Loading...
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 if (PageUptodate(page))
124 nfs_fscache_write_page(inode, page);
125 unlock_page(page);
126 }
127 nfs_release_request(req);
128}
129
130struct nfs_readdesc {
131 struct nfs_pageio_descriptor pgio;
132 struct nfs_open_context *ctx;
133};
134
135static void nfs_page_group_set_uptodate(struct nfs_page *req)
136{
137 if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
138 SetPageUptodate(req->wb_page);
139}
140
141static void nfs_read_completion(struct nfs_pgio_header *hdr)
142{
143 unsigned long bytes = 0;
144 int error;
145
146 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
147 goto out;
148 while (!list_empty(&hdr->pages)) {
149 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
150 struct page *page = req->wb_page;
151 unsigned long start = req->wb_pgbase;
152 unsigned long end = req->wb_pgbase + req->wb_bytes;
153
154 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
155 /* note: regions of the page not covered by a
156 * request are zeroed in readpage_async_filler */
157 if (bytes > hdr->good_bytes) {
158 /* nothing in this request was good, so zero
159 * the full extent of the request */
160 zero_user_segment(page, start, end);
161
162 } else if (hdr->good_bytes - bytes < req->wb_bytes) {
163 /* part of this request has good bytes, but
164 * not all. zero the bad bytes */
165 start += hdr->good_bytes - bytes;
166 WARN_ON(start < req->wb_pgbase);
167 zero_user_segment(page, start, end);
168 }
169 }
170 error = 0;
171 bytes += req->wb_bytes;
172 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
173 if (bytes <= hdr->good_bytes)
174 nfs_page_group_set_uptodate(req);
175 else {
176 error = hdr->error;
177 xchg(&nfs_req_openctx(req)->error, error);
178 }
179 } else
180 nfs_page_group_set_uptodate(req);
181 nfs_list_remove_request(req);
182 nfs_readpage_release(req, error);
183 }
184out:
185 hdr->release(hdr);
186}
187
188static void nfs_initiate_read(struct nfs_pgio_header *hdr,
189 struct rpc_message *msg,
190 const struct nfs_rpc_ops *rpc_ops,
191 struct rpc_task_setup *task_setup_data, int how)
192{
193 rpc_ops->read_setup(hdr, msg);
194 trace_nfs_initiate_read(hdr);
195}
196
197static void
198nfs_async_read_error(struct list_head *head, int error)
199{
200 struct nfs_page *req;
201
202 while (!list_empty(head)) {
203 req = nfs_list_entry(head->next);
204 nfs_list_remove_request(req);
205 nfs_readpage_release(req, error);
206 }
207}
208
209static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
210 .error_cleanup = nfs_async_read_error,
211 .completion = nfs_read_completion,
212};
213
214/*
215 * This is the callback from RPC telling us whether a reply was
216 * received or some error occurred (timeout or socket shutdown).
217 */
218static int nfs_readpage_done(struct rpc_task *task,
219 struct nfs_pgio_header *hdr,
220 struct inode *inode)
221{
222 int status = NFS_PROTO(inode)->read_done(task, hdr);
223 if (status != 0)
224 return status;
225
226 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count);
227 trace_nfs_readpage_done(task, hdr);
228
229 if (task->tk_status == -ESTALE) {
230 nfs_set_inode_stale(inode);
231 nfs_mark_for_revalidate(inode);
232 }
233 return 0;
234}
235
236static void nfs_readpage_retry(struct rpc_task *task,
237 struct nfs_pgio_header *hdr)
238{
239 struct nfs_pgio_args *argp = &hdr->args;
240 struct nfs_pgio_res *resp = &hdr->res;
241
242 /* This is a short read! */
243 nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD);
244 trace_nfs_readpage_short(task, hdr);
245
246 /* Has the server at least made some progress? */
247 if (resp->count == 0) {
248 nfs_set_pgio_error(hdr, -EIO, argp->offset);
249 return;
250 }
251
252 /* For non rpc-based layout drivers, retry-through-MDS */
253 if (!task->tk_ops) {
254 hdr->pnfs_error = -EAGAIN;
255 return;
256 }
257
258 /* Yes, so retry the read at the end of the hdr */
259 hdr->mds_offset += resp->count;
260 argp->offset += resp->count;
261 argp->pgbase += resp->count;
262 argp->count -= resp->count;
263 resp->count = 0;
264 resp->eof = 0;
265 rpc_restart_call_prepare(task);
266}
267
268static void nfs_readpage_result(struct rpc_task *task,
269 struct nfs_pgio_header *hdr)
270{
271 if (hdr->res.eof) {
272 loff_t pos = hdr->args.offset + hdr->res.count;
273 unsigned int new = pos - hdr->io_start;
274
275 if (hdr->good_bytes > new) {
276 hdr->good_bytes = new;
277 set_bit(NFS_IOHDR_EOF, &hdr->flags);
278 clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
279 }
280 } else if (hdr->res.count < hdr->args.count)
281 nfs_readpage_retry(task, hdr);
282}
283
284static int
285readpage_async_filler(struct nfs_readdesc *desc, struct page *page)
286{
287 struct inode *inode = page_file_mapping(page)->host;
288 unsigned int rsize = NFS_SERVER(inode)->rsize;
289 struct nfs_page *new;
290 unsigned int len, aligned_len;
291 int error;
292
293 len = nfs_page_length(page);
294 if (len == 0)
295 return nfs_return_empty_page(page);
296
297 aligned_len = min_t(unsigned int, ALIGN(len, rsize), PAGE_SIZE);
298
299 if (!IS_SYNC(page->mapping->host)) {
300 error = nfs_fscache_read_page(page->mapping->host, page);
301 if (error == 0)
302 goto out_unlock;
303 }
304
305 new = nfs_create_request(desc->ctx, page, 0, aligned_len);
306 if (IS_ERR(new))
307 goto out_error;
308
309 if (len < PAGE_SIZE)
310 zero_user_segment(page, len, PAGE_SIZE);
311 if (!nfs_pageio_add_request(&desc->pgio, new)) {
312 nfs_list_remove_request(new);
313 error = desc->pgio.pg_error;
314 nfs_readpage_release(new, error);
315 goto out;
316 }
317 return 0;
318out_error:
319 error = PTR_ERR(new);
320out_unlock:
321 unlock_page(page);
322out:
323 return error;
324}
325
326/*
327 * Read a page over NFS.
328 * We read the page synchronously in the following case:
329 * - The error flag is set for this page. This happens only when a
330 * previous async read operation failed.
331 */
332int nfs_read_folio(struct file *file, struct folio *folio)
333{
334 struct page *page = &folio->page;
335 struct nfs_readdesc desc;
336 struct inode *inode = page_file_mapping(page)->host;
337 int ret;
338
339 trace_nfs_aop_readpage(inode, page);
340 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
341
342 /*
343 * Try to flush any pending writes to the file..
344 *
345 * NOTE! Because we own the page lock, there cannot
346 * be any new pending writes generated at this point
347 * for this page (other pages can be written to).
348 */
349 ret = nfs_wb_page(inode, page);
350 if (ret)
351 goto out_unlock;
352 if (PageUptodate(page))
353 goto out_unlock;
354
355 ret = -ESTALE;
356 if (NFS_STALE(inode))
357 goto out_unlock;
358
359 if (file == NULL) {
360 ret = -EBADF;
361 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
362 if (desc.ctx == NULL)
363 goto out_unlock;
364 } else
365 desc.ctx = get_nfs_open_context(nfs_file_open_context(file));
366
367 xchg(&desc.ctx->error, 0);
368 nfs_pageio_init_read(&desc.pgio, inode, false,
369 &nfs_async_read_completion_ops);
370
371 ret = readpage_async_filler(&desc, page);
372 if (ret)
373 goto out;
374
375 nfs_pageio_complete_read(&desc.pgio);
376 ret = desc.pgio.pg_error < 0 ? desc.pgio.pg_error : 0;
377 if (!ret) {
378 ret = wait_on_page_locked_killable(page);
379 if (!PageUptodate(page) && !ret)
380 ret = xchg(&desc.ctx->error, 0);
381 }
382out:
383 put_nfs_open_context(desc.ctx);
384 trace_nfs_aop_readpage_done(inode, page, ret);
385 return ret;
386out_unlock:
387 unlock_page(page);
388 trace_nfs_aop_readpage_done(inode, page, ret);
389 return ret;
390}
391
392void nfs_readahead(struct readahead_control *ractl)
393{
394 unsigned int nr_pages = readahead_count(ractl);
395 struct file *file = ractl->file;
396 struct nfs_readdesc desc;
397 struct inode *inode = ractl->mapping->host;
398 struct page *page;
399 int ret;
400
401 trace_nfs_aop_readahead(inode, readahead_pos(ractl), nr_pages);
402 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
403
404 ret = -ESTALE;
405 if (NFS_STALE(inode))
406 goto out;
407
408 if (file == NULL) {
409 ret = -EBADF;
410 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
411 if (desc.ctx == NULL)
412 goto out;
413 } else
414 desc.ctx = get_nfs_open_context(nfs_file_open_context(file));
415
416 nfs_pageio_init_read(&desc.pgio, inode, false,
417 &nfs_async_read_completion_ops);
418
419 while ((page = readahead_page(ractl)) != NULL) {
420 ret = readpage_async_filler(&desc, page);
421 put_page(page);
422 if (ret)
423 break;
424 }
425
426 nfs_pageio_complete_read(&desc.pgio);
427
428 put_nfs_open_context(desc.ctx);
429out:
430 trace_nfs_aop_readahead_done(inode, nr_pages, ret);
431}
432
433int __init nfs_init_readpagecache(void)
434{
435 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
436 sizeof(struct nfs_pgio_header),
437 0, SLAB_HWCACHE_ALIGN,
438 NULL);
439 if (nfs_rdata_cachep == NULL)
440 return -ENOMEM;
441
442 return 0;
443}
444
445void nfs_destroy_readpagecache(void)
446{
447 kmem_cache_destroy(nfs_rdata_cachep);
448}
449
450static const struct nfs_rw_ops nfs_rw_read_ops = {
451 .rw_alloc_header = nfs_readhdr_alloc,
452 .rw_free_header = nfs_readhdr_free,
453 .rw_done = nfs_readpage_done,
454 .rw_result = nfs_readpage_result,
455 .rw_initiate = nfs_initiate_read,
456};
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/task_io_accounting_ops.h>
19#include <linux/pagemap.h>
20#include <linux/sunrpc/clnt.h>
21#include <linux/nfs_fs.h>
22#include <linux/nfs_page.h>
23#include <linux/module.h>
24
25#include "nfs4_fs.h"
26#include "internal.h"
27#include "iostat.h"
28#include "fscache.h"
29#include "pnfs.h"
30#include "nfstrace.h"
31#include "delegation.h"
32
33#define NFSDBG_FACILITY NFSDBG_PAGECACHE
34
35const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
36static const struct nfs_rw_ops nfs_rw_read_ops;
37
38static struct kmem_cache *nfs_rdata_cachep;
39
40static struct nfs_pgio_header *nfs_readhdr_alloc(void)
41{
42 struct nfs_pgio_header *p = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
43
44 if (p)
45 p->rw_mode = FMODE_READ;
46 return p;
47}
48
49static void nfs_readhdr_free(struct nfs_pgio_header *rhdr)
50{
51 kfree(rhdr->res.scratch);
52 kmem_cache_free(nfs_rdata_cachep, rhdr);
53}
54
55static int nfs_return_empty_folio(struct folio *folio)
56{
57 folio_zero_segment(folio, 0, folio_size(folio));
58 folio_mark_uptodate(folio);
59 folio_unlock(folio);
60 return 0;
61}
62
63void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
64 struct inode *inode, bool force_mds,
65 const struct nfs_pgio_completion_ops *compl_ops)
66{
67 struct nfs_server *server = NFS_SERVER(inode);
68 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
69
70#ifdef CONFIG_NFS_V4_1
71 if (server->pnfs_curr_ld && !force_mds)
72 pg_ops = server->pnfs_curr_ld->pg_read_ops;
73#endif
74 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops,
75 server->rsize, 0);
76}
77EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
78
79void nfs_pageio_complete_read(struct nfs_pageio_descriptor *pgio)
80{
81 struct nfs_pgio_mirror *pgm;
82 unsigned long npages;
83
84 nfs_pageio_complete(pgio);
85
86 /* It doesn't make sense to do mirrored reads! */
87 WARN_ON_ONCE(pgio->pg_mirror_count != 1);
88
89 pgm = &pgio->pg_mirrors[0];
90 NFS_I(pgio->pg_inode)->read_io += pgm->pg_bytes_written;
91 npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >> PAGE_SHIFT;
92 nfs_add_stats(pgio->pg_inode, NFSIOS_READPAGES, npages);
93}
94
95
96void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
97{
98 struct nfs_pgio_mirror *mirror;
99
100 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
101 pgio->pg_ops->pg_cleanup(pgio);
102
103 pgio->pg_ops = &nfs_pgio_rw_ops;
104
105 /* read path should never have more than one mirror */
106 WARN_ON_ONCE(pgio->pg_mirror_count != 1);
107
108 mirror = &pgio->pg_mirrors[0];
109 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
110}
111EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
112
113bool nfs_read_alloc_scratch(struct nfs_pgio_header *hdr, size_t size)
114{
115 WARN_ON(hdr->res.scratch != NULL);
116 hdr->res.scratch = kmalloc(size, GFP_KERNEL);
117 return hdr->res.scratch != NULL;
118}
119EXPORT_SYMBOL_GPL(nfs_read_alloc_scratch);
120
121static void nfs_readpage_release(struct nfs_page *req, int error)
122{
123 struct folio *folio = nfs_page_to_folio(req);
124
125 if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE))
126 if (nfs_netfs_folio_unlock(folio))
127 folio_unlock(folio);
128
129 nfs_release_request(req);
130}
131
132static void nfs_page_group_set_uptodate(struct nfs_page *req)
133{
134 if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
135 folio_mark_uptodate(nfs_page_to_folio(req));
136}
137
138static void nfs_read_completion(struct nfs_pgio_header *hdr)
139{
140 unsigned long bytes = 0;
141 int error;
142
143 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
144 goto out;
145 while (!list_empty(&hdr->pages)) {
146 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
147 struct folio *folio = nfs_page_to_folio(req);
148 unsigned long start = req->wb_pgbase;
149 unsigned long end = req->wb_pgbase + req->wb_bytes;
150
151 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
152 /* note: regions of the page not covered by a
153 * request are zeroed in nfs_read_add_folio
154 */
155 if (bytes > hdr->good_bytes) {
156 /* nothing in this request was good, so zero
157 * the full extent of the request */
158 folio_zero_segment(folio, start, end);
159
160 } else if (hdr->good_bytes - bytes < req->wb_bytes) {
161 /* part of this request has good bytes, but
162 * not all. zero the bad bytes */
163 start += hdr->good_bytes - bytes;
164 WARN_ON(start < req->wb_pgbase);
165 folio_zero_segment(folio, start, end);
166 }
167 }
168 error = 0;
169 bytes += req->wb_bytes;
170 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
171 if (bytes <= hdr->good_bytes)
172 nfs_page_group_set_uptodate(req);
173 else {
174 error = hdr->error;
175 xchg(&nfs_req_openctx(req)->error, error);
176 }
177 } else
178 nfs_page_group_set_uptodate(req);
179 nfs_list_remove_request(req);
180 nfs_readpage_release(req, error);
181 }
182 nfs_netfs_read_completion(hdr);
183
184out:
185 hdr->release(hdr);
186}
187
188static void nfs_initiate_read(struct nfs_pgio_header *hdr,
189 struct rpc_message *msg,
190 const struct nfs_rpc_ops *rpc_ops,
191 struct rpc_task_setup *task_setup_data, int how)
192{
193 rpc_ops->read_setup(hdr, msg);
194 nfs_netfs_initiate_read(hdr);
195 trace_nfs_initiate_read(hdr);
196}
197
198static void
199nfs_async_read_error(struct list_head *head, int error)
200{
201 struct nfs_page *req;
202
203 while (!list_empty(head)) {
204 req = nfs_list_entry(head->next);
205 nfs_list_remove_request(req);
206 nfs_readpage_release(req, error);
207 }
208}
209
210const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
211 .error_cleanup = nfs_async_read_error,
212 .completion = nfs_read_completion,
213};
214
215/*
216 * This is the callback from RPC telling us whether a reply was
217 * received or some error occurred (timeout or socket shutdown).
218 */
219static int nfs_readpage_done(struct rpc_task *task,
220 struct nfs_pgio_header *hdr,
221 struct inode *inode)
222{
223 int status = NFS_PROTO(inode)->read_done(task, hdr);
224 if (status != 0)
225 return status;
226
227 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count);
228 trace_nfs_readpage_done(task, hdr);
229
230 if (task->tk_status == -ESTALE) {
231 nfs_set_inode_stale(inode);
232 nfs_mark_for_revalidate(inode);
233 }
234 return 0;
235}
236
237static void nfs_readpage_retry(struct rpc_task *task,
238 struct nfs_pgio_header *hdr)
239{
240 struct nfs_pgio_args *argp = &hdr->args;
241 struct nfs_pgio_res *resp = &hdr->res;
242
243 /* This is a short read! */
244 nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD);
245 trace_nfs_readpage_short(task, hdr);
246
247 /* Has the server at least made some progress? */
248 if (resp->count == 0) {
249 nfs_set_pgio_error(hdr, -EIO, argp->offset);
250 return;
251 }
252
253 /* For non rpc-based layout drivers, retry-through-MDS */
254 if (!task->tk_ops) {
255 hdr->pnfs_error = -EAGAIN;
256 return;
257 }
258
259 /* Yes, so retry the read at the end of the hdr */
260 hdr->mds_offset += resp->count;
261 argp->offset += resp->count;
262 argp->pgbase += resp->count;
263 argp->count -= resp->count;
264 resp->count = 0;
265 resp->eof = 0;
266 rpc_restart_call_prepare(task);
267}
268
269static void nfs_readpage_result(struct rpc_task *task,
270 struct nfs_pgio_header *hdr)
271{
272 if (hdr->res.eof) {
273 loff_t pos = hdr->args.offset + hdr->res.count;
274 unsigned int new = pos - hdr->io_start;
275
276 if (hdr->good_bytes > new) {
277 hdr->good_bytes = new;
278 set_bit(NFS_IOHDR_EOF, &hdr->flags);
279 clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
280 }
281 } else if (hdr->res.count < hdr->args.count)
282 nfs_readpage_retry(task, hdr);
283}
284
285int nfs_read_add_folio(struct nfs_pageio_descriptor *pgio,
286 struct nfs_open_context *ctx,
287 struct folio *folio)
288{
289 struct inode *inode = folio->mapping->host;
290 struct nfs_server *server = NFS_SERVER(inode);
291 size_t fsize = folio_size(folio);
292 unsigned int rsize = server->rsize;
293 struct nfs_page *new;
294 unsigned int len, aligned_len;
295 int error;
296
297 len = nfs_folio_length(folio);
298 if (len == 0)
299 return nfs_return_empty_folio(folio);
300
301 aligned_len = min_t(unsigned int, ALIGN(len, rsize), fsize);
302
303 new = nfs_page_create_from_folio(ctx, folio, 0, aligned_len);
304 if (IS_ERR(new)) {
305 error = PTR_ERR(new);
306 if (nfs_netfs_folio_unlock(folio))
307 folio_unlock(folio);
308 goto out;
309 }
310
311 if (len < fsize)
312 folio_zero_segment(folio, len, fsize);
313 if (!nfs_pageio_add_request(pgio, new)) {
314 nfs_list_remove_request(new);
315 error = pgio->pg_error;
316 nfs_readpage_release(new, error);
317 goto out;
318 }
319 return 0;
320out:
321 return error;
322}
323
324/*
325 * Actually read a folio over the wire.
326 */
327static int nfs_do_read_folio(struct file *file, struct folio *folio)
328{
329 struct inode *inode = file_inode(file);
330 struct nfs_pageio_descriptor pgio;
331 struct nfs_open_context *ctx;
332 int ret;
333
334 ctx = get_nfs_open_context(nfs_file_open_context(file));
335
336 xchg(&ctx->error, 0);
337 nfs_pageio_init_read(&pgio, inode, false,
338 &nfs_async_read_completion_ops);
339
340 ret = nfs_read_add_folio(&pgio, ctx, folio);
341 if (ret)
342 goto out_put;
343
344 nfs_pageio_complete_read(&pgio);
345 nfs_update_delegated_atime(inode);
346 if (pgio.pg_error < 0) {
347 ret = pgio.pg_error;
348 goto out_put;
349 }
350
351 ret = folio_wait_locked_killable(folio);
352 if (!folio_test_uptodate(folio) && !ret)
353 ret = xchg(&ctx->error, 0);
354
355out_put:
356 put_nfs_open_context(ctx);
357 return ret;
358}
359
360/*
361 * Synchronously read a folio.
362 *
363 * This is not heavily used as most users to try an asynchronous
364 * large read through ->readahead first.
365 */
366int nfs_read_folio(struct file *file, struct folio *folio)
367{
368 struct inode *inode = file_inode(file);
369 loff_t pos = folio_pos(folio);
370 size_t len = folio_size(folio);
371 int ret;
372
373 trace_nfs_aop_readpage(inode, pos, len);
374 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
375 task_io_account_read(len);
376
377 /*
378 * Try to flush any pending writes to the file..
379 *
380 * NOTE! Because we own the folio lock, there cannot
381 * be any new pending writes generated at this point
382 * for this folio (other folios can be written to).
383 */
384 ret = nfs_wb_folio(inode, folio);
385 if (ret)
386 goto out_unlock;
387 if (folio_test_uptodate(folio))
388 goto out_unlock;
389
390 ret = -ESTALE;
391 if (NFS_STALE(inode))
392 goto out_unlock;
393
394 ret = nfs_netfs_read_folio(file, folio);
395 if (ret)
396 ret = nfs_do_read_folio(file, folio);
397out:
398 trace_nfs_aop_readpage_done(inode, pos, len, ret);
399 return ret;
400out_unlock:
401 folio_unlock(folio);
402 goto out;
403}
404
405void nfs_readahead(struct readahead_control *ractl)
406{
407 struct nfs_pageio_descriptor pgio;
408 struct nfs_open_context *ctx;
409 unsigned int nr_pages = readahead_count(ractl);
410 struct file *file = ractl->file;
411 struct inode *inode = ractl->mapping->host;
412 struct folio *folio;
413 int ret;
414
415 trace_nfs_aop_readahead(inode, readahead_pos(ractl), nr_pages);
416 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
417 task_io_account_read(readahead_length(ractl));
418
419 ret = -ESTALE;
420 if (NFS_STALE(inode))
421 goto out;
422
423 ret = nfs_netfs_readahead(ractl);
424 if (!ret)
425 goto out;
426
427 if (file == NULL) {
428 ret = -EBADF;
429 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
430 if (ctx == NULL)
431 goto out;
432 } else
433 ctx = get_nfs_open_context(nfs_file_open_context(file));
434
435 nfs_pageio_init_read(&pgio, inode, false,
436 &nfs_async_read_completion_ops);
437
438 while ((folio = readahead_folio(ractl)) != NULL) {
439 ret = nfs_read_add_folio(&pgio, ctx, folio);
440 if (ret)
441 break;
442 }
443
444 nfs_pageio_complete_read(&pgio);
445 nfs_update_delegated_atime(inode);
446
447 put_nfs_open_context(ctx);
448out:
449 trace_nfs_aop_readahead_done(inode, nr_pages, ret);
450}
451
452int __init nfs_init_readpagecache(void)
453{
454 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
455 sizeof(struct nfs_pgio_header),
456 0, SLAB_HWCACHE_ALIGN,
457 NULL);
458 if (nfs_rdata_cachep == NULL)
459 return -ENOMEM;
460
461 return 0;
462}
463
464void nfs_destroy_readpagecache(void)
465{
466 kmem_cache_destroy(nfs_rdata_cachep);
467}
468
469static const struct nfs_rw_ops nfs_rw_read_ops = {
470 .rw_alloc_header = nfs_readhdr_alloc,
471 .rw_free_header = nfs_readhdr_free,
472 .rw_done = nfs_readpage_done,
473 .rw_result = nfs_readpage_result,
474 .rw_initiate = nfs_initiate_read,
475};