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; |
1da177e4 | 38 | |
8d5658c9 | 39 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 40 | { |
b6ee8cd2 | 41 | struct nfs_read_data *p; |
3feb2d49 | 42 | |
b6ee8cd2 | 43 | p = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL); |
3feb2d49 | 44 | if (p) { |
3feb2d49 | 45 | INIT_LIST_HEAD(&p->pages); |
e9f7bee1 | 46 | p->npages = pagecount; |
0d0b5cb3 CL |
47 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
48 | p->pagevec = p->page_array; | |
3feb2d49 | 49 | else { |
93870d76 | 50 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); |
0d0b5cb3 | 51 | if (!p->pagevec) { |
b6ee8cd2 | 52 | kmem_cache_free(nfs_rdata_cachep, p); |
3feb2d49 TM |
53 | p = NULL; |
54 | } | |
55 | } | |
56 | } | |
57 | return p; | |
58 | } | |
59 | ||
1ae88b2e | 60 | void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
61 | { |
62 | if (p && (p->pagevec != &p->page_array[0])) | |
63 | kfree(p->pagevec); | |
b6ee8cd2 | 64 | kmem_cache_free(nfs_rdata_cachep, p); |
3feb2d49 TM |
65 | } |
66 | ||
493292dd | 67 | void nfs_readdata_release(struct nfs_read_data *rdata) |
1da177e4 | 68 | { |
383ba719 TM |
69 | put_nfs_open_context(rdata->args.context); |
70 | nfs_readdata_free(rdata); | |
1da177e4 LT |
71 | } |
72 | ||
1da177e4 LT |
73 | static |
74 | int nfs_return_empty_page(struct page *page) | |
75 | { | |
eebd2aa3 | 76 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
77 | SetPageUptodate(page); |
78 | unlock_page(page); | |
79 | return 0; | |
80 | } | |
81 | ||
1de3fc12 TM |
82 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
83 | { | |
84 | unsigned int remainder = data->args.count - data->res.count; | |
85 | unsigned int base = data->args.pgbase + data->res.count; | |
86 | unsigned int pglen; | |
87 | struct page **pages; | |
88 | ||
89 | if (data->res.eof == 0 || remainder == 0) | |
90 | return; | |
91 | /* | |
92 | * Note: "remainder" can never be negative, since we check for | |
93 | * this in the XDR code. | |
94 | */ | |
95 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
96 | base &= ~PAGE_CACHE_MASK; | |
97 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
98 | for (;;) { |
99 | if (remainder <= pglen) { | |
eebd2aa3 | 100 | zero_user(*pages, base, remainder); |
79558f36 TM |
101 | break; |
102 | } | |
eebd2aa3 | 103 | zero_user(*pages, base, pglen); |
79558f36 TM |
104 | pages++; |
105 | remainder -= pglen; | |
106 | pglen = PAGE_CACHE_SIZE; | |
107 | base = 0; | |
108 | } | |
1de3fc12 TM |
109 | } |
110 | ||
62e4a769 | 111 | void nfs_pageio_init_read_mds(struct nfs_pageio_descriptor *pgio, |
1751c363 TM |
112 | struct inode *inode) |
113 | { | |
114 | nfs_pageio_init(pgio, inode, &nfs_pageio_read_ops, | |
115 | NFS_SERVER(inode)->rsize, 0); | |
116 | } | |
117 | ||
493292dd TM |
118 | void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) |
119 | { | |
120 | pgio->pg_ops = &nfs_pageio_read_ops; | |
121 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize; | |
122 | } | |
1f945357 | 123 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); |
493292dd | 124 | |
1751c363 TM |
125 | static void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, |
126 | struct inode *inode) | |
127 | { | |
128 | if (!pnfs_pageio_init_read(pgio, inode)) | |
129 | nfs_pageio_init_read_mds(pgio, inode); | |
130 | } | |
131 | ||
f42b293d DH |
132 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
133 | struct page *page) | |
1da177e4 | 134 | { |
1da177e4 LT |
135 | struct nfs_page *new; |
136 | unsigned int len; | |
c76069bd | 137 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 138 | |
49a70f27 | 139 | len = nfs_page_length(page); |
1da177e4 LT |
140 | if (len == 0) |
141 | return nfs_return_empty_page(page); | |
142 | new = nfs_create_request(ctx, inode, page, 0, len); | |
143 | if (IS_ERR(new)) { | |
144 | unlock_page(page); | |
145 | return PTR_ERR(new); | |
146 | } | |
147 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 148 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 149 | |
1751c363 | 150 | nfs_pageio_init_read(&pgio, inode); |
d8007d4d | 151 | nfs_pageio_add_request(&pgio, new); |
1751c363 | 152 | nfs_pageio_complete(&pgio); |
1da177e4 LT |
153 | return 0; |
154 | } | |
155 | ||
156 | static void nfs_readpage_release(struct nfs_page *req) | |
157 | { | |
3d4ff43d | 158 | struct inode *d_inode = req->wb_context->dentry->d_inode; |
7f8e05f6 DH |
159 | |
160 | if (PageUptodate(req->wb_page)) | |
161 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
162 | ||
1da177e4 LT |
163 | unlock_page(req->wb_page); |
164 | ||
1da177e4 | 165 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
3d4ff43d AV |
166 | req->wb_context->dentry->d_inode->i_sb->s_id, |
167 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
168 | req->wb_bytes, |
169 | (long long)req_offset(req)); | |
10d2c46f | 170 | nfs_release_request(req); |
1da177e4 LT |
171 | } |
172 | ||
dc70d7b3 | 173 | int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt, |
64419a9b | 174 | const struct rpc_call_ops *call_ops) |
1da177e4 | 175 | { |
64419a9b | 176 | struct inode *inode = data->inode; |
84115e1c | 177 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
07737691 | 178 | struct rpc_task *task; |
bdc7f021 TM |
179 | struct rpc_message msg = { |
180 | .rpc_argp = &data->args, | |
181 | .rpc_resp = &data->res, | |
64419a9b | 182 | .rpc_cred = data->cred, |
bdc7f021 | 183 | }; |
84115e1c | 184 | struct rpc_task_setup task_setup_data = { |
07737691 | 185 | .task = &data->task, |
64419a9b | 186 | .rpc_client = clnt, |
bdc7f021 | 187 | .rpc_message = &msg, |
84115e1c TM |
188 | .callback_ops = call_ops, |
189 | .callback_data = data, | |
101070ca | 190 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
191 | .flags = RPC_TASK_ASYNC | swap_flags, |
192 | }; | |
1da177e4 | 193 | |
64419a9b AA |
194 | /* Set up the initial task struct. */ |
195 | NFS_PROTO(inode)->read_setup(data, &msg); | |
196 | ||
197 | dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ " | |
198 | "offset %llu)\n", | |
199 | data->task.tk_pid, | |
200 | inode->i_sb->s_id, | |
201 | (long long)NFS_FILEID(inode), | |
202 | data->args.count, | |
203 | (unsigned long long)data->args.offset); | |
204 | ||
205 | task = rpc_run_task(&task_setup_data); | |
206 | if (IS_ERR(task)) | |
207 | return PTR_ERR(task); | |
208 | rpc_put_task(task); | |
209 | return 0; | |
210 | } | |
dc70d7b3 | 211 | EXPORT_SYMBOL_GPL(nfs_initiate_read); |
64419a9b AA |
212 | |
213 | /* | |
214 | * Set up the NFS read request struct | |
215 | */ | |
6e4efd56 TM |
216 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
217 | unsigned int count, unsigned int offset) | |
64419a9b | 218 | { |
3d4ff43d | 219 | struct inode *inode = req->wb_context->dentry->d_inode; |
64419a9b | 220 | |
1da177e4 | 221 | data->req = req; |
84115e1c | 222 | data->inode = inode; |
64419a9b | 223 | data->cred = req->wb_context->cred; |
1da177e4 LT |
224 | |
225 | data->args.fh = NFS_FH(inode); | |
226 | data->args.offset = req_offset(req) + offset; | |
227 | data->args.pgbase = req->wb_pgbase + offset; | |
228 | data->args.pages = data->pagevec; | |
229 | data->args.count = count; | |
383ba719 | 230 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 231 | data->args.lock_context = req->wb_lock_context; |
1da177e4 LT |
232 | |
233 | data->res.fattr = &data->fattr; | |
234 | data->res.count = count; | |
235 | data->res.eof = 0; | |
0e574af1 | 236 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 237 | } |
1da177e4 | 238 | |
6e4efd56 | 239 | static int nfs_do_read(struct nfs_read_data *data, |
493292dd | 240 | const struct rpc_call_ops *call_ops) |
6e4efd56 | 241 | { |
5f00bcb3 | 242 | struct inode *inode = data->args.context->dentry->d_inode; |
6e4efd56 | 243 | |
64419a9b | 244 | return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops); |
1da177e4 LT |
245 | } |
246 | ||
275acaaf TM |
247 | static int |
248 | nfs_do_multiple_reads(struct list_head *head, | |
493292dd | 249 | const struct rpc_call_ops *call_ops) |
275acaaf TM |
250 | { |
251 | struct nfs_read_data *data; | |
252 | int ret = 0; | |
253 | ||
254 | while (!list_empty(head)) { | |
255 | int ret2; | |
256 | ||
257 | data = list_entry(head->next, struct nfs_read_data, list); | |
258 | list_del_init(&data->list); | |
259 | ||
493292dd | 260 | ret2 = nfs_do_read(data, call_ops); |
275acaaf TM |
261 | if (ret == 0) |
262 | ret = ret2; | |
263 | } | |
264 | return ret; | |
265 | } | |
266 | ||
1da177e4 LT |
267 | static void |
268 | nfs_async_read_error(struct list_head *head) | |
269 | { | |
270 | struct nfs_page *req; | |
271 | ||
272 | while (!list_empty(head)) { | |
273 | req = nfs_list_entry(head->next); | |
274 | nfs_list_remove_request(req); | |
1da177e4 LT |
275 | nfs_readpage_release(req); |
276 | } | |
277 | } | |
278 | ||
1da177e4 LT |
279 | /* |
280 | * Generate multiple requests to fill a single page. | |
281 | * | |
282 | * We optimize to reduce the number of read operations on the wire. If we | |
283 | * detect that we're reading a page, or an area of a page, that is past the | |
284 | * end of file, we do not generate NFS read operations but just clear the | |
285 | * parts of the page that would have come back zero from the server anyway. | |
286 | * | |
287 | * We rely on the cached value of i_size to make this determination; another | |
288 | * client can fill pages on the server past our cached end-of-file, but we | |
289 | * won't see the new data until our attribute cache is updated. This is more | |
290 | * or less conventional NFS client behavior. | |
291 | */ | |
275acaaf | 292 | static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 | 293 | { |
c76069bd | 294 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
295 | struct page *page = req->wb_page; |
296 | struct nfs_read_data *data; | |
d097971d | 297 | size_t rsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 298 | unsigned int offset; |
1da177e4 | 299 | int requests = 0; |
dbae4c73 | 300 | int ret = 0; |
1da177e4 LT |
301 | |
302 | nfs_list_remove_request(req); | |
303 | ||
275acaaf | 304 | offset = 0; |
c76069bd | 305 | nbytes = desc->pg_count; |
e9f7bee1 TM |
306 | do { |
307 | size_t len = min(nbytes,rsize); | |
308 | ||
8d5658c9 | 309 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
310 | if (!data) |
311 | goto out_bad; | |
275acaaf TM |
312 | data->pagevec[0] = page; |
313 | nfs_read_rpcsetup(req, data, len, offset); | |
314 | list_add(&data->list, res); | |
1da177e4 | 315 | requests++; |
e9f7bee1 | 316 | nbytes -= len; |
275acaaf | 317 | offset += len; |
e9f7bee1 | 318 | } while(nbytes != 0); |
1da177e4 | 319 | atomic_set(&req->wb_complete, requests); |
50828d7e | 320 | desc->pg_rpc_callops = &nfs_read_partial_ops; |
dbae4c73 | 321 | return ret; |
1da177e4 | 322 | out_bad: |
275acaaf TM |
323 | while (!list_empty(res)) { |
324 | data = list_entry(res->next, struct nfs_read_data, list); | |
6e4efd56 | 325 | list_del(&data->list); |
1da177e4 LT |
326 | nfs_readdata_free(data); |
327 | } | |
1da177e4 LT |
328 | nfs_readpage_release(req); |
329 | return -ENOMEM; | |
330 | } | |
331 | ||
275acaaf | 332 | static int nfs_pagein_one(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 LT |
333 | { |
334 | struct nfs_page *req; | |
335 | struct page **pages; | |
336 | struct nfs_read_data *data; | |
c76069bd | 337 | struct list_head *head = &desc->pg_list; |
3b609184 | 338 | int ret = 0; |
1da177e4 | 339 | |
c76069bd FI |
340 | data = nfs_readdata_alloc(nfs_page_array_len(desc->pg_base, |
341 | desc->pg_count)); | |
bae724ef FI |
342 | if (!data) { |
343 | nfs_async_read_error(head); | |
3b609184 | 344 | ret = -ENOMEM; |
bae724ef FI |
345 | goto out; |
346 | } | |
1da177e4 | 347 | |
1da177e4 | 348 | pages = data->pagevec; |
1da177e4 LT |
349 | while (!list_empty(head)) { |
350 | req = nfs_list_entry(head->next); | |
351 | nfs_list_remove_request(req); | |
352 | nfs_list_add_request(req, &data->pages); | |
1da177e4 | 353 | *pages++ = req->wb_page; |
1da177e4 LT |
354 | } |
355 | req = nfs_list_entry(data->pages.next); | |
356 | ||
6e4efd56 | 357 | nfs_read_rpcsetup(req, data, desc->pg_count, 0); |
275acaaf | 358 | list_add(&data->list, res); |
50828d7e | 359 | desc->pg_rpc_callops = &nfs_read_full_ops; |
bae724ef | 360 | out: |
dbae4c73 | 361 | return ret; |
1da177e4 LT |
362 | } |
363 | ||
493292dd TM |
364 | int nfs_generic_pagein(struct nfs_pageio_descriptor *desc, struct list_head *head) |
365 | { | |
366 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
367 | return nfs_pagein_multi(desc, head); | |
368 | return nfs_pagein_one(desc, head); | |
369 | } | |
370 | ||
371 | static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 372 | { |
275acaaf TM |
373 | LIST_HEAD(head); |
374 | int ret; | |
375 | ||
493292dd | 376 | ret = nfs_generic_pagein(desc, &head); |
50828d7e | 377 | if (ret == 0) |
493292dd | 378 | ret = nfs_do_multiple_reads(&head, desc->pg_rpc_callops); |
275acaaf | 379 | return ret; |
1751c363 | 380 | } |
1751c363 TM |
381 | |
382 | static const struct nfs_pageio_ops nfs_pageio_read_ops = { | |
383 | .pg_test = nfs_generic_pg_test, | |
384 | .pg_doio = nfs_generic_pg_readpages, | |
385 | }; | |
386 | ||
0b671301 TM |
387 | /* |
388 | * This is the callback from RPC telling us whether a reply was | |
389 | * received or some error occurred (timeout or socket shutdown). | |
390 | */ | |
391 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
392 | { | |
393 | int status; | |
394 | ||
3110ff80 | 395 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
396 | task->tk_status); |
397 | ||
398 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
399 | if (status != 0) | |
400 | return status; | |
401 | ||
402 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
403 | ||
404 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 405 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
406 | nfs_mark_for_revalidate(data->inode); |
407 | } | |
0b671301 TM |
408 | return 0; |
409 | } | |
410 | ||
fdd1e74c | 411 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
412 | { |
413 | struct nfs_readargs *argp = &data->args; | |
414 | struct nfs_readres *resp = &data->res; | |
415 | ||
416 | if (resp->eof || resp->count == argp->count) | |
d61e612a | 417 | return; |
0b671301 TM |
418 | |
419 | /* This is a short read! */ | |
420 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
421 | /* Has the server at least made some progress? */ | |
422 | if (resp->count == 0) | |
d61e612a | 423 | return; |
0b671301 TM |
424 | |
425 | /* Yes, so retry the read at the end of the data */ | |
cbdabc7f | 426 | data->mds_offset += resp->count; |
0b671301 TM |
427 | argp->offset += resp->count; |
428 | argp->pgbase += resp->count; | |
429 | argp->count -= resp->count; | |
d00c5d43 | 430 | rpc_restart_call_prepare(task); |
0b671301 TM |
431 | } |
432 | ||
1da177e4 LT |
433 | /* |
434 | * Handle a read reply that fills part of a page. | |
435 | */ | |
ec06c096 | 436 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 437 | { |
ec06c096 | 438 | struct nfs_read_data *data = calldata; |
1da177e4 | 439 | |
ec06c096 TM |
440 | if (nfs_readpage_result(task, data) != 0) |
441 | return; | |
fdd1e74c TM |
442 | if (task->tk_status < 0) |
443 | return; | |
0b671301 | 444 | |
fdd1e74c TM |
445 | nfs_readpage_truncate_uninitialised_page(data); |
446 | nfs_readpage_retry(task, data); | |
447 | } | |
448 | ||
449 | static void nfs_readpage_release_partial(void *calldata) | |
450 | { | |
451 | struct nfs_read_data *data = calldata; | |
452 | struct nfs_page *req = data->req; | |
453 | struct page *page = req->wb_page; | |
454 | int status = data->task.tk_status; | |
455 | ||
456 | if (status < 0) | |
fba73005 | 457 | set_bit(PG_PARTIAL_READ_FAILED, &req->wb_flags); |
fdd1e74c | 458 | |
1da177e4 | 459 | if (atomic_dec_and_test(&req->wb_complete)) { |
fba73005 | 460 | if (!test_bit(PG_PARTIAL_READ_FAILED, &req->wb_flags)) |
1da177e4 LT |
461 | SetPageUptodate(page); |
462 | nfs_readpage_release(req); | |
463 | } | |
fdd1e74c | 464 | nfs_readdata_release(calldata); |
1da177e4 LT |
465 | } |
466 | ||
f11c88af AA |
467 | void nfs_read_prepare(struct rpc_task *task, void *calldata) |
468 | { | |
469 | struct nfs_read_data *data = calldata; | |
ea7c3303 | 470 | NFS_PROTO(data->inode)->read_rpc_prepare(task, data); |
f11c88af | 471 | } |
f11c88af | 472 | |
ec06c096 | 473 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af | 474 | .rpc_call_prepare = nfs_read_prepare, |
ec06c096 | 475 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 476 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
477 | }; |
478 | ||
1de3fc12 TM |
479 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
480 | { | |
481 | unsigned int count = data->res.count; | |
482 | unsigned int base = data->args.pgbase; | |
483 | struct page **pages; | |
484 | ||
79558f36 TM |
485 | if (data->res.eof) |
486 | count = data->args.count; | |
1de3fc12 TM |
487 | if (unlikely(count == 0)) |
488 | return; | |
489 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
490 | base &= ~PAGE_CACHE_MASK; | |
491 | count += base; | |
492 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
493 | SetPageUptodate(*pages); | |
0b671301 TM |
494 | if (count == 0) |
495 | return; | |
496 | /* Was this a short read? */ | |
497 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
498 | SetPageUptodate(*pages); |
499 | } | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * This is the callback from RPC telling us whether a reply was | |
503 | * received or some error occurred (timeout or socket shutdown). | |
504 | */ | |
ec06c096 | 505 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 506 | { |
ec06c096 | 507 | struct nfs_read_data *data = calldata; |
1da177e4 | 508 | |
0b671301 TM |
509 | if (nfs_readpage_result(task, data) != 0) |
510 | return; | |
fdd1e74c TM |
511 | if (task->tk_status < 0) |
512 | return; | |
1de3fc12 | 513 | /* |
0b671301 | 514 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 515 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
516 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
517 | * first. | |
1de3fc12 | 518 | */ |
fdd1e74c TM |
519 | nfs_readpage_truncate_uninitialised_page(data); |
520 | nfs_readpage_set_pages_uptodate(data); | |
521 | nfs_readpage_retry(task, data); | |
522 | } | |
523 | ||
524 | static void nfs_readpage_release_full(void *calldata) | |
525 | { | |
526 | struct nfs_read_data *data = calldata; | |
527 | ||
1da177e4 LT |
528 | while (!list_empty(&data->pages)) { |
529 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 530 | |
1de3fc12 | 531 | nfs_list_remove_request(req); |
62e4a769 | 532 | nfs_readpage_release(req); |
1da177e4 | 533 | } |
fdd1e74c | 534 | nfs_readdata_release(calldata); |
1da177e4 LT |
535 | } |
536 | ||
ec06c096 | 537 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af | 538 | .rpc_call_prepare = nfs_read_prepare, |
ec06c096 | 539 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 540 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
541 | }; |
542 | ||
1da177e4 LT |
543 | /* |
544 | * Read a page over NFS. | |
545 | * We read the page synchronously in the following case: | |
546 | * - The error flag is set for this page. This happens only when a | |
547 | * previous async read operation failed. | |
548 | */ | |
549 | int nfs_readpage(struct file *file, struct page *page) | |
550 | { | |
551 | struct nfs_open_context *ctx; | |
552 | struct inode *inode = page->mapping->host; | |
553 | int error; | |
554 | ||
555 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
556 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
557 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
558 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
559 | ||
1da177e4 LT |
560 | /* |
561 | * Try to flush any pending writes to the file.. | |
562 | * | |
563 | * NOTE! Because we own the page lock, there cannot | |
564 | * be any new pending writes generated at this point | |
565 | * for this page (other pages can be written to). | |
566 | */ | |
567 | error = nfs_wb_page(inode, page); | |
568 | if (error) | |
de05a0cc TM |
569 | goto out_unlock; |
570 | if (PageUptodate(page)) | |
571 | goto out_unlock; | |
1da177e4 | 572 | |
5f004cf2 TM |
573 | error = -ESTALE; |
574 | if (NFS_STALE(inode)) | |
de05a0cc | 575 | goto out_unlock; |
5f004cf2 | 576 | |
1da177e4 | 577 | if (file == NULL) { |
cf1308ff | 578 | error = -EBADF; |
d530838b | 579 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 580 | if (ctx == NULL) |
de05a0cc | 581 | goto out_unlock; |
1da177e4 | 582 | } else |
cd3758e3 | 583 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 584 | |
9a9fc1c0 DH |
585 | if (!IS_SYNC(inode)) { |
586 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
587 | if (error == 0) | |
588 | goto out; | |
589 | } | |
590 | ||
8e0969f0 TM |
591 | error = nfs_readpage_async(ctx, inode, page); |
592 | ||
9a9fc1c0 | 593 | out: |
1da177e4 LT |
594 | put_nfs_open_context(ctx); |
595 | return error; | |
de05a0cc | 596 | out_unlock: |
1da177e4 LT |
597 | unlock_page(page); |
598 | return error; | |
599 | } | |
600 | ||
601 | struct nfs_readdesc { | |
8b09bee3 | 602 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
603 | struct nfs_open_context *ctx; |
604 | }; | |
605 | ||
606 | static int | |
607 | readpage_async_filler(void *data, struct page *page) | |
608 | { | |
609 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
610 | struct inode *inode = page->mapping->host; | |
611 | struct nfs_page *new; | |
612 | unsigned int len; | |
de05a0cc TM |
613 | int error; |
614 | ||
49a70f27 | 615 | len = nfs_page_length(page); |
1da177e4 LT |
616 | if (len == 0) |
617 | return nfs_return_empty_page(page); | |
de05a0cc | 618 | |
1da177e4 | 619 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
620 | if (IS_ERR(new)) |
621 | goto out_error; | |
622 | ||
1da177e4 | 623 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 624 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
625 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
626 | error = desc->pgio->pg_error; | |
627 | goto out_unlock; | |
628 | } | |
1da177e4 | 629 | return 0; |
de05a0cc TM |
630 | out_error: |
631 | error = PTR_ERR(new); | |
de05a0cc TM |
632 | out_unlock: |
633 | unlock_page(page); | |
634 | return error; | |
1da177e4 LT |
635 | } |
636 | ||
637 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
638 | struct list_head *pages, unsigned nr_pages) | |
639 | { | |
8b09bee3 | 640 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 641 | struct nfs_readdesc desc = { |
8b09bee3 | 642 | .pgio = &pgio, |
1da177e4 LT |
643 | }; |
644 | struct inode *inode = mapping->host; | |
8b09bee3 | 645 | unsigned long npages; |
5f004cf2 | 646 | int ret = -ESTALE; |
1da177e4 LT |
647 | |
648 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
649 | inode->i_sb->s_id, | |
650 | (long long)NFS_FILEID(inode), | |
651 | nr_pages); | |
91d5b470 | 652 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 653 | |
5f004cf2 TM |
654 | if (NFS_STALE(inode)) |
655 | goto out; | |
656 | ||
1da177e4 | 657 | if (filp == NULL) { |
d530838b | 658 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
659 | if (desc.ctx == NULL) |
660 | return -EBADF; | |
661 | } else | |
cd3758e3 | 662 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
663 | |
664 | /* attempt to read as many of the pages as possible from the cache | |
665 | * - this returns -ENOBUFS immediately if the cookie is negative | |
666 | */ | |
667 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
668 | pages, &nr_pages); | |
669 | if (ret == 0) | |
670 | goto read_complete; /* all pages were read */ | |
671 | ||
1751c363 | 672 | nfs_pageio_init_read(&pgio, inode); |
8b09bee3 | 673 | |
1da177e4 | 674 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
675 | |
676 | nfs_pageio_complete(&pgio); | |
677 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
678 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 679 | read_complete: |
1da177e4 | 680 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 681 | out: |
1da177e4 LT |
682 | return ret; |
683 | } | |
684 | ||
f7b422b1 | 685 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
686 | { |
687 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
688 | sizeof(struct nfs_read_data), | |
689 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 690 | NULL); |
1da177e4 LT |
691 | if (nfs_rdata_cachep == NULL) |
692 | return -ENOMEM; | |
693 | ||
1da177e4 LT |
694 | return 0; |
695 | } | |
696 | ||
266bee88 | 697 | void nfs_destroy_readpagecache(void) |
1da177e4 | 698 | { |
1a1d92c1 | 699 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 700 | } |