nfs_list_remove_request(req);
604 605 606 607 608 609
			nfs_list_add_request(req, &failed);
			spin_lock(cinfo.lock);
			dreq->flags = 0;
			dreq->error = -EIO;
			spin_unlock(cinfo.lock);
		}
610
		nfs_release_request(req);
611 612
	}
	nfs_pageio_complete(&desc);
613

614 615 616
	while (!list_empty(&failed)) {
		req = nfs_list_entry(failed.next);
		nfs_list_remove_request(req);
617
		nfs_unlock_and_release_request(req);
618
	}
619

620 621
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
622 623
}

624
static void nfs_direct_commit_complete(struct nfs_commit_data *data)
625
{
626
	struct nfs_direct_req *dreq = data->dreq;
627 628
	struct nfs_commit_info cinfo;
	struct nfs_page *req;
629 630
	int status = data->task.tk_status;

631
	nfs_init_cinfo_from_dreq(&cinfo, dreq);
632
	if (status < 0) {
633
		dprintk("NFS: %5u commit failed with error %d.\n",
634
			data->task.tk_pid, status);
635
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
636
	} else if (nfs_direct_cmp_commit_data_verf(dreq, data)) {
637
		dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
638
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
L
Linus Torvalds 已提交
639 640
	}

641
	dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
642 643 644 645 646 647
	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 */
			nfs_mark_request_commit(req, NULL, &cinfo);
648 649
		} else
			nfs_release_request(req);
650
		nfs_unlock_and_release_request(req);
651 652 653 654
	}

	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_direct_write_complete(dreq, data->inode);
L
Linus Torvalds 已提交
655 656
}

657 658 659 660 661 662 663 664
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,
665 666 667
};

static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
668
{
669 670 671 672 673 674 675 676 677
	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);
678
}
L
Linus Torvalds 已提交
679

680
static void nfs_direct_write_schedule_work(struct work_struct *work)
681
{
682
	struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work);
683
	int flags = dreq->flags;
L
Linus Torvalds 已提交
684

685 686 687 688
	dreq->flags = 0;
	switch (flags) {
		case NFS_ODIRECT_DO_COMMIT:
			nfs_direct_commit_schedule(dreq);
L
Linus Torvalds 已提交
689
			break;
690 691 692 693
		case NFS_ODIRECT_RESCHED_WRITES:
			nfs_direct_write_reschedule(dreq);
			break;
		default:
694
			nfs_direct_complete(dreq, true);
695 696
	}
}
L
Linus Torvalds 已提交
697

698
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
699
{
700
	schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */
701
}
702

703
#else
704 705 706
static void nfs_direct_write_schedule_work(struct work_struct *work)
{
}
L
Linus Torvalds 已提交
707

708 709
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
{
710
	nfs_direct_complete(dreq, true);
711 712
}
#endif
L
Linus Torvalds 已提交
713

714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742
static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
{
	struct nfs_direct_req *dreq = hdr->dreq;
	struct nfs_commit_info cinfo;
	int bit = -1;
	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;
	}
	if (dreq->error != 0)
		bit = NFS_IOHDR_ERROR;
	else {
		dreq->count += hdr->good_bytes;
		if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
			dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
			bit = NFS_IOHDR_NEED_RESCHED;
		} else if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
			if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES)
				bit = NFS_IOHDR_NEED_RESCHED;
			else if (dreq->flags == 0) {
743
				nfs_direct_set_hdr_verf(dreq, hdr);
744 745 746
				bit = NFS_IOHDR_NEED_COMMIT;
				dreq->flags = NFS_ODIRECT_DO_COMMIT;
			} else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) {
747 748 749
				if (nfs_direct_set_or_cmp_hdr_verf(dreq, hdr)) {
					dreq->flags =
						NFS_ODIRECT_RESCHED_WRITES;
750 751 752 753 754 755 756 757 758
					bit = NFS_IOHDR_NEED_RESCHED;
				} else
					bit = NFS_IOHDR_NEED_COMMIT;
			}
		}
	}
	spin_unlock(&dreq->lock);

	while (!list_empty(&hdr->pages)) {
759 760
		bool do_destroy = true;

761 762 763 764 765
		req = nfs_list_entry(hdr->pages.next);
		nfs_list_remove_request(req);
		switch (bit) {
		case NFS_IOHDR_NEED_RESCHED:
		case NFS_IOHDR_NEED_COMMIT:
766
			kref_get(&req->wb_kref);
767
			nfs_mark_request_commit(req, hdr->lseg, &cinfo);
768
			do_destroy = false;
769
		}
770
		nfs_unlock_and_release_request(req);
771 772 773 774 775 776 777 778
	}

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

779 780 781 782 783 784 785
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);
786
		nfs_unlock_and_release_request(req);
787 788 789
	}
}

790
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
791
	.error_cleanup = nfs_write_sync_pgio_error,
792 793 794 795
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_write_completion,
};

A
Al Viro 已提交
796 797 798 799 800 801 802 803 804 805 806 807

/*
 * 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.
 */
808
static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
809
					       struct iov_iter *iter,
A
Al Viro 已提交
810
					       loff_t pos)
811
{
812
	struct nfs_pageio_descriptor desc;
813
	struct inode *inode = dreq->inode;
814 815
	ssize_t result = 0;
	size_t requested_bytes = 0;
A
Al Viro 已提交
816
	size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE);
817

818
	nfs_pageio_init_write(&desc, inode, FLUSH_COND_STABLE, false,
819 820
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
821
	get_dreq(dreq);
822
	atomic_inc(&inode->i_dio_count);
823

A
Al Viro 已提交
824 825 826 827 828 829 830 831 832
	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);
833 834
		if (result < 0)
			break;
A
Al Viro 已提交
835 836 837 838 839 840 841 842

		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);

843
			req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
A
Al Viro 已提交
844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
						 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)
866 867
			break;
	}
868
	nfs_pageio_complete(&desc);
869

870 871 872 873 874
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
875
		inode_dio_done(inode);
876 877 878 879
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

880 881
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
882
	return 0;
883 884
}

L
Linus Torvalds 已提交
885 886 887
/**
 * nfs_file_direct_write - file direct write operation for NFS files
 * @iocb: target I/O control block
888
 * @iter: vector of user buffers from which to write data
889
 * @pos: byte offset in file where writing starts
L
Linus Torvalds 已提交
890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905
 *
 * 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.
 */
906 907
ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter,
				loff_t pos, bool uio)
L
Linus Torvalds 已提交
908
{
909
	ssize_t result = -EINVAL;
L
Linus Torvalds 已提交
910 911
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
912 913 914
	struct inode *inode = mapping->host;
	struct nfs_direct_req *dreq;
	struct nfs_lock_context *l_ctx;
915
	loff_t end;
916
	size_t count = iov_iter_count(iter);
917 918
	end = (pos + count - 1) >> PAGE_CACHE_SHIFT;

919 920
	nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);

921 922
	dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n",
		file, count, (long long) pos);
923

924 925
	result = generic_write_checks(file, &pos, &count, 0);
	if (result)
L
Linus Torvalds 已提交
926
		goto out;
927

928
	result = -EINVAL;
929
	if ((ssize_t) count < 0)
L
Linus Torvalds 已提交
930
		goto out;
931
	result = 0;
L
Linus Torvalds 已提交
932 933
	if (!count)
		goto out;
934

935 936
	mutex_lock(&inode->i_mutex);

937 938
	result = nfs_sync_mapping(mapping);
	if (result)
939 940 941 942 943 944 945 946
		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
947

948 949
	task_io_account_write(count);

950 951 952
	result = -ENOMEM;
	dreq = nfs_direct_req_alloc();
	if (!dreq)
953
		goto out_unlock;
954

955 956 957 958 959 960 961 962 963 964 965 966
	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
967
	result = nfs_direct_write_schedule_iovec(dreq, iter, pos);
968 969 970 971 972 973 974 975

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

	mutex_unlock(&inode->i_mutex);

976 977 978 979 980 981 982 983 984 985 986
	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);
		}
987
	}
988 989 990
	nfs_direct_req_release(dreq);
	return result;

991 992
out_release:
	nfs_direct_req_release(dreq);
993 994
out_unlock:
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
995
out:
996
	return result;
L
Linus Torvalds 已提交
997 998
}

999 1000 1001 1002
/**
 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 *
 */
D
David Howells 已提交
1003
int __init nfs_init_directcache(void)
L
Linus Torvalds 已提交
1004 1005 1006
{
	nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
						sizeof(struct nfs_direct_req),
1007 1008
						0, (SLAB_RECLAIM_ACCOUNT|
							SLAB_MEM_SPREAD),
1009
						NULL);
L
Linus Torvalds 已提交
1010 1011 1012 1013 1014 1015
	if (nfs_direct_cachep == NULL)
		return -ENOMEM;

	return 0;
}

1016
/**
D
David Howells 已提交
1017
 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1018 1019
 *
 */
1020
void nfs_destroy_directcache(void)
L
Linus Torvalds 已提交
1021
{
1022
	kmem_cache_destroy(nfs_direct_cachep);
L
Linus Torvalds 已提交
1023
}