xref: /kernel/linux/linux-5.10/fs/nfs/read.c (revision 8c2ecf20) 
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
77void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
78{
79	struct nfs_pgio_mirror *mirror;
80
81	if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
82		pgio->pg_ops->pg_cleanup(pgio);
83
84	pgio->pg_ops = &nfs_pgio_rw_ops;
85
86	/* read path should never have more than one mirror */
87	WARN_ON_ONCE(pgio->pg_mirror_count != 1);
88
89	mirror = &pgio->pg_mirrors[0];
90	mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
91}
92EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
93
94static void nfs_readpage_release(struct nfs_page *req, int error)
95{
96	struct inode *inode = d_inode(nfs_req_openctx(req)->dentry);
97	struct page *page = req->wb_page;
98
99	dprintk("NFS: read done (%s/%llu %d@%lld)\n", inode->i_sb->s_id,
100		(unsigned long long)NFS_FILEID(inode), req->wb_bytes,
101		(long long)req_offset(req));
102
103	if (nfs_error_is_fatal_on_server(error) && error != -ETIMEDOUT)
104		SetPageError(page);
105	if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) {
106		struct address_space *mapping = page_file_mapping(page);
107
108		if (PageUptodate(page))
109			nfs_readpage_to_fscache(inode, page, 0);
110		else if (!PageError(page) && !PagePrivate(page))
111			generic_error_remove_page(mapping, page);
112		unlock_page(page);
113	}
114	nfs_release_request(req);
115}
116
117int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
118		       struct page *page)
119{
120	struct nfs_page	*new;
121	unsigned int len;
122	struct nfs_pageio_descriptor pgio;
123	struct nfs_pgio_mirror *pgm;
124
125	len = nfs_page_length(page);
126	if (len == 0)
127		return nfs_return_empty_page(page);
128	new = nfs_create_request(ctx, page, 0, len);
129	if (IS_ERR(new)) {
130		unlock_page(page);
131		return PTR_ERR(new);
132	}
133	if (len < PAGE_SIZE)
134		zero_user_segment(page, len, PAGE_SIZE);
135
136	nfs_pageio_init_read(&pgio, inode, false,
137			     &nfs_async_read_completion_ops);
138	if (!nfs_pageio_add_request(&pgio, new)) {
139		nfs_list_remove_request(new);
140		nfs_readpage_release(new, pgio.pg_error);
141	}
142	nfs_pageio_complete(&pgio);
143
144	/* It doesn't make sense to do mirrored reads! */
145	WARN_ON_ONCE(pgio.pg_mirror_count != 1);
146
147	pgm = &pgio.pg_mirrors[0];
148	NFS_I(inode)->read_io += pgm->pg_bytes_written;
149
150	return pgio.pg_error < 0 ? pgio.pg_error : 0;
151}
152
153static void nfs_page_group_set_uptodate(struct nfs_page *req)
154{
155	if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
156		SetPageUptodate(req->wb_page);
157}
158
159static void nfs_read_completion(struct nfs_pgio_header *hdr)
160{
161	unsigned long bytes = 0;
162	int error;
163
164	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
165		goto out;
166	while (!list_empty(&hdr->pages)) {
167		struct nfs_page *req = nfs_list_entry(hdr->pages.next);
168		struct page *page = req->wb_page;
169		unsigned long start = req->wb_pgbase;
170		unsigned long end = req->wb_pgbase + req->wb_bytes;
171
172		if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
173			/* note: regions of the page not covered by a
174			 * request are zeroed in nfs_readpage_async /
175			 * readpage_async_filler */
176			if (bytes > hdr->good_bytes) {
177				/* nothing in this request was good, so zero
178				 * the full extent of the request */
179				zero_user_segment(page, start, end);
180
181			} else if (hdr->good_bytes - bytes < req->wb_bytes) {
182				/* part of this request has good bytes, but
183				 * not all. zero the bad bytes */
184				start += hdr->good_bytes - bytes;
185				WARN_ON(start < req->wb_pgbase);
186				zero_user_segment(page, start, end);
187			}
188		}
189		error = 0;
190		bytes += req->wb_bytes;
191		if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
192			if (bytes <= hdr->good_bytes)
193				nfs_page_group_set_uptodate(req);
194			else {
195				error = hdr->error;
196				xchg(&nfs_req_openctx(req)->error, error);
197			}
198		} else
199			nfs_page_group_set_uptodate(req);
200		nfs_list_remove_request(req);
201		nfs_readpage_release(req, error);
202	}
203out:
204	hdr->release(hdr);
205}
206
207static void nfs_initiate_read(struct nfs_pgio_header *hdr,
208			      struct rpc_message *msg,
209			      const struct nfs_rpc_ops *rpc_ops,
210			      struct rpc_task_setup *task_setup_data, int how)
211{
212	struct inode *inode = hdr->inode;
213	int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
214
215	task_setup_data->flags |= swap_flags;
216	rpc_ops->read_setup(hdr, msg);
217	trace_nfs_initiate_read(hdr);
218}
219
220static void
221nfs_async_read_error(struct list_head *head, int error)
222{
223	struct nfs_page	*req;
224
225	while (!list_empty(head)) {
226		req = nfs_list_entry(head->next);
227		nfs_list_remove_request(req);
228		nfs_readpage_release(req, error);
229	}
230}
231
232static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
233	.error_cleanup = nfs_async_read_error,
234	.completion = nfs_read_completion,
235};
236
237/*
238 * This is the callback from RPC telling us whether a reply was
239 * received or some error occurred (timeout or socket shutdown).
240 */
241static int nfs_readpage_done(struct rpc_task *task,
242			     struct nfs_pgio_header *hdr,
243			     struct inode *inode)
244{
245	int status = NFS_PROTO(inode)->read_done(task, hdr);
246	if (status != 0)
247		return status;
248
249	nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count);
250	trace_nfs_readpage_done(task, hdr);
251
252	if (task->tk_status == -ESTALE) {
253		nfs_set_inode_stale(inode);
254		nfs_mark_for_revalidate(inode);
255	}
256	return 0;
257}
258
259static void nfs_readpage_retry(struct rpc_task *task,
260			       struct nfs_pgio_header *hdr)
261{
262	struct nfs_pgio_args *argp = &hdr->args;
263	struct nfs_pgio_res  *resp = &hdr->res;
264
265	/* This is a short read! */
266	nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD);
267	trace_nfs_readpage_short(task, hdr);
268
269	/* Has the server at least made some progress? */
270	if (resp->count == 0) {
271		nfs_set_pgio_error(hdr, -EIO, argp->offset);
272		return;
273	}
274
275	/* For non rpc-based layout drivers, retry-through-MDS */
276	if (!task->tk_ops) {
277		hdr->pnfs_error = -EAGAIN;
278		return;
279	}
280
281	/* Yes, so retry the read at the end of the hdr */
282	hdr->mds_offset += resp->count;
283	argp->offset += resp->count;
284	argp->pgbase += resp->count;
285	argp->count -= resp->count;
286	resp->count = 0;
287	resp->eof = 0;
288	rpc_restart_call_prepare(task);
289}
290
291static void nfs_readpage_result(struct rpc_task *task,
292				struct nfs_pgio_header *hdr)
293{
294	if (hdr->res.eof) {
295		loff_t pos = hdr->args.offset + hdr->res.count;
296		unsigned int new = pos - hdr->io_start;
297
298		if (hdr->good_bytes > new) {
299			hdr->good_bytes = new;
300			set_bit(NFS_IOHDR_EOF, &hdr->flags);
301			clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
302		}
303	} else if (hdr->res.count < hdr->args.count)
304		nfs_readpage_retry(task, hdr);
305}
306
307/*
308 * Read a page over NFS.
309 * We read the page synchronously in the following case:
310 *  -	The error flag is set for this page. This happens only when a
311 *	previous async read operation failed.
312 */
313int nfs_readpage(struct file *file, struct page *page)
314{
315	struct nfs_open_context *ctx;
316	struct inode *inode = page_file_mapping(page)->host;
317	int		error;
318
319	dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
320		page, PAGE_SIZE, page_index(page));
321	nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
322	nfs_add_stats(inode, NFSIOS_READPAGES, 1);
323
324	/*
325	 * Try to flush any pending writes to the file..
326	 *
327	 * NOTE! Because we own the page lock, there cannot
328	 * be any new pending writes generated at this point
329	 * for this page (other pages can be written to).
330	 */
331	error = nfs_wb_page(inode, page);
332	if (error)
333		goto out_unlock;
334	if (PageUptodate(page))
335		goto out_unlock;
336
337	error = -ESTALE;
338	if (NFS_STALE(inode))
339		goto out_unlock;
340
341	if (file == NULL) {
342		error = -EBADF;
343		ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
344		if (ctx == NULL)
345			goto out_unlock;
346	} else
347		ctx = get_nfs_open_context(nfs_file_open_context(file));
348
349	if (!IS_SYNC(inode)) {
350		error = nfs_readpage_from_fscache(ctx, inode, page);
351		if (error == 0)
352			goto out;
353	}
354
355	xchg(&ctx->error, 0);
356	error = nfs_readpage_async(ctx, inode, page);
357	if (!error) {
358		error = wait_on_page_locked_killable(page);
359		if (!PageUptodate(page) && !error)
360			error = xchg(&ctx->error, 0);
361	}
362out:
363	put_nfs_open_context(ctx);
364	return error;
365out_unlock:
366	unlock_page(page);
367	return error;
368}
369
370struct nfs_readdesc {
371	struct nfs_pageio_descriptor *pgio;
372	struct nfs_open_context *ctx;
373};
374
375static int
376readpage_async_filler(void *data, struct page *page)
377{
378	struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
379	struct nfs_page *new;
380	unsigned int len;
381	int error;
382
383	len = nfs_page_length(page);
384	if (len == 0)
385		return nfs_return_empty_page(page);
386
387	new = nfs_create_request(desc->ctx, page, 0, len);
388	if (IS_ERR(new))
389		goto out_error;
390
391	if (len < PAGE_SIZE)
392		zero_user_segment(page, len, PAGE_SIZE);
393	if (!nfs_pageio_add_request(desc->pgio, new)) {
394		nfs_list_remove_request(new);
395		error = desc->pgio->pg_error;
396		nfs_readpage_release(new, error);
397		goto out;
398	}
399	return 0;
400out_error:
401	error = PTR_ERR(new);
402	unlock_page(page);
403out:
404	return error;
405}
406
407int nfs_readpages(struct file *filp, struct address_space *mapping,
408		struct list_head *pages, unsigned nr_pages)
409{
410	struct nfs_pageio_descriptor pgio;
411	struct nfs_pgio_mirror *pgm;
412	struct nfs_readdesc desc = {
413		.pgio = &pgio,
414	};
415	struct inode *inode = mapping->host;
416	unsigned long npages;
417	int ret = -ESTALE;
418
419	dprintk("NFS: nfs_readpages (%s/%Lu %d)\n",
420			inode->i_sb->s_id,
421			(unsigned long long)NFS_FILEID(inode),
422			nr_pages);
423	nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
424
425	if (NFS_STALE(inode))
426		goto out;
427
428	if (filp == NULL) {
429		desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
430		if (desc.ctx == NULL)
431			return -EBADF;
432	} else
433		desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
434
435	/* attempt to read as many of the pages as possible from the cache
436	 * - this returns -ENOBUFS immediately if the cookie is negative
437	 */
438	ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
439					 pages, &nr_pages);
440	if (ret == 0)
441		goto read_complete; /* all pages were read */
442
443	nfs_pageio_init_read(&pgio, inode, false,
444			     &nfs_async_read_completion_ops);
445
446	ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
447	nfs_pageio_complete(&pgio);
448
449	/* It doesn't make sense to do mirrored reads! */
450	WARN_ON_ONCE(pgio.pg_mirror_count != 1);
451
452	pgm = &pgio.pg_mirrors[0];
453	NFS_I(inode)->read_io += pgm->pg_bytes_written;
454	npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >>
455		 PAGE_SHIFT;
456	nfs_add_stats(inode, NFSIOS_READPAGES, npages);
457read_complete:
458	put_nfs_open_context(desc.ctx);
459out:
460	return ret;
461}
462
463int __init nfs_init_readpagecache(void)
464{
465	nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
466					     sizeof(struct nfs_pgio_header),
467					     0, SLAB_HWCACHE_ALIGN,
468					     NULL);
469	if (nfs_rdata_cachep == NULL)
470		return -ENOMEM;
471
472	return 0;
473}
474
475void nfs_destroy_readpagecache(void)
476{
477	kmem_cache_destroy(nfs_rdata_cachep);
478}
479
480static const struct nfs_rw_ops nfs_rw_read_ops = {
481	.rw_alloc_header	= nfs_readhdr_alloc,
482	.rw_free_header		= nfs_readhdr_free,
483	.rw_done		= nfs_readpage_done,
484	.rw_result		= nfs_readpage_result,
485	.rw_initiate		= nfs_initiate_read,
486};
487