direct.c 25.6 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 47 48 49 50 51 52 53 54 55
 *
 */

#include <linux/errno.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/file.h>
#include <linux/pagemap.h>
#include <linux/kref.h>

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

#include <asm/system.h>
#include <asm/uaccess.h>
#include <asm/atomic.h>

56
#include "internal.h"
C
Chuck Lever 已提交
57 58
#include "iostat.h"

L
Linus Torvalds 已提交
59 60
#define NFSDBG_FACILITY		NFSDBG_VFS

61
static struct kmem_cache *nfs_direct_cachep;
L
Linus Torvalds 已提交
62 63 64 65 66 67

/*
 * This represents a set of asynchronous requests that we're waiting on
 */
struct nfs_direct_req {
	struct kref		kref;		/* release manager */
68 69

	/* I/O parameters */
70
	struct nfs_open_context	*ctx;		/* file open context info */
71
	struct kiocb *		iocb;		/* controlling i/o request */
72
	struct inode *		inode;		/* target file of i/o */
73 74

	/* completion state */
75
	atomic_t		io_count;	/* i/os we're waiting for */
76 77
	spinlock_t		lock;		/* protect completion state */
	ssize_t			count,		/* bytes actually processed */
L
Linus Torvalds 已提交
78
				error;		/* any reported error */
79
	struct completion	completion;	/* wait for i/o completion */
80 81

	/* commit state */
82
	struct list_head	rewrite_list;	/* saved nfs_write_data structs */
83 84 85 86 87
	struct nfs_write_data *	commit_data;	/* special write_data for commits */
	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
88 89
};

90
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
91 92 93 94 95 96 97 98 99 100 101 102
static const struct rpc_call_ops nfs_write_direct_ops;

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);
}

L
Linus Torvalds 已提交
103
/**
104 105 106 107 108 109 110 111 112 113 114
 * 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
 * the NFS client supports direct I/O.  However, we shunt off direct
 * read and write requests before the VFS gets them, so this method
 * should never be called.
L
Linus Torvalds 已提交
115
 */
116 117 118
ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
{
	dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
119
			iocb->ki_filp->f_path.dentry->d_name.name,
120
			(long long) pos, nr_segs);
121 122 123 124

	return -EINVAL;
}

125
static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
126
{
127
	unsigned int npages;
128
	unsigned int i;
129 130 131 132 133

	if (count == 0)
		return;
	pages += (pgbase >> PAGE_SHIFT);
	npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
134 135
	for (i = 0; i < npages; i++) {
		struct page *page = pages[i];
136
		if (!PageCompound(page))
137
			set_page_dirty(page);
138
	}
139 140
}

141
static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
142
{
143
	unsigned int i;
144 145
	for (i = 0; i < npages; i++)
		page_cache_release(pages[i]);
146 147
}

148
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
L
Linus Torvalds 已提交
149
{
150 151
	struct nfs_direct_req *dreq;

152
	dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
153 154 155 156
	if (!dreq)
		return NULL;

	kref_init(&dreq->kref);
157
	kref_get(&dreq->kref);
158
	init_completion(&dreq->completion);
159
	INIT_LIST_HEAD(&dreq->rewrite_list);
160
	dreq->iocb = NULL;
161
	dreq->ctx = NULL;
162
	spin_lock_init(&dreq->lock);
163
	atomic_set(&dreq->io_count, 0);
164 165
	dreq->count = 0;
	dreq->error = 0;
166
	dreq->flags = 0;
167 168

	return dreq;
L
Linus Torvalds 已提交
169 170
}

171
static void nfs_direct_req_free(struct kref *kref)
L
Linus Torvalds 已提交
172 173
{
	struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
174 175 176

	if (dreq->ctx != NULL)
		put_nfs_open_context(dreq->ctx);
L
Linus Torvalds 已提交
177 178 179
	kmem_cache_free(nfs_direct_cachep, dreq);
}

180 181 182 183 184
static void nfs_direct_req_release(struct nfs_direct_req *dreq)
{
	kref_put(&dreq->kref, nfs_direct_req_free);
}

185 186 187 188 189
/*
 * Collects and returns the final error value/byte-count.
 */
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
190
	ssize_t result = -EIOCBQUEUED;
191 192 193 194 195

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

196
	result = wait_for_completion_interruptible(&dreq->completion);
197 198

	if (!result)
199
		result = dreq->error;
200
	if (!result)
201
		result = dreq->count;
202 203 204 205 206

out:
	return (ssize_t) result;
}

207
/*
208 209
 * 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.
210 211 212 213
 */
static void nfs_direct_complete(struct nfs_direct_req *dreq)
{
	if (dreq->iocb) {
214
		long res = (long) dreq->error;
215
		if (!res)
216
			res = (long) dreq->count;
217
		aio_complete(dreq->iocb, res, 0);
218 219
	}
	complete_all(&dreq->completion);
220

221
	nfs_direct_req_release(dreq);
222 223
}

224
/*
225 226 227
 * We must hold a reference to all the pages in this direct read request
 * until the RPCs complete.  This could be long *after* we are woken up in
 * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
228
 */
T
Trond Myklebust 已提交
229
static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
230
{
T
Trond Myklebust 已提交
231
	struct nfs_read_data *data = calldata;
L
Linus Torvalds 已提交
232 233
	struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;

T
Trond Myklebust 已提交
234 235
	if (nfs_readpage_result(task, data) != 0)
		return;
236 237

	spin_lock(&dreq->lock);
238
	if (unlikely(task->tk_status < 0)) {
239
		dreq->error = task->tk_status;
240 241 242 243 244 245 246 247 248
		spin_unlock(&dreq->lock);
	} else {
		dreq->count += data->res.count;
		spin_unlock(&dreq->lock);
		nfs_direct_dirty_pages(data->pagevec,
				data->args.pgbase,
				data->res.count);
	}
	nfs_direct_release_pages(data->pagevec, data->npages);
249 250 251

	if (put_dreq(dreq))
		nfs_direct_complete(dreq);
L
Linus Torvalds 已提交
252 253
}

T
Trond Myklebust 已提交
254 255 256 257 258
static const struct rpc_call_ops nfs_read_direct_ops = {
	.rpc_call_done = nfs_direct_read_result,
	.rpc_release = nfs_readdata_release,
};

259
/*
260 261 262 263 264
 * 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
265
 */
266
static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos)
L
Linus Torvalds 已提交
267
{
268
	struct nfs_open_context *ctx = dreq->ctx;
269
	struct inode *inode = ctx->path.dentry->d_inode;
270
	size_t rsize = NFS_SERVER(inode)->rsize;
271 272 273 274 275
	unsigned int pgbase;
	int result;
	ssize_t started = 0;

	get_dreq(dreq);
L
Linus Torvalds 已提交
276 277

	do {
278
		struct nfs_read_data *data;
279
		size_t bytes;
L
Linus Torvalds 已提交
280

281 282 283
		pgbase = user_addr & ~PAGE_MASK;
		bytes = min(rsize,count);

284
		result = -ENOMEM;
285
		data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
286 287 288 289 290 291 292
		if (unlikely(!data))
			break;

		down_read(&current->mm->mmap_sem);
		result = get_user_pages(current, current->mm, user_addr,
					data->npages, 1, 0, data->pagevec, NULL);
		up_read(&current->mm->mmap_sem);
293 294 295 296 297
		if (result < 0) {
			nfs_readdata_release(data);
			break;
		}
		if ((unsigned)result < data->npages) {
298 299 300 301 302 303 304 305
			bytes = result * PAGE_SIZE;
			if (bytes <= pgbase) {
				nfs_direct_release_pages(data->pagevec, result);
				nfs_readdata_release(data);
				break;
			}
			bytes -= pgbase;
			data->npages = result;
306 307 308
		}

		get_dreq(dreq);
309

310
		data->req = (struct nfs_page *) dreq;
L
Linus Torvalds 已提交
311 312 313 314
		data->inode = inode;
		data->cred = ctx->cred;
		data->args.fh = NFS_FH(inode);
		data->args.context = ctx;
315
		data->args.offset = pos;
L
Linus Torvalds 已提交
316
		data->args.pgbase = pgbase;
317
		data->args.pages = data->pagevec;
L
Linus Torvalds 已提交
318 319 320 321 322
		data->args.count = bytes;
		data->res.fattr = &data->fattr;
		data->res.eof = 0;
		data->res.count = bytes;

T
Trond Myklebust 已提交
323 324
		rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
				&nfs_read_direct_ops, data);
L
Linus Torvalds 已提交
325 326 327 328 329 330
		NFS_PROTO(inode)->read_setup(data);

		data->task.tk_cookie = (unsigned long) inode;

		rpc_execute(&data->task);

C
Chuck Lever 已提交
331 332
		dprintk("NFS: %5u initiated direct read call "
			"(req %s/%Ld, %zu bytes @ offset %Lu)\n",
L
Linus Torvalds 已提交
333 334 335 336 337 338
				data->task.tk_pid,
				inode->i_sb->s_id,
				(long long)NFS_FILEID(inode),
				bytes,
				(unsigned long long)data->args.offset);

339 340
		started += bytes;
		user_addr += bytes;
341
		pos += bytes;
342
		/* FIXME: Remove this unnecessary math from final patch */
L
Linus Torvalds 已提交
343 344
		pgbase += bytes;
		pgbase &= ~PAGE_MASK;
345
		BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
L
Linus Torvalds 已提交
346 347 348

		count -= bytes;
	} while (count != 0);
349 350 351 352 353 354 355

	if (put_dreq(dreq))
		nfs_direct_complete(dreq);

	if (started)
		return 0;
	return result < 0 ? (ssize_t) result : -EFAULT;
L
Linus Torvalds 已提交
356 357
}

358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
					      const struct iovec *iov,
					      unsigned long nr_segs,
					      loff_t pos)
{
	ssize_t result = -EINVAL;
	size_t requested_bytes = 0;
	unsigned long seg;

	get_dreq(dreq);

	for (seg = 0; seg < nr_segs; seg++) {
		const struct iovec *vec = &iov[seg];
		result = nfs_direct_read_schedule(dreq,
					(unsigned long)vec->iov_base,
						  vec->iov_len, pos);
		if (result < 0)
			break;
		requested_bytes += result;
		if ((size_t)result < vec->iov_len)
			break;
		pos += vec->iov_len;
	}

	if (put_dreq(dreq))
		nfs_direct_complete(dreq);

	if (requested_bytes != 0)
		return 0;

	if (result < 0)
		return result;
	return -EIO;
}

393
static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
L
Linus Torvalds 已提交
394
{
395
	ssize_t result = 0;
L
Linus Torvalds 已提交
396
	sigset_t oldset;
397
	struct inode *inode = iocb->ki_filp->f_mapping->host;
L
Linus Torvalds 已提交
398 399 400
	struct rpc_clnt *clnt = NFS_CLIENT(inode);
	struct nfs_direct_req *dreq;

401
	dreq = nfs_direct_req_alloc();
L
Linus Torvalds 已提交
402 403 404
	if (!dreq)
		return -ENOMEM;

C
Chuck Lever 已提交
405
	dreq->inode = inode;
406
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
407 408
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
409

C
Chuck Lever 已提交
410
	nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
L
Linus Torvalds 已提交
411
	rpc_clnt_sigmask(clnt, &oldset);
412 413 414
	result = nfs_direct_read_schedule(dreq, user_addr, count, pos);
	if (!result)
		result = nfs_direct_wait(dreq);
L
Linus Torvalds 已提交
415
	rpc_clnt_sigunmask(clnt, &oldset);
416
	nfs_direct_req_release(dreq);
L
Linus Torvalds 已提交
417 418 419 420

	return result;
}

421
static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
422
{
423 424
	while (!list_empty(&dreq->rewrite_list)) {
		struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
425
		list_del(&data->pages);
426
		nfs_direct_release_pages(data->pagevec, data->npages);
427 428 429
		nfs_writedata_release(data);
	}
}
L
Linus Torvalds 已提交
430

431 432 433
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
{
434 435 436
	struct inode *inode = dreq->inode;
	struct list_head *p;
	struct nfs_write_data *data;
L
Linus Torvalds 已提交
437

438
	dreq->count = 0;
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
	get_dreq(dreq);

	list_for_each(p, &dreq->rewrite_list) {
		data = list_entry(p, struct nfs_write_data, pages);

		get_dreq(dreq);

		/*
		 * Reset data->res.
		 */
		nfs_fattr_init(&data->fattr);
		data->res.count = data->args.count;
		memset(&data->verf, 0, sizeof(data->verf));

		/*
		 * Reuse data->task; data->args should not have changed
		 * since the original request was sent.
		 */
		rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
				&nfs_write_direct_ops, data);
		NFS_PROTO(inode)->write_setup(data, FLUSH_STABLE);

		data->task.tk_priority = RPC_PRIORITY_NORMAL;
		data->task.tk_cookie = (unsigned long) inode;

		/*
		 * We're called via an RPC callback, so BKL is already held.
		 */
		rpc_execute(&data->task);

		dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
				data->task.tk_pid,
				inode->i_sb->s_id,
				(long long)NFS_FILEID(inode),
				data->args.count,
				(unsigned long long)data->args.offset);
	}
476

477 478
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, inode);
479 480 481 482 483 484 485 486 487 488 489
}

static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
{
	struct nfs_write_data *data = calldata;
	struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;

	/* Call the NFS version-specific code */
	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
		return;
	if (unlikely(task->tk_status < 0)) {
490 491
		dprintk("NFS: %5u commit failed with error %d.\n",
				task->tk_pid, task->tk_status);
492
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
493
	} else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
494 495
		dprintk("NFS: %5u commit verify failed\n", task->tk_pid);
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
L
Linus Torvalds 已提交
496 497
	}

498 499
	dprintk("NFS: %5u commit returned %d\n", task->tk_pid, task->tk_status);
	nfs_direct_write_complete(dreq, data->inode);
L
Linus Torvalds 已提交
500 501
}

502 503 504 505 506 507
static const struct rpc_call_ops nfs_commit_direct_ops = {
	.rpc_call_done = nfs_direct_commit_result,
	.rpc_release = nfs_commit_release,
};

static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
508
{
509
	struct nfs_write_data *data = dreq->commit_data;
L
Linus Torvalds 已提交
510

511
	data->inode = dreq->inode;
512
	data->cred = dreq->ctx->cred;
L
Linus Torvalds 已提交
513

514
	data->args.fh = NFS_FH(data->inode);
515 516
	data->args.offset = 0;
	data->args.count = 0;
517 518 519
	data->res.count = 0;
	data->res.fattr = &data->fattr;
	data->res.verf = &data->verf;
L
Linus Torvalds 已提交
520

521 522 523
	rpc_init_task(&data->task, NFS_CLIENT(dreq->inode), RPC_TASK_ASYNC,
				&nfs_commit_direct_ops, data);
	NFS_PROTO(data->inode)->commit_setup(data, 0);
L
Linus Torvalds 已提交
524

525 526 527 528
	data->task.tk_priority = RPC_PRIORITY_NORMAL;
	data->task.tk_cookie = (unsigned long)data->inode;
	/* Note: task.tk_ops->rpc_release will free dreq->commit_data */
	dreq->commit_data = NULL;
L
Linus Torvalds 已提交
529

530
	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
L
Linus Torvalds 已提交
531

532 533
	rpc_execute(&data->task);
}
L
Linus Torvalds 已提交
534

535 536 537
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
{
	int flags = dreq->flags;
L
Linus Torvalds 已提交
538

539 540 541 542
	dreq->flags = 0;
	switch (flags) {
		case NFS_ODIRECT_DO_COMMIT:
			nfs_direct_commit_schedule(dreq);
L
Linus Torvalds 已提交
543
			break;
544 545 546 547 548 549 550
		case NFS_ODIRECT_RESCHED_WRITES:
			nfs_direct_write_reschedule(dreq);
			break;
		default:
			if (dreq->commit_data != NULL)
				nfs_commit_free(dreq->commit_data);
			nfs_direct_free_writedata(dreq);
551
			nfs_zap_mapping(inode, inode->i_mapping);
552 553 554
			nfs_direct_complete(dreq);
	}
}
L
Linus Torvalds 已提交
555

556 557
static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
{
558
	dreq->commit_data = nfs_commit_alloc();
559 560 561 562 563 564 565 566
	if (dreq->commit_data != NULL)
		dreq->commit_data->req = (struct nfs_page *) dreq;
}
#else
static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
{
	dreq->commit_data = NULL;
}
L
Linus Torvalds 已提交
567

568 569 570
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
{
	nfs_direct_free_writedata(dreq);
571
	nfs_zap_mapping(inode, inode->i_mapping);
572 573 574
	nfs_direct_complete(dreq);
}
#endif
L
Linus Torvalds 已提交
575

576
static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
577
{
578 579 580 581 582 583 584
	struct nfs_write_data *data = calldata;
	struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
	int status = task->tk_status;

	if (nfs_writeback_done(task, data) != 0)
		return;

585
	spin_lock(&dreq->lock);
L
Linus Torvalds 已提交
586

587
	if (unlikely(status < 0)) {
588
		/* An error has occurred, so we should not commit */
589
		dreq->flags = 0;
590 591
		dreq->error = status;
	}
592 593
	if (unlikely(dreq->error != 0))
		goto out_unlock;
594 595

	dreq->count += data->res.count;
L
Linus Torvalds 已提交
596

597 598 599 600 601
	if (data->res.verf->committed != NFS_FILE_SYNC) {
		switch (dreq->flags) {
			case 0:
				memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
				dreq->flags = NFS_ODIRECT_DO_COMMIT;
L
Linus Torvalds 已提交
602
				break;
603 604 605 606 607
			case NFS_ODIRECT_DO_COMMIT:
				if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
					dprintk("NFS: %5u write verify failed\n", task->tk_pid);
					dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
				}
L
Linus Torvalds 已提交
608 609
		}
	}
610
out_unlock:
611
	spin_unlock(&dreq->lock);
L
Linus Torvalds 已提交
612 613
}

614 615 616
/*
 * NB: Return the value of the first error return code.  Subsequent
 *     errors after the first one are ignored.
L
Linus Torvalds 已提交
617
 */
618
static void nfs_direct_write_release(void *calldata)
L
Linus Torvalds 已提交
619
{
620 621
	struct nfs_write_data *data = calldata;
	struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
L
Linus Torvalds 已提交
622

623 624
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, data->inode);
625 626 627 628
}

static const struct rpc_call_ops nfs_write_direct_ops = {
	.rpc_call_done = nfs_direct_write_result,
629
	.rpc_release = nfs_direct_write_release,
630 631 632
};

/*
633 634 635 636 637
 * 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.
638
 */
639
static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos, int sync)
640
{
641
	struct nfs_open_context *ctx = dreq->ctx;
642
	struct inode *inode = ctx->path.dentry->d_inode;
643
	size_t wsize = NFS_SERVER(inode)->wsize;
644 645 646
	unsigned int pgbase;
	int result;
	ssize_t started = 0;
647

648 649
	get_dreq(dreq);

L
Linus Torvalds 已提交
650
	do {
651
		struct nfs_write_data *data;
652 653
		size_t bytes;

654 655 656
		pgbase = user_addr & ~PAGE_MASK;
		bytes = min(wsize,count);

657
		result = -ENOMEM;
658
		data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
659 660 661 662 663 664 665
		if (unlikely(!data))
			break;

		down_read(&current->mm->mmap_sem);
		result = get_user_pages(current, current->mm, user_addr,
					data->npages, 0, 0, data->pagevec, NULL);
		up_read(&current->mm->mmap_sem);
666 667 668 669 670
		if (result < 0) {
			nfs_writedata_release(data);
			break;
		}
		if ((unsigned)result < data->npages) {
671 672 673 674 675 676 677 678
			bytes = result * PAGE_SIZE;
			if (bytes <= pgbase) {
				nfs_direct_release_pages(data->pagevec, result);
				nfs_writedata_release(data);
				break;
			}
			bytes -= pgbase;
			data->npages = result;
679 680 681 682
		}

		get_dreq(dreq);

683
		list_move_tail(&data->pages, &dreq->rewrite_list);
684

685
		data->req = (struct nfs_page *) dreq;
686 687 688 689
		data->inode = inode;
		data->cred = ctx->cred;
		data->args.fh = NFS_FH(inode);
		data->args.context = ctx;
690
		data->args.offset = pos;
691
		data->args.pgbase = pgbase;
692
		data->args.pages = data->pagevec;
693 694 695
		data->args.count = bytes;
		data->res.fattr = &data->fattr;
		data->res.count = bytes;
696
		data->res.verf = &data->verf;
697 698 699

		rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
				&nfs_write_direct_ops, data);
700
		NFS_PROTO(inode)->write_setup(data, sync);
L
Linus Torvalds 已提交
701

702 703
		data->task.tk_priority = RPC_PRIORITY_NORMAL;
		data->task.tk_cookie = (unsigned long) inode;
L
Linus Torvalds 已提交
704

705
		rpc_execute(&data->task);
L
Linus Torvalds 已提交
706

C
Chuck Lever 已提交
707 708
		dprintk("NFS: %5u initiated direct write call "
			"(req %s/%Ld, %zu bytes @ offset %Lu)\n",
709 710 711 712 713
				data->task.tk_pid,
				inode->i_sb->s_id,
				(long long)NFS_FILEID(inode),
				bytes,
				(unsigned long long)data->args.offset);
L
Linus Torvalds 已提交
714

715 716
		started += bytes;
		user_addr += bytes;
717
		pos += bytes;
718 719

		/* FIXME: Remove this useless math from the final patch */
720 721
		pgbase += bytes;
		pgbase &= ~PAGE_MASK;
722
		BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
L
Linus Torvalds 已提交
723

724 725
		count -= bytes;
	} while (count != 0);
726 727 728 729 730 731 732

	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, inode);

	if (started)
		return 0;
	return result < 0 ? (ssize_t) result : -EFAULT;
733
}
L
Linus Torvalds 已提交
734

735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770
static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
					       const struct iovec *iov,
					       unsigned long nr_segs,
					       loff_t pos, int sync)
{
	ssize_t result = 0;
	size_t requested_bytes = 0;
	unsigned long seg;

	get_dreq(dreq);

	for (seg = 0; seg < nr_segs; seg++) {
		const struct iovec *vec = &iov[seg];
		result = nfs_direct_write_schedule(dreq,
					(unsigned long)vec->iov_base,
						   vec->iov_len,
						   pos, sync);
		if (result < 0)
			break;
		requested_bytes += result;
		if ((size_t)result < vec->iov_len)
			break;
		pos += vec->iov_len;
	}

	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);

	if (requested_bytes != 0)
		return 0;

	if (result < 0)
		return result;
	return -EIO;
}

771
static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
772
{
773
	ssize_t result = 0;
774
	sigset_t oldset;
775
	struct inode *inode = iocb->ki_filp->f_mapping->host;
776 777
	struct rpc_clnt *clnt = NFS_CLIENT(inode);
	struct nfs_direct_req *dreq;
778 779
	size_t wsize = NFS_SERVER(inode)->wsize;
	int sync = 0;
L
Linus Torvalds 已提交
780

781
	dreq = nfs_direct_req_alloc();
782 783
	if (!dreq)
		return -ENOMEM;
784 785
	nfs_alloc_commit_data(dreq);

786 787
	if (dreq->commit_data == NULL || count < wsize)
		sync = FLUSH_STABLE;
L
Linus Torvalds 已提交
788

789
	dreq->inode = inode;
790
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
791 792
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
793

794 795
	nfs_add_stats(inode, NFSIOS_DIRECTWRITTENBYTES, count);

796
	rpc_clnt_sigmask(clnt, &oldset);
797 798 799
	result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync);
	if (!result)
		result = nfs_direct_wait(dreq);
800
	rpc_clnt_sigunmask(clnt, &oldset);
801
	nfs_direct_req_release(dreq);
L
Linus Torvalds 已提交
802 803 804 805 806 807 808

	return result;
}

/**
 * nfs_file_direct_read - file direct read operation for NFS files
 * @iocb: target I/O control block
809 810
 * @iov: vector of user buffers into which to read data
 * @nr_segs: size of iov vector
811
 * @pos: byte offset in file where reading starts
L
Linus Torvalds 已提交
812 813 814 815 816 817
 *
 * 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
818
 * READ where the file size could change.  Our preference is simply
L
Linus Torvalds 已提交
819 820
 * to do all reads the application wants, and the server will take
 * care of managing the end of file boundary.
821
 *
L
Linus Torvalds 已提交
822 823 824 825 826
 * 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.
 */
827 828
ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
				unsigned long nr_segs, loff_t pos)
L
Linus Torvalds 已提交
829 830 831 832
{
	ssize_t retval = -EINVAL;
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
833 834 835
	/* XXX: temporary */
	const char __user *buf = iov[0].iov_base;
	size_t count = iov[0].iov_len;
L
Linus Torvalds 已提交
836

837
	dprintk("nfs: direct read(%s/%s, %lu@%Ld)\n",
838 839
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name,
840
		(unsigned long) count, (long long) pos);
L
Linus Torvalds 已提交
841

842
	if (nr_segs != 1)
L
Linus Torvalds 已提交
843
		goto out;
C
Chuck Lever 已提交
844

L
Linus Torvalds 已提交
845
	retval = -EFAULT;
846
	if (!access_ok(VERIFY_WRITE, buf, count))
L
Linus Torvalds 已提交
847 848 849 850 851
		goto out;
	retval = 0;
	if (!count)
		goto out;

T
Trond Myklebust 已提交
852 853 854
	retval = nfs_sync_mapping(mapping);
	if (retval)
		goto out;
L
Linus Torvalds 已提交
855

856
	retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos);
L
Linus Torvalds 已提交
857
	if (retval > 0)
858
		iocb->ki_pos = pos + retval;
L
Linus Torvalds 已提交
859 860 861 862 863 864 865 866

out:
	return retval;
}

/**
 * nfs_file_direct_write - file direct write operation for NFS files
 * @iocb: target I/O control block
867 868
 * @iov: vector of user buffers from which to write data
 * @nr_segs: size of iov vector
869
 * @pos: byte offset in file where writing starts
L
Linus Torvalds 已提交
870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888
 *
 * 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 also avoid an unnecessary invocation of generic_osync_inode(),
 * as it is fairly meaningless to sync the metadata of an NFS file.
 *
 * 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.
 */
889 890
ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
				unsigned long nr_segs, loff_t pos)
L
Linus Torvalds 已提交
891
{
C
Chuck Lever 已提交
892
	ssize_t retval = -EINVAL;
L
Linus Torvalds 已提交
893 894
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
895 896 897
	/* XXX: temporary */
	const char __user *buf = iov[0].iov_base;
	size_t count = iov[0].iov_len;
L
Linus Torvalds 已提交
898

C
Chuck Lever 已提交
899
	dprintk("nfs: direct write(%s/%s, %lu@%Ld)\n",
900 901
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name,
902
		(unsigned long) count, (long long) pos);
L
Linus Torvalds 已提交
903

904
	if (nr_segs != 1)
C
Chuck Lever 已提交
905
		goto out;
906

907 908
	retval = generic_write_checks(file, &pos, &count, 0);
	if (retval)
L
Linus Torvalds 已提交
909
		goto out;
910 911 912

	retval = -EINVAL;
	if ((ssize_t) count < 0)
L
Linus Torvalds 已提交
913 914 915 916
		goto out;
	retval = 0;
	if (!count)
		goto out;
917 918

	retval = -EFAULT;
919
	if (!access_ok(VERIFY_READ, buf, count))
920
		goto out;
L
Linus Torvalds 已提交
921

T
Trond Myklebust 已提交
922 923 924
	retval = nfs_sync_mapping(mapping);
	if (retval)
		goto out;
L
Linus Torvalds 已提交
925

926
	retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos);
927

L
Linus Torvalds 已提交
928
	if (retval > 0)
929
		iocb->ki_pos = pos + retval;
L
Linus Torvalds 已提交
930 931 932 933 934

out:
	return retval;
}

935 936 937 938
/**
 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 *
 */
D
David Howells 已提交
939
int __init nfs_init_directcache(void)
L
Linus Torvalds 已提交
940 941 942
{
	nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
						sizeof(struct nfs_direct_req),
943 944
						0, (SLAB_RECLAIM_ACCOUNT|
							SLAB_MEM_SPREAD),
945
						NULL);
L
Linus Torvalds 已提交
946 947 948 949 950 951
	if (nfs_direct_cachep == NULL)
		return -ENOMEM;

	return 0;
}

952
/**
D
David Howells 已提交
953
 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
954 955
 *
 */
956
void nfs_destroy_directcache(void)
L
Linus Torvalds 已提交
957
{
958
	kmem_cache_destroy(nfs_direct_cachep);
L
Linus Torvalds 已提交
959
}