Loading...
1/*
2 * linux/fs/nfs/read.c
3 *
4 * Block I/O for NFS
5 *
6 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
7 * modified for async RPC by okir@monad.swb.de
8 */
9
10#include <linux/time.h>
11#include <linux/kernel.h>
12#include <linux/errno.h>
13#include <linux/fcntl.h>
14#include <linux/stat.h>
15#include <linux/mm.h>
16#include <linux/slab.h>
17#include <linux/pagemap.h>
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_page.h>
21#include <linux/module.h>
22
23#include <asm/system.h>
24#include "pnfs.h"
25
26#include "nfs4_fs.h"
27#include "internal.h"
28#include "iostat.h"
29#include "fscache.h"
30
31#define NFSDBG_FACILITY NFSDBG_PAGECACHE
32
33static const struct nfs_pageio_ops nfs_pageio_read_ops;
34static const struct rpc_call_ops nfs_read_partial_ops;
35static const struct rpc_call_ops nfs_read_full_ops;
36
37static struct kmem_cache *nfs_rdata_cachep;
38static mempool_t *nfs_rdata_mempool;
39
40#define MIN_POOL_READ (32)
41
42struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
43{
44 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL);
45
46 if (p) {
47 memset(p, 0, sizeof(*p));
48 INIT_LIST_HEAD(&p->pages);
49 p->npages = pagecount;
50 if (pagecount <= ARRAY_SIZE(p->page_array))
51 p->pagevec = p->page_array;
52 else {
53 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
54 if (!p->pagevec) {
55 mempool_free(p, nfs_rdata_mempool);
56 p = NULL;
57 }
58 }
59 }
60 return p;
61}
62
63void nfs_readdata_free(struct nfs_read_data *p)
64{
65 if (p && (p->pagevec != &p->page_array[0]))
66 kfree(p->pagevec);
67 mempool_free(p, nfs_rdata_mempool);
68}
69
70void nfs_readdata_release(struct nfs_read_data *rdata)
71{
72 put_lseg(rdata->lseg);
73 put_nfs_open_context(rdata->args.context);
74 nfs_readdata_free(rdata);
75}
76
77static
78int nfs_return_empty_page(struct page *page)
79{
80 zero_user(page, 0, PAGE_CACHE_SIZE);
81 SetPageUptodate(page);
82 unlock_page(page);
83 return 0;
84}
85
86static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
87{
88 unsigned int remainder = data->args.count - data->res.count;
89 unsigned int base = data->args.pgbase + data->res.count;
90 unsigned int pglen;
91 struct page **pages;
92
93 if (data->res.eof == 0 || remainder == 0)
94 return;
95 /*
96 * Note: "remainder" can never be negative, since we check for
97 * this in the XDR code.
98 */
99 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
100 base &= ~PAGE_CACHE_MASK;
101 pglen = PAGE_CACHE_SIZE - base;
102 for (;;) {
103 if (remainder <= pglen) {
104 zero_user(*pages, base, remainder);
105 break;
106 }
107 zero_user(*pages, base, pglen);
108 pages++;
109 remainder -= pglen;
110 pglen = PAGE_CACHE_SIZE;
111 base = 0;
112 }
113}
114
115static void nfs_pageio_init_read_mds(struct nfs_pageio_descriptor *pgio,
116 struct inode *inode)
117{
118 nfs_pageio_init(pgio, inode, &nfs_pageio_read_ops,
119 NFS_SERVER(inode)->rsize, 0);
120}
121
122void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
123{
124 pgio->pg_ops = &nfs_pageio_read_ops;
125 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
126}
127EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
128
129static void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
130 struct inode *inode)
131{
132 if (!pnfs_pageio_init_read(pgio, inode))
133 nfs_pageio_init_read_mds(pgio, inode);
134}
135
136int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
137 struct page *page)
138{
139 struct nfs_page *new;
140 unsigned int len;
141 struct nfs_pageio_descriptor pgio;
142
143 len = nfs_page_length(page);
144 if (len == 0)
145 return nfs_return_empty_page(page);
146 new = nfs_create_request(ctx, inode, page, 0, len);
147 if (IS_ERR(new)) {
148 unlock_page(page);
149 return PTR_ERR(new);
150 }
151 if (len < PAGE_CACHE_SIZE)
152 zero_user_segment(page, len, PAGE_CACHE_SIZE);
153
154 nfs_pageio_init_read(&pgio, inode);
155 nfs_pageio_add_request(&pgio, new);
156 nfs_pageio_complete(&pgio);
157 return 0;
158}
159
160static void nfs_readpage_release(struct nfs_page *req)
161{
162 struct inode *d_inode = req->wb_context->dentry->d_inode;
163
164 if (PageUptodate(req->wb_page))
165 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
166
167 unlock_page(req->wb_page);
168
169 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
170 req->wb_context->dentry->d_inode->i_sb->s_id,
171 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
172 req->wb_bytes,
173 (long long)req_offset(req));
174 nfs_release_request(req);
175}
176
177int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt,
178 const struct rpc_call_ops *call_ops)
179{
180 struct inode *inode = data->inode;
181 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
182 struct rpc_task *task;
183 struct rpc_message msg = {
184 .rpc_argp = &data->args,
185 .rpc_resp = &data->res,
186 .rpc_cred = data->cred,
187 };
188 struct rpc_task_setup task_setup_data = {
189 .task = &data->task,
190 .rpc_client = clnt,
191 .rpc_message = &msg,
192 .callback_ops = call_ops,
193 .callback_data = data,
194 .workqueue = nfsiod_workqueue,
195 .flags = RPC_TASK_ASYNC | swap_flags,
196 };
197
198 /* Set up the initial task struct. */
199 NFS_PROTO(inode)->read_setup(data, &msg);
200
201 dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
202 "offset %llu)\n",
203 data->task.tk_pid,
204 inode->i_sb->s_id,
205 (long long)NFS_FILEID(inode),
206 data->args.count,
207 (unsigned long long)data->args.offset);
208
209 task = rpc_run_task(&task_setup_data);
210 if (IS_ERR(task))
211 return PTR_ERR(task);
212 rpc_put_task(task);
213 return 0;
214}
215EXPORT_SYMBOL_GPL(nfs_initiate_read);
216
217/*
218 * Set up the NFS read request struct
219 */
220static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
221 unsigned int count, unsigned int offset)
222{
223 struct inode *inode = req->wb_context->dentry->d_inode;
224
225 data->req = req;
226 data->inode = inode;
227 data->cred = req->wb_context->cred;
228
229 data->args.fh = NFS_FH(inode);
230 data->args.offset = req_offset(req) + offset;
231 data->args.pgbase = req->wb_pgbase + offset;
232 data->args.pages = data->pagevec;
233 data->args.count = count;
234 data->args.context = get_nfs_open_context(req->wb_context);
235 data->args.lock_context = req->wb_lock_context;
236
237 data->res.fattr = &data->fattr;
238 data->res.count = count;
239 data->res.eof = 0;
240 nfs_fattr_init(&data->fattr);
241}
242
243static int nfs_do_read(struct nfs_read_data *data,
244 const struct rpc_call_ops *call_ops)
245{
246 struct inode *inode = data->args.context->dentry->d_inode;
247
248 return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops);
249}
250
251static int
252nfs_do_multiple_reads(struct list_head *head,
253 const struct rpc_call_ops *call_ops)
254{
255 struct nfs_read_data *data;
256 int ret = 0;
257
258 while (!list_empty(head)) {
259 int ret2;
260
261 data = list_entry(head->next, struct nfs_read_data, list);
262 list_del_init(&data->list);
263
264 ret2 = nfs_do_read(data, call_ops);
265 if (ret == 0)
266 ret = ret2;
267 }
268 return ret;
269}
270
271static void
272nfs_async_read_error(struct list_head *head)
273{
274 struct nfs_page *req;
275
276 while (!list_empty(head)) {
277 req = nfs_list_entry(head->next);
278 nfs_list_remove_request(req);
279 SetPageError(req->wb_page);
280 nfs_readpage_release(req);
281 }
282}
283
284/*
285 * Generate multiple requests to fill a single page.
286 *
287 * We optimize to reduce the number of read operations on the wire. If we
288 * detect that we're reading a page, or an area of a page, that is past the
289 * end of file, we do not generate NFS read operations but just clear the
290 * parts of the page that would have come back zero from the server anyway.
291 *
292 * We rely on the cached value of i_size to make this determination; another
293 * client can fill pages on the server past our cached end-of-file, but we
294 * won't see the new data until our attribute cache is updated. This is more
295 * or less conventional NFS client behavior.
296 */
297static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
298{
299 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
300 struct page *page = req->wb_page;
301 struct nfs_read_data *data;
302 size_t rsize = desc->pg_bsize, nbytes;
303 unsigned int offset;
304 int requests = 0;
305 int ret = 0;
306
307 nfs_list_remove_request(req);
308
309 offset = 0;
310 nbytes = desc->pg_count;
311 do {
312 size_t len = min(nbytes,rsize);
313
314 data = nfs_readdata_alloc(1);
315 if (!data)
316 goto out_bad;
317 data->pagevec[0] = page;
318 nfs_read_rpcsetup(req, data, len, offset);
319 list_add(&data->list, res);
320 requests++;
321 nbytes -= len;
322 offset += len;
323 } while(nbytes != 0);
324 atomic_set(&req->wb_complete, requests);
325 ClearPageError(page);
326 desc->pg_rpc_callops = &nfs_read_partial_ops;
327 return ret;
328out_bad:
329 while (!list_empty(res)) {
330 data = list_entry(res->next, struct nfs_read_data, list);
331 list_del(&data->list);
332 nfs_readdata_free(data);
333 }
334 SetPageError(page);
335 nfs_readpage_release(req);
336 return -ENOMEM;
337}
338
339static int nfs_pagein_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
340{
341 struct nfs_page *req;
342 struct page **pages;
343 struct nfs_read_data *data;
344 struct list_head *head = &desc->pg_list;
345 int ret = 0;
346
347 data = nfs_readdata_alloc(nfs_page_array_len(desc->pg_base,
348 desc->pg_count));
349 if (!data) {
350 nfs_async_read_error(head);
351 ret = -ENOMEM;
352 goto out;
353 }
354
355 pages = data->pagevec;
356 while (!list_empty(head)) {
357 req = nfs_list_entry(head->next);
358 nfs_list_remove_request(req);
359 nfs_list_add_request(req, &data->pages);
360 ClearPageError(req->wb_page);
361 *pages++ = req->wb_page;
362 }
363 req = nfs_list_entry(data->pages.next);
364
365 nfs_read_rpcsetup(req, data, desc->pg_count, 0);
366 list_add(&data->list, res);
367 desc->pg_rpc_callops = &nfs_read_full_ops;
368out:
369 return ret;
370}
371
372int nfs_generic_pagein(struct nfs_pageio_descriptor *desc, struct list_head *head)
373{
374 if (desc->pg_bsize < PAGE_CACHE_SIZE)
375 return nfs_pagein_multi(desc, head);
376 return nfs_pagein_one(desc, head);
377}
378
379static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
380{
381 LIST_HEAD(head);
382 int ret;
383
384 ret = nfs_generic_pagein(desc, &head);
385 if (ret == 0)
386 ret = nfs_do_multiple_reads(&head, desc->pg_rpc_callops);
387 return ret;
388}
389
390static const struct nfs_pageio_ops nfs_pageio_read_ops = {
391 .pg_test = nfs_generic_pg_test,
392 .pg_doio = nfs_generic_pg_readpages,
393};
394
395/*
396 * This is the callback from RPC telling us whether a reply was
397 * received or some error occurred (timeout or socket shutdown).
398 */
399int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
400{
401 int status;
402
403 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
404 task->tk_status);
405
406 status = NFS_PROTO(data->inode)->read_done(task, data);
407 if (status != 0)
408 return status;
409
410 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
411
412 if (task->tk_status == -ESTALE) {
413 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
414 nfs_mark_for_revalidate(data->inode);
415 }
416 return 0;
417}
418
419static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
420{
421 struct nfs_readargs *argp = &data->args;
422 struct nfs_readres *resp = &data->res;
423
424 if (resp->eof || resp->count == argp->count)
425 return;
426
427 /* This is a short read! */
428 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
429 /* Has the server at least made some progress? */
430 if (resp->count == 0)
431 return;
432
433 /* Yes, so retry the read at the end of the data */
434 data->mds_offset += resp->count;
435 argp->offset += resp->count;
436 argp->pgbase += resp->count;
437 argp->count -= resp->count;
438 nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client);
439}
440
441/*
442 * Handle a read reply that fills part of a page.
443 */
444static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
445{
446 struct nfs_read_data *data = calldata;
447
448 if (nfs_readpage_result(task, data) != 0)
449 return;
450 if (task->tk_status < 0)
451 return;
452
453 nfs_readpage_truncate_uninitialised_page(data);
454 nfs_readpage_retry(task, data);
455}
456
457static void nfs_readpage_release_partial(void *calldata)
458{
459 struct nfs_read_data *data = calldata;
460 struct nfs_page *req = data->req;
461 struct page *page = req->wb_page;
462 int status = data->task.tk_status;
463
464 if (status < 0)
465 SetPageError(page);
466
467 if (atomic_dec_and_test(&req->wb_complete)) {
468 if (!PageError(page))
469 SetPageUptodate(page);
470 nfs_readpage_release(req);
471 }
472 nfs_readdata_release(calldata);
473}
474
475#if defined(CONFIG_NFS_V4_1)
476void nfs_read_prepare(struct rpc_task *task, void *calldata)
477{
478 struct nfs_read_data *data = calldata;
479
480 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
481 &data->args.seq_args, &data->res.seq_res,
482 0, task))
483 return;
484 rpc_call_start(task);
485}
486#endif /* CONFIG_NFS_V4_1 */
487
488static const struct rpc_call_ops nfs_read_partial_ops = {
489#if defined(CONFIG_NFS_V4_1)
490 .rpc_call_prepare = nfs_read_prepare,
491#endif /* CONFIG_NFS_V4_1 */
492 .rpc_call_done = nfs_readpage_result_partial,
493 .rpc_release = nfs_readpage_release_partial,
494};
495
496static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
497{
498 unsigned int count = data->res.count;
499 unsigned int base = data->args.pgbase;
500 struct page **pages;
501
502 if (data->res.eof)
503 count = data->args.count;
504 if (unlikely(count == 0))
505 return;
506 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
507 base &= ~PAGE_CACHE_MASK;
508 count += base;
509 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
510 SetPageUptodate(*pages);
511 if (count == 0)
512 return;
513 /* Was this a short read? */
514 if (data->res.eof || data->res.count == data->args.count)
515 SetPageUptodate(*pages);
516}
517
518/*
519 * This is the callback from RPC telling us whether a reply was
520 * received or some error occurred (timeout or socket shutdown).
521 */
522static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
523{
524 struct nfs_read_data *data = calldata;
525
526 if (nfs_readpage_result(task, data) != 0)
527 return;
528 if (task->tk_status < 0)
529 return;
530 /*
531 * Note: nfs_readpage_retry may change the values of
532 * data->args. In the multi-page case, we therefore need
533 * to ensure that we call nfs_readpage_set_pages_uptodate()
534 * first.
535 */
536 nfs_readpage_truncate_uninitialised_page(data);
537 nfs_readpage_set_pages_uptodate(data);
538 nfs_readpage_retry(task, data);
539}
540
541static void nfs_readpage_release_full(void *calldata)
542{
543 struct nfs_read_data *data = calldata;
544
545 while (!list_empty(&data->pages)) {
546 struct nfs_page *req = nfs_list_entry(data->pages.next);
547
548 nfs_list_remove_request(req);
549 nfs_readpage_release(req);
550 }
551 nfs_readdata_release(calldata);
552}
553
554static const struct rpc_call_ops nfs_read_full_ops = {
555#if defined(CONFIG_NFS_V4_1)
556 .rpc_call_prepare = nfs_read_prepare,
557#endif /* CONFIG_NFS_V4_1 */
558 .rpc_call_done = nfs_readpage_result_full,
559 .rpc_release = nfs_readpage_release_full,
560};
561
562/*
563 * Read a page over NFS.
564 * We read the page synchronously in the following case:
565 * - The error flag is set for this page. This happens only when a
566 * previous async read operation failed.
567 */
568int nfs_readpage(struct file *file, struct page *page)
569{
570 struct nfs_open_context *ctx;
571 struct inode *inode = page->mapping->host;
572 int error;
573
574 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
575 page, PAGE_CACHE_SIZE, page->index);
576 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
577 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
578
579 /*
580 * Try to flush any pending writes to the file..
581 *
582 * NOTE! Because we own the page lock, there cannot
583 * be any new pending writes generated at this point
584 * for this page (other pages can be written to).
585 */
586 error = nfs_wb_page(inode, page);
587 if (error)
588 goto out_unlock;
589 if (PageUptodate(page))
590 goto out_unlock;
591
592 error = -ESTALE;
593 if (NFS_STALE(inode))
594 goto out_unlock;
595
596 if (file == NULL) {
597 error = -EBADF;
598 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
599 if (ctx == NULL)
600 goto out_unlock;
601 } else
602 ctx = get_nfs_open_context(nfs_file_open_context(file));
603
604 if (!IS_SYNC(inode)) {
605 error = nfs_readpage_from_fscache(ctx, inode, page);
606 if (error == 0)
607 goto out;
608 }
609
610 error = nfs_readpage_async(ctx, inode, page);
611
612out:
613 put_nfs_open_context(ctx);
614 return error;
615out_unlock:
616 unlock_page(page);
617 return error;
618}
619
620struct nfs_readdesc {
621 struct nfs_pageio_descriptor *pgio;
622 struct nfs_open_context *ctx;
623};
624
625static int
626readpage_async_filler(void *data, struct page *page)
627{
628 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
629 struct inode *inode = page->mapping->host;
630 struct nfs_page *new;
631 unsigned int len;
632 int error;
633
634 len = nfs_page_length(page);
635 if (len == 0)
636 return nfs_return_empty_page(page);
637
638 new = nfs_create_request(desc->ctx, inode, page, 0, len);
639 if (IS_ERR(new))
640 goto out_error;
641
642 if (len < PAGE_CACHE_SIZE)
643 zero_user_segment(page, len, PAGE_CACHE_SIZE);
644 if (!nfs_pageio_add_request(desc->pgio, new)) {
645 error = desc->pgio->pg_error;
646 goto out_unlock;
647 }
648 return 0;
649out_error:
650 error = PTR_ERR(new);
651 SetPageError(page);
652out_unlock:
653 unlock_page(page);
654 return error;
655}
656
657int nfs_readpages(struct file *filp, struct address_space *mapping,
658 struct list_head *pages, unsigned nr_pages)
659{
660 struct nfs_pageio_descriptor pgio;
661 struct nfs_readdesc desc = {
662 .pgio = &pgio,
663 };
664 struct inode *inode = mapping->host;
665 unsigned long npages;
666 int ret = -ESTALE;
667
668 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
669 inode->i_sb->s_id,
670 (long long)NFS_FILEID(inode),
671 nr_pages);
672 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
673
674 if (NFS_STALE(inode))
675 goto out;
676
677 if (filp == NULL) {
678 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
679 if (desc.ctx == NULL)
680 return -EBADF;
681 } else
682 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
683
684 /* attempt to read as many of the pages as possible from the cache
685 * - this returns -ENOBUFS immediately if the cookie is negative
686 */
687 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
688 pages, &nr_pages);
689 if (ret == 0)
690 goto read_complete; /* all pages were read */
691
692 nfs_pageio_init_read(&pgio, inode);
693
694 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
695
696 nfs_pageio_complete(&pgio);
697 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
698 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
699read_complete:
700 put_nfs_open_context(desc.ctx);
701out:
702 return ret;
703}
704
705int __init nfs_init_readpagecache(void)
706{
707 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
708 sizeof(struct nfs_read_data),
709 0, SLAB_HWCACHE_ALIGN,
710 NULL);
711 if (nfs_rdata_cachep == NULL)
712 return -ENOMEM;
713
714 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
715 nfs_rdata_cachep);
716 if (nfs_rdata_mempool == NULL)
717 return -ENOMEM;
718
719 return 0;
720}
721
722void nfs_destroy_readpagecache(void)
723{
724 mempool_destroy(nfs_rdata_mempool);
725 kmem_cache_destroy(nfs_rdata_cachep);
726}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/read.c
4 *
5 * Block I/O for NFS
6 *
7 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
8 * modified for async RPC by okir@monad.swb.de
9 */
10
11#include <linux/time.h>
12#include <linux/kernel.h>
13#include <linux/errno.h>
14#include <linux/fcntl.h>
15#include <linux/stat.h>
16#include <linux/mm.h>
17#include <linux/slab.h>
18#include <linux/pagemap.h>
19#include <linux/sunrpc/clnt.h>
20#include <linux/nfs_fs.h>
21#include <linux/nfs_page.h>
22#include <linux/module.h>
23
24#include "nfs4_fs.h"
25#include "internal.h"
26#include "iostat.h"
27#include "fscache.h"
28#include "pnfs.h"
29#include "nfstrace.h"
30
31#define NFSDBG_FACILITY NFSDBG_PAGECACHE
32
33static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
34static const struct nfs_rw_ops nfs_rw_read_ops;
35
36static struct kmem_cache *nfs_rdata_cachep;
37
38static struct nfs_pgio_header *nfs_readhdr_alloc(void)
39{
40 struct nfs_pgio_header *p = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
41
42 if (p)
43 p->rw_mode = FMODE_READ;
44 return p;
45}
46
47static void nfs_readhdr_free(struct nfs_pgio_header *rhdr)
48{
49 kmem_cache_free(nfs_rdata_cachep, rhdr);
50}
51
52static
53int nfs_return_empty_page(struct page *page)
54{
55 zero_user(page, 0, PAGE_SIZE);
56 SetPageUptodate(page);
57 unlock_page(page);
58 return 0;
59}
60
61void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
62 struct inode *inode, bool force_mds,
63 const struct nfs_pgio_completion_ops *compl_ops)
64{
65 struct nfs_server *server = NFS_SERVER(inode);
66 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
67
68#ifdef CONFIG_NFS_V4_1
69 if (server->pnfs_curr_ld && !force_mds)
70 pg_ops = server->pnfs_curr_ld->pg_read_ops;
71#endif
72 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops,
73 server->rsize, 0);
74}
75EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
76
77static void nfs_pageio_complete_read(struct nfs_pageio_descriptor *pgio)
78{
79 struct nfs_pgio_mirror *pgm;
80 unsigned long npages;
81
82 nfs_pageio_complete(pgio);
83
84 /* It doesn't make sense to do mirrored reads! */
85 WARN_ON_ONCE(pgio->pg_mirror_count != 1);
86
87 pgm = &pgio->pg_mirrors[0];
88 NFS_I(pgio->pg_inode)->read_io += pgm->pg_bytes_written;
89 npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >> PAGE_SHIFT;
90 nfs_add_stats(pgio->pg_inode, NFSIOS_READPAGES, npages);
91}
92
93
94void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
95{
96 struct nfs_pgio_mirror *mirror;
97
98 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
99 pgio->pg_ops->pg_cleanup(pgio);
100
101 pgio->pg_ops = &nfs_pgio_rw_ops;
102
103 /* read path should never have more than one mirror */
104 WARN_ON_ONCE(pgio->pg_mirror_count != 1);
105
106 mirror = &pgio->pg_mirrors[0];
107 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
108}
109EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
110
111static void nfs_readpage_release(struct nfs_page *req, int error)
112{
113 struct inode *inode = d_inode(nfs_req_openctx(req)->dentry);
114 struct page *page = req->wb_page;
115
116 dprintk("NFS: read done (%s/%llu %d@%lld)\n", inode->i_sb->s_id,
117 (unsigned long long)NFS_FILEID(inode), req->wb_bytes,
118 (long long)req_offset(req));
119
120 if (nfs_error_is_fatal_on_server(error) && error != -ETIMEDOUT)
121 SetPageError(page);
122 if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) {
123 struct address_space *mapping = page_file_mapping(page);
124
125 if (PageUptodate(page))
126 nfs_readpage_to_fscache(inode, page, 0);
127 else if (!PageError(page) && !PagePrivate(page))
128 generic_error_remove_page(mapping, page);
129 unlock_page(page);
130 }
131 nfs_release_request(req);
132}
133
134struct nfs_readdesc {
135 struct nfs_pageio_descriptor pgio;
136 struct nfs_open_context *ctx;
137};
138
139static void nfs_page_group_set_uptodate(struct nfs_page *req)
140{
141 if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
142 SetPageUptodate(req->wb_page);
143}
144
145static void nfs_read_completion(struct nfs_pgio_header *hdr)
146{
147 unsigned long bytes = 0;
148 int error;
149
150 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
151 goto out;
152 while (!list_empty(&hdr->pages)) {
153 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
154 struct page *page = req->wb_page;
155 unsigned long start = req->wb_pgbase;
156 unsigned long end = req->wb_pgbase + req->wb_bytes;
157
158 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
159 /* note: regions of the page not covered by a
160 * request are zeroed in readpage_async_filler */
161 if (bytes > hdr->good_bytes) {
162 /* nothing in this request was good, so zero
163 * the full extent of the request */
164 zero_user_segment(page, start, end);
165
166 } else if (hdr->good_bytes - bytes < req->wb_bytes) {
167 /* part of this request has good bytes, but
168 * not all. zero the bad bytes */
169 start += hdr->good_bytes - bytes;
170 WARN_ON(start < req->wb_pgbase);
171 zero_user_segment(page, start, end);
172 }
173 }
174 error = 0;
175 bytes += req->wb_bytes;
176 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
177 if (bytes <= hdr->good_bytes)
178 nfs_page_group_set_uptodate(req);
179 else {
180 error = hdr->error;
181 xchg(&nfs_req_openctx(req)->error, error);
182 }
183 } else
184 nfs_page_group_set_uptodate(req);
185 nfs_list_remove_request(req);
186 nfs_readpage_release(req, error);
187 }
188out:
189 hdr->release(hdr);
190}
191
192static void nfs_initiate_read(struct nfs_pgio_header *hdr,
193 struct rpc_message *msg,
194 const struct nfs_rpc_ops *rpc_ops,
195 struct rpc_task_setup *task_setup_data, int how)
196{
197 struct inode *inode = hdr->inode;
198 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
199
200 task_setup_data->flags |= swap_flags;
201 rpc_ops->read_setup(hdr, msg);
202 trace_nfs_initiate_read(hdr);
203}
204
205static void
206nfs_async_read_error(struct list_head *head, int error)
207{
208 struct nfs_page *req;
209
210 while (!list_empty(head)) {
211 req = nfs_list_entry(head->next);
212 nfs_list_remove_request(req);
213 nfs_readpage_release(req, error);
214 }
215}
216
217static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
218 .error_cleanup = nfs_async_read_error,
219 .completion = nfs_read_completion,
220};
221
222/*
223 * This is the callback from RPC telling us whether a reply was
224 * received or some error occurred (timeout or socket shutdown).
225 */
226static int nfs_readpage_done(struct rpc_task *task,
227 struct nfs_pgio_header *hdr,
228 struct inode *inode)
229{
230 int status = NFS_PROTO(inode)->read_done(task, hdr);
231 if (status != 0)
232 return status;
233
234 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count);
235 trace_nfs_readpage_done(task, hdr);
236
237 if (task->tk_status == -ESTALE) {
238 nfs_set_inode_stale(inode);
239 nfs_mark_for_revalidate(inode);
240 }
241 return 0;
242}
243
244static void nfs_readpage_retry(struct rpc_task *task,
245 struct nfs_pgio_header *hdr)
246{
247 struct nfs_pgio_args *argp = &hdr->args;
248 struct nfs_pgio_res *resp = &hdr->res;
249
250 /* This is a short read! */
251 nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD);
252 trace_nfs_readpage_short(task, hdr);
253
254 /* Has the server at least made some progress? */
255 if (resp->count == 0) {
256 nfs_set_pgio_error(hdr, -EIO, argp->offset);
257 return;
258 }
259
260 /* For non rpc-based layout drivers, retry-through-MDS */
261 if (!task->tk_ops) {
262 hdr->pnfs_error = -EAGAIN;
263 return;
264 }
265
266 /* Yes, so retry the read at the end of the hdr */
267 hdr->mds_offset += resp->count;
268 argp->offset += resp->count;
269 argp->pgbase += resp->count;
270 argp->count -= resp->count;
271 resp->count = 0;
272 resp->eof = 0;
273 rpc_restart_call_prepare(task);
274}
275
276static void nfs_readpage_result(struct rpc_task *task,
277 struct nfs_pgio_header *hdr)
278{
279 if (hdr->res.eof) {
280 loff_t pos = hdr->args.offset + hdr->res.count;
281 unsigned int new = pos - hdr->io_start;
282
283 if (hdr->good_bytes > new) {
284 hdr->good_bytes = new;
285 set_bit(NFS_IOHDR_EOF, &hdr->flags);
286 clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
287 }
288 } else if (hdr->res.count < hdr->args.count)
289 nfs_readpage_retry(task, hdr);
290}
291
292static int
293readpage_async_filler(void *data, struct page *page)
294{
295 struct nfs_readdesc *desc = data;
296 struct nfs_page *new;
297 unsigned int len;
298 int error;
299
300 len = nfs_page_length(page);
301 if (len == 0)
302 return nfs_return_empty_page(page);
303
304 new = nfs_create_request(desc->ctx, page, 0, len);
305 if (IS_ERR(new))
306 goto out_error;
307
308 if (len < PAGE_SIZE)
309 zero_user_segment(page, len, PAGE_SIZE);
310 if (!nfs_pageio_add_request(&desc->pgio, new)) {
311 nfs_list_remove_request(new);
312 error = desc->pgio.pg_error;
313 nfs_readpage_release(new, error);
314 goto out;
315 }
316 return 0;
317out_error:
318 error = PTR_ERR(new);
319 unlock_page(page);
320out:
321 return error;
322}
323
324/*
325 * Read a page over NFS.
326 * We read the page synchronously in the following case:
327 * - The error flag is set for this page. This happens only when a
328 * previous async read operation failed.
329 */
330int nfs_readpage(struct file *file, struct page *page)
331{
332 struct nfs_readdesc desc;
333 struct inode *inode = page_file_mapping(page)->host;
334 int ret;
335
336 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
337 page, PAGE_SIZE, page_index(page));
338 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
339
340 /*
341 * Try to flush any pending writes to the file..
342 *
343 * NOTE! Because we own the page lock, there cannot
344 * be any new pending writes generated at this point
345 * for this page (other pages can be written to).
346 */
347 ret = nfs_wb_page(inode, page);
348 if (ret)
349 goto out_unlock;
350 if (PageUptodate(page))
351 goto out_unlock;
352
353 ret = -ESTALE;
354 if (NFS_STALE(inode))
355 goto out_unlock;
356
357 if (file == NULL) {
358 ret = -EBADF;
359 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
360 if (desc.ctx == NULL)
361 goto out_unlock;
362 } else
363 desc.ctx = get_nfs_open_context(nfs_file_open_context(file));
364
365 xchg(&desc.ctx->error, 0);
366 if (!IS_SYNC(inode)) {
367 ret = nfs_readpage_from_fscache(desc.ctx, inode, page);
368 if (ret == 0)
369 goto out_wait;
370 }
371
372 nfs_pageio_init_read(&desc.pgio, inode, false,
373 &nfs_async_read_completion_ops);
374
375 ret = readpage_async_filler(&desc, page);
376 if (ret)
377 goto out;
378
379 nfs_pageio_complete_read(&desc.pgio);
380 ret = desc.pgio.pg_error < 0 ? desc.pgio.pg_error : 0;
381out_wait:
382 if (!ret) {
383 ret = wait_on_page_locked_killable(page);
384 if (!PageUptodate(page) && !ret)
385 ret = xchg(&desc.ctx->error, 0);
386 }
387out:
388 put_nfs_open_context(desc.ctx);
389 return ret;
390out_unlock:
391 unlock_page(page);
392 return ret;
393}
394
395int nfs_readpages(struct file *file, struct address_space *mapping,
396 struct list_head *pages, unsigned nr_pages)
397{
398 struct nfs_readdesc desc;
399 struct inode *inode = mapping->host;
400 int ret;
401
402 dprintk("NFS: nfs_readpages (%s/%Lu %d)\n",
403 inode->i_sb->s_id,
404 (unsigned long long)NFS_FILEID(inode),
405 nr_pages);
406 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
407
408 ret = -ESTALE;
409 if (NFS_STALE(inode))
410 goto out;
411
412 if (file == NULL) {
413 ret = -EBADF;
414 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
415 if (desc.ctx == NULL)
416 goto out;
417 } else
418 desc.ctx = get_nfs_open_context(nfs_file_open_context(file));
419
420 /* attempt to read as many of the pages as possible from the cache
421 * - this returns -ENOBUFS immediately if the cookie is negative
422 */
423 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
424 pages, &nr_pages);
425 if (ret == 0)
426 goto read_complete; /* all pages were read */
427
428 nfs_pageio_init_read(&desc.pgio, inode, false,
429 &nfs_async_read_completion_ops);
430
431 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
432
433 nfs_pageio_complete_read(&desc.pgio);
434
435read_complete:
436 put_nfs_open_context(desc.ctx);
437out:
438 return ret;
439}
440
441int __init nfs_init_readpagecache(void)
442{
443 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
444 sizeof(struct nfs_pgio_header),
445 0, SLAB_HWCACHE_ALIGN,
446 NULL);
447 if (nfs_rdata_cachep == NULL)
448 return -ENOMEM;
449
450 return 0;
451}
452
453void nfs_destroy_readpagecache(void)
454{
455 kmem_cache_destroy(nfs_rdata_cachep);
456}
457
458static const struct nfs_rw_ops nfs_rw_read_ops = {
459 .rw_alloc_header = nfs_readhdr_alloc,
460 .rw_free_header = nfs_readhdr_free,
461 .rw_done = nfs_readpage_done,
462 .rw_result = nfs_readpage_result,
463 .rw_initiate = nfs_initiate_read,
464};