]>
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> | |
21 | #include <linux/smp_lock.h> | |
22 | ||
23 | #include <asm/system.h> | |
24 | ||
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 | ||
5e4424af | 63 | static 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 | ||
963d8fe5 | 70 | void nfs_readdata_release(void *data) |
1da177e4 | 71 | { |
383ba719 TM |
72 | struct nfs_read_data *rdata = data; |
73 | ||
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 LT |
118 | { |
119 | LIST_HEAD(one_request); | |
120 | struct nfs_page *new; | |
121 | unsigned int len; | |
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 | |
1da177e4 | 134 | nfs_list_add_request(new, &one_request); |
bcb71bba | 135 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
8d5658c9 | 136 | nfs_pagein_multi(inode, &one_request, 1, len, 0); |
bcb71bba | 137 | else |
8d5658c9 | 138 | nfs_pagein_one(inode, &one_request, 1, len, 0); |
1da177e4 LT |
139 | return 0; |
140 | } | |
141 | ||
142 | static void nfs_readpage_release(struct nfs_page *req) | |
143 | { | |
7f8e05f6 DH |
144 | struct inode *d_inode = req->wb_context->path.dentry->d_inode; |
145 | ||
146 | if (PageUptodate(req->wb_page)) | |
147 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
148 | ||
1da177e4 LT |
149 | unlock_page(req->wb_page); |
150 | ||
1da177e4 | 151 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
88be9f99 TM |
152 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
153 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
154 | req->wb_bytes, |
155 | (long long)req_offset(req)); | |
10d2c46f NW |
156 | nfs_clear_request(req); |
157 | nfs_release_request(req); | |
1da177e4 LT |
158 | } |
159 | ||
160 | /* | |
161 | * Set up the NFS read request struct | |
162 | */ | |
dbae4c73 | 163 | static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
ec06c096 | 164 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
165 | unsigned int count, unsigned int offset) |
166 | { | |
84115e1c TM |
167 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
168 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; | |
07737691 | 169 | struct rpc_task *task; |
bdc7f021 TM |
170 | struct rpc_message msg = { |
171 | .rpc_argp = &data->args, | |
172 | .rpc_resp = &data->res, | |
173 | .rpc_cred = req->wb_context->cred, | |
174 | }; | |
84115e1c | 175 | struct rpc_task_setup task_setup_data = { |
07737691 | 176 | .task = &data->task, |
84115e1c | 177 | .rpc_client = NFS_CLIENT(inode), |
bdc7f021 | 178 | .rpc_message = &msg, |
84115e1c TM |
179 | .callback_ops = call_ops, |
180 | .callback_data = data, | |
101070ca | 181 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
182 | .flags = RPC_TASK_ASYNC | swap_flags, |
183 | }; | |
1da177e4 LT |
184 | |
185 | data->req = req; | |
84115e1c | 186 | data->inode = inode; |
bdc7f021 | 187 | data->cred = msg.rpc_cred; |
1da177e4 LT |
188 | |
189 | data->args.fh = NFS_FH(inode); | |
190 | data->args.offset = req_offset(req) + offset; | |
191 | data->args.pgbase = req->wb_pgbase + offset; | |
192 | data->args.pages = data->pagevec; | |
193 | data->args.count = count; | |
383ba719 | 194 | data->args.context = get_nfs_open_context(req->wb_context); |
1da177e4 LT |
195 | |
196 | data->res.fattr = &data->fattr; | |
197 | data->res.count = count; | |
198 | data->res.eof = 0; | |
0e574af1 | 199 | nfs_fattr_init(&data->fattr); |
1da177e4 | 200 | |
ec06c096 | 201 | /* Set up the initial task struct. */ |
bdc7f021 | 202 | NFS_PROTO(inode)->read_setup(data, &msg); |
1da177e4 | 203 | |
a3f565b1 | 204 | dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", |
1da177e4 LT |
205 | data->task.tk_pid, |
206 | inode->i_sb->s_id, | |
207 | (long long)NFS_FILEID(inode), | |
208 | count, | |
209 | (unsigned long long)data->args.offset); | |
bdc7f021 | 210 | |
07737691 | 211 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
212 | if (IS_ERR(task)) |
213 | return PTR_ERR(task); | |
214 | rpc_put_task(task); | |
215 | return 0; | |
1da177e4 LT |
216 | } |
217 | ||
218 | static void | |
219 | nfs_async_read_error(struct list_head *head) | |
220 | { | |
221 | struct nfs_page *req; | |
222 | ||
223 | while (!list_empty(head)) { | |
224 | req = nfs_list_entry(head->next); | |
225 | nfs_list_remove_request(req); | |
226 | SetPageError(req->wb_page); | |
227 | nfs_readpage_release(req); | |
228 | } | |
229 | } | |
230 | ||
1da177e4 LT |
231 | /* |
232 | * Generate multiple requests to fill a single page. | |
233 | * | |
234 | * We optimize to reduce the number of read operations on the wire. If we | |
235 | * detect that we're reading a page, or an area of a page, that is past the | |
236 | * end of file, we do not generate NFS read operations but just clear the | |
237 | * parts of the page that would have come back zero from the server anyway. | |
238 | * | |
239 | * We rely on the cached value of i_size to make this determination; another | |
240 | * client can fill pages on the server past our cached end-of-file, but we | |
241 | * won't see the new data until our attribute cache is updated. This is more | |
242 | * or less conventional NFS client behavior. | |
243 | */ | |
8d5658c9 | 244 | static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
1da177e4 LT |
245 | { |
246 | struct nfs_page *req = nfs_list_entry(head->next); | |
247 | struct page *page = req->wb_page; | |
248 | struct nfs_read_data *data; | |
e9f7bee1 TM |
249 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
250 | unsigned int offset; | |
1da177e4 | 251 | int requests = 0; |
dbae4c73 | 252 | int ret = 0; |
1da177e4 LT |
253 | LIST_HEAD(list); |
254 | ||
255 | nfs_list_remove_request(req); | |
256 | ||
bcb71bba | 257 | nbytes = count; |
e9f7bee1 TM |
258 | do { |
259 | size_t len = min(nbytes,rsize); | |
260 | ||
8d5658c9 | 261 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
262 | if (!data) |
263 | goto out_bad; | |
1da177e4 LT |
264 | list_add(&data->pages, &list); |
265 | requests++; | |
e9f7bee1 TM |
266 | nbytes -= len; |
267 | } while(nbytes != 0); | |
1da177e4 LT |
268 | atomic_set(&req->wb_complete, requests); |
269 | ||
270 | ClearPageError(page); | |
271 | offset = 0; | |
bcb71bba | 272 | nbytes = count; |
1da177e4 | 273 | do { |
dbae4c73 TM |
274 | int ret2; |
275 | ||
1da177e4 LT |
276 | data = list_entry(list.next, struct nfs_read_data, pages); |
277 | list_del_init(&data->pages); | |
278 | ||
279 | data->pagevec[0] = page; | |
1da177e4 | 280 | |
bcb71bba TM |
281 | if (nbytes < rsize) |
282 | rsize = nbytes; | |
dbae4c73 | 283 | ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
bcb71bba | 284 | rsize, offset); |
dbae4c73 TM |
285 | if (ret == 0) |
286 | ret = ret2; | |
bcb71bba TM |
287 | offset += rsize; |
288 | nbytes -= rsize; | |
1da177e4 LT |
289 | } while (nbytes != 0); |
290 | ||
dbae4c73 | 291 | return ret; |
1da177e4 LT |
292 | |
293 | out_bad: | |
294 | while (!list_empty(&list)) { | |
295 | data = list_entry(list.next, struct nfs_read_data, pages); | |
296 | list_del(&data->pages); | |
297 | nfs_readdata_free(data); | |
298 | } | |
299 | SetPageError(page); | |
300 | nfs_readpage_release(req); | |
301 | return -ENOMEM; | |
302 | } | |
303 | ||
8d5658c9 | 304 | static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
1da177e4 LT |
305 | { |
306 | struct nfs_page *req; | |
307 | struct page **pages; | |
308 | struct nfs_read_data *data; | |
dbae4c73 | 309 | int ret = -ENOMEM; |
1da177e4 | 310 | |
8d5658c9 | 311 | data = nfs_readdata_alloc(npages); |
1da177e4 LT |
312 | if (!data) |
313 | goto out_bad; | |
314 | ||
1da177e4 | 315 | pages = data->pagevec; |
1da177e4 LT |
316 | while (!list_empty(head)) { |
317 | req = nfs_list_entry(head->next); | |
318 | nfs_list_remove_request(req); | |
319 | nfs_list_add_request(req, &data->pages); | |
320 | ClearPageError(req->wb_page); | |
321 | *pages++ = req->wb_page; | |
1da177e4 LT |
322 | } |
323 | req = nfs_list_entry(data->pages.next); | |
324 | ||
dbae4c73 | 325 | return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
1da177e4 LT |
326 | out_bad: |
327 | nfs_async_read_error(head); | |
dbae4c73 | 328 | return ret; |
1da177e4 LT |
329 | } |
330 | ||
0b671301 TM |
331 | /* |
332 | * This is the callback from RPC telling us whether a reply was | |
333 | * received or some error occurred (timeout or socket shutdown). | |
334 | */ | |
335 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
336 | { | |
337 | int status; | |
338 | ||
3110ff80 | 339 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
340 | task->tk_status); |
341 | ||
342 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
343 | if (status != 0) | |
344 | return status; | |
345 | ||
346 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
347 | ||
348 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 349 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
350 | nfs_mark_for_revalidate(data->inode); |
351 | } | |
0b671301 TM |
352 | return 0; |
353 | } | |
354 | ||
fdd1e74c | 355 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
356 | { |
357 | struct nfs_readargs *argp = &data->args; | |
358 | struct nfs_readres *resp = &data->res; | |
359 | ||
360 | if (resp->eof || resp->count == argp->count) | |
fdd1e74c | 361 | return; |
0b671301 TM |
362 | |
363 | /* This is a short read! */ | |
364 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
365 | /* Has the server at least made some progress? */ | |
366 | if (resp->count == 0) | |
fdd1e74c | 367 | return; |
0b671301 TM |
368 | |
369 | /* Yes, so retry the read at the end of the data */ | |
370 | argp->offset += resp->count; | |
371 | argp->pgbase += resp->count; | |
372 | argp->count -= resp->count; | |
373 | rpc_restart_call(task); | |
0b671301 TM |
374 | } |
375 | ||
1da177e4 LT |
376 | /* |
377 | * Handle a read reply that fills part of a page. | |
378 | */ | |
ec06c096 | 379 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 380 | { |
ec06c096 | 381 | struct nfs_read_data *data = calldata; |
1da177e4 | 382 | |
ec06c096 TM |
383 | if (nfs_readpage_result(task, data) != 0) |
384 | return; | |
fdd1e74c TM |
385 | if (task->tk_status < 0) |
386 | return; | |
0b671301 | 387 | |
fdd1e74c TM |
388 | nfs_readpage_truncate_uninitialised_page(data); |
389 | nfs_readpage_retry(task, data); | |
390 | } | |
391 | ||
392 | static void nfs_readpage_release_partial(void *calldata) | |
393 | { | |
394 | struct nfs_read_data *data = calldata; | |
395 | struct nfs_page *req = data->req; | |
396 | struct page *page = req->wb_page; | |
397 | int status = data->task.tk_status; | |
398 | ||
399 | if (status < 0) | |
0b671301 | 400 | SetPageError(page); |
fdd1e74c | 401 | |
1da177e4 LT |
402 | if (atomic_dec_and_test(&req->wb_complete)) { |
403 | if (!PageError(page)) | |
404 | SetPageUptodate(page); | |
405 | nfs_readpage_release(req); | |
406 | } | |
fdd1e74c | 407 | nfs_readdata_release(calldata); |
1da177e4 LT |
408 | } |
409 | ||
ec06c096 TM |
410 | static const struct rpc_call_ops nfs_read_partial_ops = { |
411 | .rpc_call_done = nfs_readpage_result_partial, | |
fdd1e74c | 412 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
413 | }; |
414 | ||
1de3fc12 TM |
415 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
416 | { | |
417 | unsigned int count = data->res.count; | |
418 | unsigned int base = data->args.pgbase; | |
419 | struct page **pages; | |
420 | ||
79558f36 TM |
421 | if (data->res.eof) |
422 | count = data->args.count; | |
1de3fc12 TM |
423 | if (unlikely(count == 0)) |
424 | return; | |
425 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
426 | base &= ~PAGE_CACHE_MASK; | |
427 | count += base; | |
428 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
429 | SetPageUptodate(*pages); | |
0b671301 TM |
430 | if (count == 0) |
431 | return; | |
432 | /* Was this a short read? */ | |
433 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
434 | SetPageUptodate(*pages); |
435 | } | |
436 | ||
1da177e4 LT |
437 | /* |
438 | * This is the callback from RPC telling us whether a reply was | |
439 | * received or some error occurred (timeout or socket shutdown). | |
440 | */ | |
ec06c096 | 441 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 442 | { |
ec06c096 | 443 | struct nfs_read_data *data = calldata; |
1da177e4 | 444 | |
0b671301 TM |
445 | if (nfs_readpage_result(task, data) != 0) |
446 | return; | |
fdd1e74c TM |
447 | if (task->tk_status < 0) |
448 | return; | |
1de3fc12 | 449 | /* |
0b671301 | 450 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 451 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
452 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
453 | * first. | |
1de3fc12 | 454 | */ |
fdd1e74c TM |
455 | nfs_readpage_truncate_uninitialised_page(data); |
456 | nfs_readpage_set_pages_uptodate(data); | |
457 | nfs_readpage_retry(task, data); | |
458 | } | |
459 | ||
460 | static void nfs_readpage_release_full(void *calldata) | |
461 | { | |
462 | struct nfs_read_data *data = calldata; | |
463 | ||
1da177e4 LT |
464 | while (!list_empty(&data->pages)) { |
465 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 466 | |
1de3fc12 | 467 | nfs_list_remove_request(req); |
1da177e4 LT |
468 | nfs_readpage_release(req); |
469 | } | |
fdd1e74c | 470 | nfs_readdata_release(calldata); |
1da177e4 LT |
471 | } |
472 | ||
ec06c096 TM |
473 | static const struct rpc_call_ops nfs_read_full_ops = { |
474 | .rpc_call_done = nfs_readpage_result_full, | |
fdd1e74c | 475 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
476 | }; |
477 | ||
1da177e4 LT |
478 | /* |
479 | * Read a page over NFS. | |
480 | * We read the page synchronously in the following case: | |
481 | * - The error flag is set for this page. This happens only when a | |
482 | * previous async read operation failed. | |
483 | */ | |
484 | int nfs_readpage(struct file *file, struct page *page) | |
485 | { | |
486 | struct nfs_open_context *ctx; | |
487 | struct inode *inode = page->mapping->host; | |
488 | int error; | |
489 | ||
490 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
491 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
492 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
493 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
494 | ||
1da177e4 LT |
495 | /* |
496 | * Try to flush any pending writes to the file.. | |
497 | * | |
498 | * NOTE! Because we own the page lock, there cannot | |
499 | * be any new pending writes generated at this point | |
500 | * for this page (other pages can be written to). | |
501 | */ | |
502 | error = nfs_wb_page(inode, page); | |
503 | if (error) | |
de05a0cc TM |
504 | goto out_unlock; |
505 | if (PageUptodate(page)) | |
506 | goto out_unlock; | |
1da177e4 | 507 | |
5f004cf2 TM |
508 | error = -ESTALE; |
509 | if (NFS_STALE(inode)) | |
de05a0cc | 510 | goto out_unlock; |
5f004cf2 | 511 | |
1da177e4 | 512 | if (file == NULL) { |
cf1308ff | 513 | error = -EBADF; |
d530838b | 514 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 515 | if (ctx == NULL) |
de05a0cc | 516 | goto out_unlock; |
1da177e4 | 517 | } else |
cd3758e3 | 518 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 519 | |
9a9fc1c0 DH |
520 | if (!IS_SYNC(inode)) { |
521 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
522 | if (error == 0) | |
523 | goto out; | |
524 | } | |
525 | ||
8e0969f0 TM |
526 | error = nfs_readpage_async(ctx, inode, page); |
527 | ||
9a9fc1c0 | 528 | out: |
1da177e4 LT |
529 | put_nfs_open_context(ctx); |
530 | return error; | |
de05a0cc | 531 | out_unlock: |
1da177e4 LT |
532 | unlock_page(page); |
533 | return error; | |
534 | } | |
535 | ||
536 | struct nfs_readdesc { | |
8b09bee3 | 537 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
538 | struct nfs_open_context *ctx; |
539 | }; | |
540 | ||
541 | static int | |
542 | readpage_async_filler(void *data, struct page *page) | |
543 | { | |
544 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
545 | struct inode *inode = page->mapping->host; | |
546 | struct nfs_page *new; | |
547 | unsigned int len; | |
de05a0cc TM |
548 | int error; |
549 | ||
49a70f27 | 550 | len = nfs_page_length(page); |
1da177e4 LT |
551 | if (len == 0) |
552 | return nfs_return_empty_page(page); | |
de05a0cc | 553 | |
1da177e4 | 554 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
555 | if (IS_ERR(new)) |
556 | goto out_error; | |
557 | ||
1da177e4 | 558 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 559 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
560 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
561 | error = desc->pgio->pg_error; | |
562 | goto out_unlock; | |
563 | } | |
1da177e4 | 564 | return 0; |
de05a0cc TM |
565 | out_error: |
566 | error = PTR_ERR(new); | |
567 | SetPageError(page); | |
568 | out_unlock: | |
569 | unlock_page(page); | |
570 | return error; | |
1da177e4 LT |
571 | } |
572 | ||
573 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
574 | struct list_head *pages, unsigned nr_pages) | |
575 | { | |
8b09bee3 | 576 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 577 | struct nfs_readdesc desc = { |
8b09bee3 | 578 | .pgio = &pgio, |
1da177e4 LT |
579 | }; |
580 | struct inode *inode = mapping->host; | |
581 | struct nfs_server *server = NFS_SERVER(inode); | |
8b09bee3 TM |
582 | size_t rsize = server->rsize; |
583 | unsigned long npages; | |
5f004cf2 | 584 | int ret = -ESTALE; |
1da177e4 LT |
585 | |
586 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
587 | inode->i_sb->s_id, | |
588 | (long long)NFS_FILEID(inode), | |
589 | nr_pages); | |
91d5b470 | 590 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 591 | |
5f004cf2 TM |
592 | if (NFS_STALE(inode)) |
593 | goto out; | |
594 | ||
1da177e4 | 595 | if (filp == NULL) { |
d530838b | 596 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
597 | if (desc.ctx == NULL) |
598 | return -EBADF; | |
599 | } else | |
cd3758e3 | 600 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
601 | |
602 | /* attempt to read as many of the pages as possible from the cache | |
603 | * - this returns -ENOBUFS immediately if the cookie is negative | |
604 | */ | |
605 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
606 | pages, &nr_pages); | |
607 | if (ret == 0) | |
608 | goto read_complete; /* all pages were read */ | |
609 | ||
8b09bee3 TM |
610 | if (rsize < PAGE_CACHE_SIZE) |
611 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); | |
612 | else | |
613 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); | |
614 | ||
1da177e4 | 615 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
616 | |
617 | nfs_pageio_complete(&pgio); | |
618 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
619 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 620 | read_complete: |
1da177e4 | 621 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 622 | out: |
1da177e4 LT |
623 | return ret; |
624 | } | |
625 | ||
f7b422b1 | 626 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
627 | { |
628 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
629 | sizeof(struct nfs_read_data), | |
630 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 631 | NULL); |
1da177e4 LT |
632 | if (nfs_rdata_cachep == NULL) |
633 | return -ENOMEM; | |
634 | ||
93d2341c MD |
635 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
636 | nfs_rdata_cachep); | |
1da177e4 LT |
637 | if (nfs_rdata_mempool == NULL) |
638 | return -ENOMEM; | |
639 | ||
640 | return 0; | |
641 | } | |
642 | ||
266bee88 | 643 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
644 | { |
645 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 646 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 647 | } |