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 | ||
8aca67f0 | 68 | static void nfs_readdata_rcu_free(struct rcu_head *head) |
3feb2d49 | 69 | { |
8aca67f0 | 70 | struct nfs_read_data *p = container_of(head, struct nfs_read_data, task.u.tk_rcu); |
3feb2d49 TM |
71 | if (p && (p->pagevec != &p->page_array[0])) |
72 | kfree(p->pagevec); | |
73 | mempool_free(p, nfs_rdata_mempool); | |
74 | } | |
75 | ||
8aca67f0 TM |
76 | static void nfs_readdata_free(struct nfs_read_data *rdata) |
77 | { | |
78 | call_rcu_bh(&rdata->task.u.tk_rcu, nfs_readdata_rcu_free); | |
79 | } | |
80 | ||
963d8fe5 | 81 | void nfs_readdata_release(void *data) |
1da177e4 | 82 | { |
1da177e4 LT |
83 | nfs_readdata_free(data); |
84 | } | |
85 | ||
86 | static | |
87 | unsigned int nfs_page_length(struct inode *inode, struct page *page) | |
88 | { | |
89 | loff_t i_size = i_size_read(inode); | |
90 | unsigned long idx; | |
91 | ||
92 | if (i_size <= 0) | |
93 | return 0; | |
94 | idx = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
95 | if (page->index > idx) | |
96 | return 0; | |
97 | if (page->index != idx) | |
98 | return PAGE_CACHE_SIZE; | |
99 | return 1 + ((i_size - 1) & (PAGE_CACHE_SIZE - 1)); | |
100 | } | |
101 | ||
102 | static | |
103 | int nfs_return_empty_page(struct page *page) | |
104 | { | |
105 | memclear_highpage_flush(page, 0, PAGE_CACHE_SIZE); | |
106 | SetPageUptodate(page); | |
107 | unlock_page(page); | |
108 | return 0; | |
109 | } | |
110 | ||
1de3fc12 TM |
111 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
112 | { | |
113 | unsigned int remainder = data->args.count - data->res.count; | |
114 | unsigned int base = data->args.pgbase + data->res.count; | |
115 | unsigned int pglen; | |
116 | struct page **pages; | |
117 | ||
118 | if (data->res.eof == 0 || remainder == 0) | |
119 | return; | |
120 | /* | |
121 | * Note: "remainder" can never be negative, since we check for | |
122 | * this in the XDR code. | |
123 | */ | |
124 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
125 | base &= ~PAGE_CACHE_MASK; | |
126 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
127 | for (;;) { |
128 | if (remainder <= pglen) { | |
129 | memclear_highpage_flush(*pages, base, remainder); | |
130 | break; | |
131 | } | |
1de3fc12 | 132 | memclear_highpage_flush(*pages, base, pglen); |
79558f36 TM |
133 | pages++; |
134 | remainder -= pglen; | |
135 | pglen = PAGE_CACHE_SIZE; | |
136 | base = 0; | |
137 | } | |
1de3fc12 TM |
138 | } |
139 | ||
1da177e4 LT |
140 | /* |
141 | * Read a page synchronously. | |
142 | */ | |
143 | static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode, | |
144 | struct page *page) | |
145 | { | |
146 | unsigned int rsize = NFS_SERVER(inode)->rsize; | |
147 | unsigned int count = PAGE_CACHE_SIZE; | |
148 | int result; | |
149 | struct nfs_read_data *rdata; | |
150 | ||
e9f7bee1 | 151 | rdata = nfs_readdata_alloc(count); |
1da177e4 LT |
152 | if (!rdata) |
153 | return -ENOMEM; | |
154 | ||
155 | memset(rdata, 0, sizeof(*rdata)); | |
156 | rdata->flags = (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0); | |
157 | rdata->cred = ctx->cred; | |
158 | rdata->inode = inode; | |
159 | INIT_LIST_HEAD(&rdata->pages); | |
160 | rdata->args.fh = NFS_FH(inode); | |
161 | rdata->args.context = ctx; | |
162 | rdata->args.pages = &page; | |
163 | rdata->args.pgbase = 0UL; | |
164 | rdata->args.count = rsize; | |
165 | rdata->res.fattr = &rdata->fattr; | |
166 | ||
167 | dprintk("NFS: nfs_readpage_sync(%p)\n", page); | |
168 | ||
169 | /* | |
170 | * This works now because the socket layer never tries to DMA | |
171 | * into this buffer directly. | |
172 | */ | |
173 | do { | |
174 | if (count < rsize) | |
175 | rdata->args.count = count; | |
176 | rdata->res.count = rdata->args.count; | |
177 | rdata->args.offset = page_offset(page) + rdata->args.pgbase; | |
178 | ||
179 | dprintk("NFS: nfs_proc_read(%s, (%s/%Ld), %Lu, %u)\n", | |
54ceac45 | 180 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
181 | inode->i_sb->s_id, |
182 | (long long)NFS_FILEID(inode), | |
183 | (unsigned long long)rdata->args.pgbase, | |
184 | rdata->args.count); | |
185 | ||
186 | lock_kernel(); | |
187 | result = NFS_PROTO(inode)->read(rdata); | |
188 | unlock_kernel(); | |
189 | ||
190 | /* | |
191 | * Even if we had a partial success we can't mark the page | |
192 | * cache valid. | |
193 | */ | |
194 | if (result < 0) { | |
195 | if (result == -EISDIR) | |
196 | result = -EINVAL; | |
197 | goto io_error; | |
198 | } | |
199 | count -= result; | |
200 | rdata->args.pgbase += result; | |
91d5b470 CL |
201 | nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, result); |
202 | ||
1da177e4 LT |
203 | /* Note: result == 0 should only happen if we're caching |
204 | * a write that extends the file and punches a hole. | |
205 | */ | |
206 | if (rdata->res.eof != 0 || result == 0) | |
207 | break; | |
208 | } while (count); | |
dc59250c | 209 | spin_lock(&inode->i_lock); |
55296809 | 210 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 211 | spin_unlock(&inode->i_lock); |
1da177e4 | 212 | |
7a524111 | 213 | if (rdata->res.eof || rdata->res.count == rdata->args.count) { |
1de3fc12 | 214 | SetPageUptodate(page); |
7a524111 TM |
215 | if (rdata->res.eof && count != 0) |
216 | memclear_highpage_flush(page, rdata->args.pgbase, count); | |
217 | } | |
1da177e4 LT |
218 | result = 0; |
219 | ||
220 | io_error: | |
221 | unlock_page(page); | |
222 | nfs_readdata_free(rdata); | |
223 | return result; | |
224 | } | |
225 | ||
226 | static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, | |
227 | struct page *page) | |
228 | { | |
229 | LIST_HEAD(one_request); | |
230 | struct nfs_page *new; | |
231 | unsigned int len; | |
232 | ||
233 | len = nfs_page_length(inode, page); | |
234 | if (len == 0) | |
235 | return nfs_return_empty_page(page); | |
236 | new = nfs_create_request(ctx, inode, page, 0, len); | |
237 | if (IS_ERR(new)) { | |
238 | unlock_page(page); | |
239 | return PTR_ERR(new); | |
240 | } | |
241 | if (len < PAGE_CACHE_SIZE) | |
242 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
243 | ||
1da177e4 LT |
244 | nfs_list_add_request(new, &one_request); |
245 | nfs_pagein_one(&one_request, inode); | |
246 | return 0; | |
247 | } | |
248 | ||
249 | static void nfs_readpage_release(struct nfs_page *req) | |
250 | { | |
251 | unlock_page(req->wb_page); | |
252 | ||
1da177e4 LT |
253 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
254 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
255 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
256 | req->wb_bytes, | |
257 | (long long)req_offset(req)); | |
10d2c46f NW |
258 | nfs_clear_request(req); |
259 | nfs_release_request(req); | |
1da177e4 LT |
260 | } |
261 | ||
262 | /* | |
263 | * Set up the NFS read request struct | |
264 | */ | |
265 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, | |
ec06c096 | 266 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
267 | unsigned int count, unsigned int offset) |
268 | { | |
269 | struct inode *inode; | |
ec06c096 | 270 | int flags; |
1da177e4 LT |
271 | |
272 | data->req = req; | |
273 | data->inode = inode = req->wb_context->dentry->d_inode; | |
274 | data->cred = req->wb_context->cred; | |
275 | ||
276 | data->args.fh = NFS_FH(inode); | |
277 | data->args.offset = req_offset(req) + offset; | |
278 | data->args.pgbase = req->wb_pgbase + offset; | |
279 | data->args.pages = data->pagevec; | |
280 | data->args.count = count; | |
281 | data->args.context = req->wb_context; | |
282 | ||
283 | data->res.fattr = &data->fattr; | |
284 | data->res.count = count; | |
285 | data->res.eof = 0; | |
0e574af1 | 286 | nfs_fattr_init(&data->fattr); |
1da177e4 | 287 | |
ec06c096 TM |
288 | /* Set up the initial task struct. */ |
289 | flags = RPC_TASK_ASYNC | (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0); | |
290 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); | |
1da177e4 LT |
291 | NFS_PROTO(inode)->read_setup(data); |
292 | ||
293 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 LT |
294 | |
295 | dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", | |
296 | data->task.tk_pid, | |
297 | inode->i_sb->s_id, | |
298 | (long long)NFS_FILEID(inode), | |
299 | count, | |
300 | (unsigned long long)data->args.offset); | |
301 | } | |
302 | ||
303 | static void | |
304 | nfs_async_read_error(struct list_head *head) | |
305 | { | |
306 | struct nfs_page *req; | |
307 | ||
308 | while (!list_empty(head)) { | |
309 | req = nfs_list_entry(head->next); | |
310 | nfs_list_remove_request(req); | |
311 | SetPageError(req->wb_page); | |
312 | nfs_readpage_release(req); | |
313 | } | |
314 | } | |
315 | ||
316 | /* | |
317 | * Start an async read operation | |
318 | */ | |
319 | static void nfs_execute_read(struct nfs_read_data *data) | |
320 | { | |
321 | struct rpc_clnt *clnt = NFS_CLIENT(data->inode); | |
322 | sigset_t oldset; | |
323 | ||
324 | rpc_clnt_sigmask(clnt, &oldset); | |
325 | lock_kernel(); | |
326 | rpc_execute(&data->task); | |
327 | unlock_kernel(); | |
328 | rpc_clnt_sigunmask(clnt, &oldset); | |
329 | } | |
330 | ||
331 | /* | |
332 | * Generate multiple requests to fill a single page. | |
333 | * | |
334 | * We optimize to reduce the number of read operations on the wire. If we | |
335 | * detect that we're reading a page, or an area of a page, that is past the | |
336 | * end of file, we do not generate NFS read operations but just clear the | |
337 | * parts of the page that would have come back zero from the server anyway. | |
338 | * | |
339 | * We rely on the cached value of i_size to make this determination; another | |
340 | * client can fill pages on the server past our cached end-of-file, but we | |
341 | * won't see the new data until our attribute cache is updated. This is more | |
342 | * or less conventional NFS client behavior. | |
343 | */ | |
344 | static int nfs_pagein_multi(struct list_head *head, struct inode *inode) | |
345 | { | |
346 | struct nfs_page *req = nfs_list_entry(head->next); | |
347 | struct page *page = req->wb_page; | |
348 | struct nfs_read_data *data; | |
e9f7bee1 TM |
349 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
350 | unsigned int offset; | |
1da177e4 LT |
351 | int requests = 0; |
352 | LIST_HEAD(list); | |
353 | ||
354 | nfs_list_remove_request(req); | |
355 | ||
356 | nbytes = req->wb_bytes; | |
e9f7bee1 TM |
357 | do { |
358 | size_t len = min(nbytes,rsize); | |
359 | ||
360 | data = nfs_readdata_alloc(len); | |
1da177e4 LT |
361 | if (!data) |
362 | goto out_bad; | |
363 | INIT_LIST_HEAD(&data->pages); | |
364 | list_add(&data->pages, &list); | |
365 | requests++; | |
e9f7bee1 TM |
366 | nbytes -= len; |
367 | } while(nbytes != 0); | |
1da177e4 LT |
368 | atomic_set(&req->wb_complete, requests); |
369 | ||
370 | ClearPageError(page); | |
371 | offset = 0; | |
372 | nbytes = req->wb_bytes; | |
373 | do { | |
374 | data = list_entry(list.next, struct nfs_read_data, pages); | |
375 | list_del_init(&data->pages); | |
376 | ||
377 | data->pagevec[0] = page; | |
1da177e4 LT |
378 | |
379 | if (nbytes > rsize) { | |
ec06c096 TM |
380 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
381 | rsize, offset); | |
1da177e4 LT |
382 | offset += rsize; |
383 | nbytes -= rsize; | |
384 | } else { | |
ec06c096 TM |
385 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
386 | nbytes, offset); | |
1da177e4 LT |
387 | nbytes = 0; |
388 | } | |
389 | nfs_execute_read(data); | |
390 | } while (nbytes != 0); | |
391 | ||
392 | return 0; | |
393 | ||
394 | out_bad: | |
395 | while (!list_empty(&list)) { | |
396 | data = list_entry(list.next, struct nfs_read_data, pages); | |
397 | list_del(&data->pages); | |
398 | nfs_readdata_free(data); | |
399 | } | |
400 | SetPageError(page); | |
401 | nfs_readpage_release(req); | |
402 | return -ENOMEM; | |
403 | } | |
404 | ||
405 | static int nfs_pagein_one(struct list_head *head, struct inode *inode) | |
406 | { | |
407 | struct nfs_page *req; | |
408 | struct page **pages; | |
409 | struct nfs_read_data *data; | |
410 | unsigned int count; | |
411 | ||
412 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) | |
413 | return nfs_pagein_multi(head, inode); | |
414 | ||
e9f7bee1 | 415 | data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize); |
1da177e4 LT |
416 | if (!data) |
417 | goto out_bad; | |
418 | ||
419 | INIT_LIST_HEAD(&data->pages); | |
420 | pages = data->pagevec; | |
421 | count = 0; | |
422 | while (!list_empty(head)) { | |
423 | req = nfs_list_entry(head->next); | |
424 | nfs_list_remove_request(req); | |
425 | nfs_list_add_request(req, &data->pages); | |
426 | ClearPageError(req->wb_page); | |
427 | *pages++ = req->wb_page; | |
428 | count += req->wb_bytes; | |
429 | } | |
430 | req = nfs_list_entry(data->pages.next); | |
431 | ||
ec06c096 | 432 | nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
1da177e4 LT |
433 | |
434 | nfs_execute_read(data); | |
435 | return 0; | |
436 | out_bad: | |
437 | nfs_async_read_error(head); | |
438 | return -ENOMEM; | |
439 | } | |
440 | ||
441 | static int | |
442 | nfs_pagein_list(struct list_head *head, int rpages) | |
443 | { | |
444 | LIST_HEAD(one_request); | |
445 | struct nfs_page *req; | |
446 | int error = 0; | |
447 | unsigned int pages = 0; | |
448 | ||
449 | while (!list_empty(head)) { | |
450 | pages += nfs_coalesce_requests(head, &one_request, rpages); | |
451 | req = nfs_list_entry(one_request.next); | |
452 | error = nfs_pagein_one(&one_request, req->wb_context->dentry->d_inode); | |
453 | if (error < 0) | |
454 | break; | |
455 | } | |
456 | if (error >= 0) | |
457 | return pages; | |
458 | ||
459 | nfs_async_read_error(head); | |
460 | return error; | |
461 | } | |
462 | ||
0b671301 TM |
463 | /* |
464 | * This is the callback from RPC telling us whether a reply was | |
465 | * received or some error occurred (timeout or socket shutdown). | |
466 | */ | |
467 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
468 | { | |
469 | int status; | |
470 | ||
471 | dprintk("%s: %4d, (status %d)\n", __FUNCTION__, task->tk_pid, | |
472 | task->tk_status); | |
473 | ||
474 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
475 | if (status != 0) | |
476 | return status; | |
477 | ||
478 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
479 | ||
480 | if (task->tk_status == -ESTALE) { | |
481 | set_bit(NFS_INO_STALE, &NFS_FLAGS(data->inode)); | |
482 | nfs_mark_for_revalidate(data->inode); | |
483 | } | |
484 | spin_lock(&data->inode->i_lock); | |
485 | NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME; | |
486 | spin_unlock(&data->inode->i_lock); | |
487 | return 0; | |
488 | } | |
489 | ||
490 | static int nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) | |
491 | { | |
492 | struct nfs_readargs *argp = &data->args; | |
493 | struct nfs_readres *resp = &data->res; | |
494 | ||
495 | if (resp->eof || resp->count == argp->count) | |
496 | return 0; | |
497 | ||
498 | /* This is a short read! */ | |
499 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
500 | /* Has the server at least made some progress? */ | |
501 | if (resp->count == 0) | |
502 | return 0; | |
503 | ||
504 | /* Yes, so retry the read at the end of the data */ | |
505 | argp->offset += resp->count; | |
506 | argp->pgbase += resp->count; | |
507 | argp->count -= resp->count; | |
508 | rpc_restart_call(task); | |
509 | return -EAGAIN; | |
510 | } | |
511 | ||
1da177e4 LT |
512 | /* |
513 | * Handle a read reply that fills part of a page. | |
514 | */ | |
ec06c096 | 515 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 516 | { |
ec06c096 | 517 | struct nfs_read_data *data = calldata; |
1da177e4 LT |
518 | struct nfs_page *req = data->req; |
519 | struct page *page = req->wb_page; | |
520 | ||
ec06c096 TM |
521 | if (nfs_readpage_result(task, data) != 0) |
522 | return; | |
0b671301 TM |
523 | |
524 | if (likely(task->tk_status >= 0)) { | |
525 | nfs_readpage_truncate_uninitialised_page(data); | |
526 | if (nfs_readpage_retry(task, data) != 0) | |
527 | return; | |
528 | } | |
529 | if (unlikely(task->tk_status < 0)) | |
530 | SetPageError(page); | |
1da177e4 LT |
531 | if (atomic_dec_and_test(&req->wb_complete)) { |
532 | if (!PageError(page)) | |
533 | SetPageUptodate(page); | |
534 | nfs_readpage_release(req); | |
535 | } | |
536 | } | |
537 | ||
ec06c096 TM |
538 | static const struct rpc_call_ops nfs_read_partial_ops = { |
539 | .rpc_call_done = nfs_readpage_result_partial, | |
540 | .rpc_release = nfs_readdata_release, | |
541 | }; | |
542 | ||
1de3fc12 TM |
543 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
544 | { | |
545 | unsigned int count = data->res.count; | |
546 | unsigned int base = data->args.pgbase; | |
547 | struct page **pages; | |
548 | ||
79558f36 TM |
549 | if (data->res.eof) |
550 | count = data->args.count; | |
1de3fc12 TM |
551 | if (unlikely(count == 0)) |
552 | return; | |
553 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
554 | base &= ~PAGE_CACHE_MASK; | |
555 | count += base; | |
556 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
557 | SetPageUptodate(*pages); | |
0b671301 TM |
558 | if (count == 0) |
559 | return; | |
560 | /* Was this a short read? */ | |
561 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
562 | SetPageUptodate(*pages); |
563 | } | |
564 | ||
1da177e4 LT |
565 | /* |
566 | * This is the callback from RPC telling us whether a reply was | |
567 | * received or some error occurred (timeout or socket shutdown). | |
568 | */ | |
ec06c096 | 569 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 570 | { |
ec06c096 | 571 | struct nfs_read_data *data = calldata; |
1da177e4 | 572 | |
0b671301 TM |
573 | if (nfs_readpage_result(task, data) != 0) |
574 | return; | |
1de3fc12 | 575 | /* |
0b671301 | 576 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 577 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
578 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
579 | * first. | |
1de3fc12 TM |
580 | */ |
581 | if (likely(task->tk_status >= 0)) { | |
582 | nfs_readpage_truncate_uninitialised_page(data); | |
583 | nfs_readpage_set_pages_uptodate(data); | |
0b671301 TM |
584 | if (nfs_readpage_retry(task, data) != 0) |
585 | return; | |
586 | } | |
1da177e4 LT |
587 | while (!list_empty(&data->pages)) { |
588 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 589 | |
1de3fc12 | 590 | nfs_list_remove_request(req); |
1da177e4 LT |
591 | nfs_readpage_release(req); |
592 | } | |
593 | } | |
594 | ||
ec06c096 TM |
595 | static const struct rpc_call_ops nfs_read_full_ops = { |
596 | .rpc_call_done = nfs_readpage_result_full, | |
597 | .rpc_release = nfs_readdata_release, | |
598 | }; | |
599 | ||
1da177e4 LT |
600 | /* |
601 | * Read a page over NFS. | |
602 | * We read the page synchronously in the following case: | |
603 | * - The error flag is set for this page. This happens only when a | |
604 | * previous async read operation failed. | |
605 | */ | |
606 | int nfs_readpage(struct file *file, struct page *page) | |
607 | { | |
608 | struct nfs_open_context *ctx; | |
609 | struct inode *inode = page->mapping->host; | |
610 | int error; | |
611 | ||
612 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
613 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
614 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
615 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
616 | ||
1da177e4 LT |
617 | /* |
618 | * Try to flush any pending writes to the file.. | |
619 | * | |
620 | * NOTE! Because we own the page lock, there cannot | |
621 | * be any new pending writes generated at this point | |
622 | * for this page (other pages can be written to). | |
623 | */ | |
624 | error = nfs_wb_page(inode, page); | |
625 | if (error) | |
626 | goto out_error; | |
627 | ||
5f004cf2 TM |
628 | error = -ESTALE; |
629 | if (NFS_STALE(inode)) | |
630 | goto out_error; | |
631 | ||
1da177e4 | 632 | if (file == NULL) { |
d530838b | 633 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
634 | if (ctx == NULL) |
635 | return -EBADF; | |
636 | } else | |
637 | ctx = get_nfs_open_context((struct nfs_open_context *) | |
638 | file->private_data); | |
639 | if (!IS_SYNC(inode)) { | |
640 | error = nfs_readpage_async(ctx, inode, page); | |
641 | goto out; | |
642 | } | |
643 | ||
644 | error = nfs_readpage_sync(ctx, inode, page); | |
645 | if (error < 0 && IS_SWAPFILE(inode)) | |
646 | printk("Aiee.. nfs swap-in of page failed!\n"); | |
647 | out: | |
648 | put_nfs_open_context(ctx); | |
649 | return error; | |
650 | ||
651 | out_error: | |
652 | unlock_page(page); | |
653 | return error; | |
654 | } | |
655 | ||
656 | struct nfs_readdesc { | |
657 | struct list_head *head; | |
658 | struct nfs_open_context *ctx; | |
659 | }; | |
660 | ||
661 | static int | |
662 | readpage_async_filler(void *data, struct page *page) | |
663 | { | |
664 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
665 | struct inode *inode = page->mapping->host; | |
666 | struct nfs_page *new; | |
667 | unsigned int len; | |
668 | ||
669 | nfs_wb_page(inode, page); | |
670 | len = nfs_page_length(inode, page); | |
671 | if (len == 0) | |
672 | return nfs_return_empty_page(page); | |
673 | new = nfs_create_request(desc->ctx, inode, page, 0, len); | |
674 | if (IS_ERR(new)) { | |
675 | SetPageError(page); | |
676 | unlock_page(page); | |
677 | return PTR_ERR(new); | |
678 | } | |
679 | if (len < PAGE_CACHE_SIZE) | |
680 | memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len); | |
1da177e4 LT |
681 | nfs_list_add_request(new, desc->head); |
682 | return 0; | |
683 | } | |
684 | ||
685 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
686 | struct list_head *pages, unsigned nr_pages) | |
687 | { | |
688 | LIST_HEAD(head); | |
689 | struct nfs_readdesc desc = { | |
690 | .head = &head, | |
691 | }; | |
692 | struct inode *inode = mapping->host; | |
693 | struct nfs_server *server = NFS_SERVER(inode); | |
5f004cf2 | 694 | int ret = -ESTALE; |
1da177e4 LT |
695 | |
696 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
697 | inode->i_sb->s_id, | |
698 | (long long)NFS_FILEID(inode), | |
699 | nr_pages); | |
91d5b470 | 700 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 701 | |
5f004cf2 TM |
702 | if (NFS_STALE(inode)) |
703 | goto out; | |
704 | ||
1da177e4 | 705 | if (filp == NULL) { |
d530838b | 706 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
707 | if (desc.ctx == NULL) |
708 | return -EBADF; | |
709 | } else | |
710 | desc.ctx = get_nfs_open_context((struct nfs_open_context *) | |
711 | filp->private_data); | |
712 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); | |
713 | if (!list_empty(&head)) { | |
714 | int err = nfs_pagein_list(&head, server->rpages); | |
715 | if (!ret) | |
91d5b470 | 716 | nfs_add_stats(inode, NFSIOS_READPAGES, err); |
1da177e4 LT |
717 | ret = err; |
718 | } | |
719 | put_nfs_open_context(desc.ctx); | |
5f004cf2 | 720 | out: |
1da177e4 LT |
721 | return ret; |
722 | } | |
723 | ||
f7b422b1 | 724 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
725 | { |
726 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
727 | sizeof(struct nfs_read_data), | |
728 | 0, SLAB_HWCACHE_ALIGN, | |
729 | NULL, NULL); | |
730 | if (nfs_rdata_cachep == NULL) | |
731 | return -ENOMEM; | |
732 | ||
93d2341c MD |
733 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
734 | nfs_rdata_cachep); | |
1da177e4 LT |
735 | if (nfs_rdata_mempool == NULL) |
736 | return -ENOMEM; | |
737 | ||
738 | return 0; | |
739 | } | |
740 | ||
266bee88 | 741 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
742 | { |
743 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 744 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 745 | } |