]>
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" |
e5e94017 | 28 | #include "pnfs.h" |
91d5b470 | 29 | |
1da177e4 LT |
30 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
31 | ||
8d5658c9 TM |
32 | static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int); |
33 | static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int); | |
ec06c096 TM |
34 | static const struct rpc_call_ops nfs_read_partial_ops; |
35 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 | 36 | |
e18b890b | 37 | static struct kmem_cache *nfs_rdata_cachep; |
3feb2d49 | 38 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
39 | |
40 | #define MIN_POOL_READ (32) | |
41 | ||
8d5658c9 | 42 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 43 | { |
93870d76 | 44 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL); |
3feb2d49 TM |
45 | |
46 | if (p) { | |
47 | memset(p, 0, sizeof(*p)); | |
48 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 49 | p->npages = pagecount; |
0d0b5cb3 CL |
50 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
51 | p->pagevec = p->page_array; | |
3feb2d49 | 52 | else { |
93870d76 | 53 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); |
0d0b5cb3 | 54 | if (!p->pagevec) { |
3feb2d49 TM |
55 | mempool_free(p, nfs_rdata_mempool); |
56 | p = NULL; | |
57 | } | |
58 | } | |
59 | } | |
60 | return p; | |
61 | } | |
62 | ||
1ae88b2e | 63 | void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
64 | { |
65 | if (p && (p->pagevec != &p->page_array[0])) | |
66 | kfree(p->pagevec); | |
67 | mempool_free(p, nfs_rdata_mempool); | |
68 | } | |
69 | ||
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); | |
e5e94017 | 124 | pnfs_update_layout(inode, ctx, IOMODE_READ); |
1da177e4 LT |
125 | new = nfs_create_request(ctx, inode, page, 0, len); |
126 | if (IS_ERR(new)) { | |
127 | unlock_page(page); | |
128 | return PTR_ERR(new); | |
129 | } | |
130 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 131 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 132 | |
1da177e4 | 133 | nfs_list_add_request(new, &one_request); |
bcb71bba | 134 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
8d5658c9 | 135 | nfs_pagein_multi(inode, &one_request, 1, len, 0); |
bcb71bba | 136 | else |
8d5658c9 | 137 | nfs_pagein_one(inode, &one_request, 1, len, 0); |
1da177e4 LT |
138 | return 0; |
139 | } | |
140 | ||
141 | static void nfs_readpage_release(struct nfs_page *req) | |
142 | { | |
7f8e05f6 DH |
143 | struct inode *d_inode = req->wb_context->path.dentry->d_inode; |
144 | ||
145 | if (PageUptodate(req->wb_page)) | |
146 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
147 | ||
1da177e4 LT |
148 | unlock_page(req->wb_page); |
149 | ||
1da177e4 | 150 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
88be9f99 TM |
151 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
152 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
153 | req->wb_bytes, |
154 | (long long)req_offset(req)); | |
10d2c46f NW |
155 | nfs_clear_request(req); |
156 | nfs_release_request(req); | |
1da177e4 LT |
157 | } |
158 | ||
159 | /* | |
160 | * Set up the NFS read request struct | |
161 | */ | |
dbae4c73 | 162 | static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
ec06c096 | 163 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
164 | unsigned int count, unsigned int offset) |
165 | { | |
84115e1c TM |
166 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
167 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; | |
07737691 | 168 | struct rpc_task *task; |
bdc7f021 TM |
169 | struct rpc_message msg = { |
170 | .rpc_argp = &data->args, | |
171 | .rpc_resp = &data->res, | |
172 | .rpc_cred = req->wb_context->cred, | |
173 | }; | |
84115e1c | 174 | struct rpc_task_setup task_setup_data = { |
07737691 | 175 | .task = &data->task, |
84115e1c | 176 | .rpc_client = NFS_CLIENT(inode), |
bdc7f021 | 177 | .rpc_message = &msg, |
84115e1c TM |
178 | .callback_ops = call_ops, |
179 | .callback_data = data, | |
101070ca | 180 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
181 | .flags = RPC_TASK_ASYNC | swap_flags, |
182 | }; | |
1da177e4 LT |
183 | |
184 | data->req = req; | |
84115e1c | 185 | data->inode = inode; |
bdc7f021 | 186 | data->cred = msg.rpc_cred; |
1da177e4 LT |
187 | |
188 | data->args.fh = NFS_FH(inode); | |
189 | data->args.offset = req_offset(req) + offset; | |
190 | data->args.pgbase = req->wb_pgbase + offset; | |
191 | data->args.pages = data->pagevec; | |
192 | data->args.count = count; | |
383ba719 | 193 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 194 | data->args.lock_context = req->wb_lock_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) | |
d61e612a | 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) | |
d61e612a | 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; | |
0110ee15 | 373 | nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client); |
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 | ||
f11c88af AA |
410 | #if defined(CONFIG_NFS_V4_1) |
411 | void nfs_read_prepare(struct rpc_task *task, void *calldata) | |
412 | { | |
413 | struct nfs_read_data *data = calldata; | |
414 | ||
035168ab | 415 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
f11c88af AA |
416 | &data->args.seq_args, &data->res.seq_res, |
417 | 0, task)) | |
418 | return; | |
419 | rpc_call_start(task); | |
420 | } | |
421 | #endif /* CONFIG_NFS_V4_1 */ | |
422 | ||
ec06c096 | 423 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af AA |
424 | #if defined(CONFIG_NFS_V4_1) |
425 | .rpc_call_prepare = nfs_read_prepare, | |
426 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 427 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 428 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
429 | }; |
430 | ||
1de3fc12 TM |
431 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
432 | { | |
433 | unsigned int count = data->res.count; | |
434 | unsigned int base = data->args.pgbase; | |
435 | struct page **pages; | |
436 | ||
79558f36 TM |
437 | if (data->res.eof) |
438 | count = data->args.count; | |
1de3fc12 TM |
439 | if (unlikely(count == 0)) |
440 | return; | |
441 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
442 | base &= ~PAGE_CACHE_MASK; | |
443 | count += base; | |
444 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
445 | SetPageUptodate(*pages); | |
0b671301 TM |
446 | if (count == 0) |
447 | return; | |
448 | /* Was this a short read? */ | |
449 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
450 | SetPageUptodate(*pages); |
451 | } | |
452 | ||
1da177e4 LT |
453 | /* |
454 | * This is the callback from RPC telling us whether a reply was | |
455 | * received or some error occurred (timeout or socket shutdown). | |
456 | */ | |
ec06c096 | 457 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 458 | { |
ec06c096 | 459 | struct nfs_read_data *data = calldata; |
1da177e4 | 460 | |
0b671301 TM |
461 | if (nfs_readpage_result(task, data) != 0) |
462 | return; | |
fdd1e74c TM |
463 | if (task->tk_status < 0) |
464 | return; | |
1de3fc12 | 465 | /* |
0b671301 | 466 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 467 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
468 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
469 | * first. | |
1de3fc12 | 470 | */ |
fdd1e74c TM |
471 | nfs_readpage_truncate_uninitialised_page(data); |
472 | nfs_readpage_set_pages_uptodate(data); | |
473 | nfs_readpage_retry(task, data); | |
474 | } | |
475 | ||
476 | static void nfs_readpage_release_full(void *calldata) | |
477 | { | |
478 | struct nfs_read_data *data = calldata; | |
479 | ||
1da177e4 LT |
480 | while (!list_empty(&data->pages)) { |
481 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 482 | |
1de3fc12 | 483 | nfs_list_remove_request(req); |
1da177e4 LT |
484 | nfs_readpage_release(req); |
485 | } | |
fdd1e74c | 486 | nfs_readdata_release(calldata); |
1da177e4 LT |
487 | } |
488 | ||
ec06c096 | 489 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af AA |
490 | #if defined(CONFIG_NFS_V4_1) |
491 | .rpc_call_prepare = nfs_read_prepare, | |
492 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 493 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 494 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
495 | }; |
496 | ||
1da177e4 LT |
497 | /* |
498 | * Read a page over NFS. | |
499 | * We read the page synchronously in the following case: | |
500 | * - The error flag is set for this page. This happens only when a | |
501 | * previous async read operation failed. | |
502 | */ | |
503 | int nfs_readpage(struct file *file, struct page *page) | |
504 | { | |
505 | struct nfs_open_context *ctx; | |
506 | struct inode *inode = page->mapping->host; | |
507 | int error; | |
508 | ||
509 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
510 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
511 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
512 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
513 | ||
1da177e4 LT |
514 | /* |
515 | * Try to flush any pending writes to the file.. | |
516 | * | |
517 | * NOTE! Because we own the page lock, there cannot | |
518 | * be any new pending writes generated at this point | |
519 | * for this page (other pages can be written to). | |
520 | */ | |
521 | error = nfs_wb_page(inode, page); | |
522 | if (error) | |
de05a0cc TM |
523 | goto out_unlock; |
524 | if (PageUptodate(page)) | |
525 | goto out_unlock; | |
1da177e4 | 526 | |
5f004cf2 TM |
527 | error = -ESTALE; |
528 | if (NFS_STALE(inode)) | |
de05a0cc | 529 | goto out_unlock; |
5f004cf2 | 530 | |
1da177e4 | 531 | if (file == NULL) { |
cf1308ff | 532 | error = -EBADF; |
d530838b | 533 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 534 | if (ctx == NULL) |
de05a0cc | 535 | goto out_unlock; |
1da177e4 | 536 | } else |
cd3758e3 | 537 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 538 | |
9a9fc1c0 DH |
539 | if (!IS_SYNC(inode)) { |
540 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
541 | if (error == 0) | |
542 | goto out; | |
543 | } | |
544 | ||
8e0969f0 TM |
545 | error = nfs_readpage_async(ctx, inode, page); |
546 | ||
9a9fc1c0 | 547 | out: |
1da177e4 LT |
548 | put_nfs_open_context(ctx); |
549 | return error; | |
de05a0cc | 550 | out_unlock: |
1da177e4 LT |
551 | unlock_page(page); |
552 | return error; | |
553 | } | |
554 | ||
555 | struct nfs_readdesc { | |
8b09bee3 | 556 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
557 | struct nfs_open_context *ctx; |
558 | }; | |
559 | ||
560 | static int | |
561 | readpage_async_filler(void *data, struct page *page) | |
562 | { | |
563 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
564 | struct inode *inode = page->mapping->host; | |
565 | struct nfs_page *new; | |
566 | unsigned int len; | |
de05a0cc TM |
567 | int error; |
568 | ||
49a70f27 | 569 | len = nfs_page_length(page); |
1da177e4 LT |
570 | if (len == 0) |
571 | return nfs_return_empty_page(page); | |
de05a0cc | 572 | |
1da177e4 | 573 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
574 | if (IS_ERR(new)) |
575 | goto out_error; | |
576 | ||
1da177e4 | 577 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 578 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
579 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
580 | error = desc->pgio->pg_error; | |
581 | goto out_unlock; | |
582 | } | |
1da177e4 | 583 | return 0; |
de05a0cc TM |
584 | out_error: |
585 | error = PTR_ERR(new); | |
586 | SetPageError(page); | |
587 | out_unlock: | |
588 | unlock_page(page); | |
589 | return error; | |
1da177e4 LT |
590 | } |
591 | ||
592 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
593 | struct list_head *pages, unsigned nr_pages) | |
594 | { | |
8b09bee3 | 595 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 596 | struct nfs_readdesc desc = { |
8b09bee3 | 597 | .pgio = &pgio, |
1da177e4 LT |
598 | }; |
599 | struct inode *inode = mapping->host; | |
600 | struct nfs_server *server = NFS_SERVER(inode); | |
8b09bee3 TM |
601 | size_t rsize = server->rsize; |
602 | unsigned long npages; | |
5f004cf2 | 603 | int ret = -ESTALE; |
1da177e4 LT |
604 | |
605 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
606 | inode->i_sb->s_id, | |
607 | (long long)NFS_FILEID(inode), | |
608 | nr_pages); | |
91d5b470 | 609 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 610 | |
5f004cf2 TM |
611 | if (NFS_STALE(inode)) |
612 | goto out; | |
613 | ||
1da177e4 | 614 | if (filp == NULL) { |
d530838b | 615 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
616 | if (desc.ctx == NULL) |
617 | return -EBADF; | |
618 | } else | |
cd3758e3 | 619 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
620 | |
621 | /* attempt to read as many of the pages as possible from the cache | |
622 | * - this returns -ENOBUFS immediately if the cookie is negative | |
623 | */ | |
624 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
625 | pages, &nr_pages); | |
626 | if (ret == 0) | |
627 | goto read_complete; /* all pages were read */ | |
628 | ||
e5e94017 | 629 | pnfs_update_layout(inode, desc.ctx, IOMODE_READ); |
8b09bee3 TM |
630 | if (rsize < PAGE_CACHE_SIZE) |
631 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); | |
632 | else | |
633 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); | |
634 | ||
1da177e4 | 635 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
636 | |
637 | nfs_pageio_complete(&pgio); | |
638 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
639 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 640 | read_complete: |
1da177e4 | 641 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 642 | out: |
1da177e4 LT |
643 | return ret; |
644 | } | |
645 | ||
f7b422b1 | 646 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
647 | { |
648 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
649 | sizeof(struct nfs_read_data), | |
650 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 651 | NULL); |
1da177e4 LT |
652 | if (nfs_rdata_cachep == NULL) |
653 | return -ENOMEM; | |
654 | ||
93d2341c MD |
655 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
656 | nfs_rdata_cachep); | |
1da177e4 LT |
657 | if (nfs_rdata_mempool == NULL) |
658 | return -ENOMEM; | |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
266bee88 | 663 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
664 | { |
665 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 666 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 667 | } |