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> | |
1da177e4 LT |
21 | |
22 | #include <asm/system.h> | |
23 | ||
f11c88af | 24 | #include "nfs4_fs.h" |
49a70f27 | 25 | #include "internal.h" |
91d5b470 | 26 | #include "iostat.h" |
9a9fc1c0 | 27 | #include "fscache.h" |
91d5b470 | 28 | |
1da177e4 LT |
29 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
30 | ||
8d5658c9 TM |
31 | static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int); |
32 | static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int); | |
ec06c096 TM |
33 | static const struct rpc_call_ops nfs_read_partial_ops; |
34 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 | 35 | |
e18b890b | 36 | static struct kmem_cache *nfs_rdata_cachep; |
3feb2d49 | 37 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
38 | |
39 | #define MIN_POOL_READ (32) | |
40 | ||
8d5658c9 | 41 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 42 | { |
e6b4f8da | 43 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS); |
3feb2d49 TM |
44 | |
45 | if (p) { | |
46 | memset(p, 0, sizeof(*p)); | |
47 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 48 | p->npages = pagecount; |
5f7dbd5c | 49 | p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE; |
0d0b5cb3 CL |
50 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
51 | p->pagevec = p->page_array; | |
3feb2d49 | 52 | else { |
0d0b5cb3 CL |
53 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
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 | ||
1ae88b2e | 70 | static void nfs_readdata_release(struct nfs_read_data *rdata) |
1da177e4 | 71 | { |
383ba719 TM |
72 | put_nfs_open_context(rdata->args.context); |
73 | nfs_readdata_free(rdata); | |
1da177e4 LT |
74 | } |
75 | ||
1da177e4 LT |
76 | static |
77 | int nfs_return_empty_page(struct page *page) | |
78 | { | |
eebd2aa3 | 79 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
80 | SetPageUptodate(page); |
81 | unlock_page(page); | |
82 | return 0; | |
83 | } | |
84 | ||
1de3fc12 TM |
85 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
86 | { | |
87 | unsigned int remainder = data->args.count - data->res.count; | |
88 | unsigned int base = data->args.pgbase + data->res.count; | |
89 | unsigned int pglen; | |
90 | struct page **pages; | |
91 | ||
92 | if (data->res.eof == 0 || remainder == 0) | |
93 | return; | |
94 | /* | |
95 | * Note: "remainder" can never be negative, since we check for | |
96 | * this in the XDR code. | |
97 | */ | |
98 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
99 | base &= ~PAGE_CACHE_MASK; | |
100 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
101 | for (;;) { |
102 | if (remainder <= pglen) { | |
eebd2aa3 | 103 | zero_user(*pages, base, remainder); |
79558f36 TM |
104 | break; |
105 | } | |
eebd2aa3 | 106 | zero_user(*pages, base, pglen); |
79558f36 TM |
107 | pages++; |
108 | remainder -= pglen; | |
109 | pglen = PAGE_CACHE_SIZE; | |
110 | base = 0; | |
111 | } | |
1de3fc12 TM |
112 | } |
113 | ||
f42b293d DH |
114 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
115 | struct page *page) | |
1da177e4 LT |
116 | { |
117 | LIST_HEAD(one_request); | |
118 | struct nfs_page *new; | |
119 | unsigned int len; | |
120 | ||
49a70f27 | 121 | len = nfs_page_length(page); |
1da177e4 LT |
122 | if (len == 0) |
123 | return nfs_return_empty_page(page); | |
124 | new = nfs_create_request(ctx, inode, page, 0, len); | |
125 | if (IS_ERR(new)) { | |
126 | unlock_page(page); | |
127 | return PTR_ERR(new); | |
128 | } | |
129 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 130 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 131 | |
1da177e4 | 132 | nfs_list_add_request(new, &one_request); |
bcb71bba | 133 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
8d5658c9 | 134 | nfs_pagein_multi(inode, &one_request, 1, len, 0); |
bcb71bba | 135 | else |
8d5658c9 | 136 | nfs_pagein_one(inode, &one_request, 1, len, 0); |
1da177e4 LT |
137 | return 0; |
138 | } | |
139 | ||
140 | static void nfs_readpage_release(struct nfs_page *req) | |
141 | { | |
7f8e05f6 DH |
142 | struct inode *d_inode = req->wb_context->path.dentry->d_inode; |
143 | ||
144 | if (PageUptodate(req->wb_page)) | |
145 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
146 | ||
1da177e4 LT |
147 | unlock_page(req->wb_page); |
148 | ||
1da177e4 | 149 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
88be9f99 TM |
150 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
151 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
152 | req->wb_bytes, |
153 | (long long)req_offset(req)); | |
10d2c46f NW |
154 | nfs_clear_request(req); |
155 | nfs_release_request(req); | |
1da177e4 LT |
156 | } |
157 | ||
158 | /* | |
159 | * Set up the NFS read request struct | |
160 | */ | |
dbae4c73 | 161 | static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
ec06c096 | 162 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
163 | unsigned int count, unsigned int offset) |
164 | { | |
84115e1c TM |
165 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
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, | |
171 | .rpc_cred = req->wb_context->cred, | |
172 | }; | |
84115e1c | 173 | struct rpc_task_setup task_setup_data = { |
07737691 | 174 | .task = &data->task, |
84115e1c | 175 | .rpc_client = NFS_CLIENT(inode), |
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 LT |
182 | |
183 | data->req = req; | |
84115e1c | 184 | data->inode = inode; |
bdc7f021 | 185 | data->cred = msg.rpc_cred; |
1da177e4 LT |
186 | |
187 | data->args.fh = NFS_FH(inode); | |
188 | data->args.offset = req_offset(req) + offset; | |
189 | data->args.pgbase = req->wb_pgbase + offset; | |
190 | data->args.pages = data->pagevec; | |
191 | data->args.count = count; | |
383ba719 | 192 | data->args.context = get_nfs_open_context(req->wb_context); |
1da177e4 LT |
193 | |
194 | data->res.fattr = &data->fattr; | |
195 | data->res.count = count; | |
196 | data->res.eof = 0; | |
0e574af1 | 197 | nfs_fattr_init(&data->fattr); |
1da177e4 | 198 | |
ec06c096 | 199 | /* Set up the initial task struct. */ |
bdc7f021 | 200 | NFS_PROTO(inode)->read_setup(data, &msg); |
1da177e4 | 201 | |
a3f565b1 | 202 | dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", |
1da177e4 LT |
203 | data->task.tk_pid, |
204 | inode->i_sb->s_id, | |
205 | (long long)NFS_FILEID(inode), | |
206 | count, | |
207 | (unsigned long long)data->args.offset); | |
bdc7f021 | 208 | |
07737691 | 209 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
210 | if (IS_ERR(task)) |
211 | return PTR_ERR(task); | |
212 | rpc_put_task(task); | |
213 | return 0; | |
1da177e4 LT |
214 | } |
215 | ||
216 | static void | |
217 | nfs_async_read_error(struct list_head *head) | |
218 | { | |
219 | struct nfs_page *req; | |
220 | ||
221 | while (!list_empty(head)) { | |
222 | req = nfs_list_entry(head->next); | |
223 | nfs_list_remove_request(req); | |
224 | SetPageError(req->wb_page); | |
225 | nfs_readpage_release(req); | |
226 | } | |
227 | } | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * Generate multiple requests to fill a single page. | |
231 | * | |
232 | * We optimize to reduce the number of read operations on the wire. If we | |
233 | * detect that we're reading a page, or an area of a page, that is past the | |
234 | * end of file, we do not generate NFS read operations but just clear the | |
235 | * parts of the page that would have come back zero from the server anyway. | |
236 | * | |
237 | * We rely on the cached value of i_size to make this determination; another | |
238 | * client can fill pages on the server past our cached end-of-file, but we | |
239 | * won't see the new data until our attribute cache is updated. This is more | |
240 | * or less conventional NFS client behavior. | |
241 | */ | |
8d5658c9 | 242 | static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
1da177e4 LT |
243 | { |
244 | struct nfs_page *req = nfs_list_entry(head->next); | |
245 | struct page *page = req->wb_page; | |
246 | struct nfs_read_data *data; | |
e9f7bee1 TM |
247 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
248 | unsigned int offset; | |
1da177e4 | 249 | int requests = 0; |
dbae4c73 | 250 | int ret = 0; |
1da177e4 LT |
251 | LIST_HEAD(list); |
252 | ||
253 | nfs_list_remove_request(req); | |
254 | ||
bcb71bba | 255 | nbytes = count; |
e9f7bee1 TM |
256 | do { |
257 | size_t len = min(nbytes,rsize); | |
258 | ||
8d5658c9 | 259 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
260 | if (!data) |
261 | goto out_bad; | |
1da177e4 LT |
262 | list_add(&data->pages, &list); |
263 | requests++; | |
e9f7bee1 TM |
264 | nbytes -= len; |
265 | } while(nbytes != 0); | |
1da177e4 LT |
266 | atomic_set(&req->wb_complete, requests); |
267 | ||
268 | ClearPageError(page); | |
269 | offset = 0; | |
bcb71bba | 270 | nbytes = count; |
1da177e4 | 271 | do { |
dbae4c73 TM |
272 | int ret2; |
273 | ||
1da177e4 LT |
274 | data = list_entry(list.next, struct nfs_read_data, pages); |
275 | list_del_init(&data->pages); | |
276 | ||
277 | data->pagevec[0] = page; | |
1da177e4 | 278 | |
bcb71bba TM |
279 | if (nbytes < rsize) |
280 | rsize = nbytes; | |
dbae4c73 | 281 | ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
bcb71bba | 282 | rsize, offset); |
dbae4c73 TM |
283 | if (ret == 0) |
284 | ret = ret2; | |
bcb71bba TM |
285 | offset += rsize; |
286 | nbytes -= rsize; | |
1da177e4 LT |
287 | } while (nbytes != 0); |
288 | ||
dbae4c73 | 289 | return ret; |
1da177e4 LT |
290 | |
291 | out_bad: | |
292 | while (!list_empty(&list)) { | |
293 | data = list_entry(list.next, struct nfs_read_data, pages); | |
294 | list_del(&data->pages); | |
295 | nfs_readdata_free(data); | |
296 | } | |
297 | SetPageError(page); | |
298 | nfs_readpage_release(req); | |
299 | return -ENOMEM; | |
300 | } | |
301 | ||
8d5658c9 | 302 | static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
1da177e4 LT |
303 | { |
304 | struct nfs_page *req; | |
305 | struct page **pages; | |
306 | struct nfs_read_data *data; | |
dbae4c73 | 307 | int ret = -ENOMEM; |
1da177e4 | 308 | |
8d5658c9 | 309 | data = nfs_readdata_alloc(npages); |
1da177e4 LT |
310 | if (!data) |
311 | goto out_bad; | |
312 | ||
1da177e4 | 313 | pages = data->pagevec; |
1da177e4 LT |
314 | while (!list_empty(head)) { |
315 | req = nfs_list_entry(head->next); | |
316 | nfs_list_remove_request(req); | |
317 | nfs_list_add_request(req, &data->pages); | |
318 | ClearPageError(req->wb_page); | |
319 | *pages++ = req->wb_page; | |
1da177e4 LT |
320 | } |
321 | req = nfs_list_entry(data->pages.next); | |
322 | ||
dbae4c73 | 323 | return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
1da177e4 LT |
324 | out_bad: |
325 | nfs_async_read_error(head); | |
dbae4c73 | 326 | return ret; |
1da177e4 LT |
327 | } |
328 | ||
0b671301 TM |
329 | /* |
330 | * This is the callback from RPC telling us whether a reply was | |
331 | * received or some error occurred (timeout or socket shutdown). | |
332 | */ | |
333 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
334 | { | |
335 | int status; | |
336 | ||
3110ff80 | 337 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
338 | task->tk_status); |
339 | ||
340 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
341 | if (status != 0) | |
342 | return status; | |
343 | ||
344 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
345 | ||
346 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 347 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
348 | nfs_mark_for_revalidate(data->inode); |
349 | } | |
0b671301 TM |
350 | return 0; |
351 | } | |
352 | ||
fdd1e74c | 353 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
354 | { |
355 | struct nfs_readargs *argp = &data->args; | |
356 | struct nfs_readres *resp = &data->res; | |
357 | ||
358 | if (resp->eof || resp->count == argp->count) | |
f11c88af | 359 | goto out; |
0b671301 TM |
360 | |
361 | /* This is a short read! */ | |
362 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
363 | /* Has the server at least made some progress? */ | |
364 | if (resp->count == 0) | |
f11c88af | 365 | goto out; |
0b671301 TM |
366 | |
367 | /* Yes, so retry the read at the end of the data */ | |
368 | argp->offset += resp->count; | |
369 | argp->pgbase += resp->count; | |
370 | argp->count -= resp->count; | |
e608e79f AA |
371 | nfs4_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client, |
372 | &data->res.seq_res); | |
f11c88af AA |
373 | return; |
374 | out: | |
375 | nfs4_sequence_free_slot(NFS_SERVER(data->inode)->nfs_client, | |
376 | &data->res.seq_res); | |
377 | return; | |
378 | ||
0b671301 TM |
379 | } |
380 | ||
1da177e4 LT |
381 | /* |
382 | * Handle a read reply that fills part of a page. | |
383 | */ | |
ec06c096 | 384 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 385 | { |
ec06c096 | 386 | struct nfs_read_data *data = calldata; |
1da177e4 | 387 | |
ec06c096 TM |
388 | if (nfs_readpage_result(task, data) != 0) |
389 | return; | |
fdd1e74c TM |
390 | if (task->tk_status < 0) |
391 | return; | |
0b671301 | 392 | |
fdd1e74c TM |
393 | nfs_readpage_truncate_uninitialised_page(data); |
394 | nfs_readpage_retry(task, data); | |
395 | } | |
396 | ||
397 | static void nfs_readpage_release_partial(void *calldata) | |
398 | { | |
399 | struct nfs_read_data *data = calldata; | |
400 | struct nfs_page *req = data->req; | |
401 | struct page *page = req->wb_page; | |
402 | int status = data->task.tk_status; | |
403 | ||
404 | if (status < 0) | |
0b671301 | 405 | SetPageError(page); |
fdd1e74c | 406 | |
1da177e4 LT |
407 | if (atomic_dec_and_test(&req->wb_complete)) { |
408 | if (!PageError(page)) | |
409 | SetPageUptodate(page); | |
410 | nfs_readpage_release(req); | |
411 | } | |
fdd1e74c | 412 | nfs_readdata_release(calldata); |
1da177e4 LT |
413 | } |
414 | ||
f11c88af AA |
415 | #if defined(CONFIG_NFS_V4_1) |
416 | void nfs_read_prepare(struct rpc_task *task, void *calldata) | |
417 | { | |
418 | struct nfs_read_data *data = calldata; | |
419 | ||
420 | if (nfs4_setup_sequence(NFS_SERVER(data->inode)->nfs_client, | |
421 | &data->args.seq_args, &data->res.seq_res, | |
422 | 0, task)) | |
423 | return; | |
424 | rpc_call_start(task); | |
425 | } | |
426 | #endif /* CONFIG_NFS_V4_1 */ | |
427 | ||
ec06c096 | 428 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af AA |
429 | #if defined(CONFIG_NFS_V4_1) |
430 | .rpc_call_prepare = nfs_read_prepare, | |
431 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 432 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 433 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
434 | }; |
435 | ||
1de3fc12 TM |
436 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
437 | { | |
438 | unsigned int count = data->res.count; | |
439 | unsigned int base = data->args.pgbase; | |
440 | struct page **pages; | |
441 | ||
79558f36 TM |
442 | if (data->res.eof) |
443 | count = data->args.count; | |
1de3fc12 TM |
444 | if (unlikely(count == 0)) |
445 | return; | |
446 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
447 | base &= ~PAGE_CACHE_MASK; | |
448 | count += base; | |
449 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
450 | SetPageUptodate(*pages); | |
0b671301 TM |
451 | if (count == 0) |
452 | return; | |
453 | /* Was this a short read? */ | |
454 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
455 | SetPageUptodate(*pages); |
456 | } | |
457 | ||
1da177e4 LT |
458 | /* |
459 | * This is the callback from RPC telling us whether a reply was | |
460 | * received or some error occurred (timeout or socket shutdown). | |
461 | */ | |
ec06c096 | 462 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 463 | { |
ec06c096 | 464 | struct nfs_read_data *data = calldata; |
1da177e4 | 465 | |
0b671301 TM |
466 | if (nfs_readpage_result(task, data) != 0) |
467 | return; | |
fdd1e74c TM |
468 | if (task->tk_status < 0) |
469 | return; | |
1de3fc12 | 470 | /* |
0b671301 | 471 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 472 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
473 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
474 | * first. | |
1de3fc12 | 475 | */ |
fdd1e74c TM |
476 | nfs_readpage_truncate_uninitialised_page(data); |
477 | nfs_readpage_set_pages_uptodate(data); | |
478 | nfs_readpage_retry(task, data); | |
479 | } | |
480 | ||
481 | static void nfs_readpage_release_full(void *calldata) | |
482 | { | |
483 | struct nfs_read_data *data = calldata; | |
484 | ||
1da177e4 LT |
485 | while (!list_empty(&data->pages)) { |
486 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 487 | |
1de3fc12 | 488 | nfs_list_remove_request(req); |
1da177e4 LT |
489 | nfs_readpage_release(req); |
490 | } | |
fdd1e74c | 491 | nfs_readdata_release(calldata); |
1da177e4 LT |
492 | } |
493 | ||
ec06c096 | 494 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af AA |
495 | #if defined(CONFIG_NFS_V4_1) |
496 | .rpc_call_prepare = nfs_read_prepare, | |
497 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 498 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 499 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
500 | }; |
501 | ||
1da177e4 LT |
502 | /* |
503 | * Read a page over NFS. | |
504 | * We read the page synchronously in the following case: | |
505 | * - The error flag is set for this page. This happens only when a | |
506 | * previous async read operation failed. | |
507 | */ | |
508 | int nfs_readpage(struct file *file, struct page *page) | |
509 | { | |
510 | struct nfs_open_context *ctx; | |
511 | struct inode *inode = page->mapping->host; | |
512 | int error; | |
513 | ||
514 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
515 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
516 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
517 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
518 | ||
1da177e4 LT |
519 | /* |
520 | * Try to flush any pending writes to the file.. | |
521 | * | |
522 | * NOTE! Because we own the page lock, there cannot | |
523 | * be any new pending writes generated at this point | |
524 | * for this page (other pages can be written to). | |
525 | */ | |
526 | error = nfs_wb_page(inode, page); | |
527 | if (error) | |
de05a0cc TM |
528 | goto out_unlock; |
529 | if (PageUptodate(page)) | |
530 | goto out_unlock; | |
1da177e4 | 531 | |
5f004cf2 TM |
532 | error = -ESTALE; |
533 | if (NFS_STALE(inode)) | |
de05a0cc | 534 | goto out_unlock; |
5f004cf2 | 535 | |
1da177e4 | 536 | if (file == NULL) { |
cf1308ff | 537 | error = -EBADF; |
d530838b | 538 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 539 | if (ctx == NULL) |
de05a0cc | 540 | goto out_unlock; |
1da177e4 | 541 | } else |
cd3758e3 | 542 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 543 | |
9a9fc1c0 DH |
544 | if (!IS_SYNC(inode)) { |
545 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
546 | if (error == 0) | |
547 | goto out; | |
548 | } | |
549 | ||
8e0969f0 TM |
550 | error = nfs_readpage_async(ctx, inode, page); |
551 | ||
9a9fc1c0 | 552 | out: |
1da177e4 LT |
553 | put_nfs_open_context(ctx); |
554 | return error; | |
de05a0cc | 555 | out_unlock: |
1da177e4 LT |
556 | unlock_page(page); |
557 | return error; | |
558 | } | |
559 | ||
560 | struct nfs_readdesc { | |
8b09bee3 | 561 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
562 | struct nfs_open_context *ctx; |
563 | }; | |
564 | ||
565 | static int | |
566 | readpage_async_filler(void *data, struct page *page) | |
567 | { | |
568 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
569 | struct inode *inode = page->mapping->host; | |
570 | struct nfs_page *new; | |
571 | unsigned int len; | |
de05a0cc TM |
572 | int error; |
573 | ||
49a70f27 | 574 | len = nfs_page_length(page); |
1da177e4 LT |
575 | if (len == 0) |
576 | return nfs_return_empty_page(page); | |
de05a0cc | 577 | |
1da177e4 | 578 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
579 | if (IS_ERR(new)) |
580 | goto out_error; | |
581 | ||
1da177e4 | 582 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 583 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
584 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
585 | error = desc->pgio->pg_error; | |
586 | goto out_unlock; | |
587 | } | |
1da177e4 | 588 | return 0; |
de05a0cc TM |
589 | out_error: |
590 | error = PTR_ERR(new); | |
591 | SetPageError(page); | |
592 | out_unlock: | |
593 | unlock_page(page); | |
594 | return error; | |
1da177e4 LT |
595 | } |
596 | ||
597 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
598 | struct list_head *pages, unsigned nr_pages) | |
599 | { | |
8b09bee3 | 600 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 601 | struct nfs_readdesc desc = { |
8b09bee3 | 602 | .pgio = &pgio, |
1da177e4 LT |
603 | }; |
604 | struct inode *inode = mapping->host; | |
605 | struct nfs_server *server = NFS_SERVER(inode); | |
8b09bee3 TM |
606 | size_t rsize = server->rsize; |
607 | unsigned long npages; | |
5f004cf2 | 608 | int ret = -ESTALE; |
1da177e4 LT |
609 | |
610 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
611 | inode->i_sb->s_id, | |
612 | (long long)NFS_FILEID(inode), | |
613 | nr_pages); | |
91d5b470 | 614 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 615 | |
5f004cf2 TM |
616 | if (NFS_STALE(inode)) |
617 | goto out; | |
618 | ||
1da177e4 | 619 | if (filp == NULL) { |
d530838b | 620 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
621 | if (desc.ctx == NULL) |
622 | return -EBADF; | |
623 | } else | |
cd3758e3 | 624 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
625 | |
626 | /* attempt to read as many of the pages as possible from the cache | |
627 | * - this returns -ENOBUFS immediately if the cookie is negative | |
628 | */ | |
629 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
630 | pages, &nr_pages); | |
631 | if (ret == 0) | |
632 | goto read_complete; /* all pages were read */ | |
633 | ||
8b09bee3 TM |
634 | if (rsize < PAGE_CACHE_SIZE) |
635 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); | |
636 | else | |
637 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); | |
638 | ||
1da177e4 | 639 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
640 | |
641 | nfs_pageio_complete(&pgio); | |
642 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
643 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 644 | read_complete: |
1da177e4 | 645 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 646 | out: |
1da177e4 LT |
647 | return ret; |
648 | } | |
649 | ||
f7b422b1 | 650 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
651 | { |
652 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
653 | sizeof(struct nfs_read_data), | |
654 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 655 | NULL); |
1da177e4 LT |
656 | if (nfs_rdata_cachep == NULL) |
657 | return -ENOMEM; | |
658 | ||
93d2341c MD |
659 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
660 | nfs_rdata_cachep); | |
1da177e4 LT |
661 | if (nfs_rdata_mempool == NULL) |
662 | return -ENOMEM; | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
266bee88 | 667 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
668 | { |
669 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 670 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 671 | } |