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 | ||
c76069bd FI |
33 | static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc); |
34 | static int nfs_pagein_one(struct nfs_pageio_descriptor *desc); | |
ec06c096 TM |
35 | static const struct rpc_call_ops nfs_read_partial_ops; |
36 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 | 37 | |
e18b890b | 38 | static struct kmem_cache *nfs_rdata_cachep; |
3feb2d49 | 39 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
40 | |
41 | #define MIN_POOL_READ (32) | |
42 | ||
8d5658c9 | 43 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 44 | { |
93870d76 | 45 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL); |
3feb2d49 TM |
46 | |
47 | if (p) { | |
48 | memset(p, 0, sizeof(*p)); | |
49 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 50 | p->npages = pagecount; |
0d0b5cb3 CL |
51 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
52 | p->pagevec = p->page_array; | |
3feb2d49 | 53 | else { |
93870d76 | 54 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); |
0d0b5cb3 | 55 | if (!p->pagevec) { |
3feb2d49 TM |
56 | mempool_free(p, nfs_rdata_mempool); |
57 | p = NULL; | |
58 | } | |
59 | } | |
60 | } | |
61 | return p; | |
62 | } | |
63 | ||
1ae88b2e | 64 | void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
65 | { |
66 | if (p && (p->pagevec != &p->page_array[0])) | |
67 | kfree(p->pagevec); | |
68 | mempool_free(p, nfs_rdata_mempool); | |
69 | } | |
70 | ||
1ae88b2e | 71 | static void nfs_readdata_release(struct nfs_read_data *rdata) |
1da177e4 | 72 | { |
bae724ef | 73 | put_lseg(rdata->lseg); |
383ba719 TM |
74 | put_nfs_open_context(rdata->args.context); |
75 | nfs_readdata_free(rdata); | |
1da177e4 LT |
76 | } |
77 | ||
1da177e4 LT |
78 | static |
79 | int nfs_return_empty_page(struct page *page) | |
80 | { | |
eebd2aa3 | 81 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
82 | SetPageUptodate(page); |
83 | unlock_page(page); | |
84 | return 0; | |
85 | } | |
86 | ||
1de3fc12 TM |
87 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
88 | { | |
89 | unsigned int remainder = data->args.count - data->res.count; | |
90 | unsigned int base = data->args.pgbase + data->res.count; | |
91 | unsigned int pglen; | |
92 | struct page **pages; | |
93 | ||
94 | if (data->res.eof == 0 || remainder == 0) | |
95 | return; | |
96 | /* | |
97 | * Note: "remainder" can never be negative, since we check for | |
98 | * this in the XDR code. | |
99 | */ | |
100 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
101 | base &= ~PAGE_CACHE_MASK; | |
102 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
103 | for (;;) { |
104 | if (remainder <= pglen) { | |
eebd2aa3 | 105 | zero_user(*pages, base, remainder); |
79558f36 TM |
106 | break; |
107 | } | |
eebd2aa3 | 108 | zero_user(*pages, base, pglen); |
79558f36 TM |
109 | pages++; |
110 | remainder -= pglen; | |
111 | pglen = PAGE_CACHE_SIZE; | |
112 | base = 0; | |
113 | } | |
1de3fc12 TM |
114 | } |
115 | ||
f42b293d DH |
116 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
117 | struct page *page) | |
1da177e4 | 118 | { |
1da177e4 LT |
119 | struct nfs_page *new; |
120 | unsigned int len; | |
c76069bd | 121 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 122 | |
49a70f27 | 123 | len = nfs_page_length(page); |
1da177e4 LT |
124 | if (len == 0) |
125 | return nfs_return_empty_page(page); | |
126 | new = nfs_create_request(ctx, inode, page, 0, len); | |
127 | if (IS_ERR(new)) { | |
128 | unlock_page(page); | |
129 | return PTR_ERR(new); | |
130 | } | |
131 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 132 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 133 | |
c76069bd FI |
134 | nfs_pageio_init(&pgio, inode, NULL, 0, 0); |
135 | nfs_list_add_request(new, &pgio.pg_list); | |
136 | pgio.pg_count = len; | |
137 | ||
bcb71bba | 138 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
c76069bd | 139 | nfs_pagein_multi(&pgio); |
bcb71bba | 140 | else |
c76069bd | 141 | nfs_pagein_one(&pgio); |
1da177e4 LT |
142 | return 0; |
143 | } | |
144 | ||
145 | static void nfs_readpage_release(struct nfs_page *req) | |
146 | { | |
7f8e05f6 DH |
147 | struct inode *d_inode = req->wb_context->path.dentry->d_inode; |
148 | ||
149 | if (PageUptodate(req->wb_page)) | |
150 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
151 | ||
1da177e4 LT |
152 | unlock_page(req->wb_page); |
153 | ||
1da177e4 | 154 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
88be9f99 TM |
155 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
156 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
157 | req->wb_bytes, |
158 | (long long)req_offset(req)); | |
10d2c46f | 159 | nfs_release_request(req); |
1da177e4 LT |
160 | } |
161 | ||
dc70d7b3 | 162 | int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt, |
64419a9b | 163 | const struct rpc_call_ops *call_ops) |
1da177e4 | 164 | { |
64419a9b | 165 | struct inode *inode = data->inode; |
84115e1c | 166 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
07737691 | 167 | struct rpc_task *task; |
bdc7f021 TM |
168 | struct rpc_message msg = { |
169 | .rpc_argp = &data->args, | |
170 | .rpc_resp = &data->res, | |
64419a9b | 171 | .rpc_cred = data->cred, |
bdc7f021 | 172 | }; |
84115e1c | 173 | struct rpc_task_setup task_setup_data = { |
07737691 | 174 | .task = &data->task, |
64419a9b | 175 | .rpc_client = clnt, |
bdc7f021 | 176 | .rpc_message = &msg, |
84115e1c TM |
177 | .callback_ops = call_ops, |
178 | .callback_data = data, | |
101070ca | 179 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
180 | .flags = RPC_TASK_ASYNC | swap_flags, |
181 | }; | |
1da177e4 | 182 | |
64419a9b AA |
183 | /* Set up the initial task struct. */ |
184 | NFS_PROTO(inode)->read_setup(data, &msg); | |
185 | ||
186 | dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ " | |
187 | "offset %llu)\n", | |
188 | data->task.tk_pid, | |
189 | inode->i_sb->s_id, | |
190 | (long long)NFS_FILEID(inode), | |
191 | data->args.count, | |
192 | (unsigned long long)data->args.offset); | |
193 | ||
194 | task = rpc_run_task(&task_setup_data); | |
195 | if (IS_ERR(task)) | |
196 | return PTR_ERR(task); | |
197 | rpc_put_task(task); | |
198 | return 0; | |
199 | } | |
dc70d7b3 | 200 | EXPORT_SYMBOL_GPL(nfs_initiate_read); |
64419a9b AA |
201 | |
202 | /* | |
203 | * Set up the NFS read request struct | |
204 | */ | |
205 | static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, | |
206 | const struct rpc_call_ops *call_ops, | |
207 | unsigned int count, unsigned int offset, | |
208 | struct pnfs_layout_segment *lseg) | |
209 | { | |
210 | struct inode *inode = req->wb_context->path.dentry->d_inode; | |
211 | ||
1da177e4 | 212 | data->req = req; |
84115e1c | 213 | data->inode = inode; |
64419a9b | 214 | data->cred = req->wb_context->cred; |
bae724ef | 215 | data->lseg = get_lseg(lseg); |
1da177e4 LT |
216 | |
217 | data->args.fh = NFS_FH(inode); | |
218 | data->args.offset = req_offset(req) + offset; | |
219 | data->args.pgbase = req->wb_pgbase + offset; | |
220 | data->args.pages = data->pagevec; | |
221 | data->args.count = count; | |
383ba719 | 222 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 223 | data->args.lock_context = req->wb_lock_context; |
1da177e4 LT |
224 | |
225 | data->res.fattr = &data->fattr; | |
226 | data->res.count = count; | |
227 | data->res.eof = 0; | |
0e574af1 | 228 | nfs_fattr_init(&data->fattr); |
1da177e4 | 229 | |
64419a9b AA |
230 | if (data->lseg && |
231 | (pnfs_try_to_read_data(data, call_ops) == PNFS_ATTEMPTED)) | |
232 | return 0; | |
bdc7f021 | 233 | |
64419a9b | 234 | return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops); |
1da177e4 LT |
235 | } |
236 | ||
237 | static void | |
238 | nfs_async_read_error(struct list_head *head) | |
239 | { | |
240 | struct nfs_page *req; | |
241 | ||
242 | while (!list_empty(head)) { | |
243 | req = nfs_list_entry(head->next); | |
244 | nfs_list_remove_request(req); | |
245 | SetPageError(req->wb_page); | |
246 | nfs_readpage_release(req); | |
247 | } | |
248 | } | |
249 | ||
1da177e4 LT |
250 | /* |
251 | * Generate multiple requests to fill a single page. | |
252 | * | |
253 | * We optimize to reduce the number of read operations on the wire. If we | |
254 | * detect that we're reading a page, or an area of a page, that is past the | |
255 | * end of file, we do not generate NFS read operations but just clear the | |
256 | * parts of the page that would have come back zero from the server anyway. | |
257 | * | |
258 | * We rely on the cached value of i_size to make this determination; another | |
259 | * client can fill pages on the server past our cached end-of-file, but we | |
260 | * won't see the new data until our attribute cache is updated. This is more | |
261 | * or less conventional NFS client behavior. | |
262 | */ | |
c76069bd | 263 | static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc) |
1da177e4 | 264 | { |
c76069bd | 265 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
266 | struct page *page = req->wb_page; |
267 | struct nfs_read_data *data; | |
c76069bd | 268 | size_t rsize = NFS_SERVER(desc->pg_inode)->rsize, nbytes; |
e9f7bee1 | 269 | unsigned int offset; |
1da177e4 | 270 | int requests = 0; |
dbae4c73 | 271 | int ret = 0; |
c76069bd | 272 | struct pnfs_layout_segment *lseg; |
1da177e4 LT |
273 | LIST_HEAD(list); |
274 | ||
275 | nfs_list_remove_request(req); | |
276 | ||
c76069bd | 277 | nbytes = desc->pg_count; |
e9f7bee1 TM |
278 | do { |
279 | size_t len = min(nbytes,rsize); | |
280 | ||
8d5658c9 | 281 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
282 | if (!data) |
283 | goto out_bad; | |
1da177e4 LT |
284 | list_add(&data->pages, &list); |
285 | requests++; | |
e9f7bee1 TM |
286 | nbytes -= len; |
287 | } while(nbytes != 0); | |
1da177e4 LT |
288 | atomic_set(&req->wb_complete, requests); |
289 | ||
c76069bd | 290 | BUG_ON(desc->pg_lseg != NULL); |
fb3296eb BH |
291 | lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, |
292 | req_offset(req), desc->pg_count, | |
293 | IOMODE_READ, GFP_KERNEL); | |
1da177e4 LT |
294 | ClearPageError(page); |
295 | offset = 0; | |
c76069bd | 296 | nbytes = desc->pg_count; |
1da177e4 | 297 | do { |
dbae4c73 TM |
298 | int ret2; |
299 | ||
1da177e4 LT |
300 | data = list_entry(list.next, struct nfs_read_data, pages); |
301 | list_del_init(&data->pages); | |
302 | ||
303 | data->pagevec[0] = page; | |
1da177e4 | 304 | |
bcb71bba TM |
305 | if (nbytes < rsize) |
306 | rsize = nbytes; | |
dbae4c73 | 307 | ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
bae724ef | 308 | rsize, offset, lseg); |
dbae4c73 TM |
309 | if (ret == 0) |
310 | ret = ret2; | |
bcb71bba TM |
311 | offset += rsize; |
312 | nbytes -= rsize; | |
1da177e4 | 313 | } while (nbytes != 0); |
bae724ef | 314 | put_lseg(lseg); |
36fe432d | 315 | desc->pg_lseg = NULL; |
1da177e4 | 316 | |
dbae4c73 | 317 | return ret; |
1da177e4 LT |
318 | |
319 | out_bad: | |
320 | while (!list_empty(&list)) { | |
321 | data = list_entry(list.next, struct nfs_read_data, pages); | |
322 | list_del(&data->pages); | |
323 | nfs_readdata_free(data); | |
324 | } | |
325 | SetPageError(page); | |
326 | nfs_readpage_release(req); | |
327 | return -ENOMEM; | |
328 | } | |
329 | ||
c76069bd | 330 | static int nfs_pagein_one(struct nfs_pageio_descriptor *desc) |
1da177e4 LT |
331 | { |
332 | struct nfs_page *req; | |
333 | struct page **pages; | |
334 | struct nfs_read_data *data; | |
c76069bd FI |
335 | struct list_head *head = &desc->pg_list; |
336 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
dbae4c73 | 337 | int ret = -ENOMEM; |
1da177e4 | 338 | |
c76069bd FI |
339 | data = nfs_readdata_alloc(nfs_page_array_len(desc->pg_base, |
340 | desc->pg_count)); | |
bae724ef FI |
341 | if (!data) { |
342 | nfs_async_read_error(head); | |
343 | goto out; | |
344 | } | |
1da177e4 | 345 | |
1da177e4 | 346 | pages = data->pagevec; |
1da177e4 LT |
347 | while (!list_empty(head)) { |
348 | req = nfs_list_entry(head->next); | |
349 | nfs_list_remove_request(req); | |
350 | nfs_list_add_request(req, &data->pages); | |
351 | ClearPageError(req->wb_page); | |
352 | *pages++ = req->wb_page; | |
1da177e4 LT |
353 | } |
354 | req = nfs_list_entry(data->pages.next); | |
bae724ef | 355 | if ((!lseg) && list_is_singular(&data->pages)) |
fb3296eb BH |
356 | lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, |
357 | req_offset(req), desc->pg_count, | |
358 | IOMODE_READ, GFP_KERNEL); | |
1da177e4 | 359 | |
c76069bd FI |
360 | ret = nfs_read_rpcsetup(req, data, &nfs_read_full_ops, desc->pg_count, |
361 | 0, lseg); | |
bae724ef FI |
362 | out: |
363 | put_lseg(lseg); | |
36fe432d | 364 | desc->pg_lseg = NULL; |
dbae4c73 | 365 | return ret; |
1da177e4 LT |
366 | } |
367 | ||
0b671301 TM |
368 | /* |
369 | * This is the callback from RPC telling us whether a reply was | |
370 | * received or some error occurred (timeout or socket shutdown). | |
371 | */ | |
372 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
373 | { | |
374 | int status; | |
375 | ||
3110ff80 | 376 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
377 | task->tk_status); |
378 | ||
379 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
380 | if (status != 0) | |
381 | return status; | |
382 | ||
383 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
384 | ||
385 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 386 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
387 | nfs_mark_for_revalidate(data->inode); |
388 | } | |
0b671301 TM |
389 | return 0; |
390 | } | |
391 | ||
fdd1e74c | 392 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
393 | { |
394 | struct nfs_readargs *argp = &data->args; | |
395 | struct nfs_readres *resp = &data->res; | |
396 | ||
397 | if (resp->eof || resp->count == argp->count) | |
d61e612a | 398 | return; |
0b671301 TM |
399 | |
400 | /* This is a short read! */ | |
401 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
402 | /* Has the server at least made some progress? */ | |
403 | if (resp->count == 0) | |
d61e612a | 404 | return; |
0b671301 TM |
405 | |
406 | /* Yes, so retry the read at the end of the data */ | |
cbdabc7f | 407 | data->mds_offset += resp->count; |
0b671301 TM |
408 | argp->offset += resp->count; |
409 | argp->pgbase += resp->count; | |
410 | argp->count -= resp->count; | |
0110ee15 | 411 | nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client); |
0b671301 TM |
412 | } |
413 | ||
1da177e4 LT |
414 | /* |
415 | * Handle a read reply that fills part of a page. | |
416 | */ | |
ec06c096 | 417 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 418 | { |
ec06c096 | 419 | struct nfs_read_data *data = calldata; |
1da177e4 | 420 | |
ec06c096 TM |
421 | if (nfs_readpage_result(task, data) != 0) |
422 | return; | |
fdd1e74c TM |
423 | if (task->tk_status < 0) |
424 | return; | |
0b671301 | 425 | |
fdd1e74c TM |
426 | nfs_readpage_truncate_uninitialised_page(data); |
427 | nfs_readpage_retry(task, data); | |
428 | } | |
429 | ||
430 | static void nfs_readpage_release_partial(void *calldata) | |
431 | { | |
432 | struct nfs_read_data *data = calldata; | |
433 | struct nfs_page *req = data->req; | |
434 | struct page *page = req->wb_page; | |
435 | int status = data->task.tk_status; | |
436 | ||
437 | if (status < 0) | |
0b671301 | 438 | SetPageError(page); |
fdd1e74c | 439 | |
1da177e4 LT |
440 | if (atomic_dec_and_test(&req->wb_complete)) { |
441 | if (!PageError(page)) | |
442 | SetPageUptodate(page); | |
443 | nfs_readpage_release(req); | |
444 | } | |
fdd1e74c | 445 | nfs_readdata_release(calldata); |
1da177e4 LT |
446 | } |
447 | ||
f11c88af AA |
448 | #if defined(CONFIG_NFS_V4_1) |
449 | void nfs_read_prepare(struct rpc_task *task, void *calldata) | |
450 | { | |
451 | struct nfs_read_data *data = calldata; | |
452 | ||
035168ab | 453 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
f11c88af AA |
454 | &data->args.seq_args, &data->res.seq_res, |
455 | 0, task)) | |
456 | return; | |
457 | rpc_call_start(task); | |
458 | } | |
459 | #endif /* CONFIG_NFS_V4_1 */ | |
460 | ||
ec06c096 | 461 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af AA |
462 | #if defined(CONFIG_NFS_V4_1) |
463 | .rpc_call_prepare = nfs_read_prepare, | |
464 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 465 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 466 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
467 | }; |
468 | ||
1de3fc12 TM |
469 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
470 | { | |
471 | unsigned int count = data->res.count; | |
472 | unsigned int base = data->args.pgbase; | |
473 | struct page **pages; | |
474 | ||
79558f36 TM |
475 | if (data->res.eof) |
476 | count = data->args.count; | |
1de3fc12 TM |
477 | if (unlikely(count == 0)) |
478 | return; | |
479 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
480 | base &= ~PAGE_CACHE_MASK; | |
481 | count += base; | |
482 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
483 | SetPageUptodate(*pages); | |
0b671301 TM |
484 | if (count == 0) |
485 | return; | |
486 | /* Was this a short read? */ | |
487 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
488 | SetPageUptodate(*pages); |
489 | } | |
490 | ||
1da177e4 LT |
491 | /* |
492 | * This is the callback from RPC telling us whether a reply was | |
493 | * received or some error occurred (timeout or socket shutdown). | |
494 | */ | |
ec06c096 | 495 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 496 | { |
ec06c096 | 497 | struct nfs_read_data *data = calldata; |
1da177e4 | 498 | |
0b671301 TM |
499 | if (nfs_readpage_result(task, data) != 0) |
500 | return; | |
fdd1e74c TM |
501 | if (task->tk_status < 0) |
502 | return; | |
1de3fc12 | 503 | /* |
0b671301 | 504 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 505 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
506 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
507 | * first. | |
1de3fc12 | 508 | */ |
fdd1e74c TM |
509 | nfs_readpage_truncate_uninitialised_page(data); |
510 | nfs_readpage_set_pages_uptodate(data); | |
511 | nfs_readpage_retry(task, data); | |
512 | } | |
513 | ||
514 | static void nfs_readpage_release_full(void *calldata) | |
515 | { | |
516 | struct nfs_read_data *data = calldata; | |
517 | ||
1da177e4 LT |
518 | while (!list_empty(&data->pages)) { |
519 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 520 | |
1de3fc12 | 521 | nfs_list_remove_request(req); |
1da177e4 LT |
522 | nfs_readpage_release(req); |
523 | } | |
fdd1e74c | 524 | nfs_readdata_release(calldata); |
1da177e4 LT |
525 | } |
526 | ||
ec06c096 | 527 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af AA |
528 | #if defined(CONFIG_NFS_V4_1) |
529 | .rpc_call_prepare = nfs_read_prepare, | |
530 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 531 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 532 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
533 | }; |
534 | ||
1da177e4 LT |
535 | /* |
536 | * Read a page over NFS. | |
537 | * We read the page synchronously in the following case: | |
538 | * - The error flag is set for this page. This happens only when a | |
539 | * previous async read operation failed. | |
540 | */ | |
541 | int nfs_readpage(struct file *file, struct page *page) | |
542 | { | |
543 | struct nfs_open_context *ctx; | |
544 | struct inode *inode = page->mapping->host; | |
545 | int error; | |
546 | ||
547 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
548 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
549 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
550 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
551 | ||
1da177e4 LT |
552 | /* |
553 | * Try to flush any pending writes to the file.. | |
554 | * | |
555 | * NOTE! Because we own the page lock, there cannot | |
556 | * be any new pending writes generated at this point | |
557 | * for this page (other pages can be written to). | |
558 | */ | |
559 | error = nfs_wb_page(inode, page); | |
560 | if (error) | |
de05a0cc TM |
561 | goto out_unlock; |
562 | if (PageUptodate(page)) | |
563 | goto out_unlock; | |
1da177e4 | 564 | |
5f004cf2 TM |
565 | error = -ESTALE; |
566 | if (NFS_STALE(inode)) | |
de05a0cc | 567 | goto out_unlock; |
5f004cf2 | 568 | |
1da177e4 | 569 | if (file == NULL) { |
cf1308ff | 570 | error = -EBADF; |
d530838b | 571 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 572 | if (ctx == NULL) |
de05a0cc | 573 | goto out_unlock; |
1da177e4 | 574 | } else |
cd3758e3 | 575 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 576 | |
9a9fc1c0 DH |
577 | if (!IS_SYNC(inode)) { |
578 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
579 | if (error == 0) | |
580 | goto out; | |
581 | } | |
582 | ||
8e0969f0 TM |
583 | error = nfs_readpage_async(ctx, inode, page); |
584 | ||
9a9fc1c0 | 585 | out: |
1da177e4 LT |
586 | put_nfs_open_context(ctx); |
587 | return error; | |
de05a0cc | 588 | out_unlock: |
1da177e4 LT |
589 | unlock_page(page); |
590 | return error; | |
591 | } | |
592 | ||
593 | struct nfs_readdesc { | |
8b09bee3 | 594 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
595 | struct nfs_open_context *ctx; |
596 | }; | |
597 | ||
598 | static int | |
599 | readpage_async_filler(void *data, struct page *page) | |
600 | { | |
601 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
602 | struct inode *inode = page->mapping->host; | |
603 | struct nfs_page *new; | |
604 | unsigned int len; | |
de05a0cc TM |
605 | int error; |
606 | ||
49a70f27 | 607 | len = nfs_page_length(page); |
1da177e4 LT |
608 | if (len == 0) |
609 | return nfs_return_empty_page(page); | |
de05a0cc | 610 | |
1da177e4 | 611 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
612 | if (IS_ERR(new)) |
613 | goto out_error; | |
614 | ||
1da177e4 | 615 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 616 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
617 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
618 | error = desc->pgio->pg_error; | |
619 | goto out_unlock; | |
620 | } | |
1da177e4 | 621 | return 0; |
de05a0cc TM |
622 | out_error: |
623 | error = PTR_ERR(new); | |
624 | SetPageError(page); | |
625 | out_unlock: | |
626 | unlock_page(page); | |
627 | return error; | |
1da177e4 LT |
628 | } |
629 | ||
630 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
631 | struct list_head *pages, unsigned nr_pages) | |
632 | { | |
8b09bee3 | 633 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 634 | struct nfs_readdesc desc = { |
8b09bee3 | 635 | .pgio = &pgio, |
1da177e4 LT |
636 | }; |
637 | struct inode *inode = mapping->host; | |
638 | struct nfs_server *server = NFS_SERVER(inode); | |
8b09bee3 TM |
639 | size_t rsize = server->rsize; |
640 | unsigned long npages; | |
5f004cf2 | 641 | int ret = -ESTALE; |
1da177e4 LT |
642 | |
643 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
644 | inode->i_sb->s_id, | |
645 | (long long)NFS_FILEID(inode), | |
646 | nr_pages); | |
91d5b470 | 647 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 648 | |
5f004cf2 TM |
649 | if (NFS_STALE(inode)) |
650 | goto out; | |
651 | ||
1da177e4 | 652 | if (filp == NULL) { |
d530838b | 653 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
654 | if (desc.ctx == NULL) |
655 | return -EBADF; | |
656 | } else | |
cd3758e3 | 657 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
658 | |
659 | /* attempt to read as many of the pages as possible from the cache | |
660 | * - this returns -ENOBUFS immediately if the cookie is negative | |
661 | */ | |
662 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
663 | pages, &nr_pages); | |
664 | if (ret == 0) | |
665 | goto read_complete; /* all pages were read */ | |
666 | ||
8b09bee3 TM |
667 | if (rsize < PAGE_CACHE_SIZE) |
668 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); | |
669 | else | |
670 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); | |
671 | ||
1da177e4 | 672 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
673 | |
674 | nfs_pageio_complete(&pgio); | |
675 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
676 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 677 | read_complete: |
1da177e4 | 678 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 679 | out: |
1da177e4 LT |
680 | return ret; |
681 | } | |
682 | ||
f7b422b1 | 683 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
684 | { |
685 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
686 | sizeof(struct nfs_read_data), | |
687 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 688 | NULL); |
1da177e4 LT |
689 | if (nfs_rdata_cachep == NULL) |
690 | return -ENOMEM; | |
691 | ||
93d2341c MD |
692 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
693 | nfs_rdata_cachep); | |
1da177e4 LT |
694 | if (nfs_rdata_mempool == NULL) |
695 | return -ENOMEM; | |
696 | ||
697 | return 0; | |
698 | } | |
699 | ||
266bee88 | 700 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
701 | { |
702 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 703 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 704 | } |