read.c 17.7 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
/*
 * linux/fs/nfs/read.c
 *
 * Block I/O for NFS
 *
 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
 * modified for async RPC by okir@monad.swb.de
 */

#include <linux/time.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/fcntl.h>
#include <linux/stat.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_page.h>
A
Andy Adamson 已提交
21
#include <linux/module.h>
L
Linus Torvalds 已提交
22

23
#include "nfs4_fs.h"
24
#include "internal.h"
C
Chuck Lever 已提交
25
#include "iostat.h"
26
#include "fscache.h"
27
#include "pnfs.h"
C
Chuck Lever 已提交
28

L
Linus Torvalds 已提交
29 30
#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

31
static const struct nfs_pageio_ops nfs_pageio_read_ops;
32
static const struct rpc_call_ops nfs_read_common_ops;
33
static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
L
Linus Torvalds 已提交
34

35
static struct kmem_cache *nfs_rdata_cachep;
L
Linus Torvalds 已提交
36

37
struct nfs_rw_header *nfs_readhdr_alloc(void)
38
{
39
	struct nfs_rw_header *rhdr;
40

41 42 43
	rhdr = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
	if (rhdr) {
		struct nfs_pgio_header *hdr = &rhdr->header;
44 45

		INIT_LIST_HEAD(&hdr->pages);
46 47 48 49 50 51
		INIT_LIST_HEAD(&hdr->rpc_list);
		spin_lock_init(&hdr->lock);
		atomic_set(&hdr->refcnt, 0);
	}
	return rhdr;
}
B
Bryan Schumaker 已提交
52
EXPORT_SYMBOL_GPL(nfs_readhdr_alloc);
53

54
static struct nfs_pgio_data *nfs_readdata_alloc(struct nfs_pgio_header *hdr,
55
						unsigned int pagecount)
56
{
57
	struct nfs_pgio_data *data, *prealloc;
58

59
	prealloc = &container_of(hdr, struct nfs_rw_header, header)->rpc_data;
60 61 62 63 64 65 66 67
	if (prealloc->header == NULL)
		data = prealloc;
	else
		data = kzalloc(sizeof(*data), GFP_KERNEL);
	if (!data)
		goto out;

	if (nfs_pgarray_set(&data->pages, pagecount)) {
68
		data->header = hdr;
69 70 71 72 73
		atomic_inc(&hdr->refcnt);
	} else {
		if (data != prealloc)
			kfree(data);
		data = NULL;
74
	}
75 76
out:
	return data;
77 78
}

79
void nfs_readhdr_free(struct nfs_pgio_header *hdr)
80
{
81
	struct nfs_rw_header *rhdr = container_of(hdr, struct nfs_rw_header, header);
82 83

	kmem_cache_free(nfs_rdata_cachep, rhdr);
84
}
B
Bryan Schumaker 已提交
85
EXPORT_SYMBOL_GPL(nfs_readhdr_free);
86

87
void nfs_readdata_release(struct nfs_pgio_data *rdata)
L
Linus Torvalds 已提交
88
{
89
	struct nfs_pgio_header *hdr = rdata->header;
90
	struct nfs_rw_header *read_header = container_of(hdr, struct nfs_rw_header, header);
91

92
	put_nfs_open_context(rdata->args.context);
F
Fred Isaman 已提交
93 94
	if (rdata->pages.pagevec != rdata->pages.page_array)
		kfree(rdata->pages.pagevec);
95
	if (rdata == &read_header->rpc_data) {
96
		rdata->header = NULL;
97 98
		rdata = NULL;
	}
99
	if (atomic_dec_and_test(&hdr->refcnt))
100
		hdr->completion_ops->completion(hdr);
101 102 103 104
	/* Note: we only free the rpc_task after callbacks are done.
	 * See the comment in rpc_free_task() for why
	 */
	kfree(rdata);
L
Linus Torvalds 已提交
105
}
B
Bryan Schumaker 已提交
106
EXPORT_SYMBOL_GPL(nfs_readdata_release);
L
Linus Torvalds 已提交
107 108 109 110

static
int nfs_return_empty_page(struct page *page)
{
111
	zero_user(page, 0, PAGE_CACHE_SIZE);
L
Linus Torvalds 已提交
112 113 114 115 116
	SetPageUptodate(page);
	unlock_page(page);
	return 0;
}

117
void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
118
			      struct inode *inode, bool force_mds,
119
			      const struct nfs_pgio_completion_ops *compl_ops)
120
{
121 122 123 124 125 126 127 128
	struct nfs_server *server = NFS_SERVER(inode);
	const struct nfs_pageio_ops *pg_ops = &nfs_pageio_read_ops;

#ifdef CONFIG_NFS_V4_1
	if (server->pnfs_curr_ld && !force_mds)
		pg_ops = server->pnfs_curr_ld->pg_read_ops;
#endif
	nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->rsize, 0);
129
}
B
Bryan Schumaker 已提交
130
EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
131

132 133 134 135 136
void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
{
	pgio->pg_ops = &nfs_pageio_read_ops;
	pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
}
137
EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
138

139 140
int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
		       struct page *page)
L
Linus Torvalds 已提交
141 142 143
{
	struct nfs_page	*new;
	unsigned int len;
F
Fred Isaman 已提交
144
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
145

146
	len = nfs_page_length(page);
L
Linus Torvalds 已提交
147 148 149 150 151 152 153 154
	if (len == 0)
		return nfs_return_empty_page(page);
	new = nfs_create_request(ctx, inode, page, 0, len);
	if (IS_ERR(new)) {
		unlock_page(page);
		return PTR_ERR(new);
	}
	if (len < PAGE_CACHE_SIZE)
155
		zero_user_segment(page, len, PAGE_CACHE_SIZE);
L
Linus Torvalds 已提交
156

157 158
	nfs_pageio_init_read(&pgio, inode, false,
			     &nfs_async_read_completion_ops);
159
	nfs_pageio_add_request(&pgio, new);
160
	nfs_pageio_complete(&pgio);
161
	NFS_I(inode)->read_io += pgio.pg_bytes_written;
L
Linus Torvalds 已提交
162 163 164 165 166
	return 0;
}

static void nfs_readpage_release(struct nfs_page *req)
{
167
	struct inode *d_inode = req->wb_context->dentry->d_inode;
168 169 170 171

	if (PageUptodate(req->wb_page))
		nfs_readpage_to_fscache(d_inode, req->wb_page, 0);

L
Linus Torvalds 已提交
172 173
	unlock_page(req->wb_page);

174
	dprintk("NFS: read done (%s/%Lu %d@%Ld)\n",
175
			req->wb_context->dentry->d_inode->i_sb->s_id,
176
			(unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
L
Linus Torvalds 已提交
177 178
			req->wb_bytes,
			(long long)req_offset(req));
179
	nfs_release_request(req);
L
Linus Torvalds 已提交
180 181
}

182
/* Note io was page aligned */
183
static void nfs_read_completion(struct nfs_pgio_header *hdr)
184 185 186 187 188
{
	unsigned long bytes = 0;

	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
		goto out;
189 190 191 192 193 194 195 196 197 198 199
	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);
200
		}
201 202
		bytes += req->wb_bytes;
		if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
203
			if (bytes <= hdr->good_bytes)
204 205 206 207 208
				SetPageUptodate(page);
		} else
			SetPageUptodate(page);
		nfs_list_remove_request(req);
		nfs_readpage_release(req);
209 210 211 212 213
	}
out:
	hdr->release(hdr);
}

214
int nfs_initiate_read(struct rpc_clnt *clnt,
215
		      struct nfs_pgio_data *data,
216
		      const struct rpc_call_ops *call_ops, int flags)
L
Linus Torvalds 已提交
217
{
218
	struct inode *inode = data->header->inode;
219
	int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
220
	struct rpc_task *task;
221 222 223
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
224
		.rpc_cred = data->header->cred,
225
	};
226
	struct rpc_task_setup task_setup_data = {
227
		.task = &data->task,
A
Andy Adamson 已提交
228
		.rpc_client = clnt,
229
		.rpc_message = &msg,
230 231
		.callback_ops = call_ops,
		.callback_data = data,
232
		.workqueue = nfsiod_workqueue,
233
		.flags = RPC_TASK_ASYNC | swap_flags | flags,
234
	};
L
Linus Torvalds 已提交
235

A
Andy Adamson 已提交
236 237 238
	/* Set up the initial task struct. */
	NFS_PROTO(inode)->read_setup(data, &msg);

239
	dprintk("NFS: %5u initiated read call (req %s/%llu, %u bytes @ "
A
Andy Adamson 已提交
240 241 242
			"offset %llu)\n",
			data->task.tk_pid,
			inode->i_sb->s_id,
243
			(unsigned long long)NFS_FILEID(inode),
A
Andy Adamson 已提交
244 245 246 247 248 249 250 251 252
			data->args.count,
			(unsigned long long)data->args.offset);

	task = rpc_run_task(&task_setup_data);
	if (IS_ERR(task))
		return PTR_ERR(task);
	rpc_put_task(task);
	return 0;
}
A
Andy Adamson 已提交
253
EXPORT_SYMBOL_GPL(nfs_initiate_read);
A
Andy Adamson 已提交
254 255 256 257

/*
 * Set up the NFS read request struct
 */
258
static void nfs_read_rpcsetup(struct nfs_pgio_data *data,
259
		unsigned int count, unsigned int offset)
A
Andy Adamson 已提交
260
{
261
	struct nfs_page *req = data->header->req;
L
Linus Torvalds 已提交
262

263
	data->args.fh     = NFS_FH(data->header->inode);
L
Linus Torvalds 已提交
264 265
	data->args.offset = req_offset(req) + offset;
	data->args.pgbase = req->wb_pgbase + offset;
F
Fred Isaman 已提交
266
	data->args.pages  = data->pages.pagevec;
L
Linus Torvalds 已提交
267
	data->args.count  = count;
268
	data->args.context = get_nfs_open_context(req->wb_context);
269
	data->args.lock_context = req->wb_lock_context;
L
Linus Torvalds 已提交
270 271 272 273

	data->res.fattr   = &data->fattr;
	data->res.count   = count;
	data->res.eof     = 0;
274
	nfs_fattr_init(&data->fattr);
275
}
L
Linus Torvalds 已提交
276

277
static int nfs_do_read(struct nfs_pgio_data *data,
278
		const struct rpc_call_ops *call_ops)
279
{
280
	struct inode *inode = data->header->inode;
281

282
	return nfs_initiate_read(NFS_CLIENT(inode), data, call_ops, 0);
L
Linus Torvalds 已提交
283 284
}

285 286
static int
nfs_do_multiple_reads(struct list_head *head,
287
		const struct rpc_call_ops *call_ops)
288
{
289
	struct nfs_pgio_data *data;
290 291 292 293 294
	int ret = 0;

	while (!list_empty(head)) {
		int ret2;

295
		data = list_first_entry(head, struct nfs_pgio_data, list);
296 297
		list_del_init(&data->list);

298
		ret2 = nfs_do_read(data, call_ops);
299 300 301 302 303 304
		if (ret == 0)
			ret = ret2;
	}
	return ret;
}

305
static void
L
Linus Torvalds 已提交
306 307 308 309 310 311 312 313 314 315 316
nfs_async_read_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_readpage_release(req);
	}
}

317 318 319 320 321
static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
	.error_cleanup = nfs_async_read_error,
	.completion = nfs_read_completion,
};

322 323 324 325 326
static void nfs_pagein_error(struct nfs_pageio_descriptor *desc,
		struct nfs_pgio_header *hdr)
{
	set_bit(NFS_IOHDR_REDO, &hdr->flags);
	while (!list_empty(&hdr->rpc_list)) {
327 328
		struct nfs_pgio_data *data = list_first_entry(&hdr->rpc_list,
				struct nfs_pgio_data, list);
329 330 331 332 333 334
		list_del(&data->list);
		nfs_readdata_release(data);
	}
	desc->pg_completion_ops->error_cleanup(&desc->pg_list);
}

L
Linus Torvalds 已提交
335 336 337 338 339 340 341 342 343 344 345 346 347
/*
 * Generate multiple requests to fill a single page.
 *
 * We optimize to reduce the number of read operations on the wire.  If we
 * detect that we're reading a page, or an area of a page, that is past the
 * end of file, we do not generate NFS read operations but just clear the
 * parts of the page that would have come back zero from the server anyway.
 *
 * We rely on the cached value of i_size to make this determination; another
 * client can fill pages on the server past our cached end-of-file, but we
 * won't see the new data until our attribute cache is updated.  This is more
 * or less conventional NFS client behavior.
 */
348 349
static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc,
			    struct nfs_pgio_header *hdr)
L
Linus Torvalds 已提交
350
{
351
	struct nfs_page *req = hdr->req;
L
Linus Torvalds 已提交
352
	struct page *page = req->wb_page;
353
	struct nfs_pgio_data *data;
354
	size_t rsize = desc->pg_bsize, nbytes;
355
	unsigned int offset;
L
Linus Torvalds 已提交
356

357
	offset = 0;
F
Fred Isaman 已提交
358
	nbytes = desc->pg_count;
359 360 361
	do {
		size_t len = min(nbytes,rsize);

362
		data = nfs_readdata_alloc(hdr, 1);
363 364 365 366
		if (!data) {
			nfs_pagein_error(desc, hdr);
			return -ENOMEM;
		}
F
Fred Isaman 已提交
367
		data->pages.pagevec[0] = page;
368 369
		nfs_read_rpcsetup(data, len, offset);
		list_add(&data->list, &hdr->rpc_list);
370
		nbytes -= len;
371
		offset += len;
T
Trond Myklebust 已提交
372
	} while (nbytes != 0);
373 374 375

	nfs_list_remove_request(req);
	nfs_list_add_request(req, &hdr->pages);
376
	desc->pg_rpc_callops = &nfs_read_common_ops;
T
Trond Myklebust 已提交
377
	return 0;
L
Linus Torvalds 已提交
378 379
}

380 381
static int nfs_pagein_one(struct nfs_pageio_descriptor *desc,
			  struct nfs_pgio_header *hdr)
L
Linus Torvalds 已提交
382 383 384
{
	struct nfs_page		*req;
	struct page		**pages;
385
	struct nfs_pgio_data	*data;
F
Fred Isaman 已提交
386
	struct list_head *head = &desc->pg_list;
L
Linus Torvalds 已提交
387

388 389 390
	data = nfs_readdata_alloc(hdr, nfs_page_array_len(desc->pg_base,
							  desc->pg_count));
	if (!data) {
391
		nfs_pagein_error(desc, hdr);
T
Trond Myklebust 已提交
392
		return -ENOMEM;
393
	}
L
Linus Torvalds 已提交
394

F
Fred Isaman 已提交
395
	pages = data->pages.pagevec;
L
Linus Torvalds 已提交
396 397 398
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
399
		nfs_list_add_request(req, &hdr->pages);
L
Linus Torvalds 已提交
400 401 402
		*pages++ = req->wb_page;
	}

403 404 405
	nfs_read_rpcsetup(data, desc->pg_count, 0);
	list_add(&data->list, &hdr->rpc_list);
	desc->pg_rpc_callops = &nfs_read_common_ops;
T
Trond Myklebust 已提交
406
	return 0;
L
Linus Torvalds 已提交
407 408
}

409 410
int nfs_generic_pagein(struct nfs_pageio_descriptor *desc,
		       struct nfs_pgio_header *hdr)
411 412
{
	if (desc->pg_bsize < PAGE_CACHE_SIZE)
413 414
		return nfs_pagein_multi(desc, hdr);
	return nfs_pagein_one(desc, hdr);
415
}
B
Bryan Schumaker 已提交
416
EXPORT_SYMBOL_GPL(nfs_generic_pagein);
417 418

static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
419
{
420
	struct nfs_rw_header *rhdr;
421
	struct nfs_pgio_header *hdr;
422 423
	int ret;

424 425
	rhdr = nfs_readhdr_alloc();
	if (!rhdr) {
426
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
427 428 429 430 431 432
		return -ENOMEM;
	}
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, nfs_readhdr_free);
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_pagein(desc, hdr);
433
	if (ret == 0)
434 435 436
		ret = nfs_do_multiple_reads(&hdr->rpc_list,
					    desc->pg_rpc_callops);
	if (atomic_dec_and_test(&hdr->refcnt))
437
		hdr->completion_ops->completion(hdr);
438
	return ret;
439 440 441 442 443 444 445
}

static const struct nfs_pageio_ops nfs_pageio_read_ops = {
	.pg_test = nfs_generic_pg_test,
	.pg_doio = nfs_generic_pg_readpages,
};

446 447 448 449
/*
 * This is the callback from RPC telling us whether a reply was
 * received or some error occurred (timeout or socket shutdown).
 */
450
int nfs_readpage_result(struct rpc_task *task, struct nfs_pgio_data *data)
451
{
452
	struct inode *inode = data->header->inode;
453 454
	int status;

455
	dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
456 457
			task->tk_status);

458
	status = NFS_PROTO(inode)->read_done(task, data);
459 460 461
	if (status != 0)
		return status;

462
	nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, data->res.count);
463 464

	if (task->tk_status == -ESTALE) {
465 466
		set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
		nfs_mark_for_revalidate(inode);
467 468 469 470
	}
	return 0;
}

471
static void nfs_readpage_retry(struct rpc_task *task, struct nfs_pgio_data *data)
472
{
473
	struct nfs_pgio_args *argp = &data->args;
474
	struct nfs_pgio_res  *resp = &data->res;
475 476

	/* This is a short read! */
477
	nfs_inc_stats(data->header->inode, NFSIOS_SHORTREAD);
478
	/* Has the server at least made some progress? */
479 480
	if (resp->count == 0) {
		nfs_set_pgio_error(data->header, -EIO, argp->offset);
481
		return;
482
	}
483
	/* Yes, so retry the read at the end of the data */
484
	data->mds_offset += resp->count;
485 486 487
	argp->offset += resp->count;
	argp->pgbase += resp->count;
	argp->count -= resp->count;
488
	rpc_restart_call_prepare(task);
489 490
}

491
static void nfs_readpage_result_common(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
492
{
493
	struct nfs_pgio_data *data = calldata;
494 495 496
	struct nfs_pgio_header *hdr = data->header;

	/* Note the only returns of nfs_readpage_result are 0 and -EAGAIN */
T
Trond Myklebust 已提交
497 498
	if (nfs_readpage_result(task, data) != 0)
		return;
499
	if (task->tk_status < 0)
500 501 502 503 504 505 506 507 508 509 510 511 512 513
		nfs_set_pgio_error(hdr, task->tk_status, data->args.offset);
	else if (data->res.eof) {
		loff_t bound;

		bound = data->args.offset + data->res.count;
		spin_lock(&hdr->lock);
		if (bound < hdr->io_start + hdr->good_bytes) {
			set_bit(NFS_IOHDR_EOF, &hdr->flags);
			clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
			hdr->good_bytes = bound - hdr->io_start;
		}
		spin_unlock(&hdr->lock);
	} else if (data->res.count != data->args.count)
		nfs_readpage_retry(task, data);
514 515
}

516
static void nfs_readpage_release_common(void *calldata)
517
{
518
	nfs_readdata_release(calldata);
L
Linus Torvalds 已提交
519 520
}

521 522
void nfs_read_prepare(struct rpc_task *task, void *calldata)
{
523
	struct nfs_pgio_data *data = calldata;
524 525 526 527
	int err;
	err = NFS_PROTO(data->header->inode)->read_rpc_prepare(task, data);
	if (err)
		rpc_exit(task, err);
528 529
}

530
static const struct rpc_call_ops nfs_read_common_ops = {
531
	.rpc_call_prepare = nfs_read_prepare,
532 533
	.rpc_call_done = nfs_readpage_result_common,
	.rpc_release = nfs_readpage_release_common,
T
Trond Myklebust 已提交
534 535
};

L
Linus Torvalds 已提交
536 537 538 539 540 541 542 543 544
/*
 * Read a page over NFS.
 * We read the page synchronously in the following case:
 *  -	The error flag is set for this page. This happens only when a
 *	previous async read operation failed.
 */
int nfs_readpage(struct file *file, struct page *page)
{
	struct nfs_open_context *ctx;
545
	struct inode *inode = page_file_mapping(page)->host;
L
Linus Torvalds 已提交
546 547 548
	int		error;

	dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
549
		page, PAGE_CACHE_SIZE, page_file_index(page));
C
Chuck Lever 已提交
550 551 552
	nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
	nfs_add_stats(inode, NFSIOS_READPAGES, 1);

L
Linus Torvalds 已提交
553 554 555 556 557 558 559 560 561
	/*
	 * Try to flush any pending writes to the file..
	 *
	 * NOTE! Because we own the page lock, there cannot
	 * be any new pending writes generated at this point
	 * for this page (other pages can be written to).
	 */
	error = nfs_wb_page(inode, page);
	if (error)
562 563 564
		goto out_unlock;
	if (PageUptodate(page))
		goto out_unlock;
L
Linus Torvalds 已提交
565

566 567
	error = -ESTALE;
	if (NFS_STALE(inode))
568
		goto out_unlock;
569

L
Linus Torvalds 已提交
570
	if (file == NULL) {
571
		error = -EBADF;
572
		ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
L
Linus Torvalds 已提交
573
		if (ctx == NULL)
574
			goto out_unlock;
L
Linus Torvalds 已提交
575
	} else
576
		ctx = get_nfs_open_context(nfs_file_open_context(file));
L
Linus Torvalds 已提交
577

578 579 580 581 582 583
	if (!IS_SYNC(inode)) {
		error = nfs_readpage_from_fscache(ctx, inode, page);
		if (error == 0)
			goto out;
	}

584 585
	error = nfs_readpage_async(ctx, inode, page);

586
out:
L
Linus Torvalds 已提交
587 588
	put_nfs_open_context(ctx);
	return error;
589
out_unlock:
L
Linus Torvalds 已提交
590 591 592 593 594
	unlock_page(page);
	return error;
}

struct nfs_readdesc {
595
	struct nfs_pageio_descriptor *pgio;
L
Linus Torvalds 已提交
596 597 598 599 600 601 602
	struct nfs_open_context *ctx;
};

static int
readpage_async_filler(void *data, struct page *page)
{
	struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
603
	struct inode *inode = page_file_mapping(page)->host;
L
Linus Torvalds 已提交
604 605
	struct nfs_page *new;
	unsigned int len;
606 607
	int error;

608
	len = nfs_page_length(page);
L
Linus Torvalds 已提交
609 610
	if (len == 0)
		return nfs_return_empty_page(page);
611

L
Linus Torvalds 已提交
612
	new = nfs_create_request(desc->ctx, inode, page, 0, len);
613 614 615
	if (IS_ERR(new))
		goto out_error;

L
Linus Torvalds 已提交
616
	if (len < PAGE_CACHE_SIZE)
617
		zero_user_segment(page, len, PAGE_CACHE_SIZE);
618 619 620 621
	if (!nfs_pageio_add_request(desc->pgio, new)) {
		error = desc->pgio->pg_error;
		goto out_unlock;
	}
L
Linus Torvalds 已提交
622
	return 0;
623 624 625 626 627
out_error:
	error = PTR_ERR(new);
out_unlock:
	unlock_page(page);
	return error;
L
Linus Torvalds 已提交
628 629 630 631 632
}

int nfs_readpages(struct file *filp, struct address_space *mapping,
		struct list_head *pages, unsigned nr_pages)
{
633
	struct nfs_pageio_descriptor pgio;
L
Linus Torvalds 已提交
634
	struct nfs_readdesc desc = {
635
		.pgio = &pgio,
L
Linus Torvalds 已提交
636 637
	};
	struct inode *inode = mapping->host;
638
	unsigned long npages;
639
	int ret = -ESTALE;
L
Linus Torvalds 已提交
640

641
	dprintk("NFS: nfs_readpages (%s/%Lu %d)\n",
L
Linus Torvalds 已提交
642
			inode->i_sb->s_id,
643
			(unsigned long long)NFS_FILEID(inode),
L
Linus Torvalds 已提交
644
			nr_pages);
C
Chuck Lever 已提交
645
	nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
L
Linus Torvalds 已提交
646

647 648 649
	if (NFS_STALE(inode))
		goto out;

L
Linus Torvalds 已提交
650
	if (filp == NULL) {
651
		desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
L
Linus Torvalds 已提交
652 653 654
		if (desc.ctx == NULL)
			return -EBADF;
	} else
655
		desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
656 657 658 659 660 661 662 663 664

	/* attempt to read as many of the pages as possible from the cache
	 * - this returns -ENOBUFS immediately if the cookie is negative
	 */
	ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
					 pages, &nr_pages);
	if (ret == 0)
		goto read_complete; /* all pages were read */

665 666
	nfs_pageio_init_read(&pgio, inode, false,
			     &nfs_async_read_completion_ops);
667

L
Linus Torvalds 已提交
668
	ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
669 670

	nfs_pageio_complete(&pgio);
671
	NFS_I(inode)->read_io += pgio.pg_bytes_written;
672 673
	npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
	nfs_add_stats(inode, NFSIOS_READPAGES, npages);
674
read_complete:
L
Linus Torvalds 已提交
675
	put_nfs_open_context(desc.ctx);
676
out:
L
Linus Torvalds 已提交
677 678 679
	return ret;
}

D
David Howells 已提交
680
int __init nfs_init_readpagecache(void)
L
Linus Torvalds 已提交
681 682
{
	nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
683
					     sizeof(struct nfs_rw_header),
L
Linus Torvalds 已提交
684
					     0, SLAB_HWCACHE_ALIGN,
685
					     NULL);
L
Linus Torvalds 已提交
686 687 688 689 690 691
	if (nfs_rdata_cachep == NULL)
		return -ENOMEM;

	return 0;
}

692
void nfs_destroy_readpagecache(void)
L
Linus Torvalds 已提交
693
{
694
	kmem_cache_destroy(nfs_rdata_cachep);
L
Linus Torvalds 已提交
695
}