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
32#define NFSDBG_FACILITY NFSDBG_PAGECACHE
33
34const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
35static const struct nfs_rw_ops nfs_rw_read_ops;
36
37static struct kmem_cache *nfs_rdata_cachep;
38
39static struct nfs_pgio_header *nfs_readhdr_alloc(void)
40{
41 struct nfs_pgio_header *p = kmem_cache_zalloc(k: nfs_rdata_cachep, GFP_KERNEL);
42
43 if (p)
44 p->rw_mode = FMODE_READ;
45 return p;
46}
47
48static void nfs_readhdr_free(struct nfs_pgio_header *rhdr)
49{
50 if (rhdr->res.scratch != NULL)
51 kfree(objp: rhdr->res.scratch);
52 kmem_cache_free(s: nfs_rdata_cachep, objp: rhdr);
53}
54
55static int nfs_return_empty_folio(struct folio *folio)
56{
57 folio_zero_segment(folio, start: 0, xend: 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(desc: pgio, inode, pg_ops, compl_ops, rw_ops: &nfs_rw_read_ops,
75 bsize: server->rsize, how: 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(desc: 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(inode: pgio->pg_inode)->read_io += pgm->pg_bytes_written;
91 npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >> PAGE_SHIFT;
92 nfs_add_stats(inode: pgio->pg_inode, stat: NFSIOS_READPAGES, addend: 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(inode: 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_error_is_fatal_on_server(err: error) && error != -ETIMEDOUT)
126 folio_set_error(folio);
127 if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE))
128 if (nfs_netfs_folio_unlock(folio))
129 folio_unlock(folio);
130
131 nfs_release_request(req);
132}
133
134static void nfs_page_group_set_uptodate(struct nfs_page *req)
135{
136 if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
137 folio_mark_uptodate(folio: nfs_page_to_folio(req));
138}
139
140static void nfs_read_completion(struct nfs_pgio_header *hdr)
141{
142 unsigned long bytes = 0;
143 int error;
144
145 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
146 goto out;
147 while (!list_empty(head: &hdr->pages)) {
148 struct nfs_page *req = nfs_list_entry(head: hdr->pages.next);
149 struct folio *folio = nfs_page_to_folio(req);
150 unsigned long start = req->wb_pgbase;
151 unsigned long end = req->wb_pgbase + req->wb_bytes;
152
153 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
154 /* note: regions of the page not covered by a
155 * request are zeroed in nfs_read_add_folio
156 */
157 if (bytes > hdr->good_bytes) {
158 /* nothing in this request was good, so zero
159 * the full extent of the request */
160 folio_zero_segment(folio, start, xend: 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 folio_zero_segment(folio, start, xend: 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 }
184 nfs_netfs_read_completion(hdr);
185
186out:
187 hdr->release(hdr);
188}
189
190static void nfs_initiate_read(struct nfs_pgio_header *hdr,
191 struct rpc_message *msg,
192 const struct nfs_rpc_ops *rpc_ops,
193 struct rpc_task_setup *task_setup_data, int how)
194{
195 rpc_ops->read_setup(hdr, msg);
196 nfs_netfs_initiate_read(hdr);
197 trace_nfs_initiate_read(hdr);
198}
199
200static void
201nfs_async_read_error(struct list_head *head, int error)
202{
203 struct nfs_page *req;
204
205 while (!list_empty(head)) {
206 req = nfs_list_entry(head: head->next);
207 nfs_list_remove_request(req);
208 nfs_readpage_release(req, error);
209 }
210}
211
212const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
213 .error_cleanup = nfs_async_read_error,
214 .completion = nfs_read_completion,
215};
216
217/*
218 * This is the callback from RPC telling us whether a reply was
219 * received or some error occurred (timeout or socket shutdown).
220 */
221static int nfs_readpage_done(struct rpc_task *task,
222 struct nfs_pgio_header *hdr,
223 struct inode *inode)
224{
225 int status = NFS_PROTO(inode)->read_done(task, hdr);
226 if (status != 0)
227 return status;
228
229 nfs_add_stats(inode, stat: NFSIOS_SERVERREADBYTES, addend: hdr->res.count);
230 trace_nfs_readpage_done(task, hdr);
231
232 if (task->tk_status == -ESTALE) {
233 nfs_set_inode_stale(inode);
234 nfs_mark_for_revalidate(inode);
235 }
236 return 0;
237}
238
239static void nfs_readpage_retry(struct rpc_task *task,
240 struct nfs_pgio_header *hdr)
241{
242 struct nfs_pgio_args *argp = &hdr->args;
243 struct nfs_pgio_res *resp = &hdr->res;
244
245 /* This is a short read! */
246 nfs_inc_stats(inode: hdr->inode, stat: NFSIOS_SHORTREAD);
247 trace_nfs_readpage_short(task, hdr);
248
249 /* Has the server at least made some progress? */
250 if (resp->count == 0) {
251 nfs_set_pgio_error(hdr, error: -EIO, pos: argp->offset);
252 return;
253 }
254
255 /* For non rpc-based layout drivers, retry-through-MDS */
256 if (!task->tk_ops) {
257 hdr->pnfs_error = -EAGAIN;
258 return;
259 }
260
261 /* Yes, so retry the read at the end of the hdr */
262 hdr->mds_offset += resp->count;
263 argp->offset += resp->count;
264 argp->pgbase += resp->count;
265 argp->count -= resp->count;
266 resp->count = 0;
267 resp->eof = 0;
268 rpc_restart_call_prepare(task);
269}
270
271static void nfs_readpage_result(struct rpc_task *task,
272 struct nfs_pgio_header *hdr)
273{
274 if (hdr->res.eof) {
275 loff_t pos = hdr->args.offset + hdr->res.count;
276 unsigned int new = pos - hdr->io_start;
277
278 if (hdr->good_bytes > new) {
279 hdr->good_bytes = new;
280 set_bit(nr: NFS_IOHDR_EOF, addr: &hdr->flags);
281 clear_bit(nr: NFS_IOHDR_ERROR, addr: &hdr->flags);
282 }
283 } else if (hdr->res.count < hdr->args.count)
284 nfs_readpage_retry(task, hdr);
285}
286
287int nfs_read_add_folio(struct nfs_pageio_descriptor *pgio,
288 struct nfs_open_context *ctx,
289 struct folio *folio)
290{
291 struct inode *inode = folio_file_mapping(folio)->host;
292 struct nfs_server *server = NFS_SERVER(inode);
293 size_t fsize = folio_size(folio);
294 unsigned int rsize = server->rsize;
295 struct nfs_page *new;
296 unsigned int len, aligned_len;
297 int error;
298
299 len = nfs_folio_length(folio);
300 if (len == 0)
301 return nfs_return_empty_folio(folio);
302
303 aligned_len = min_t(unsigned int, ALIGN(len, rsize), fsize);
304
305 new = nfs_page_create_from_folio(ctx, folio, offset: 0, count: aligned_len);
306 if (IS_ERR(ptr: new)) {
307 error = PTR_ERR(ptr: new);
308 if (nfs_netfs_folio_unlock(folio))
309 folio_unlock(folio);
310 goto out;
311 }
312
313 if (len < fsize)
314 folio_zero_segment(folio, start: len, xend: fsize);
315 if (!nfs_pageio_add_request(pgio, new)) {
316 nfs_list_remove_request(req: new);
317 error = pgio->pg_error;
318 nfs_readpage_release(req: new, error);
319 goto out;
320 }
321 return 0;
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 inode *inode = file_inode(f: file);
335 struct nfs_pageio_descriptor pgio;
336 struct nfs_open_context *ctx;
337 int ret;
338
339 trace_nfs_aop_readpage(inode, folio);
340 nfs_inc_stats(inode, stat: NFSIOS_VFSREADPAGE);
341 task_io_account_read(bytes: folio_size(folio));
342
343 /*
344 * Try to flush any pending writes to the file..
345 *
346 * NOTE! Because we own the folio lock, there cannot
347 * be any new pending writes generated at this point
348 * for this folio (other folios can be written to).
349 */
350 ret = nfs_wb_folio(inode, folio);
351 if (ret)
352 goto out_unlock;
353 if (folio_test_uptodate(folio))
354 goto out_unlock;
355
356 ret = -ESTALE;
357 if (NFS_STALE(inode))
358 goto out_unlock;
359
360 ret = nfs_netfs_read_folio(file, folio);
361 if (!ret)
362 goto out;
363
364 ctx = get_nfs_open_context(ctx: nfs_file_open_context(filp: file));
365
366 xchg(&ctx->error, 0);
367 nfs_pageio_init_read(&pgio, inode, false,
368 &nfs_async_read_completion_ops);
369
370 ret = nfs_read_add_folio(pgio: &pgio, ctx, folio);
371 if (ret)
372 goto out_put;
373
374 nfs_pageio_complete_read(pgio: &pgio);
375 ret = pgio.pg_error < 0 ? pgio.pg_error : 0;
376 if (!ret) {
377 ret = folio_wait_locked_killable(folio);
378 if (!folio_test_uptodate(folio) && !ret)
379 ret = xchg(&ctx->error, 0);
380 }
381out_put:
382 put_nfs_open_context(ctx);
383out:
384 trace_nfs_aop_readpage_done(inode, folio, ret);
385 return ret;
386out_unlock:
387 folio_unlock(folio);
388 goto out;
389}
390
391void nfs_readahead(struct readahead_control *ractl)
392{
393 struct nfs_pageio_descriptor pgio;
394 struct nfs_open_context *ctx;
395 unsigned int nr_pages = readahead_count(rac: ractl);
396 struct file *file = ractl->file;
397 struct inode *inode = ractl->mapping->host;
398 struct folio *folio;
399 int ret;
400
401 trace_nfs_aop_readahead(inode, pos: readahead_pos(rac: ractl), nr_pages);
402 nfs_inc_stats(inode, stat: NFSIOS_VFSREADPAGES);
403 task_io_account_read(bytes: readahead_length(rac: ractl));
404
405 ret = -ESTALE;
406 if (NFS_STALE(inode))
407 goto out;
408
409 ret = nfs_netfs_readahead(ractl);
410 if (!ret)
411 goto out;
412
413 if (file == NULL) {
414 ret = -EBADF;
415 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
416 if (ctx == NULL)
417 goto out;
418 } else
419 ctx = get_nfs_open_context(ctx: nfs_file_open_context(filp: file));
420
421 nfs_pageio_init_read(&pgio, inode, false,
422 &nfs_async_read_completion_ops);
423
424 while ((folio = readahead_folio(ractl)) != NULL) {
425 ret = nfs_read_add_folio(pgio: &pgio, ctx, folio);
426 if (ret)
427 break;
428 }
429
430 nfs_pageio_complete_read(pgio: &pgio);
431
432 put_nfs_open_context(ctx);
433out:
434 trace_nfs_aop_readahead_done(inode, nr_pages, ret);
435}
436
437int __init nfs_init_readpagecache(void)
438{
439 nfs_rdata_cachep = kmem_cache_create(name: "nfs_read_data",
440 size: sizeof(struct nfs_pgio_header),
441 align: 0, SLAB_HWCACHE_ALIGN,
442 NULL);
443 if (nfs_rdata_cachep == NULL)
444 return -ENOMEM;
445
446 return 0;
447}
448
449void nfs_destroy_readpagecache(void)
450{
451 kmem_cache_destroy(s: nfs_rdata_cachep);
452}
453
454static const struct nfs_rw_ops nfs_rw_read_ops = {
455 .rw_alloc_header = nfs_readhdr_alloc,
456 .rw_free_header = nfs_readhdr_free,
457 .rw_done = nfs_readpage_done,
458 .rw_result = nfs_readpage_result,
459 .rw_initiate = nfs_initiate_read,
460};
461

source code of linux/fs/nfs/read.c