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", | |
174 | NFS_SERVER(inode)->hostname, | |
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 | |
1de3fc12 TM |
207 | nfs_readpage_truncate_uninitialised_page(rdata); |
208 | if (rdata->res.eof || rdata->res.count == rdata->args.count) | |
209 | SetPageUptodate(page); | |
1da177e4 LT |
210 | result = 0; |
211 | ||
212 | io_error: | |
213 | unlock_page(page); | |
214 | nfs_readdata_free(rdata); | |
215 | return result; | |
216 | } | |
217 | ||
218 | static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, | |
219 | struct page *page) | |
220 | { | |
221 | LIST_HEAD(one_request); | |
222 | struct nfs_page *new; | |
223 | unsigned int len; | |
224 | ||
225 | len = nfs_page_length(inode, page); | |
226 | if (len == 0) | |
227 | return nfs_return_empty_page(page); | |
228 | new = nfs_create_request(ctx, inode, page, 0, len); | |
229 | if (IS_ERR(new)) { | |
230 | unlock_page(page); | |
231 | return PTR_ERR(new); | |
232 | } | |
233 | if (len < PAGE_CACHE_SIZE) | |
234 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
235 | ||
1da177e4 LT |
236 | nfs_list_add_request(new, &one_request); |
237 | nfs_pagein_one(&one_request, inode); | |
238 | return 0; | |
239 | } | |
240 | ||
241 | static void nfs_readpage_release(struct nfs_page *req) | |
242 | { | |
243 | unlock_page(req->wb_page); | |
244 | ||
1da177e4 LT |
245 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
246 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
247 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
248 | req->wb_bytes, | |
249 | (long long)req_offset(req)); | |
10d2c46f NW |
250 | nfs_clear_request(req); |
251 | nfs_release_request(req); | |
1da177e4 LT |
252 | } |
253 | ||
254 | /* | |
255 | * Set up the NFS read request struct | |
256 | */ | |
257 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, | |
ec06c096 | 258 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
259 | unsigned int count, unsigned int offset) |
260 | { | |
261 | struct inode *inode; | |
ec06c096 | 262 | int flags; |
1da177e4 LT |
263 | |
264 | data->req = req; | |
265 | data->inode = inode = req->wb_context->dentry->d_inode; | |
266 | data->cred = req->wb_context->cred; | |
267 | ||
268 | data->args.fh = NFS_FH(inode); | |
269 | data->args.offset = req_offset(req) + offset; | |
270 | data->args.pgbase = req->wb_pgbase + offset; | |
271 | data->args.pages = data->pagevec; | |
272 | data->args.count = count; | |
273 | data->args.context = req->wb_context; | |
274 | ||
275 | data->res.fattr = &data->fattr; | |
276 | data->res.count = count; | |
277 | data->res.eof = 0; | |
0e574af1 | 278 | nfs_fattr_init(&data->fattr); |
1da177e4 | 279 | |
ec06c096 TM |
280 | /* Set up the initial task struct. */ |
281 | flags = RPC_TASK_ASYNC | (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0); | |
282 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); | |
1da177e4 LT |
283 | NFS_PROTO(inode)->read_setup(data); |
284 | ||
285 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 LT |
286 | |
287 | dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", | |
288 | data->task.tk_pid, | |
289 | inode->i_sb->s_id, | |
290 | (long long)NFS_FILEID(inode), | |
291 | count, | |
292 | (unsigned long long)data->args.offset); | |
293 | } | |
294 | ||
295 | static void | |
296 | nfs_async_read_error(struct list_head *head) | |
297 | { | |
298 | struct nfs_page *req; | |
299 | ||
300 | while (!list_empty(head)) { | |
301 | req = nfs_list_entry(head->next); | |
302 | nfs_list_remove_request(req); | |
303 | SetPageError(req->wb_page); | |
304 | nfs_readpage_release(req); | |
305 | } | |
306 | } | |
307 | ||
308 | /* | |
309 | * Start an async read operation | |
310 | */ | |
311 | static void nfs_execute_read(struct nfs_read_data *data) | |
312 | { | |
313 | struct rpc_clnt *clnt = NFS_CLIENT(data->inode); | |
314 | sigset_t oldset; | |
315 | ||
316 | rpc_clnt_sigmask(clnt, &oldset); | |
317 | lock_kernel(); | |
318 | rpc_execute(&data->task); | |
319 | unlock_kernel(); | |
320 | rpc_clnt_sigunmask(clnt, &oldset); | |
321 | } | |
322 | ||
323 | /* | |
324 | * Generate multiple requests to fill a single page. | |
325 | * | |
326 | * We optimize to reduce the number of read operations on the wire. If we | |
327 | * detect that we're reading a page, or an area of a page, that is past the | |
328 | * end of file, we do not generate NFS read operations but just clear the | |
329 | * parts of the page that would have come back zero from the server anyway. | |
330 | * | |
331 | * We rely on the cached value of i_size to make this determination; another | |
332 | * client can fill pages on the server past our cached end-of-file, but we | |
333 | * won't see the new data until our attribute cache is updated. This is more | |
334 | * or less conventional NFS client behavior. | |
335 | */ | |
336 | static int nfs_pagein_multi(struct list_head *head, struct inode *inode) | |
337 | { | |
338 | struct nfs_page *req = nfs_list_entry(head->next); | |
339 | struct page *page = req->wb_page; | |
340 | struct nfs_read_data *data; | |
e9f7bee1 TM |
341 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
342 | unsigned int offset; | |
1da177e4 LT |
343 | int requests = 0; |
344 | LIST_HEAD(list); | |
345 | ||
346 | nfs_list_remove_request(req); | |
347 | ||
348 | nbytes = req->wb_bytes; | |
e9f7bee1 TM |
349 | do { |
350 | size_t len = min(nbytes,rsize); | |
351 | ||
352 | data = nfs_readdata_alloc(len); | |
1da177e4 LT |
353 | if (!data) |
354 | goto out_bad; | |
355 | INIT_LIST_HEAD(&data->pages); | |
356 | list_add(&data->pages, &list); | |
357 | requests++; | |
e9f7bee1 TM |
358 | nbytes -= len; |
359 | } while(nbytes != 0); | |
1da177e4 LT |
360 | atomic_set(&req->wb_complete, requests); |
361 | ||
362 | ClearPageError(page); | |
363 | offset = 0; | |
364 | nbytes = req->wb_bytes; | |
365 | do { | |
366 | data = list_entry(list.next, struct nfs_read_data, pages); | |
367 | list_del_init(&data->pages); | |
368 | ||
369 | data->pagevec[0] = page; | |
1da177e4 LT |
370 | |
371 | if (nbytes > rsize) { | |
ec06c096 TM |
372 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
373 | rsize, offset); | |
1da177e4 LT |
374 | offset += rsize; |
375 | nbytes -= rsize; | |
376 | } else { | |
ec06c096 TM |
377 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
378 | nbytes, offset); | |
1da177e4 LT |
379 | nbytes = 0; |
380 | } | |
381 | nfs_execute_read(data); | |
382 | } while (nbytes != 0); | |
383 | ||
384 | return 0; | |
385 | ||
386 | out_bad: | |
387 | while (!list_empty(&list)) { | |
388 | data = list_entry(list.next, struct nfs_read_data, pages); | |
389 | list_del(&data->pages); | |
390 | nfs_readdata_free(data); | |
391 | } | |
392 | SetPageError(page); | |
393 | nfs_readpage_release(req); | |
394 | return -ENOMEM; | |
395 | } | |
396 | ||
397 | static int nfs_pagein_one(struct list_head *head, struct inode *inode) | |
398 | { | |
399 | struct nfs_page *req; | |
400 | struct page **pages; | |
401 | struct nfs_read_data *data; | |
402 | unsigned int count; | |
403 | ||
404 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) | |
405 | return nfs_pagein_multi(head, inode); | |
406 | ||
e9f7bee1 | 407 | data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize); |
1da177e4 LT |
408 | if (!data) |
409 | goto out_bad; | |
410 | ||
411 | INIT_LIST_HEAD(&data->pages); | |
412 | pages = data->pagevec; | |
413 | count = 0; | |
414 | while (!list_empty(head)) { | |
415 | req = nfs_list_entry(head->next); | |
416 | nfs_list_remove_request(req); | |
417 | nfs_list_add_request(req, &data->pages); | |
418 | ClearPageError(req->wb_page); | |
419 | *pages++ = req->wb_page; | |
420 | count += req->wb_bytes; | |
421 | } | |
422 | req = nfs_list_entry(data->pages.next); | |
423 | ||
ec06c096 | 424 | nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
1da177e4 LT |
425 | |
426 | nfs_execute_read(data); | |
427 | return 0; | |
428 | out_bad: | |
429 | nfs_async_read_error(head); | |
430 | return -ENOMEM; | |
431 | } | |
432 | ||
433 | static int | |
434 | nfs_pagein_list(struct list_head *head, int rpages) | |
435 | { | |
436 | LIST_HEAD(one_request); | |
437 | struct nfs_page *req; | |
438 | int error = 0; | |
439 | unsigned int pages = 0; | |
440 | ||
441 | while (!list_empty(head)) { | |
442 | pages += nfs_coalesce_requests(head, &one_request, rpages); | |
443 | req = nfs_list_entry(one_request.next); | |
444 | error = nfs_pagein_one(&one_request, req->wb_context->dentry->d_inode); | |
445 | if (error < 0) | |
446 | break; | |
447 | } | |
448 | if (error >= 0) | |
449 | return pages; | |
450 | ||
451 | nfs_async_read_error(head); | |
452 | return error; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Handle a read reply that fills part of a page. | |
457 | */ | |
ec06c096 | 458 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 459 | { |
ec06c096 | 460 | struct nfs_read_data *data = calldata; |
1da177e4 LT |
461 | struct nfs_page *req = data->req; |
462 | struct page *page = req->wb_page; | |
463 | ||
1de3fc12 TM |
464 | if (likely(task->tk_status >= 0)) |
465 | nfs_readpage_truncate_uninitialised_page(data); | |
466 | else | |
467 | SetPageError(page); | |
ec06c096 TM |
468 | if (nfs_readpage_result(task, data) != 0) |
469 | return; | |
1da177e4 LT |
470 | if (atomic_dec_and_test(&req->wb_complete)) { |
471 | if (!PageError(page)) | |
472 | SetPageUptodate(page); | |
473 | nfs_readpage_release(req); | |
474 | } | |
475 | } | |
476 | ||
ec06c096 TM |
477 | static const struct rpc_call_ops nfs_read_partial_ops = { |
478 | .rpc_call_done = nfs_readpage_result_partial, | |
479 | .rpc_release = nfs_readdata_release, | |
480 | }; | |
481 | ||
1de3fc12 TM |
482 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
483 | { | |
484 | unsigned int count = data->res.count; | |
485 | unsigned int base = data->args.pgbase; | |
486 | struct page **pages; | |
487 | ||
79558f36 TM |
488 | if (data->res.eof) |
489 | count = data->args.count; | |
1de3fc12 TM |
490 | if (unlikely(count == 0)) |
491 | return; | |
492 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
493 | base &= ~PAGE_CACHE_MASK; | |
494 | count += base; | |
495 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
496 | SetPageUptodate(*pages); | |
79558f36 | 497 | if (count != 0) |
1de3fc12 TM |
498 | SetPageUptodate(*pages); |
499 | } | |
500 | ||
501 | static void nfs_readpage_set_pages_error(struct nfs_read_data *data) | |
502 | { | |
503 | unsigned int count = data->args.count; | |
504 | unsigned int base = data->args.pgbase; | |
505 | struct page **pages; | |
506 | ||
507 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
508 | base &= ~PAGE_CACHE_MASK; | |
509 | count += base; | |
510 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
511 | SetPageError(*pages); | |
79558f36 TM |
512 | if (count != 0) |
513 | SetPageError(*pages); | |
1de3fc12 TM |
514 | } |
515 | ||
1da177e4 LT |
516 | /* |
517 | * This is the callback from RPC telling us whether a reply was | |
518 | * received or some error occurred (timeout or socket shutdown). | |
519 | */ | |
ec06c096 | 520 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 521 | { |
ec06c096 | 522 | struct nfs_read_data *data = calldata; |
1da177e4 | 523 | |
1de3fc12 TM |
524 | /* |
525 | * Note: nfs_readpage_result may change the values of | |
526 | * data->args. In the multi-page case, we therefore need | |
527 | * to ensure that we call the next nfs_readpage_set_page_uptodate() | |
528 | * first in the multi-page case. | |
529 | */ | |
530 | if (likely(task->tk_status >= 0)) { | |
531 | nfs_readpage_truncate_uninitialised_page(data); | |
532 | nfs_readpage_set_pages_uptodate(data); | |
533 | } else | |
534 | nfs_readpage_set_pages_error(data); | |
ec06c096 TM |
535 | if (nfs_readpage_result(task, data) != 0) |
536 | return; | |
1da177e4 LT |
537 | while (!list_empty(&data->pages)) { |
538 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 539 | |
1de3fc12 | 540 | nfs_list_remove_request(req); |
1da177e4 LT |
541 | nfs_readpage_release(req); |
542 | } | |
543 | } | |
544 | ||
ec06c096 TM |
545 | static const struct rpc_call_ops nfs_read_full_ops = { |
546 | .rpc_call_done = nfs_readpage_result_full, | |
547 | .rpc_release = nfs_readdata_release, | |
548 | }; | |
549 | ||
1da177e4 LT |
550 | /* |
551 | * This is the callback from RPC telling us whether a reply was | |
552 | * received or some error occurred (timeout or socket shutdown). | |
553 | */ | |
ec06c096 | 554 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) |
1da177e4 | 555 | { |
1da177e4 LT |
556 | struct nfs_readargs *argp = &data->args; |
557 | struct nfs_readres *resp = &data->res; | |
ec06c096 | 558 | int status; |
1da177e4 LT |
559 | |
560 | dprintk("NFS: %4d nfs_readpage_result, (status %d)\n", | |
ec06c096 TM |
561 | task->tk_pid, task->tk_status); |
562 | ||
563 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
564 | if (status != 0) | |
565 | return status; | |
1da177e4 | 566 | |
91d5b470 CL |
567 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, resp->count); |
568 | ||
1da177e4 LT |
569 | /* Is this a short read? */ |
570 | if (task->tk_status >= 0 && resp->count < argp->count && !resp->eof) { | |
91d5b470 | 571 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); |
1da177e4 LT |
572 | /* Has the server at least made some progress? */ |
573 | if (resp->count != 0) { | |
574 | /* Yes, so retry the read at the end of the data */ | |
575 | argp->offset += resp->count; | |
576 | argp->pgbase += resp->count; | |
577 | argp->count -= resp->count; | |
578 | rpc_restart_call(task); | |
ec06c096 | 579 | return -EAGAIN; |
1da177e4 LT |
580 | } |
581 | task->tk_status = -EIO; | |
582 | } | |
dc59250c | 583 | spin_lock(&data->inode->i_lock); |
55296809 | 584 | NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 585 | spin_unlock(&data->inode->i_lock); |
ec06c096 | 586 | return 0; |
1da177e4 LT |
587 | } |
588 | ||
589 | /* | |
590 | * Read a page over NFS. | |
591 | * We read the page synchronously in the following case: | |
592 | * - The error flag is set for this page. This happens only when a | |
593 | * previous async read operation failed. | |
594 | */ | |
595 | int nfs_readpage(struct file *file, struct page *page) | |
596 | { | |
597 | struct nfs_open_context *ctx; | |
598 | struct inode *inode = page->mapping->host; | |
599 | int error; | |
600 | ||
601 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
602 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
603 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
604 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
605 | ||
1da177e4 LT |
606 | /* |
607 | * Try to flush any pending writes to the file.. | |
608 | * | |
609 | * NOTE! Because we own the page lock, there cannot | |
610 | * be any new pending writes generated at this point | |
611 | * for this page (other pages can be written to). | |
612 | */ | |
613 | error = nfs_wb_page(inode, page); | |
614 | if (error) | |
615 | goto out_error; | |
616 | ||
617 | if (file == NULL) { | |
d530838b | 618 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
619 | if (ctx == NULL) |
620 | return -EBADF; | |
621 | } else | |
622 | ctx = get_nfs_open_context((struct nfs_open_context *) | |
623 | file->private_data); | |
624 | if (!IS_SYNC(inode)) { | |
625 | error = nfs_readpage_async(ctx, inode, page); | |
626 | goto out; | |
627 | } | |
628 | ||
629 | error = nfs_readpage_sync(ctx, inode, page); | |
630 | if (error < 0 && IS_SWAPFILE(inode)) | |
631 | printk("Aiee.. nfs swap-in of page failed!\n"); | |
632 | out: | |
633 | put_nfs_open_context(ctx); | |
634 | return error; | |
635 | ||
636 | out_error: | |
637 | unlock_page(page); | |
638 | return error; | |
639 | } | |
640 | ||
641 | struct nfs_readdesc { | |
642 | struct list_head *head; | |
643 | struct nfs_open_context *ctx; | |
644 | }; | |
645 | ||
646 | static int | |
647 | readpage_async_filler(void *data, struct page *page) | |
648 | { | |
649 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
650 | struct inode *inode = page->mapping->host; | |
651 | struct nfs_page *new; | |
652 | unsigned int len; | |
653 | ||
654 | nfs_wb_page(inode, page); | |
655 | len = nfs_page_length(inode, page); | |
656 | if (len == 0) | |
657 | return nfs_return_empty_page(page); | |
658 | new = nfs_create_request(desc->ctx, inode, page, 0, len); | |
659 | if (IS_ERR(new)) { | |
660 | SetPageError(page); | |
661 | unlock_page(page); | |
662 | return PTR_ERR(new); | |
663 | } | |
664 | if (len < PAGE_CACHE_SIZE) | |
665 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
1da177e4 LT |
666 | nfs_list_add_request(new, desc->head); |
667 | return 0; | |
668 | } | |
669 | ||
670 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
671 | struct list_head *pages, unsigned nr_pages) | |
672 | { | |
673 | LIST_HEAD(head); | |
674 | struct nfs_readdesc desc = { | |
675 | .head = &head, | |
676 | }; | |
677 | struct inode *inode = mapping->host; | |
678 | struct nfs_server *server = NFS_SERVER(inode); | |
679 | int ret; | |
680 | ||
681 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
682 | inode->i_sb->s_id, | |
683 | (long long)NFS_FILEID(inode), | |
684 | nr_pages); | |
91d5b470 | 685 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 LT |
686 | |
687 | if (filp == NULL) { | |
d530838b | 688 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
689 | if (desc.ctx == NULL) |
690 | return -EBADF; | |
691 | } else | |
692 | desc.ctx = get_nfs_open_context((struct nfs_open_context *) | |
693 | filp->private_data); | |
694 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); | |
695 | if (!list_empty(&head)) { | |
696 | int err = nfs_pagein_list(&head, server->rpages); | |
697 | if (!ret) | |
91d5b470 | 698 | nfs_add_stats(inode, NFSIOS_READPAGES, err); |
1da177e4 LT |
699 | ret = err; |
700 | } | |
701 | put_nfs_open_context(desc.ctx); | |
702 | return ret; | |
703 | } | |
704 | ||
f7b422b1 | 705 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
706 | { |
707 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
708 | sizeof(struct nfs_read_data), | |
709 | 0, SLAB_HWCACHE_ALIGN, | |
710 | NULL, NULL); | |
711 | if (nfs_rdata_cachep == NULL) | |
712 | return -ENOMEM; | |
713 | ||
93d2341c MD |
714 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
715 | nfs_rdata_cachep); | |
1da177e4 LT |
716 | if (nfs_rdata_mempool == NULL) |
717 | return -ENOMEM; | |
718 | ||
719 | return 0; | |
720 | } | |
721 | ||
266bee88 | 722 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
723 | { |
724 | mempool_destroy(nfs_rdata_mempool); | |
725 | if (kmem_cache_destroy(nfs_rdata_cachep)) | |
726 | printk(KERN_INFO "nfs_read_data: not all structures were freed\n"); | |
727 | } |