direct.c 27.0 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>
L
Linus Torvalds 已提交
49 50 51 52 53 54

#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
55
#include <linux/atomic.h>
L
Linus Torvalds 已提交
56

57
#include "internal.h"
C
Chuck Lever 已提交
58
#include "iostat.h"
59
#include "pnfs.h"
C
Chuck Lever 已提交
60

L
Linus Torvalds 已提交
61 62
#define NFSDBG_FACILITY		NFSDBG_VFS

63
static struct kmem_cache *nfs_direct_cachep;
L
Linus Torvalds 已提交
64 65 66 67 68 69

/*
 * This represents a set of asynchronous requests that we're waiting on
 */
struct nfs_direct_req {
	struct kref		kref;		/* release manager */
70 71

	/* I/O parameters */
72
	struct nfs_open_context	*ctx;		/* file open context info */
73
	struct nfs_lock_context *l_ctx;		/* Lock context info */
74
	struct kiocb *		iocb;		/* controlling i/o request */
75
	struct inode *		inode;		/* target file of i/o */
76 77

	/* completion state */
78
	atomic_t		io_count;	/* i/os we're waiting for */
79 80
	spinlock_t		lock;		/* protect completion state */
	ssize_t			count,		/* bytes actually processed */
L
Linus Torvalds 已提交
81
				error;		/* any reported error */
82
	struct completion	completion;	/* wait for i/o completion */
83 84

	/* commit state */
85 86 87
	struct nfs_mds_commit_info mds_cinfo;	/* Storage for cinfo */
	struct pnfs_ds_commit_info ds_cinfo;	/* Storage for cinfo */
	struct work_struct	work;
88 89 90 91
	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
92 93
};

94 95
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops;
static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops;
96
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
97
static void nfs_direct_write_schedule_work(struct work_struct *work);
98 99 100 101 102 103 104 105 106 107 108

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
109
/**
110 111 112 113 114 115 116 117 118 119 120
 * 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
121
 */
122 123 124
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",
125
			iocb->ki_filp->f_path.dentry->d_name.name,
126
			(long long) pos, nr_segs);
127 128 129 130

	return -EINVAL;
}

131
static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
132
{
133
	unsigned int i;
134 135
	for (i = 0; i < npages; i++)
		page_cache_release(pages[i]);
136 137
}

138 139 140 141 142 143 144 145 146 147
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;
}

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);
159 160 161 162 163
	dreq->mds_cinfo.ncommit = 0;
	atomic_set(&dreq->mds_cinfo.rpcs_out, 0);
	INIT_LIST_HEAD(&dreq->mds_cinfo.list);
	INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
	memset(&dreq->ds_cinfo, 0, sizeof(dreq->ds_cinfo));
164
	dreq->iocb = NULL;
165
	dreq->ctx = NULL;
166
	dreq->l_ctx = NULL;
167
	spin_lock_init(&dreq->lock);
168
	atomic_set(&dreq->io_count, 0);
169 170
	dreq->count = 0;
	dreq->error = 0;
171
	dreq->flags = 0;
172 173

	return dreq;
L
Linus Torvalds 已提交
174 175
}

176
static void nfs_direct_req_free(struct kref *kref)
L
Linus Torvalds 已提交
177 178
{
	struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
179

180 181
	if (dreq->l_ctx != NULL)
		nfs_put_lock_context(dreq->l_ctx);
182 183
	if (dreq->ctx != NULL)
		put_nfs_open_context(dreq->ctx);
L
Linus Torvalds 已提交
184 185 186
	kmem_cache_free(nfs_direct_cachep, dreq);
}

187 188 189 190 191
static void nfs_direct_req_release(struct nfs_direct_req *dreq)
{
	kref_put(&dreq->kref, nfs_direct_req_free);
}

192 193 194 195 196
/*
 * Collects and returns the final error value/byte-count.
 */
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
197
	ssize_t result = -EIOCBQUEUED;
198 199 200 201 202

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

203
	result = wait_for_completion_killable(&dreq->completion);
204 205

	if (!result)
206
		result = dreq->error;
207
	if (!result)
208
		result = dreq->count;
209 210 211 212 213

out:
	return (ssize_t) result;
}

214
/*
215 216
 * 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.
217 218 219 220
 */
static void nfs_direct_complete(struct nfs_direct_req *dreq)
{
	if (dreq->iocb) {
221
		long res = (long) dreq->error;
222
		if (!res)
223
			res = (long) dreq->count;
224
		aio_complete(dreq->iocb, res, 0);
225 226
	}
	complete_all(&dreq->completion);
227

228
	nfs_direct_req_release(dreq);
229 230
}

231
void nfs_direct_readpage_release(struct nfs_page *req)
L
Linus Torvalds 已提交
232
{
233 234 235 236 237 238
	dprintk("NFS: direct read done (%s/%lld %d@%lld)\n",
		req->wb_context->dentry->d_inode->i_sb->s_id,
		(long long)NFS_FILEID(req->wb_context->dentry->d_inode),
		req->wb_bytes,
		(long long)req_offset(req));
	nfs_release_request(req);
239 240
}

241
static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
242
{
243 244
	unsigned long bytes = 0;
	struct nfs_direct_req *dreq = hdr->dreq;
245

246 247
	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out_put;
248 249

	spin_lock(&dreq->lock);
250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
	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);

	if (!test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
		while (!list_empty(&hdr->pages)) {
			struct nfs_page *req = nfs_list_entry(hdr->pages.next);
			struct page *page = req->wb_page;

			if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
				if (bytes > hdr->good_bytes)
					zero_user(page, 0, PAGE_SIZE);
				else if (hdr->good_bytes - bytes < PAGE_SIZE)
					zero_user_segment(page,
						hdr->good_bytes & ~PAGE_MASK,
						PAGE_SIZE);
			}
			bytes += req->wb_bytes;
			nfs_list_remove_request(req);
			if (!PageCompound(page))
				set_page_dirty(page);
273
			nfs_direct_readpage_release(req);
274
		}
275
	} else {
276 277 278 279 280 281 282 283 284 285
		while (!list_empty(&hdr->pages)) {
			struct nfs_page *req = nfs_list_entry(hdr->pages.next);

			if (bytes < hdr->good_bytes)
				if (!PageCompound(req->wb_page))
					set_page_dirty(req->wb_page);
			bytes += req->wb_bytes;
			nfs_list_remove_request(req);
			nfs_direct_readpage_release(req);
		}
286
	}
287
out_put:
288 289
	if (put_dreq(dreq))
		nfs_direct_complete(dreq);
290
	hdr->release(hdr);
L
Linus Torvalds 已提交
291 292
}

293
static void nfs_read_sync_pgio_error(struct list_head *head)
294
{
295
	struct nfs_page *req;
296

297 298 299 300 301
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_release_request(req);
	}
302 303
}

304 305 306 307 308 309
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 = {
310
	.error_cleanup = nfs_read_sync_pgio_error,
311 312 313 314
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_read_completion,
};

315
/*
316 317 318 319 320
 * 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
321
 */
322
static ssize_t nfs_direct_read_schedule_segment(struct nfs_pageio_descriptor *desc,
323 324
						const struct iovec *iov,
						loff_t pos)
L
Linus Torvalds 已提交
325
{
326
	struct nfs_direct_req *dreq = desc->pg_dreq;
327
	struct nfs_open_context *ctx = dreq->ctx;
328
	struct inode *inode = ctx->dentry->d_inode;
329 330
	unsigned long user_addr = (unsigned long)iov->iov_base;
	size_t count = iov->iov_len;
331
	size_t rsize = NFS_SERVER(inode)->rsize;
332 333 334
	unsigned int pgbase;
	int result;
	ssize_t started = 0;
335 336
	struct page **pagevec = NULL;
	unsigned int npages;
337

L
Linus Torvalds 已提交
338
	do {
339
		size_t bytes;
340
		int i;
L
Linus Torvalds 已提交
341

342
		pgbase = user_addr & ~PAGE_MASK;
343
		bytes = min(max(rsize, PAGE_SIZE), count);
344

345
		result = -ENOMEM;
346 347 348 349 350
		npages = nfs_page_array_len(pgbase, bytes);
		if (!pagevec)
			pagevec = kmalloc(npages * sizeof(struct page *),
					  GFP_KERNEL);
		if (!pagevec)
351
			break;
352 353
		down_read(&current->mm->mmap_sem);
		result = get_user_pages(current, current->mm, user_addr,
354
					npages, 1, 0, pagevec, NULL);
355
		up_read(&current->mm->mmap_sem);
356
		if (result < 0)
357
			break;
358
		if ((unsigned)result < npages) {
359 360
			bytes = result * PAGE_SIZE;
			if (bytes <= pgbase) {
361
				nfs_direct_release_pages(pagevec, result);
362 363 364
				break;
			}
			bytes -= pgbase;
365
			npages = result;
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
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
			unsigned int req_len = min(bytes, PAGE_SIZE - pgbase);
			/* XXX do we need to do the eof zeroing found in async_filler? */
			req = nfs_create_request(dreq->ctx, dreq->inode,
						 pagevec[i],
						 pgbase, req_len);
			if (IS_ERR(req)) {
				result = PTR_ERR(req);
				break;
			}
			req->wb_index = pos >> PAGE_SHIFT;
			req->wb_offset = pos & ~PAGE_MASK;
			if (!nfs_pageio_add_request(desc, req)) {
				result = desc->pg_error;
				nfs_release_request(req);
				break;
			}
			pgbase = 0;
			bytes -= req_len;
			started += req_len;
			user_addr += req_len;
			pos += req_len;
			count -= req_len;
		}
393 394
		/* The nfs_page now hold references to these pages */
		nfs_direct_release_pages(pagevec, npages);
395
	} while (count != 0 && result >= 0);
396

397 398
	kfree(pagevec);

399
	if (started)
400
		return started;
401
	return result < 0 ? (ssize_t) result : -EFAULT;
L
Linus Torvalds 已提交
402 403
}

404 405 406 407 408
static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
					      const struct iovec *iov,
					      unsigned long nr_segs,
					      loff_t pos)
{
409
	struct nfs_pageio_descriptor desc;
410 411 412 413
	ssize_t result = -EINVAL;
	size_t requested_bytes = 0;
	unsigned long seg;

414 415
	nfs_pageio_init_read(&desc, dreq->inode,
			     &nfs_direct_read_completion_ops);
416
	get_dreq(dreq);
417
	desc.pg_dreq = dreq;
418 419 420

	for (seg = 0; seg < nr_segs; seg++) {
		const struct iovec *vec = &iov[seg];
421
		result = nfs_direct_read_schedule_segment(&desc, vec, pos);
422 423 424 425 426 427 428 429
		if (result < 0)
			break;
		requested_bytes += result;
		if ((size_t)result < vec->iov_len)
			break;
		pos += vec->iov_len;
	}

430 431
	nfs_pageio_complete(&desc);

432 433 434 435 436 437 438 439 440
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

441 442
	if (put_dreq(dreq))
		nfs_direct_complete(dreq);
443
	return 0;
444 445
}

446 447
static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
			       unsigned long nr_segs, loff_t pos)
L
Linus Torvalds 已提交
448
{
449
	ssize_t result = -ENOMEM;
450
	struct inode *inode = iocb->ki_filp->f_mapping->host;
L
Linus Torvalds 已提交
451 452
	struct nfs_direct_req *dreq;

453
	dreq = nfs_direct_req_alloc();
454 455
	if (dreq == NULL)
		goto out;
L
Linus Torvalds 已提交
456

C
Chuck Lever 已提交
457
	dreq->inode = inode;
458
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
459 460 461
	dreq->l_ctx = nfs_get_lock_context(dreq->ctx);
	if (dreq->l_ctx == NULL)
		goto out_release;
462 463
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
464

465
	result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
466 467
	if (!result)
		result = nfs_direct_wait(dreq);
468
out_release:
469
	nfs_direct_req_release(dreq);
470
out:
L
Linus Torvalds 已提交
471 472 473
	return result;
}

474 475 476
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
{
477 478 479 480 481 482 483 484 485 486 487
	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
488

489
	dreq->count = 0;
490 491
	get_dreq(dreq);

492 493 494
	nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE,
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
495

496 497 498 499 500 501 502 503 504 505
	list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
		if (!nfs_pageio_add_request(&desc, req)) {
			nfs_list_add_request(req, &failed);
			spin_lock(cinfo.lock);
			dreq->flags = 0;
			dreq->error = -EIO;
			spin_unlock(cinfo.lock);
		}
	}
	nfs_pageio_complete(&desc);
506

507 508 509 510
	while (!list_empty(&failed)) {
		nfs_release_request(req);
		nfs_unlock_request(req);
	}
511

512 513
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
514 515
}

516
static void nfs_direct_commit_complete(struct nfs_commit_data *data)
517
{
518
	struct nfs_direct_req *dreq = data->dreq;
519 520
	struct nfs_commit_info cinfo;
	struct nfs_page *req;
521 522
	int status = data->task.tk_status;

523
	nfs_init_cinfo_from_dreq(&cinfo, dreq);
524
	if (status < 0) {
525
		dprintk("NFS: %5u commit failed with error %d.\n",
526
			data->task.tk_pid, status);
527
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
528
	} else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
529
		dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
530
		dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
L
Linus Torvalds 已提交
531 532
	}

533
	dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
534 535 536 537 538 539
	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);
540
		} else
541 542 543 544 545 546
			nfs_release_request(req);
		nfs_unlock_request(req);
	}

	if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
		nfs_direct_write_complete(dreq, data->inode);
L
Linus Torvalds 已提交
547 548
}

549 550 551 552 553 554 555 556
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,
557 558 559
};

static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
L
Linus Torvalds 已提交
560
{
561 562 563 564 565 566 567 568 569
	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);
570
}
L
Linus Torvalds 已提交
571

572
static void nfs_direct_write_schedule_work(struct work_struct *work)
573
{
574
	struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work);
575
	int flags = dreq->flags;
L
Linus Torvalds 已提交
576

577 578 579 580
	dreq->flags = 0;
	switch (flags) {
		case NFS_ODIRECT_DO_COMMIT:
			nfs_direct_commit_schedule(dreq);
L
Linus Torvalds 已提交
581
			break;
582 583 584 585
		case NFS_ODIRECT_RESCHED_WRITES:
			nfs_direct_write_reschedule(dreq);
			break;
		default:
586
			nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping);
587 588 589
			nfs_direct_complete(dreq);
	}
}
L
Linus Torvalds 已提交
590

591
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
592
{
593
	schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */
594
}
595

596
#else
597 598 599
static void nfs_direct_write_schedule_work(struct work_struct *work)
{
}
L
Linus Torvalds 已提交
600

601 602
static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
{
603
	nfs_zap_mapping(inode, inode->i_mapping);
604 605 606
	nfs_direct_complete(dreq);
}
#endif
L
Linus Torvalds 已提交
607

608 609 610 611
/*
 * NB: Return the value of the first error return code.  Subsequent
 *     errors after the first one are ignored.
 */
612
/*
613 614 615 616 617
 * 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.
618
 */
619
static ssize_t nfs_direct_write_schedule_segment(struct nfs_pageio_descriptor *desc,
620
						 const struct iovec *iov,
621
						 loff_t pos)
622
{
623
	struct nfs_direct_req *dreq = desc->pg_dreq;
624
	struct nfs_open_context *ctx = dreq->ctx;
625
	struct inode *inode = ctx->dentry->d_inode;
626 627
	unsigned long user_addr = (unsigned long)iov->iov_base;
	size_t count = iov->iov_len;
628
	size_t wsize = NFS_SERVER(inode)->wsize;
629 630 631
	unsigned int pgbase;
	int result;
	ssize_t started = 0;
632 633
	struct page **pagevec = NULL;
	unsigned int npages;
634

L
Linus Torvalds 已提交
635
	do {
636
		size_t bytes;
637
		int i;
638

639
		pgbase = user_addr & ~PAGE_MASK;
640
		bytes = min(max(wsize, PAGE_SIZE), count);
641

642
		result = -ENOMEM;
643 644 645 646
		npages = nfs_page_array_len(pgbase, bytes);
		if (!pagevec)
			pagevec = kmalloc(npages * sizeof(struct page *), GFP_KERNEL);
		if (!pagevec)
647 648 649 650
			break;

		down_read(&current->mm->mmap_sem);
		result = get_user_pages(current, current->mm, user_addr,
651
					npages, 0, 0, pagevec, NULL);
652
		up_read(&current->mm->mmap_sem);
653
		if (result < 0)
654
			break;
655 656

		if ((unsigned)result < npages) {
657 658
			bytes = result * PAGE_SIZE;
			if (bytes <= pgbase) {
659
				nfs_direct_release_pages(pagevec, result);
660 661 662
				break;
			}
			bytes -= pgbase;
663
			npages = result;
664 665
		}

666 667 668
		for (i = 0; i < npages; i++) {
			struct nfs_page *req;
			unsigned int req_len = min(bytes, PAGE_SIZE - pgbase);
L
Linus Torvalds 已提交
669

670 671 672 673 674 675 676 677 678 679 680 681 682 683
			req = nfs_create_request(dreq->ctx, dreq->inode,
						 pagevec[i],
						 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_request(req);
				nfs_release_request(req);
684
				break;
685 686 687 688 689 690 691 692
			}
			pgbase = 0;
			bytes -= req_len;
			started += req_len;
			user_addr += req_len;
			pos += req_len;
			count -= req_len;
		}
693 694
		/* The nfs_page now hold references to these pages */
		nfs_direct_release_pages(pagevec, npages);
695
	} while (count != 0 && result >= 0);
696

697 698
	kfree(pagevec);

699
	if (started)
700
		return started;
701
	return result < 0 ? (ssize_t) result : -EFAULT;
702
}
L
Linus Torvalds 已提交
703

704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767
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) {
				memcpy(&dreq->verf, &req->wb_verf,
				       sizeof(dreq->verf));
				bit = NFS_IOHDR_NEED_COMMIT;
				dreq->flags = NFS_ODIRECT_DO_COMMIT;
			} else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) {
				if (memcmp(&dreq->verf, &req->wb_verf, sizeof(dreq->verf))) {
					dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
					bit = NFS_IOHDR_NEED_RESCHED;
				} else
					bit = NFS_IOHDR_NEED_COMMIT;
			}
		}
	}
	spin_unlock(&dreq->lock);

	while (!list_empty(&hdr->pages)) {
		req = nfs_list_entry(hdr->pages.next);
		nfs_list_remove_request(req);
		switch (bit) {
		case NFS_IOHDR_NEED_RESCHED:
		case NFS_IOHDR_NEED_COMMIT:
			nfs_mark_request_commit(req, hdr->lseg, &cinfo);
			break;
		default:
			nfs_release_request(req);
		}
		nfs_unlock_request(req);
	}

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

768 769 770 771 772 773 774 775 776 777 778 779
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);
		nfs_release_request(req);
		nfs_unlock_request(req);
	}
}

780
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
781
	.error_cleanup = nfs_write_sync_pgio_error,
782 783 784 785
	.init_hdr = nfs_direct_pgio_init,
	.completion = nfs_direct_write_completion,
};

786 787 788
static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
					       const struct iovec *iov,
					       unsigned long nr_segs,
789
					       loff_t pos)
790
{
791
	struct nfs_pageio_descriptor desc;
792 793 794 795
	ssize_t result = 0;
	size_t requested_bytes = 0;
	unsigned long seg;

796 797 798
	nfs_pageio_init_write(&desc, dreq->inode, FLUSH_COND_STABLE,
			      &nfs_direct_write_completion_ops);
	desc.pg_dreq = dreq;
799 800 801 802
	get_dreq(dreq);

	for (seg = 0; seg < nr_segs; seg++) {
		const struct iovec *vec = &iov[seg];
803
		result = nfs_direct_write_schedule_segment(&desc, vec, pos);
804 805 806 807 808 809 810
		if (result < 0)
			break;
		requested_bytes += result;
		if ((size_t)result < vec->iov_len)
			break;
		pos += vec->iov_len;
	}
811
	nfs_pageio_complete(&desc);
812

813 814 815 816 817 818 819 820 821
	/*
	 * If no bytes were started, return the error, and let the
	 * generic layer handle the completion.
	 */
	if (requested_bytes == 0) {
		nfs_direct_req_release(dreq);
		return result < 0 ? result : -EIO;
	}

822 823
	if (put_dreq(dreq))
		nfs_direct_write_complete(dreq, dreq->inode);
824
	return 0;
825 826
}

827 828 829
static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
				unsigned long nr_segs, loff_t pos,
				size_t count)
830
{
831
	ssize_t result = -ENOMEM;
832
	struct inode *inode = iocb->ki_filp->f_mapping->host;
833
	struct nfs_direct_req *dreq;
L
Linus Torvalds 已提交
834

835
	dreq = nfs_direct_req_alloc();
836
	if (!dreq)
837
		goto out;
L
Linus Torvalds 已提交
838

839
	dreq->inode = inode;
840
	dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
841
	dreq->l_ctx = nfs_get_lock_context(dreq->ctx);
842
	if (dreq->l_ctx == NULL)
843
		goto out_release;
844 845
	if (!is_sync_kiocb(iocb))
		dreq->iocb = iocb;
L
Linus Torvalds 已提交
846

847
	result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos);
848 849
	if (!result)
		result = nfs_direct_wait(dreq);
850
out_release:
851
	nfs_direct_req_release(dreq);
852
out:
L
Linus Torvalds 已提交
853 854 855 856 857 858
	return result;
}

/**
 * nfs_file_direct_read - file direct read operation for NFS files
 * @iocb: target I/O control block
859 860
 * @iov: vector of user buffers into which to read data
 * @nr_segs: size of iov vector
861
 * @pos: byte offset in file where reading starts
L
Linus Torvalds 已提交
862 863 864 865 866 867
 *
 * 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
868
 * READ where the file size could change.  Our preference is simply
L
Linus Torvalds 已提交
869 870
 * to do all reads the application wants, and the server will take
 * care of managing the end of file boundary.
871
 *
L
Linus Torvalds 已提交
872 873 874 875 876
 * 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.
 */
877 878
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
879 880 881 882
{
	ssize_t retval = -EINVAL;
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
883 884 885 886
	size_t count;

	count = iov_length(iov, nr_segs);
	nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
L
Linus Torvalds 已提交
887

C
Chuck Lever 已提交
888
	dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
889 890
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name,
891
		count, (long long) pos);
L
Linus Torvalds 已提交
892 893 894 895 896

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

T
Trond Myklebust 已提交
897 898 899
	retval = nfs_sync_mapping(mapping);
	if (retval)
		goto out;
L
Linus Torvalds 已提交
900

901 902
	task_io_account_read(count);

903
	retval = nfs_direct_read(iocb, iov, nr_segs, pos);
L
Linus Torvalds 已提交
904
	if (retval > 0)
905
		iocb->ki_pos = pos + retval;
L
Linus Torvalds 已提交
906 907 908 909 910 911 912 913

out:
	return retval;
}

/**
 * nfs_file_direct_write - file direct write operation for NFS files
 * @iocb: target I/O control block
914 915
 * @iov: vector of user buffers from which to write data
 * @nr_segs: size of iov vector
916
 * @pos: byte offset in file where writing starts
L
Linus Torvalds 已提交
917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932
 *
 * 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.
 */
933 934
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
935
{
C
Chuck Lever 已提交
936
	ssize_t retval = -EINVAL;
L
Linus Torvalds 已提交
937 938
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
939
	size_t count;
L
Linus Torvalds 已提交
940

941 942 943
	count = iov_length(iov, nr_segs);
	nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);

C
Chuck Lever 已提交
944
	dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
945 946
		file->f_path.dentry->d_parent->d_name.name,
		file->f_path.dentry->d_name.name,
947
		count, (long long) pos);
948

949 950
	retval = generic_write_checks(file, &pos, &count, 0);
	if (retval)
L
Linus Torvalds 已提交
951
		goto out;
952 953 954

	retval = -EINVAL;
	if ((ssize_t) count < 0)
L
Linus Torvalds 已提交
955 956 957 958
		goto out;
	retval = 0;
	if (!count)
		goto out;
959

T
Trond Myklebust 已提交
960 961 962
	retval = nfs_sync_mapping(mapping);
	if (retval)
		goto out;
L
Linus Torvalds 已提交
963

964 965
	task_io_account_write(count);

966
	retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
967 968
	if (retval > 0) {
		struct inode *inode = mapping->host;
969

970
		iocb->ki_pos = pos + retval;
971 972 973 974 975
		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);
	}
L
Linus Torvalds 已提交
976 977 978 979
out:
	return retval;
}

980 981 982 983
/**
 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 *
 */
D
David Howells 已提交
984
int __init nfs_init_directcache(void)
L
Linus Torvalds 已提交
985 986 987
{
	nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
						sizeof(struct nfs_direct_req),
988 989
						0, (SLAB_RECLAIM_ACCOUNT|
							SLAB_MEM_SPREAD),
990
						NULL);
L
Linus Torvalds 已提交
991 992 993 994 995 996
	if (nfs_direct_cachep == NULL)
		return -ENOMEM;

	return 0;
}

997
/**
D
David Howells 已提交
998
 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
999 1000
 *
 */
1001
void nfs_destroy_directcache(void)
L
Linus Torvalds 已提交
1002
{
1003
	kmem_cache_destroy(nfs_direct_cachep);
L
Linus Torvalds 已提交
1004
}