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 | |
8 | * | |
9 | * We do an ugly hack here in order to return proper error codes to the | |
10 | * user program when a read request failed: since generic_file_read | |
11 | * only checks the return value of inode->i_op->readpage() which is always 0 | |
12 | * for async RPC, we set the error bit of the page to 1 when an error occurs, | |
13 | * and make nfs_readpage transmit requests synchronously when encountering this. | |
14 | * This is only a small problem, though, since we now retry all operations | |
15 | * within the RPC code when root squashing is suspected. | |
16 | */ | |
17 | ||
1da177e4 LT |
18 | #include <linux/time.h> |
19 | #include <linux/kernel.h> | |
20 | #include <linux/errno.h> | |
21 | #include <linux/fcntl.h> | |
22 | #include <linux/stat.h> | |
23 | #include <linux/mm.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/sunrpc/clnt.h> | |
27 | #include <linux/nfs_fs.h> | |
28 | #include <linux/nfs_page.h> | |
29 | #include <linux/smp_lock.h> | |
30 | ||
31 | #include <asm/system.h> | |
32 | ||
91d5b470 CL |
33 | #include "iostat.h" |
34 | ||
1da177e4 LT |
35 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
36 | ||
37 | static int nfs_pagein_one(struct list_head *, struct inode *); | |
ec06c096 TM |
38 | static const struct rpc_call_ops nfs_read_partial_ops; |
39 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 LT |
40 | |
41 | static kmem_cache_t *nfs_rdata_cachep; | |
3feb2d49 | 42 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
43 | |
44 | #define MIN_POOL_READ (32) | |
45 | ||
e9f7bee1 | 46 | struct nfs_read_data *nfs_readdata_alloc(size_t len) |
3feb2d49 | 47 | { |
e9f7bee1 | 48 | unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
3feb2d49 TM |
49 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, SLAB_NOFS); |
50 | ||
51 | if (p) { | |
52 | memset(p, 0, sizeof(*p)); | |
53 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 54 | p->npages = pagecount; |
0d0b5cb3 CL |
55 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
56 | p->pagevec = p->page_array; | |
3feb2d49 | 57 | else { |
0d0b5cb3 CL |
58 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
59 | if (!p->pagevec) { | |
3feb2d49 TM |
60 | mempool_free(p, nfs_rdata_mempool); |
61 | p = NULL; | |
62 | } | |
63 | } | |
64 | } | |
65 | return p; | |
66 | } | |
67 | ||
e4e20512 | 68 | static void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
69 | { |
70 | if (p && (p->pagevec != &p->page_array[0])) | |
71 | kfree(p->pagevec); | |
72 | mempool_free(p, nfs_rdata_mempool); | |
73 | } | |
74 | ||
963d8fe5 | 75 | void nfs_readdata_release(void *data) |
1da177e4 | 76 | { |
1da177e4 LT |
77 | nfs_readdata_free(data); |
78 | } | |
79 | ||
80 | static | |
81 | unsigned int nfs_page_length(struct inode *inode, struct page *page) | |
82 | { | |
83 | loff_t i_size = i_size_read(inode); | |
84 | unsigned long idx; | |
85 | ||
86 | if (i_size <= 0) | |
87 | return 0; | |
88 | idx = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
89 | if (page->index > idx) | |
90 | return 0; | |
91 | if (page->index != idx) | |
92 | return PAGE_CACHE_SIZE; | |
93 | return 1 + ((i_size - 1) & (PAGE_CACHE_SIZE - 1)); | |
94 | } | |
95 | ||
96 | static | |
97 | int nfs_return_empty_page(struct page *page) | |
98 | { | |
99 | memclear_highpage_flush(page, 0, PAGE_CACHE_SIZE); | |
100 | SetPageUptodate(page); | |
101 | unlock_page(page); | |
102 | return 0; | |
103 | } | |
104 | ||
1de3fc12 TM |
105 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
106 | { | |
107 | unsigned int remainder = data->args.count - data->res.count; | |
108 | unsigned int base = data->args.pgbase + data->res.count; | |
109 | unsigned int pglen; | |
110 | struct page **pages; | |
111 | ||
112 | if (data->res.eof == 0 || remainder == 0) | |
113 | return; | |
114 | /* | |
115 | * Note: "remainder" can never be negative, since we check for | |
116 | * this in the XDR code. | |
117 | */ | |
118 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
119 | base &= ~PAGE_CACHE_MASK; | |
120 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
121 | for (;;) { |
122 | if (remainder <= pglen) { | |
123 | memclear_highpage_flush(*pages, base, remainder); | |
124 | break; | |
125 | } | |
1de3fc12 | 126 | memclear_highpage_flush(*pages, base, pglen); |
79558f36 TM |
127 | pages++; |
128 | remainder -= pglen; | |
129 | pglen = PAGE_CACHE_SIZE; | |
130 | base = 0; | |
131 | } | |
1de3fc12 TM |
132 | } |
133 | ||
1da177e4 LT |
134 | /* |
135 | * Read a page synchronously. | |
136 | */ | |
137 | static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode, | |
138 | struct page *page) | |
139 | { | |
140 | unsigned int rsize = NFS_SERVER(inode)->rsize; | |
141 | unsigned int count = PAGE_CACHE_SIZE; | |
142 | int result; | |
143 | struct nfs_read_data *rdata; | |
144 | ||
e9f7bee1 | 145 | rdata = nfs_readdata_alloc(count); |
1da177e4 LT |
146 | if (!rdata) |
147 | return -ENOMEM; | |
148 | ||
149 | memset(rdata, 0, sizeof(*rdata)); | |
150 | rdata->flags = (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0); | |
151 | rdata->cred = ctx->cred; | |
152 | rdata->inode = inode; | |
153 | INIT_LIST_HEAD(&rdata->pages); | |
154 | rdata->args.fh = NFS_FH(inode); | |
155 | rdata->args.context = ctx; | |
156 | rdata->args.pages = &page; | |
157 | rdata->args.pgbase = 0UL; | |
158 | rdata->args.count = rsize; | |
159 | rdata->res.fattr = &rdata->fattr; | |
160 | ||
161 | dprintk("NFS: nfs_readpage_sync(%p)\n", page); | |
162 | ||
163 | /* | |
164 | * This works now because the socket layer never tries to DMA | |
165 | * into this buffer directly. | |
166 | */ | |
167 | do { | |
168 | if (count < rsize) | |
169 | rdata->args.count = count; | |
170 | rdata->res.count = rdata->args.count; | |
171 | rdata->args.offset = page_offset(page) + rdata->args.pgbase; | |
172 | ||
173 | dprintk("NFS: nfs_proc_read(%s, (%s/%Ld), %Lu, %u)\n", | |
54ceac45 | 174 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
175 | inode->i_sb->s_id, |
176 | (long long)NFS_FILEID(inode), | |
177 | (unsigned long long)rdata->args.pgbase, | |
178 | rdata->args.count); | |
179 | ||
180 | lock_kernel(); | |
181 | result = NFS_PROTO(inode)->read(rdata); | |
182 | unlock_kernel(); | |
183 | ||
184 | /* | |
185 | * Even if we had a partial success we can't mark the page | |
186 | * cache valid. | |
187 | */ | |
188 | if (result < 0) { | |
189 | if (result == -EISDIR) | |
190 | result = -EINVAL; | |
191 | goto io_error; | |
192 | } | |
193 | count -= result; | |
194 | rdata->args.pgbase += result; | |
91d5b470 CL |
195 | nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, result); |
196 | ||
1da177e4 LT |
197 | /* Note: result == 0 should only happen if we're caching |
198 | * a write that extends the file and punches a hole. | |
199 | */ | |
200 | if (rdata->res.eof != 0 || result == 0) | |
201 | break; | |
202 | } while (count); | |
dc59250c | 203 | spin_lock(&inode->i_lock); |
55296809 | 204 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 205 | spin_unlock(&inode->i_lock); |
1da177e4 | 206 | |
7a524111 | 207 | if (rdata->res.eof || rdata->res.count == rdata->args.count) { |
1de3fc12 | 208 | SetPageUptodate(page); |
7a524111 TM |
209 | if (rdata->res.eof && count != 0) |
210 | memclear_highpage_flush(page, rdata->args.pgbase, count); | |
211 | } | |
1da177e4 LT |
212 | result = 0; |
213 | ||
214 | io_error: | |
215 | unlock_page(page); | |
216 | nfs_readdata_free(rdata); | |
217 | return result; | |
218 | } | |
219 | ||
220 | static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, | |
221 | struct page *page) | |
222 | { | |
223 | LIST_HEAD(one_request); | |
224 | struct nfs_page *new; | |
225 | unsigned int len; | |
226 | ||
227 | len = nfs_page_length(inode, page); | |
228 | if (len == 0) | |
229 | return nfs_return_empty_page(page); | |
230 | new = nfs_create_request(ctx, inode, page, 0, len); | |
231 | if (IS_ERR(new)) { | |
232 | unlock_page(page); | |
233 | return PTR_ERR(new); | |
234 | } | |
235 | if (len < PAGE_CACHE_SIZE) | |
236 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
237 | ||
1da177e4 LT |
238 | nfs_list_add_request(new, &one_request); |
239 | nfs_pagein_one(&one_request, inode); | |
240 | return 0; | |
241 | } | |
242 | ||
243 | static void nfs_readpage_release(struct nfs_page *req) | |
244 | { | |
245 | unlock_page(req->wb_page); | |
246 | ||
1da177e4 LT |
247 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
248 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
249 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
250 | req->wb_bytes, | |
251 | (long long)req_offset(req)); | |
10d2c46f NW |
252 | nfs_clear_request(req); |
253 | nfs_release_request(req); | |
1da177e4 LT |
254 | } |
255 | ||
256 | /* | |
257 | * Set up the NFS read request struct | |
258 | */ | |
259 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, | |
ec06c096 | 260 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
261 | unsigned int count, unsigned int offset) |
262 | { | |
263 | struct inode *inode; | |
ec06c096 | 264 | int flags; |
1da177e4 LT |
265 | |
266 | data->req = req; | |
267 | data->inode = inode = req->wb_context->dentry->d_inode; | |
268 | data->cred = req->wb_context->cred; | |
269 | ||
270 | data->args.fh = NFS_FH(inode); | |
271 | data->args.offset = req_offset(req) + offset; | |
272 | data->args.pgbase = req->wb_pgbase + offset; | |
273 | data->args.pages = data->pagevec; | |
274 | data->args.count = count; | |
275 | data->args.context = req->wb_context; | |
276 | ||
277 | data->res.fattr = &data->fattr; | |
278 | data->res.count = count; | |
279 | data->res.eof = 0; | |
0e574af1 | 280 | nfs_fattr_init(&data->fattr); |
1da177e4 | 281 | |
ec06c096 TM |
282 | /* Set up the initial task struct. */ |
283 | flags = RPC_TASK_ASYNC | (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0); | |
284 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); | |
1da177e4 LT |
285 | NFS_PROTO(inode)->read_setup(data); |
286 | ||
287 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 LT |
288 | |
289 | dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", | |
290 | data->task.tk_pid, | |
291 | inode->i_sb->s_id, | |
292 | (long long)NFS_FILEID(inode), | |
293 | count, | |
294 | (unsigned long long)data->args.offset); | |
295 | } | |
296 | ||
297 | static void | |
298 | nfs_async_read_error(struct list_head *head) | |
299 | { | |
300 | struct nfs_page *req; | |
301 | ||
302 | while (!list_empty(head)) { | |
303 | req = nfs_list_entry(head->next); | |
304 | nfs_list_remove_request(req); | |
305 | SetPageError(req->wb_page); | |
306 | nfs_readpage_release(req); | |
307 | } | |
308 | } | |
309 | ||
310 | /* | |
311 | * Start an async read operation | |
312 | */ | |
313 | static void nfs_execute_read(struct nfs_read_data *data) | |
314 | { | |
315 | struct rpc_clnt *clnt = NFS_CLIENT(data->inode); | |
316 | sigset_t oldset; | |
317 | ||
318 | rpc_clnt_sigmask(clnt, &oldset); | |
319 | lock_kernel(); | |
320 | rpc_execute(&data->task); | |
321 | unlock_kernel(); | |
322 | rpc_clnt_sigunmask(clnt, &oldset); | |
323 | } | |
324 | ||
325 | /* | |
326 | * Generate multiple requests to fill a single page. | |
327 | * | |
328 | * We optimize to reduce the number of read operations on the wire. If we | |
329 | * detect that we're reading a page, or an area of a page, that is past the | |
330 | * end of file, we do not generate NFS read operations but just clear the | |
331 | * parts of the page that would have come back zero from the server anyway. | |
332 | * | |
333 | * We rely on the cached value of i_size to make this determination; another | |
334 | * client can fill pages on the server past our cached end-of-file, but we | |
335 | * won't see the new data until our attribute cache is updated. This is more | |
336 | * or less conventional NFS client behavior. | |
337 | */ | |
338 | static int nfs_pagein_multi(struct list_head *head, struct inode *inode) | |
339 | { | |
340 | struct nfs_page *req = nfs_list_entry(head->next); | |
341 | struct page *page = req->wb_page; | |
342 | struct nfs_read_data *data; | |
e9f7bee1 TM |
343 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
344 | unsigned int offset; | |
1da177e4 LT |
345 | int requests = 0; |
346 | LIST_HEAD(list); | |
347 | ||
348 | nfs_list_remove_request(req); | |
349 | ||
350 | nbytes = req->wb_bytes; | |
e9f7bee1 TM |
351 | do { |
352 | size_t len = min(nbytes,rsize); | |
353 | ||
354 | data = nfs_readdata_alloc(len); | |
1da177e4 LT |
355 | if (!data) |
356 | goto out_bad; | |
357 | INIT_LIST_HEAD(&data->pages); | |
358 | list_add(&data->pages, &list); | |
359 | requests++; | |
e9f7bee1 TM |
360 | nbytes -= len; |
361 | } while(nbytes != 0); | |
1da177e4 LT |
362 | atomic_set(&req->wb_complete, requests); |
363 | ||
364 | ClearPageError(page); | |
365 | offset = 0; | |
366 | nbytes = req->wb_bytes; | |
367 | do { | |
368 | data = list_entry(list.next, struct nfs_read_data, pages); | |
369 | list_del_init(&data->pages); | |
370 | ||
371 | data->pagevec[0] = page; | |
1da177e4 LT |
372 | |
373 | if (nbytes > rsize) { | |
ec06c096 TM |
374 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
375 | rsize, offset); | |
1da177e4 LT |
376 | offset += rsize; |
377 | nbytes -= rsize; | |
378 | } else { | |
ec06c096 TM |
379 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
380 | nbytes, offset); | |
1da177e4 LT |
381 | nbytes = 0; |
382 | } | |
383 | nfs_execute_read(data); | |
384 | } while (nbytes != 0); | |
385 | ||
386 | return 0; | |
387 | ||
388 | out_bad: | |
389 | while (!list_empty(&list)) { | |
390 | data = list_entry(list.next, struct nfs_read_data, pages); | |
391 | list_del(&data->pages); | |
392 | nfs_readdata_free(data); | |
393 | } | |
394 | SetPageError(page); | |
395 | nfs_readpage_release(req); | |
396 | return -ENOMEM; | |
397 | } | |
398 | ||
399 | static int nfs_pagein_one(struct list_head *head, struct inode *inode) | |
400 | { | |
401 | struct nfs_page *req; | |
402 | struct page **pages; | |
403 | struct nfs_read_data *data; | |
404 | unsigned int count; | |
405 | ||
406 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) | |
407 | return nfs_pagein_multi(head, inode); | |
408 | ||
e9f7bee1 | 409 | data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize); |
1da177e4 LT |
410 | if (!data) |
411 | goto out_bad; | |
412 | ||
413 | INIT_LIST_HEAD(&data->pages); | |
414 | pages = data->pagevec; | |
415 | count = 0; | |
416 | while (!list_empty(head)) { | |
417 | req = nfs_list_entry(head->next); | |
418 | nfs_list_remove_request(req); | |
419 | nfs_list_add_request(req, &data->pages); | |
420 | ClearPageError(req->wb_page); | |
421 | *pages++ = req->wb_page; | |
422 | count += req->wb_bytes; | |
423 | } | |
424 | req = nfs_list_entry(data->pages.next); | |
425 | ||
ec06c096 | 426 | nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
1da177e4 LT |
427 | |
428 | nfs_execute_read(data); | |
429 | return 0; | |
430 | out_bad: | |
431 | nfs_async_read_error(head); | |
432 | return -ENOMEM; | |
433 | } | |
434 | ||
435 | static int | |
436 | nfs_pagein_list(struct list_head *head, int rpages) | |
437 | { | |
438 | LIST_HEAD(one_request); | |
439 | struct nfs_page *req; | |
440 | int error = 0; | |
441 | unsigned int pages = 0; | |
442 | ||
443 | while (!list_empty(head)) { | |
444 | pages += nfs_coalesce_requests(head, &one_request, rpages); | |
445 | req = nfs_list_entry(one_request.next); | |
446 | error = nfs_pagein_one(&one_request, req->wb_context->dentry->d_inode); | |
447 | if (error < 0) | |
448 | break; | |
449 | } | |
450 | if (error >= 0) | |
451 | return pages; | |
452 | ||
453 | nfs_async_read_error(head); | |
454 | return error; | |
455 | } | |
456 | ||
457 | /* | |
458 | * Handle a read reply that fills part of a page. | |
459 | */ | |
ec06c096 | 460 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 461 | { |
ec06c096 | 462 | struct nfs_read_data *data = calldata; |
1da177e4 LT |
463 | struct nfs_page *req = data->req; |
464 | struct page *page = req->wb_page; | |
465 | ||
1de3fc12 TM |
466 | if (likely(task->tk_status >= 0)) |
467 | nfs_readpage_truncate_uninitialised_page(data); | |
468 | else | |
469 | SetPageError(page); | |
ec06c096 TM |
470 | if (nfs_readpage_result(task, data) != 0) |
471 | return; | |
1da177e4 LT |
472 | if (atomic_dec_and_test(&req->wb_complete)) { |
473 | if (!PageError(page)) | |
474 | SetPageUptodate(page); | |
475 | nfs_readpage_release(req); | |
476 | } | |
477 | } | |
478 | ||
ec06c096 TM |
479 | static const struct rpc_call_ops nfs_read_partial_ops = { |
480 | .rpc_call_done = nfs_readpage_result_partial, | |
481 | .rpc_release = nfs_readdata_release, | |
482 | }; | |
483 | ||
1de3fc12 TM |
484 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
485 | { | |
486 | unsigned int count = data->res.count; | |
487 | unsigned int base = data->args.pgbase; | |
488 | struct page **pages; | |
489 | ||
79558f36 TM |
490 | if (data->res.eof) |
491 | count = data->args.count; | |
1de3fc12 TM |
492 | if (unlikely(count == 0)) |
493 | return; | |
494 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
495 | base &= ~PAGE_CACHE_MASK; | |
496 | count += base; | |
497 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
498 | SetPageUptodate(*pages); | |
79558f36 | 499 | if (count != 0) |
1de3fc12 TM |
500 | SetPageUptodate(*pages); |
501 | } | |
502 | ||
503 | static void nfs_readpage_set_pages_error(struct nfs_read_data *data) | |
504 | { | |
505 | unsigned int count = data->args.count; | |
506 | unsigned int base = data->args.pgbase; | |
507 | struct page **pages; | |
508 | ||
509 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
510 | base &= ~PAGE_CACHE_MASK; | |
511 | count += base; | |
512 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
513 | SetPageError(*pages); | |
79558f36 TM |
514 | if (count != 0) |
515 | SetPageError(*pages); | |
1de3fc12 TM |
516 | } |
517 | ||
1da177e4 LT |
518 | /* |
519 | * This is the callback from RPC telling us whether a reply was | |
520 | * received or some error occurred (timeout or socket shutdown). | |
521 | */ | |
ec06c096 | 522 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 523 | { |
ec06c096 | 524 | struct nfs_read_data *data = calldata; |
1da177e4 | 525 | |
1de3fc12 TM |
526 | /* |
527 | * Note: nfs_readpage_result may change the values of | |
528 | * data->args. In the multi-page case, we therefore need | |
529 | * to ensure that we call the next nfs_readpage_set_page_uptodate() | |
530 | * first in the multi-page case. | |
531 | */ | |
532 | if (likely(task->tk_status >= 0)) { | |
533 | nfs_readpage_truncate_uninitialised_page(data); | |
534 | nfs_readpage_set_pages_uptodate(data); | |
535 | } else | |
536 | nfs_readpage_set_pages_error(data); | |
ec06c096 TM |
537 | if (nfs_readpage_result(task, data) != 0) |
538 | return; | |
1da177e4 LT |
539 | while (!list_empty(&data->pages)) { |
540 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 541 | |
1de3fc12 | 542 | nfs_list_remove_request(req); |
1da177e4 LT |
543 | nfs_readpage_release(req); |
544 | } | |
545 | } | |
546 | ||
ec06c096 TM |
547 | static const struct rpc_call_ops nfs_read_full_ops = { |
548 | .rpc_call_done = nfs_readpage_result_full, | |
549 | .rpc_release = nfs_readdata_release, | |
550 | }; | |
551 | ||
1da177e4 LT |
552 | /* |
553 | * This is the callback from RPC telling us whether a reply was | |
554 | * received or some error occurred (timeout or socket shutdown). | |
555 | */ | |
ec06c096 | 556 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) |
1da177e4 | 557 | { |
1da177e4 LT |
558 | struct nfs_readargs *argp = &data->args; |
559 | struct nfs_readres *resp = &data->res; | |
ec06c096 | 560 | int status; |
1da177e4 LT |
561 | |
562 | dprintk("NFS: %4d nfs_readpage_result, (status %d)\n", | |
ec06c096 TM |
563 | task->tk_pid, task->tk_status); |
564 | ||
565 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
566 | if (status != 0) | |
567 | return status; | |
1da177e4 | 568 | |
91d5b470 CL |
569 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, resp->count); |
570 | ||
1da177e4 LT |
571 | /* Is this a short read? */ |
572 | if (task->tk_status >= 0 && resp->count < argp->count && !resp->eof) { | |
91d5b470 | 573 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); |
1da177e4 LT |
574 | /* Has the server at least made some progress? */ |
575 | if (resp->count != 0) { | |
576 | /* Yes, so retry the read at the end of the data */ | |
577 | argp->offset += resp->count; | |
578 | argp->pgbase += resp->count; | |
579 | argp->count -= resp->count; | |
580 | rpc_restart_call(task); | |
ec06c096 | 581 | return -EAGAIN; |
1da177e4 LT |
582 | } |
583 | task->tk_status = -EIO; | |
584 | } | |
dc59250c | 585 | spin_lock(&data->inode->i_lock); |
55296809 | 586 | NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 587 | spin_unlock(&data->inode->i_lock); |
ec06c096 | 588 | return 0; |
1da177e4 LT |
589 | } |
590 | ||
591 | /* | |
592 | * Read a page over NFS. | |
593 | * We read the page synchronously in the following case: | |
594 | * - The error flag is set for this page. This happens only when a | |
595 | * previous async read operation failed. | |
596 | */ | |
597 | int nfs_readpage(struct file *file, struct page *page) | |
598 | { | |
599 | struct nfs_open_context *ctx; | |
600 | struct inode *inode = page->mapping->host; | |
601 | int error; | |
602 | ||
603 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
604 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
605 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
606 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
607 | ||
1da177e4 LT |
608 | /* |
609 | * Try to flush any pending writes to the file.. | |
610 | * | |
611 | * NOTE! Because we own the page lock, there cannot | |
612 | * be any new pending writes generated at this point | |
613 | * for this page (other pages can be written to). | |
614 | */ | |
615 | error = nfs_wb_page(inode, page); | |
616 | if (error) | |
617 | goto out_error; | |
618 | ||
619 | if (file == NULL) { | |
d530838b | 620 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
621 | if (ctx == NULL) |
622 | return -EBADF; | |
623 | } else | |
624 | ctx = get_nfs_open_context((struct nfs_open_context *) | |
625 | file->private_data); | |
626 | if (!IS_SYNC(inode)) { | |
627 | error = nfs_readpage_async(ctx, inode, page); | |
628 | goto out; | |
629 | } | |
630 | ||
631 | error = nfs_readpage_sync(ctx, inode, page); | |
632 | if (error < 0 && IS_SWAPFILE(inode)) | |
633 | printk("Aiee.. nfs swap-in of page failed!\n"); | |
634 | out: | |
635 | put_nfs_open_context(ctx); | |
636 | return error; | |
637 | ||
638 | out_error: | |
639 | unlock_page(page); | |
640 | return error; | |
641 | } | |
642 | ||
643 | struct nfs_readdesc { | |
644 | struct list_head *head; | |
645 | struct nfs_open_context *ctx; | |
646 | }; | |
647 | ||
648 | static int | |
649 | readpage_async_filler(void *data, struct page *page) | |
650 | { | |
651 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
652 | struct inode *inode = page->mapping->host; | |
653 | struct nfs_page *new; | |
654 | unsigned int len; | |
655 | ||
656 | nfs_wb_page(inode, page); | |
657 | len = nfs_page_length(inode, page); | |
658 | if (len == 0) | |
659 | return nfs_return_empty_page(page); | |
660 | new = nfs_create_request(desc->ctx, inode, page, 0, len); | |
661 | if (IS_ERR(new)) { | |
662 | SetPageError(page); | |
663 | unlock_page(page); | |
664 | return PTR_ERR(new); | |
665 | } | |
666 | if (len < PAGE_CACHE_SIZE) | |
667 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
1da177e4 LT |
668 | nfs_list_add_request(new, desc->head); |
669 | return 0; | |
670 | } | |
671 | ||
672 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
673 | struct list_head *pages, unsigned nr_pages) | |
674 | { | |
675 | LIST_HEAD(head); | |
676 | struct nfs_readdesc desc = { | |
677 | .head = &head, | |
678 | }; | |
679 | struct inode *inode = mapping->host; | |
680 | struct nfs_server *server = NFS_SERVER(inode); | |
681 | int ret; | |
682 | ||
683 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
684 | inode->i_sb->s_id, | |
685 | (long long)NFS_FILEID(inode), | |
686 | nr_pages); | |
91d5b470 | 687 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 LT |
688 | |
689 | if (filp == NULL) { | |
d530838b | 690 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
691 | if (desc.ctx == NULL) |
692 | return -EBADF; | |
693 | } else | |
694 | desc.ctx = get_nfs_open_context((struct nfs_open_context *) | |
695 | filp->private_data); | |
696 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); | |
697 | if (!list_empty(&head)) { | |
698 | int err = nfs_pagein_list(&head, server->rpages); | |
699 | if (!ret) | |
91d5b470 | 700 | nfs_add_stats(inode, NFSIOS_READPAGES, err); |
1da177e4 LT |
701 | ret = err; |
702 | } | |
703 | put_nfs_open_context(desc.ctx); | |
704 | return ret; | |
705 | } | |
706 | ||
f7b422b1 | 707 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
708 | { |
709 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
710 | sizeof(struct nfs_read_data), | |
711 | 0, SLAB_HWCACHE_ALIGN, | |
712 | NULL, NULL); | |
713 | if (nfs_rdata_cachep == NULL) | |
714 | return -ENOMEM; | |
715 | ||
93d2341c MD |
716 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
717 | nfs_rdata_cachep); | |
1da177e4 LT |
718 | if (nfs_rdata_mempool == NULL) |
719 | return -ENOMEM; | |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
266bee88 | 724 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
725 | { |
726 | mempool_destroy(nfs_rdata_mempool); | |
727 | if (kmem_cache_destroy(nfs_rdata_cachep)) | |
728 | printk(KERN_INFO "nfs_read_data: not all structures were freed\n"); | |
729 | } |