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