Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
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 | |
1da177e4 LT |
9 | */ |
10 | ||
1da177e4 LT |
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> | |
9c88ea00 | 18 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
19 | #include <linux/pagemap.h> |
20 | #include <linux/sunrpc/clnt.h> | |
21 | #include <linux/nfs_fs.h> | |
22 | #include <linux/nfs_page.h> | |
64419a9b | 23 | #include <linux/module.h> |
1da177e4 | 24 | |
f11c88af | 25 | #include "nfs4_fs.h" |
49a70f27 | 26 | #include "internal.h" |
91d5b470 | 27 | #include "iostat.h" |
9a9fc1c0 | 28 | #include "fscache.h" |
fab5fc25 | 29 | #include "pnfs.h" |
8224b273 | 30 | #include "nfstrace.h" |
91d5b470 | 31 | |
1da177e4 LT |
32 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
33 | ||
000dbe0b | 34 | const struct nfs_pgio_completion_ops nfs_async_read_completion_ops; |
4a0de55c | 35 | static const struct nfs_rw_ops nfs_rw_read_ops; |
1da177e4 | 36 | |
e18b890b | 37 | static struct kmem_cache *nfs_rdata_cachep; |
1da177e4 | 38 | |
1e7f3a48 | 39 | static struct nfs_pgio_header *nfs_readhdr_alloc(void) |
3feb2d49 | 40 | { |
fbe77c30 BC |
41 | struct nfs_pgio_header *p = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL); |
42 | ||
43 | if (p) | |
44 | p->rw_mode = FMODE_READ; | |
45 | return p; | |
4db6e0b7 FI |
46 | } |
47 | ||
1e7f3a48 | 48 | static void nfs_readhdr_free(struct nfs_pgio_header *rhdr) |
3feb2d49 | 49 | { |
cd841605 | 50 | kmem_cache_free(nfs_rdata_cachep, rhdr); |
3feb2d49 TM |
51 | } |
52 | ||
ab75bff1 | 53 | static int nfs_return_empty_folio(struct folio *folio) |
1da177e4 | 54 | { |
ab75bff1 TM |
55 | folio_zero_segment(folio, 0, folio_size(folio)); |
56 | folio_mark_uptodate(folio); | |
57 | folio_unlock(folio); | |
1da177e4 LT |
58 | return 0; |
59 | } | |
60 | ||
1abb5088 | 61 | void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, |
fab5fc25 | 62 | struct inode *inode, bool force_mds, |
061ae2ed | 63 | const struct nfs_pgio_completion_ops *compl_ops) |
1751c363 | 64 | { |
fab5fc25 | 65 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 66 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
fab5fc25 CH |
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 | |
4a0de55c | 72 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops, |
3bde7afd | 73 | server->rsize, 0); |
1751c363 | 74 | } |
ddda8e0a | 75 | EXPORT_SYMBOL_GPL(nfs_pageio_init_read); |
1751c363 | 76 | |
000dbe0b | 77 | void nfs_pageio_complete_read(struct nfs_pageio_descriptor *pgio) |
1e83b173 DW |
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]; | |
b42ad64f | 88 | NFS_I(pgio->pg_inode)->read_io += pgm->pg_bytes_written; |
1e83b173 | 89 | npages = (pgm->pg_bytes_written + PAGE_SIZE - 1) >> PAGE_SHIFT; |
b42ad64f | 90 | nfs_add_stats(pgio->pg_inode, NFSIOS_READPAGES, npages); |
1e83b173 DW |
91 | } |
92 | ||
93 | ||
493292dd TM |
94 | void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) |
95 | { | |
a7d42ddb WAA |
96 | struct nfs_pgio_mirror *mirror; |
97 | ||
6f29b9bb KM |
98 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
99 | pgio->pg_ops->pg_cleanup(pgio); | |
100 | ||
41d8d5b7 | 101 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
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; | |
493292dd | 108 | } |
1f945357 | 109 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); |
493292dd | 110 | |
8f54c7a4 | 111 | static void nfs_readpage_release(struct nfs_page *req, int error) |
0bcbf039 | 112 | { |
ab75bff1 | 113 | struct folio *folio = nfs_page_to_folio(req); |
0bcbf039 | 114 | |
8f54c7a4 | 115 | if (nfs_error_is_fatal_on_server(error) && error != -ETIMEDOUT) |
ab75bff1 | 116 | folio_set_error(folio); |
000dbe0b DW |
117 | if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) |
118 | if (nfs_netfs_folio_unlock(folio)) | |
119 | folio_unlock(folio); | |
120 | ||
0bcbf039 PT |
121 | nfs_release_request(req); |
122 | } | |
123 | ||
67d0338e WAA |
124 | static void nfs_page_group_set_uptodate(struct nfs_page *req) |
125 | { | |
126 | if (nfs_page_group_sync_on_bit(req, PG_UPTODATE)) | |
ab75bff1 | 127 | folio_mark_uptodate(nfs_page_to_folio(req)); |
67d0338e WAA |
128 | } |
129 | ||
061ae2ed | 130 | static void nfs_read_completion(struct nfs_pgio_header *hdr) |
4db6e0b7 FI |
131 | { |
132 | unsigned long bytes = 0; | |
8f54c7a4 | 133 | int error; |
4db6e0b7 FI |
134 | |
135 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
136 | goto out; | |
4bd8b010 TM |
137 | while (!list_empty(&hdr->pages)) { |
138 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
ab75bff1 | 139 | struct folio *folio = nfs_page_to_folio(req); |
68072992 WAA |
140 | unsigned long start = req->wb_pgbase; |
141 | unsigned long end = req->wb_pgbase + req->wb_bytes; | |
4bd8b010 TM |
142 | |
143 | if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) { | |
68072992 | 144 | /* note: regions of the page not covered by a |
01c3a400 DW |
145 | * request are zeroed in nfs_read_add_folio |
146 | */ | |
68072992 WAA |
147 | if (bytes > hdr->good_bytes) { |
148 | /* nothing in this request was good, so zero | |
149 | * the full extent of the request */ | |
ab75bff1 | 150 | folio_zero_segment(folio, start, end); |
68072992 WAA |
151 | |
152 | } else if (hdr->good_bytes - bytes < req->wb_bytes) { | |
153 | /* part of this request has good bytes, but | |
154 | * not all. zero the bad bytes */ | |
155 | start += hdr->good_bytes - bytes; | |
156 | WARN_ON(start < req->wb_pgbase); | |
ab75bff1 | 157 | folio_zero_segment(folio, start, end); |
68072992 | 158 | } |
4db6e0b7 | 159 | } |
8f54c7a4 | 160 | error = 0; |
4bd8b010 TM |
161 | bytes += req->wb_bytes; |
162 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
4db6e0b7 | 163 | if (bytes <= hdr->good_bytes) |
67d0338e | 164 | nfs_page_group_set_uptodate(req); |
8f54c7a4 TM |
165 | else { |
166 | error = hdr->error; | |
167 | xchg(&nfs_req_openctx(req)->error, error); | |
168 | } | |
4bd8b010 | 169 | } else |
67d0338e | 170 | nfs_page_group_set_uptodate(req); |
4bd8b010 | 171 | nfs_list_remove_request(req); |
8f54c7a4 | 172 | nfs_readpage_release(req, error); |
4db6e0b7 | 173 | } |
000dbe0b DW |
174 | nfs_netfs_read_completion(hdr); |
175 | ||
4db6e0b7 FI |
176 | out: |
177 | hdr->release(hdr); | |
178 | } | |
179 | ||
d45f60c6 WAA |
180 | static void nfs_initiate_read(struct nfs_pgio_header *hdr, |
181 | struct rpc_message *msg, | |
abde71f4 | 182 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 183 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 184 | { |
abde71f4 | 185 | rpc_ops->read_setup(hdr, msg); |
000dbe0b | 186 | nfs_netfs_initiate_read(hdr); |
2343172d | 187 | trace_nfs_initiate_read(hdr); |
64419a9b AA |
188 | } |
189 | ||
061ae2ed | 190 | static void |
df3accb8 | 191 | nfs_async_read_error(struct list_head *head, int error) |
1da177e4 LT |
192 | { |
193 | struct nfs_page *req; | |
194 | ||
195 | while (!list_empty(head)) { | |
196 | req = nfs_list_entry(head->next); | |
197 | nfs_list_remove_request(req); | |
8f54c7a4 | 198 | nfs_readpage_release(req, error); |
1da177e4 LT |
199 | } |
200 | } | |
201 | ||
000dbe0b | 202 | const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = { |
061ae2ed FI |
203 | .error_cleanup = nfs_async_read_error, |
204 | .completion = nfs_read_completion, | |
205 | }; | |
206 | ||
0b671301 TM |
207 | /* |
208 | * This is the callback from RPC telling us whether a reply was | |
209 | * received or some error occurred (timeout or socket shutdown). | |
210 | */ | |
d45f60c6 WAA |
211 | static int nfs_readpage_done(struct rpc_task *task, |
212 | struct nfs_pgio_header *hdr, | |
0eecb214 | 213 | struct inode *inode) |
0b671301 | 214 | { |
d45f60c6 | 215 | int status = NFS_PROTO(inode)->read_done(task, hdr); |
0b671301 TM |
216 | if (status != 0) |
217 | return status; | |
218 | ||
d45f60c6 | 219 | nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count); |
2343172d | 220 | trace_nfs_readpage_done(task, hdr); |
0b671301 TM |
221 | |
222 | if (task->tk_status == -ESTALE) { | |
93ce4af7 | 223 | nfs_set_inode_stale(inode); |
cd841605 | 224 | nfs_mark_for_revalidate(inode); |
0b671301 | 225 | } |
0b671301 TM |
226 | return 0; |
227 | } | |
228 | ||
d45f60c6 WAA |
229 | static void nfs_readpage_retry(struct rpc_task *task, |
230 | struct nfs_pgio_header *hdr) | |
0b671301 | 231 | { |
d45f60c6 WAA |
232 | struct nfs_pgio_args *argp = &hdr->args; |
233 | struct nfs_pgio_res *resp = &hdr->res; | |
0b671301 | 234 | |
0b671301 | 235 | /* This is a short read! */ |
d45f60c6 | 236 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD); |
fd2b6121 CL |
237 | trace_nfs_readpage_short(task, hdr); |
238 | ||
0b671301 | 239 | /* Has the server at least made some progress? */ |
4db6e0b7 | 240 | if (resp->count == 0) { |
d45f60c6 | 241 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
d61e612a | 242 | return; |
4db6e0b7 | 243 | } |
f8417b48 KM |
244 | |
245 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
246 | if (!task->tk_ops) { | |
247 | hdr->pnfs_error = -EAGAIN; | |
248 | return; | |
249 | } | |
250 | ||
d45f60c6 WAA |
251 | /* Yes, so retry the read at the end of the hdr */ |
252 | hdr->mds_offset += resp->count; | |
0b671301 TM |
253 | argp->offset += resp->count; |
254 | argp->pgbase += resp->count; | |
255 | argp->count -= resp->count; | |
8c9cb714 TM |
256 | resp->count = 0; |
257 | resp->eof = 0; | |
d00c5d43 | 258 | rpc_restart_call_prepare(task); |
0b671301 TM |
259 | } |
260 | ||
d45f60c6 WAA |
261 | static void nfs_readpage_result(struct rpc_task *task, |
262 | struct nfs_pgio_header *hdr) | |
1da177e4 | 263 | { |
d45f60c6 | 264 | if (hdr->res.eof) { |
1c6c4b74 TM |
265 | loff_t pos = hdr->args.offset + hdr->res.count; |
266 | unsigned int new = pos - hdr->io_start; | |
4db6e0b7 | 267 | |
1c6c4b74 TM |
268 | if (hdr->good_bytes > new) { |
269 | hdr->good_bytes = new; | |
4db6e0b7 FI |
270 | set_bit(NFS_IOHDR_EOF, &hdr->flags); |
271 | clear_bit(NFS_IOHDR_ERROR, &hdr->flags); | |
4db6e0b7 | 272 | } |
f8417b48 | 273 | } else if (hdr->res.count < hdr->args.count) |
d45f60c6 | 274 | nfs_readpage_retry(task, hdr); |
fdd1e74c TM |
275 | } |
276 | ||
000dbe0b DW |
277 | int nfs_read_add_folio(struct nfs_pageio_descriptor *pgio, |
278 | struct nfs_open_context *ctx, | |
279 | struct folio *folio) | |
1e83b173 | 280 | { |
ab75bff1 TM |
281 | struct inode *inode = folio_file_mapping(folio)->host; |
282 | struct nfs_server *server = NFS_SERVER(inode); | |
283 | size_t fsize = folio_size(folio); | |
284 | unsigned int rsize = server->rsize; | |
1e83b173 | 285 | struct nfs_page *new; |
8cfb9015 | 286 | unsigned int len, aligned_len; |
1e83b173 DW |
287 | int error; |
288 | ||
ab75bff1 | 289 | len = nfs_folio_length(folio); |
1e83b173 | 290 | if (len == 0) |
ab75bff1 | 291 | return nfs_return_empty_folio(folio); |
1e83b173 | 292 | |
ab75bff1 | 293 | aligned_len = min_t(unsigned int, ALIGN(len, rsize), fsize); |
8cfb9015 | 294 | |
01c3a400 | 295 | new = nfs_page_create_from_folio(ctx, folio, 0, aligned_len); |
000dbe0b DW |
296 | if (IS_ERR(new)) { |
297 | error = PTR_ERR(new); | |
298 | goto out; | |
299 | } | |
1e83b173 | 300 | |
ab75bff1 TM |
301 | if (len < fsize) |
302 | folio_zero_segment(folio, len, fsize); | |
01c3a400 | 303 | if (!nfs_pageio_add_request(pgio, new)) { |
1e83b173 | 304 | nfs_list_remove_request(new); |
01c3a400 | 305 | error = pgio->pg_error; |
1e83b173 DW |
306 | nfs_readpage_release(new, error); |
307 | goto out; | |
308 | } | |
309 | return 0; | |
1e83b173 DW |
310 | out: |
311 | return error; | |
312 | } | |
313 | ||
1da177e4 LT |
314 | /* |
315 | * Read a page over NFS. | |
316 | * We read the page synchronously in the following case: | |
317 | * - The error flag is set for this page. This happens only when a | |
318 | * previous async read operation failed. | |
319 | */ | |
65d023af | 320 | int nfs_read_folio(struct file *file, struct folio *folio) |
1da177e4 | 321 | { |
ab75bff1 | 322 | struct inode *inode = file_inode(file); |
01c3a400 DW |
323 | struct nfs_pageio_descriptor pgio; |
324 | struct nfs_open_context *ctx; | |
49dee700 | 325 | int ret; |
1da177e4 | 326 | |
ab75bff1 | 327 | trace_nfs_aop_readpage(inode, folio); |
91d5b470 | 328 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
9c88ea00 | 329 | task_io_account_read(folio_size(folio)); |
91d5b470 | 330 | |
1da177e4 LT |
331 | /* |
332 | * Try to flush any pending writes to the file.. | |
333 | * | |
ab75bff1 | 334 | * NOTE! Because we own the folio lock, there cannot |
1da177e4 | 335 | * be any new pending writes generated at this point |
ab75bff1 | 336 | * for this folio (other folios can be written to). |
1da177e4 | 337 | */ |
ab75bff1 | 338 | ret = nfs_wb_folio(inode, folio); |
49dee700 | 339 | if (ret) |
de05a0cc | 340 | goto out_unlock; |
ab75bff1 | 341 | if (folio_test_uptodate(folio)) |
de05a0cc | 342 | goto out_unlock; |
1da177e4 | 343 | |
49dee700 | 344 | ret = -ESTALE; |
5f004cf2 | 345 | if (NFS_STALE(inode)) |
de05a0cc | 346 | goto out_unlock; |
5f004cf2 | 347 | |
000dbe0b DW |
348 | ret = nfs_netfs_read_folio(file, folio); |
349 | if (!ret) | |
350 | goto out; | |
351 | ||
01c3a400 | 352 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 353 | |
01c3a400 DW |
354 | xchg(&ctx->error, 0); |
355 | nfs_pageio_init_read(&pgio, inode, false, | |
1e83b173 DW |
356 | &nfs_async_read_completion_ops); |
357 | ||
01c3a400 | 358 | ret = nfs_read_add_folio(&pgio, ctx, folio); |
e0340f16 | 359 | if (ret) |
000dbe0b | 360 | goto out_put; |
1e83b173 | 361 | |
01c3a400 DW |
362 | nfs_pageio_complete_read(&pgio); |
363 | ret = pgio.pg_error < 0 ? pgio.pg_error : 0; | |
49dee700 | 364 | if (!ret) { |
ab75bff1 TM |
365 | ret = folio_wait_locked_killable(folio); |
366 | if (!folio_test_uptodate(folio) && !ret) | |
01c3a400 | 367 | ret = xchg(&ctx->error, 0); |
8f54c7a4 | 368 | } |
000dbe0b | 369 | out_put: |
01c3a400 | 370 | put_nfs_open_context(ctx); |
000dbe0b | 371 | out: |
ab75bff1 | 372 | trace_nfs_aop_readpage_done(inode, folio, ret); |
49dee700 | 373 | return ret; |
de05a0cc | 374 | out_unlock: |
ab75bff1 | 375 | folio_unlock(folio); |
000dbe0b | 376 | goto out; |
1da177e4 LT |
377 | } |
378 | ||
8786fde8 | 379 | void nfs_readahead(struct readahead_control *ractl) |
1da177e4 | 380 | { |
01c3a400 DW |
381 | struct nfs_pageio_descriptor pgio; |
382 | struct nfs_open_context *ctx; | |
8786fde8 MWO |
383 | unsigned int nr_pages = readahead_count(ractl); |
384 | struct file *file = ractl->file; | |
8786fde8 | 385 | struct inode *inode = ractl->mapping->host; |
ab75bff1 | 386 | struct folio *folio; |
49dee700 | 387 | int ret; |
1da177e4 | 388 | |
8786fde8 | 389 | trace_nfs_aop_readahead(inode, readahead_pos(ractl), nr_pages); |
91d5b470 | 390 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
9c88ea00 | 391 | task_io_account_read(readahead_length(ractl)); |
1da177e4 | 392 | |
49dee700 | 393 | ret = -ESTALE; |
5f004cf2 TM |
394 | if (NFS_STALE(inode)) |
395 | goto out; | |
396 | ||
000dbe0b DW |
397 | ret = nfs_netfs_readahead(ractl); |
398 | if (!ret) | |
399 | goto out; | |
400 | ||
49dee700 DW |
401 | if (file == NULL) { |
402 | ret = -EBADF; | |
01c3a400 DW |
403 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
404 | if (ctx == NULL) | |
49dee700 | 405 | goto out; |
1da177e4 | 406 | } else |
01c3a400 | 407 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
9a9fc1c0 | 408 | |
01c3a400 | 409 | nfs_pageio_init_read(&pgio, inode, false, |
fab5fc25 | 410 | &nfs_async_read_completion_ops); |
8b09bee3 | 411 | |
ab75bff1 | 412 | while ((folio = readahead_folio(ractl)) != NULL) { |
01c3a400 | 413 | ret = nfs_read_add_folio(&pgio, ctx, folio); |
8786fde8 MWO |
414 | if (ret) |
415 | break; | |
416 | } | |
a7d42ddb | 417 | |
01c3a400 | 418 | nfs_pageio_complete_read(&pgio); |
a7d42ddb | 419 | |
01c3a400 | 420 | put_nfs_open_context(ctx); |
5f004cf2 | 421 | out: |
d9f87743 | 422 | trace_nfs_aop_readahead_done(inode, nr_pages, ret); |
1da177e4 LT |
423 | } |
424 | ||
f7b422b1 | 425 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
426 | { |
427 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
1e7f3a48 | 428 | sizeof(struct nfs_pgio_header), |
1da177e4 | 429 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 430 | NULL); |
1da177e4 LT |
431 | if (nfs_rdata_cachep == NULL) |
432 | return -ENOMEM; | |
433 | ||
1da177e4 LT |
434 | return 0; |
435 | } | |
436 | ||
266bee88 | 437 | void nfs_destroy_readpagecache(void) |
1da177e4 | 438 | { |
1a1d92c1 | 439 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 440 | } |
4a0de55c AS |
441 | |
442 | static const struct nfs_rw_ops nfs_rw_read_ops = { | |
443 | .rw_alloc_header = nfs_readhdr_alloc, | |
444 | .rw_free_header = nfs_readhdr_free, | |
0eecb214 AS |
445 | .rw_done = nfs_readpage_done, |
446 | .rw_result = nfs_readpage_result, | |
1ed26f33 | 447 | .rw_initiate = nfs_initiate_read, |
4a0de55c | 448 | }; |