]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
1da177e4 LT |
8 | */ |
9 | ||
1da177e4 LT |
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> | |
64419a9b | 21 | #include <linux/module.h> |
1da177e4 LT |
22 | |
23 | #include <asm/system.h> | |
bae724ef | 24 | #include "pnfs.h" |
1da177e4 | 25 | |
f11c88af | 26 | #include "nfs4_fs.h" |
49a70f27 | 27 | #include "internal.h" |
91d5b470 | 28 | #include "iostat.h" |
9a9fc1c0 | 29 | #include "fscache.h" |
91d5b470 | 30 | |
1da177e4 LT |
31 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
32 | ||
1751c363 | 33 | static const struct nfs_pageio_ops nfs_pageio_read_ops; |
ec06c096 TM |
34 | static const struct rpc_call_ops nfs_read_partial_ops; |
35 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 | 36 | |
e18b890b | 37 | static struct kmem_cache *nfs_rdata_cachep; |
3feb2d49 | 38 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
39 | |
40 | #define MIN_POOL_READ (32) | |
41 | ||
8d5658c9 | 42 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 43 | { |
93870d76 | 44 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL); |
3feb2d49 TM |
45 | |
46 | if (p) { | |
47 | memset(p, 0, sizeof(*p)); | |
48 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 49 | p->npages = pagecount; |
0d0b5cb3 CL |
50 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
51 | p->pagevec = p->page_array; | |
3feb2d49 | 52 | else { |
93870d76 | 53 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); |
0d0b5cb3 | 54 | if (!p->pagevec) { |
3feb2d49 TM |
55 | mempool_free(p, nfs_rdata_mempool); |
56 | p = NULL; | |
57 | } | |
58 | } | |
59 | } | |
60 | return p; | |
61 | } | |
62 | ||
1ae88b2e | 63 | void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
64 | { |
65 | if (p && (p->pagevec != &p->page_array[0])) | |
66 | kfree(p->pagevec); | |
67 | mempool_free(p, nfs_rdata_mempool); | |
68 | } | |
69 | ||
493292dd | 70 | void nfs_readdata_release(struct nfs_read_data *rdata) |
1da177e4 | 71 | { |
bae724ef | 72 | put_lseg(rdata->lseg); |
383ba719 TM |
73 | put_nfs_open_context(rdata->args.context); |
74 | nfs_readdata_free(rdata); | |
1da177e4 LT |
75 | } |
76 | ||
1da177e4 LT |
77 | static |
78 | int nfs_return_empty_page(struct page *page) | |
79 | { | |
eebd2aa3 | 80 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
81 | SetPageUptodate(page); |
82 | unlock_page(page); | |
83 | return 0; | |
84 | } | |
85 | ||
1de3fc12 TM |
86 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
87 | { | |
88 | unsigned int remainder = data->args.count - data->res.count; | |
89 | unsigned int base = data->args.pgbase + data->res.count; | |
90 | unsigned int pglen; | |
91 | struct page **pages; | |
92 | ||
93 | if (data->res.eof == 0 || remainder == 0) | |
94 | return; | |
95 | /* | |
96 | * Note: "remainder" can never be negative, since we check for | |
97 | * this in the XDR code. | |
98 | */ | |
99 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
100 | base &= ~PAGE_CACHE_MASK; | |
101 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
102 | for (;;) { |
103 | if (remainder <= pglen) { | |
eebd2aa3 | 104 | zero_user(*pages, base, remainder); |
79558f36 TM |
105 | break; |
106 | } | |
eebd2aa3 | 107 | zero_user(*pages, base, pglen); |
79558f36 TM |
108 | pages++; |
109 | remainder -= pglen; | |
110 | pglen = PAGE_CACHE_SIZE; | |
111 | base = 0; | |
112 | } | |
1de3fc12 TM |
113 | } |
114 | ||
1f945357 | 115 | static void nfs_pageio_init_read_mds(struct nfs_pageio_descriptor *pgio, |
1751c363 TM |
116 | struct inode *inode) |
117 | { | |
118 | nfs_pageio_init(pgio, inode, &nfs_pageio_read_ops, | |
119 | NFS_SERVER(inode)->rsize, 0); | |
120 | } | |
121 | ||
493292dd TM |
122 | void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) |
123 | { | |
124 | pgio->pg_ops = &nfs_pageio_read_ops; | |
125 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize; | |
126 | } | |
1f945357 | 127 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); |
493292dd | 128 | |
1751c363 TM |
129 | static void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, |
130 | struct inode *inode) | |
131 | { | |
132 | if (!pnfs_pageio_init_read(pgio, inode)) | |
133 | nfs_pageio_init_read_mds(pgio, inode); | |
134 | } | |
135 | ||
f42b293d DH |
136 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
137 | struct page *page) | |
1da177e4 | 138 | { |
1da177e4 LT |
139 | struct nfs_page *new; |
140 | unsigned int len; | |
c76069bd | 141 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 142 | |
49a70f27 | 143 | len = nfs_page_length(page); |
1da177e4 LT |
144 | if (len == 0) |
145 | return nfs_return_empty_page(page); | |
146 | new = nfs_create_request(ctx, inode, page, 0, len); | |
147 | if (IS_ERR(new)) { | |
148 | unlock_page(page); | |
149 | return PTR_ERR(new); | |
150 | } | |
151 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 152 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 153 | |
1751c363 | 154 | nfs_pageio_init_read(&pgio, inode); |
d8007d4d | 155 | nfs_pageio_add_request(&pgio, new); |
1751c363 | 156 | nfs_pageio_complete(&pgio); |
1da177e4 LT |
157 | return 0; |
158 | } | |
159 | ||
160 | static void nfs_readpage_release(struct nfs_page *req) | |
161 | { | |
3d4ff43d | 162 | struct inode *d_inode = req->wb_context->dentry->d_inode; |
7f8e05f6 DH |
163 | |
164 | if (PageUptodate(req->wb_page)) | |
165 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
166 | ||
1da177e4 LT |
167 | unlock_page(req->wb_page); |
168 | ||
1da177e4 | 169 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
3d4ff43d AV |
170 | req->wb_context->dentry->d_inode->i_sb->s_id, |
171 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
172 | req->wb_bytes, |
173 | (long long)req_offset(req)); | |
10d2c46f | 174 | nfs_release_request(req); |
1da177e4 LT |
175 | } |
176 | ||
dc70d7b3 | 177 | int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt, |
64419a9b | 178 | const struct rpc_call_ops *call_ops) |
1da177e4 | 179 | { |
64419a9b | 180 | struct inode *inode = data->inode; |
84115e1c | 181 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
07737691 | 182 | struct rpc_task *task; |
bdc7f021 TM |
183 | struct rpc_message msg = { |
184 | .rpc_argp = &data->args, | |
185 | .rpc_resp = &data->res, | |
64419a9b | 186 | .rpc_cred = data->cred, |
bdc7f021 | 187 | }; |
84115e1c | 188 | struct rpc_task_setup task_setup_data = { |
07737691 | 189 | .task = &data->task, |
64419a9b | 190 | .rpc_client = clnt, |
bdc7f021 | 191 | .rpc_message = &msg, |
84115e1c TM |
192 | .callback_ops = call_ops, |
193 | .callback_data = data, | |
101070ca | 194 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
195 | .flags = RPC_TASK_ASYNC | swap_flags, |
196 | }; | |
1da177e4 | 197 | |
64419a9b AA |
198 | /* Set up the initial task struct. */ |
199 | NFS_PROTO(inode)->read_setup(data, &msg); | |
200 | ||
201 | dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ " | |
202 | "offset %llu)\n", | |
203 | data->task.tk_pid, | |
204 | inode->i_sb->s_id, | |
205 | (long long)NFS_FILEID(inode), | |
206 | data->args.count, | |
207 | (unsigned long long)data->args.offset); | |
208 | ||
209 | task = rpc_run_task(&task_setup_data); | |
210 | if (IS_ERR(task)) | |
211 | return PTR_ERR(task); | |
212 | rpc_put_task(task); | |
213 | return 0; | |
214 | } | |
dc70d7b3 | 215 | EXPORT_SYMBOL_GPL(nfs_initiate_read); |
64419a9b AA |
216 | |
217 | /* | |
218 | * Set up the NFS read request struct | |
219 | */ | |
6e4efd56 TM |
220 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
221 | unsigned int count, unsigned int offset) | |
64419a9b | 222 | { |
3d4ff43d | 223 | struct inode *inode = req->wb_context->dentry->d_inode; |
64419a9b | 224 | |
1da177e4 | 225 | data->req = req; |
84115e1c | 226 | data->inode = inode; |
64419a9b | 227 | data->cred = req->wb_context->cred; |
1da177e4 LT |
228 | |
229 | data->args.fh = NFS_FH(inode); | |
230 | data->args.offset = req_offset(req) + offset; | |
231 | data->args.pgbase = req->wb_pgbase + offset; | |
232 | data->args.pages = data->pagevec; | |
233 | data->args.count = count; | |
383ba719 | 234 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 235 | data->args.lock_context = req->wb_lock_context; |
1da177e4 LT |
236 | |
237 | data->res.fattr = &data->fattr; | |
238 | data->res.count = count; | |
239 | data->res.eof = 0; | |
0e574af1 | 240 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 241 | } |
1da177e4 | 242 | |
6e4efd56 | 243 | static int nfs_do_read(struct nfs_read_data *data, |
493292dd | 244 | const struct rpc_call_ops *call_ops) |
6e4efd56 | 245 | { |
5f00bcb3 | 246 | struct inode *inode = data->args.context->dentry->d_inode; |
6e4efd56 | 247 | |
64419a9b | 248 | return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops); |
1da177e4 LT |
249 | } |
250 | ||
275acaaf TM |
251 | static int |
252 | nfs_do_multiple_reads(struct list_head *head, | |
493292dd | 253 | const struct rpc_call_ops *call_ops) |
275acaaf TM |
254 | { |
255 | struct nfs_read_data *data; | |
256 | int ret = 0; | |
257 | ||
258 | while (!list_empty(head)) { | |
259 | int ret2; | |
260 | ||
261 | data = list_entry(head->next, struct nfs_read_data, list); | |
262 | list_del_init(&data->list); | |
263 | ||
493292dd | 264 | ret2 = nfs_do_read(data, call_ops); |
275acaaf TM |
265 | if (ret == 0) |
266 | ret = ret2; | |
267 | } | |
268 | return ret; | |
269 | } | |
270 | ||
1da177e4 LT |
271 | static void |
272 | nfs_async_read_error(struct list_head *head) | |
273 | { | |
274 | struct nfs_page *req; | |
275 | ||
276 | while (!list_empty(head)) { | |
277 | req = nfs_list_entry(head->next); | |
278 | nfs_list_remove_request(req); | |
279 | SetPageError(req->wb_page); | |
280 | nfs_readpage_release(req); | |
281 | } | |
282 | } | |
283 | ||
1da177e4 LT |
284 | /* |
285 | * Generate multiple requests to fill a single page. | |
286 | * | |
287 | * We optimize to reduce the number of read operations on the wire. If we | |
288 | * detect that we're reading a page, or an area of a page, that is past the | |
289 | * end of file, we do not generate NFS read operations but just clear the | |
290 | * parts of the page that would have come back zero from the server anyway. | |
291 | * | |
292 | * We rely on the cached value of i_size to make this determination; another | |
293 | * client can fill pages on the server past our cached end-of-file, but we | |
294 | * won't see the new data until our attribute cache is updated. This is more | |
295 | * or less conventional NFS client behavior. | |
296 | */ | |
275acaaf | 297 | static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 | 298 | { |
c76069bd | 299 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
300 | struct page *page = req->wb_page; |
301 | struct nfs_read_data *data; | |
d097971d | 302 | size_t rsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 303 | unsigned int offset; |
1da177e4 | 304 | int requests = 0; |
dbae4c73 | 305 | int ret = 0; |
1da177e4 LT |
306 | |
307 | nfs_list_remove_request(req); | |
308 | ||
275acaaf | 309 | offset = 0; |
c76069bd | 310 | nbytes = desc->pg_count; |
e9f7bee1 TM |
311 | do { |
312 | size_t len = min(nbytes,rsize); | |
313 | ||
8d5658c9 | 314 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
315 | if (!data) |
316 | goto out_bad; | |
275acaaf TM |
317 | data->pagevec[0] = page; |
318 | nfs_read_rpcsetup(req, data, len, offset); | |
319 | list_add(&data->list, res); | |
1da177e4 | 320 | requests++; |
e9f7bee1 | 321 | nbytes -= len; |
275acaaf | 322 | offset += len; |
e9f7bee1 | 323 | } while(nbytes != 0); |
1da177e4 | 324 | atomic_set(&req->wb_complete, requests); |
1da177e4 | 325 | ClearPageError(page); |
50828d7e | 326 | desc->pg_rpc_callops = &nfs_read_partial_ops; |
dbae4c73 | 327 | return ret; |
1da177e4 | 328 | out_bad: |
275acaaf TM |
329 | while (!list_empty(res)) { |
330 | data = list_entry(res->next, struct nfs_read_data, list); | |
6e4efd56 | 331 | list_del(&data->list); |
1da177e4 LT |
332 | nfs_readdata_free(data); |
333 | } | |
334 | SetPageError(page); | |
335 | nfs_readpage_release(req); | |
336 | return -ENOMEM; | |
337 | } | |
338 | ||
275acaaf | 339 | static int nfs_pagein_one(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 LT |
340 | { |
341 | struct nfs_page *req; | |
342 | struct page **pages; | |
343 | struct nfs_read_data *data; | |
c76069bd | 344 | struct list_head *head = &desc->pg_list; |
3b609184 | 345 | int ret = 0; |
1da177e4 | 346 | |
c76069bd FI |
347 | data = nfs_readdata_alloc(nfs_page_array_len(desc->pg_base, |
348 | desc->pg_count)); | |
bae724ef FI |
349 | if (!data) { |
350 | nfs_async_read_error(head); | |
3b609184 | 351 | ret = -ENOMEM; |
bae724ef FI |
352 | goto out; |
353 | } | |
1da177e4 | 354 | |
1da177e4 | 355 | pages = data->pagevec; |
1da177e4 LT |
356 | while (!list_empty(head)) { |
357 | req = nfs_list_entry(head->next); | |
358 | nfs_list_remove_request(req); | |
359 | nfs_list_add_request(req, &data->pages); | |
360 | ClearPageError(req->wb_page); | |
361 | *pages++ = req->wb_page; | |
1da177e4 LT |
362 | } |
363 | req = nfs_list_entry(data->pages.next); | |
364 | ||
6e4efd56 | 365 | nfs_read_rpcsetup(req, data, desc->pg_count, 0); |
275acaaf | 366 | list_add(&data->list, res); |
50828d7e | 367 | desc->pg_rpc_callops = &nfs_read_full_ops; |
bae724ef | 368 | out: |
dbae4c73 | 369 | return ret; |
1da177e4 LT |
370 | } |
371 | ||
493292dd TM |
372 | int nfs_generic_pagein(struct nfs_pageio_descriptor *desc, struct list_head *head) |
373 | { | |
374 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
375 | return nfs_pagein_multi(desc, head); | |
376 | return nfs_pagein_one(desc, head); | |
377 | } | |
378 | ||
379 | static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 380 | { |
275acaaf TM |
381 | LIST_HEAD(head); |
382 | int ret; | |
383 | ||
493292dd | 384 | ret = nfs_generic_pagein(desc, &head); |
50828d7e | 385 | if (ret == 0) |
493292dd | 386 | ret = nfs_do_multiple_reads(&head, desc->pg_rpc_callops); |
275acaaf | 387 | return ret; |
1751c363 | 388 | } |
1751c363 TM |
389 | |
390 | static const struct nfs_pageio_ops nfs_pageio_read_ops = { | |
391 | .pg_test = nfs_generic_pg_test, | |
392 | .pg_doio = nfs_generic_pg_readpages, | |
393 | }; | |
394 | ||
0b671301 TM |
395 | /* |
396 | * This is the callback from RPC telling us whether a reply was | |
397 | * received or some error occurred (timeout or socket shutdown). | |
398 | */ | |
399 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
400 | { | |
401 | int status; | |
402 | ||
3110ff80 | 403 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
404 | task->tk_status); |
405 | ||
406 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
407 | if (status != 0) | |
408 | return status; | |
409 | ||
410 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
411 | ||
412 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 413 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
414 | nfs_mark_for_revalidate(data->inode); |
415 | } | |
0b671301 TM |
416 | return 0; |
417 | } | |
418 | ||
fdd1e74c | 419 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
420 | { |
421 | struct nfs_readargs *argp = &data->args; | |
422 | struct nfs_readres *resp = &data->res; | |
423 | ||
424 | if (resp->eof || resp->count == argp->count) | |
d61e612a | 425 | return; |
0b671301 TM |
426 | |
427 | /* This is a short read! */ | |
428 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
429 | /* Has the server at least made some progress? */ | |
430 | if (resp->count == 0) | |
d61e612a | 431 | return; |
0b671301 TM |
432 | |
433 | /* Yes, so retry the read at the end of the data */ | |
cbdabc7f | 434 | data->mds_offset += resp->count; |
0b671301 TM |
435 | argp->offset += resp->count; |
436 | argp->pgbase += resp->count; | |
437 | argp->count -= resp->count; | |
d00c5d43 | 438 | rpc_restart_call_prepare(task); |
0b671301 TM |
439 | } |
440 | ||
1da177e4 LT |
441 | /* |
442 | * Handle a read reply that fills part of a page. | |
443 | */ | |
ec06c096 | 444 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 445 | { |
ec06c096 | 446 | struct nfs_read_data *data = calldata; |
1da177e4 | 447 | |
ec06c096 TM |
448 | if (nfs_readpage_result(task, data) != 0) |
449 | return; | |
fdd1e74c TM |
450 | if (task->tk_status < 0) |
451 | return; | |
0b671301 | 452 | |
fdd1e74c TM |
453 | nfs_readpage_truncate_uninitialised_page(data); |
454 | nfs_readpage_retry(task, data); | |
455 | } | |
456 | ||
457 | static void nfs_readpage_release_partial(void *calldata) | |
458 | { | |
459 | struct nfs_read_data *data = calldata; | |
460 | struct nfs_page *req = data->req; | |
461 | struct page *page = req->wb_page; | |
462 | int status = data->task.tk_status; | |
463 | ||
464 | if (status < 0) | |
0b671301 | 465 | SetPageError(page); |
fdd1e74c | 466 | |
1da177e4 LT |
467 | if (atomic_dec_and_test(&req->wb_complete)) { |
468 | if (!PageError(page)) | |
469 | SetPageUptodate(page); | |
470 | nfs_readpage_release(req); | |
471 | } | |
fdd1e74c | 472 | nfs_readdata_release(calldata); |
1da177e4 LT |
473 | } |
474 | ||
f11c88af AA |
475 | #if defined(CONFIG_NFS_V4_1) |
476 | void nfs_read_prepare(struct rpc_task *task, void *calldata) | |
477 | { | |
478 | struct nfs_read_data *data = calldata; | |
479 | ||
035168ab | 480 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
f11c88af AA |
481 | &data->args.seq_args, &data->res.seq_res, |
482 | 0, task)) | |
483 | return; | |
484 | rpc_call_start(task); | |
485 | } | |
486 | #endif /* CONFIG_NFS_V4_1 */ | |
487 | ||
ec06c096 | 488 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af AA |
489 | #if defined(CONFIG_NFS_V4_1) |
490 | .rpc_call_prepare = nfs_read_prepare, | |
491 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 492 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 493 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
494 | }; |
495 | ||
1de3fc12 TM |
496 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
497 | { | |
498 | unsigned int count = data->res.count; | |
499 | unsigned int base = data->args.pgbase; | |
500 | struct page **pages; | |
501 | ||
79558f36 TM |
502 | if (data->res.eof) |
503 | count = data->args.count; | |
1de3fc12 TM |
504 | if (unlikely(count == 0)) |
505 | return; | |
506 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
507 | base &= ~PAGE_CACHE_MASK; | |
508 | count += base; | |
509 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
510 | SetPageUptodate(*pages); | |
0b671301 TM |
511 | if (count == 0) |
512 | return; | |
513 | /* Was this a short read? */ | |
514 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
515 | SetPageUptodate(*pages); |
516 | } | |
517 | ||
1da177e4 LT |
518 | /* |
519 | * This is the callback from RPC telling us whether a reply was | |
520 | * received or some error occurred (timeout or socket shutdown). | |
521 | */ | |
ec06c096 | 522 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 523 | { |
ec06c096 | 524 | struct nfs_read_data *data = calldata; |
1da177e4 | 525 | |
0b671301 TM |
526 | if (nfs_readpage_result(task, data) != 0) |
527 | return; | |
fdd1e74c TM |
528 | if (task->tk_status < 0) |
529 | return; | |
1de3fc12 | 530 | /* |
0b671301 | 531 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 532 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
533 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
534 | * first. | |
1de3fc12 | 535 | */ |
fdd1e74c TM |
536 | nfs_readpage_truncate_uninitialised_page(data); |
537 | nfs_readpage_set_pages_uptodate(data); | |
538 | nfs_readpage_retry(task, data); | |
539 | } | |
540 | ||
541 | static void nfs_readpage_release_full(void *calldata) | |
542 | { | |
543 | struct nfs_read_data *data = calldata; | |
9b7eecdc | 544 | struct nfs_pageio_descriptor pgio; |
fdd1e74c | 545 | |
9b7eecdc PT |
546 | if (data->pnfs_error) { |
547 | nfs_pageio_init_read_mds(&pgio, data->inode); | |
548 | pgio.pg_recoalesce = 1; | |
549 | } | |
1da177e4 LT |
550 | while (!list_empty(&data->pages)) { |
551 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 552 | |
1de3fc12 | 553 | nfs_list_remove_request(req); |
9b7eecdc PT |
554 | if (!data->pnfs_error) |
555 | nfs_readpage_release(req); | |
556 | else | |
557 | nfs_pageio_add_request(&pgio, req); | |
1da177e4 | 558 | } |
9b7eecdc PT |
559 | if (data->pnfs_error) |
560 | nfs_pageio_complete(&pgio); | |
fdd1e74c | 561 | nfs_readdata_release(calldata); |
1da177e4 LT |
562 | } |
563 | ||
ec06c096 | 564 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af AA |
565 | #if defined(CONFIG_NFS_V4_1) |
566 | .rpc_call_prepare = nfs_read_prepare, | |
567 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 568 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 569 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
570 | }; |
571 | ||
1da177e4 LT |
572 | /* |
573 | * Read a page over NFS. | |
574 | * We read the page synchronously in the following case: | |
575 | * - The error flag is set for this page. This happens only when a | |
576 | * previous async read operation failed. | |
577 | */ | |
578 | int nfs_readpage(struct file *file, struct page *page) | |
579 | { | |
580 | struct nfs_open_context *ctx; | |
581 | struct inode *inode = page->mapping->host; | |
582 | int error; | |
583 | ||
584 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
585 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
586 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
587 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
588 | ||
1da177e4 LT |
589 | /* |
590 | * Try to flush any pending writes to the file.. | |
591 | * | |
592 | * NOTE! Because we own the page lock, there cannot | |
593 | * be any new pending writes generated at this point | |
594 | * for this page (other pages can be written to). | |
595 | */ | |
596 | error = nfs_wb_page(inode, page); | |
597 | if (error) | |
de05a0cc TM |
598 | goto out_unlock; |
599 | if (PageUptodate(page)) | |
600 | goto out_unlock; | |
1da177e4 | 601 | |
5f004cf2 TM |
602 | error = -ESTALE; |
603 | if (NFS_STALE(inode)) | |
de05a0cc | 604 | goto out_unlock; |
5f004cf2 | 605 | |
1da177e4 | 606 | if (file == NULL) { |
cf1308ff | 607 | error = -EBADF; |
d530838b | 608 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 609 | if (ctx == NULL) |
de05a0cc | 610 | goto out_unlock; |
1da177e4 | 611 | } else |
cd3758e3 | 612 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 613 | |
9a9fc1c0 DH |
614 | if (!IS_SYNC(inode)) { |
615 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
616 | if (error == 0) | |
617 | goto out; | |
618 | } | |
619 | ||
8e0969f0 TM |
620 | error = nfs_readpage_async(ctx, inode, page); |
621 | ||
9a9fc1c0 | 622 | out: |
1da177e4 LT |
623 | put_nfs_open_context(ctx); |
624 | return error; | |
de05a0cc | 625 | out_unlock: |
1da177e4 LT |
626 | unlock_page(page); |
627 | return error; | |
628 | } | |
629 | ||
630 | struct nfs_readdesc { | |
8b09bee3 | 631 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
632 | struct nfs_open_context *ctx; |
633 | }; | |
634 | ||
635 | static int | |
636 | readpage_async_filler(void *data, struct page *page) | |
637 | { | |
638 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
639 | struct inode *inode = page->mapping->host; | |
640 | struct nfs_page *new; | |
641 | unsigned int len; | |
de05a0cc TM |
642 | int error; |
643 | ||
49a70f27 | 644 | len = nfs_page_length(page); |
1da177e4 LT |
645 | if (len == 0) |
646 | return nfs_return_empty_page(page); | |
de05a0cc | 647 | |
1da177e4 | 648 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
649 | if (IS_ERR(new)) |
650 | goto out_error; | |
651 | ||
1da177e4 | 652 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 653 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
654 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
655 | error = desc->pgio->pg_error; | |
656 | goto out_unlock; | |
657 | } | |
1da177e4 | 658 | return 0; |
de05a0cc TM |
659 | out_error: |
660 | error = PTR_ERR(new); | |
661 | SetPageError(page); | |
662 | out_unlock: | |
663 | unlock_page(page); | |
664 | return error; | |
1da177e4 LT |
665 | } |
666 | ||
667 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
668 | struct list_head *pages, unsigned nr_pages) | |
669 | { | |
8b09bee3 | 670 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 671 | struct nfs_readdesc desc = { |
8b09bee3 | 672 | .pgio = &pgio, |
1da177e4 LT |
673 | }; |
674 | struct inode *inode = mapping->host; | |
8b09bee3 | 675 | unsigned long npages; |
5f004cf2 | 676 | int ret = -ESTALE; |
1da177e4 LT |
677 | |
678 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
679 | inode->i_sb->s_id, | |
680 | (long long)NFS_FILEID(inode), | |
681 | nr_pages); | |
91d5b470 | 682 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 683 | |
5f004cf2 TM |
684 | if (NFS_STALE(inode)) |
685 | goto out; | |
686 | ||
1da177e4 | 687 | if (filp == NULL) { |
d530838b | 688 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
689 | if (desc.ctx == NULL) |
690 | return -EBADF; | |
691 | } else | |
cd3758e3 | 692 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
693 | |
694 | /* attempt to read as many of the pages as possible from the cache | |
695 | * - this returns -ENOBUFS immediately if the cookie is negative | |
696 | */ | |
697 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
698 | pages, &nr_pages); | |
699 | if (ret == 0) | |
700 | goto read_complete; /* all pages were read */ | |
701 | ||
1751c363 | 702 | nfs_pageio_init_read(&pgio, inode); |
8b09bee3 | 703 | |
1da177e4 | 704 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
705 | |
706 | nfs_pageio_complete(&pgio); | |
707 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
708 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 709 | read_complete: |
1da177e4 | 710 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 711 | out: |
1da177e4 LT |
712 | return ret; |
713 | } | |
714 | ||
f7b422b1 | 715 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
716 | { |
717 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
718 | sizeof(struct nfs_read_data), | |
719 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 720 | NULL); |
1da177e4 LT |
721 | if (nfs_rdata_cachep == NULL) |
722 | return -ENOMEM; | |
723 | ||
93d2341c MD |
724 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
725 | nfs_rdata_cachep); | |
1da177e4 LT |
726 | if (nfs_rdata_mempool == NULL) |
727 | return -ENOMEM; | |
728 | ||
729 | return 0; | |
730 | } | |
731 | ||
266bee88 | 732 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
733 | { |
734 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 735 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 736 | } |