Linux 2.6.28-rc5
[cris-mirror.git] / fs / nfs / read.c
blob40d17987d0e847a87fa4596a9a4c9f39a464288e
1 /*
2 * linux/fs/nfs/read.c
4 * Block I/O for NFS
6 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
7 * modified for async RPC by okir@monad.swb.de
8 */
10 #include <linux/time.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/fcntl.h>
14 #include <linux/stat.h>
15 #include <linux/mm.h>
16 #include <linux/slab.h>
17 #include <linux/pagemap.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_page.h>
21 #include <linux/smp_lock.h>
23 #include <asm/system.h>
25 #include "internal.h"
26 #include "iostat.h"
28 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
30 static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int);
31 static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int);
32 static const struct rpc_call_ops nfs_read_partial_ops;
33 static const struct rpc_call_ops nfs_read_full_ops;
35 static struct kmem_cache *nfs_rdata_cachep;
36 static mempool_t *nfs_rdata_mempool;
38 #define MIN_POOL_READ (32)
40 struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
42 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
44 if (p) {
45 memset(p, 0, sizeof(*p));
46 INIT_LIST_HEAD(&p->pages);
47 p->npages = pagecount;
48 if (pagecount <= ARRAY_SIZE(p->page_array))
49 p->pagevec = p->page_array;
50 else {
51 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
52 if (!p->pagevec) {
53 mempool_free(p, nfs_rdata_mempool);
54 p = NULL;
58 return p;
61 static void nfs_readdata_free(struct nfs_read_data *p)
63 if (p && (p->pagevec != &p->page_array[0]))
64 kfree(p->pagevec);
65 mempool_free(p, nfs_rdata_mempool);
68 void nfs_readdata_release(void *data)
70 struct nfs_read_data *rdata = data;
72 put_nfs_open_context(rdata->args.context);
73 nfs_readdata_free(rdata);
76 static
77 int nfs_return_empty_page(struct page *page)
79 zero_user(page, 0, PAGE_CACHE_SIZE);
80 SetPageUptodate(page);
81 unlock_page(page);
82 return 0;
85 static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
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;
92 if (data->res.eof == 0 || remainder == 0)
93 return;
95 * Note: "remainder" can never be negative, since we check for
96 * this in the XDR code.
98 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
99 base &= ~PAGE_CACHE_MASK;
100 pglen = PAGE_CACHE_SIZE - base;
101 for (;;) {
102 if (remainder <= pglen) {
103 zero_user(*pages, base, remainder);
104 break;
106 zero_user(*pages, base, pglen);
107 pages++;
108 remainder -= pglen;
109 pglen = PAGE_CACHE_SIZE;
110 base = 0;
114 static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
115 struct page *page)
117 LIST_HEAD(one_request);
118 struct nfs_page *new;
119 unsigned int len;
121 len = nfs_page_length(page);
122 if (len == 0)
123 return nfs_return_empty_page(page);
124 new = nfs_create_request(ctx, inode, page, 0, len);
125 if (IS_ERR(new)) {
126 unlock_page(page);
127 return PTR_ERR(new);
129 if (len < PAGE_CACHE_SIZE)
130 zero_user_segment(page, len, PAGE_CACHE_SIZE);
132 nfs_list_add_request(new, &one_request);
133 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
134 nfs_pagein_multi(inode, &one_request, 1, len, 0);
135 else
136 nfs_pagein_one(inode, &one_request, 1, len, 0);
137 return 0;
140 static void nfs_readpage_release(struct nfs_page *req)
142 unlock_page(req->wb_page);
144 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
145 req->wb_context->path.dentry->d_inode->i_sb->s_id,
146 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
147 req->wb_bytes,
148 (long long)req_offset(req));
149 nfs_clear_request(req);
150 nfs_release_request(req);
154 * Set up the NFS read request struct
156 static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
157 const struct rpc_call_ops *call_ops,
158 unsigned int count, unsigned int offset)
160 struct inode *inode = req->wb_context->path.dentry->d_inode;
161 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
162 struct rpc_task *task;
163 struct rpc_message msg = {
164 .rpc_argp = &data->args,
165 .rpc_resp = &data->res,
166 .rpc_cred = req->wb_context->cred,
168 struct rpc_task_setup task_setup_data = {
169 .task = &data->task,
170 .rpc_client = NFS_CLIENT(inode),
171 .rpc_message = &msg,
172 .callback_ops = call_ops,
173 .callback_data = data,
174 .workqueue = nfsiod_workqueue,
175 .flags = RPC_TASK_ASYNC | swap_flags,
178 data->req = req;
179 data->inode = inode;
180 data->cred = msg.rpc_cred;
182 data->args.fh = NFS_FH(inode);
183 data->args.offset = req_offset(req) + offset;
184 data->args.pgbase = req->wb_pgbase + offset;
185 data->args.pages = data->pagevec;
186 data->args.count = count;
187 data->args.context = get_nfs_open_context(req->wb_context);
189 data->res.fattr = &data->fattr;
190 data->res.count = count;
191 data->res.eof = 0;
192 nfs_fattr_init(&data->fattr);
194 /* Set up the initial task struct. */
195 NFS_PROTO(inode)->read_setup(data, &msg);
197 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
198 data->task.tk_pid,
199 inode->i_sb->s_id,
200 (long long)NFS_FILEID(inode),
201 count,
202 (unsigned long long)data->args.offset);
204 task = rpc_run_task(&task_setup_data);
205 if (IS_ERR(task))
206 return PTR_ERR(task);
207 rpc_put_task(task);
208 return 0;
211 static void
212 nfs_async_read_error(struct list_head *head)
214 struct nfs_page *req;
216 while (!list_empty(head)) {
217 req = nfs_list_entry(head->next);
218 nfs_list_remove_request(req);
219 SetPageError(req->wb_page);
220 nfs_readpage_release(req);
225 * Generate multiple requests to fill a single page.
227 * We optimize to reduce the number of read operations on the wire. If we
228 * detect that we're reading a page, or an area of a page, that is past the
229 * end of file, we do not generate NFS read operations but just clear the
230 * parts of the page that would have come back zero from the server anyway.
232 * We rely on the cached value of i_size to make this determination; another
233 * client can fill pages on the server past our cached end-of-file, but we
234 * won't see the new data until our attribute cache is updated. This is more
235 * or less conventional NFS client behavior.
237 static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
239 struct nfs_page *req = nfs_list_entry(head->next);
240 struct page *page = req->wb_page;
241 struct nfs_read_data *data;
242 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
243 unsigned int offset;
244 int requests = 0;
245 int ret = 0;
246 LIST_HEAD(list);
248 nfs_list_remove_request(req);
250 nbytes = count;
251 do {
252 size_t len = min(nbytes,rsize);
254 data = nfs_readdata_alloc(1);
255 if (!data)
256 goto out_bad;
257 list_add(&data->pages, &list);
258 requests++;
259 nbytes -= len;
260 } while(nbytes != 0);
261 atomic_set(&req->wb_complete, requests);
263 ClearPageError(page);
264 offset = 0;
265 nbytes = count;
266 do {
267 int ret2;
269 data = list_entry(list.next, struct nfs_read_data, pages);
270 list_del_init(&data->pages);
272 data->pagevec[0] = page;
274 if (nbytes < rsize)
275 rsize = nbytes;
276 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
277 rsize, offset);
278 if (ret == 0)
279 ret = ret2;
280 offset += rsize;
281 nbytes -= rsize;
282 } while (nbytes != 0);
284 return ret;
286 out_bad:
287 while (!list_empty(&list)) {
288 data = list_entry(list.next, struct nfs_read_data, pages);
289 list_del(&data->pages);
290 nfs_readdata_free(data);
292 SetPageError(page);
293 nfs_readpage_release(req);
294 return -ENOMEM;
297 static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
299 struct nfs_page *req;
300 struct page **pages;
301 struct nfs_read_data *data;
302 int ret = -ENOMEM;
304 data = nfs_readdata_alloc(npages);
305 if (!data)
306 goto out_bad;
308 pages = data->pagevec;
309 while (!list_empty(head)) {
310 req = nfs_list_entry(head->next);
311 nfs_list_remove_request(req);
312 nfs_list_add_request(req, &data->pages);
313 ClearPageError(req->wb_page);
314 *pages++ = req->wb_page;
316 req = nfs_list_entry(data->pages.next);
318 return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
319 out_bad:
320 nfs_async_read_error(head);
321 return ret;
325 * This is the callback from RPC telling us whether a reply was
326 * received or some error occurred (timeout or socket shutdown).
328 int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
330 int status;
332 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
333 task->tk_status);
335 status = NFS_PROTO(data->inode)->read_done(task, data);
336 if (status != 0)
337 return status;
339 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
341 if (task->tk_status == -ESTALE) {
342 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
343 nfs_mark_for_revalidate(data->inode);
345 return 0;
348 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
350 struct nfs_readargs *argp = &data->args;
351 struct nfs_readres *resp = &data->res;
353 if (resp->eof || resp->count == argp->count)
354 return;
356 /* This is a short read! */
357 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
358 /* Has the server at least made some progress? */
359 if (resp->count == 0)
360 return;
362 /* Yes, so retry the read at the end of the data */
363 argp->offset += resp->count;
364 argp->pgbase += resp->count;
365 argp->count -= resp->count;
366 rpc_restart_call(task);
370 * Handle a read reply that fills part of a page.
372 static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
374 struct nfs_read_data *data = calldata;
376 if (nfs_readpage_result(task, data) != 0)
377 return;
378 if (task->tk_status < 0)
379 return;
381 nfs_readpage_truncate_uninitialised_page(data);
382 nfs_readpage_retry(task, data);
385 static void nfs_readpage_release_partial(void *calldata)
387 struct nfs_read_data *data = calldata;
388 struct nfs_page *req = data->req;
389 struct page *page = req->wb_page;
390 int status = data->task.tk_status;
392 if (status < 0)
393 SetPageError(page);
395 if (atomic_dec_and_test(&req->wb_complete)) {
396 if (!PageError(page))
397 SetPageUptodate(page);
398 nfs_readpage_release(req);
400 nfs_readdata_release(calldata);
403 static const struct rpc_call_ops nfs_read_partial_ops = {
404 .rpc_call_done = nfs_readpage_result_partial,
405 .rpc_release = nfs_readpage_release_partial,
408 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
410 unsigned int count = data->res.count;
411 unsigned int base = data->args.pgbase;
412 struct page **pages;
414 if (data->res.eof)
415 count = data->args.count;
416 if (unlikely(count == 0))
417 return;
418 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
419 base &= ~PAGE_CACHE_MASK;
420 count += base;
421 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
422 SetPageUptodate(*pages);
423 if (count == 0)
424 return;
425 /* Was this a short read? */
426 if (data->res.eof || data->res.count == data->args.count)
427 SetPageUptodate(*pages);
431 * This is the callback from RPC telling us whether a reply was
432 * received or some error occurred (timeout or socket shutdown).
434 static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
436 struct nfs_read_data *data = calldata;
438 if (nfs_readpage_result(task, data) != 0)
439 return;
440 if (task->tk_status < 0)
441 return;
443 * Note: nfs_readpage_retry may change the values of
444 * data->args. In the multi-page case, we therefore need
445 * to ensure that we call nfs_readpage_set_pages_uptodate()
446 * first.
448 nfs_readpage_truncate_uninitialised_page(data);
449 nfs_readpage_set_pages_uptodate(data);
450 nfs_readpage_retry(task, data);
453 static void nfs_readpage_release_full(void *calldata)
455 struct nfs_read_data *data = calldata;
457 while (!list_empty(&data->pages)) {
458 struct nfs_page *req = nfs_list_entry(data->pages.next);
460 nfs_list_remove_request(req);
461 nfs_readpage_release(req);
463 nfs_readdata_release(calldata);
466 static const struct rpc_call_ops nfs_read_full_ops = {
467 .rpc_call_done = nfs_readpage_result_full,
468 .rpc_release = nfs_readpage_release_full,
472 * Read a page over NFS.
473 * We read the page synchronously in the following case:
474 * - The error flag is set for this page. This happens only when a
475 * previous async read operation failed.
477 int nfs_readpage(struct file *file, struct page *page)
479 struct nfs_open_context *ctx;
480 struct inode *inode = page->mapping->host;
481 int error;
483 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
484 page, PAGE_CACHE_SIZE, page->index);
485 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
486 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
489 * Try to flush any pending writes to the file..
491 * NOTE! Because we own the page lock, there cannot
492 * be any new pending writes generated at this point
493 * for this page (other pages can be written to).
495 error = nfs_wb_page(inode, page);
496 if (error)
497 goto out_unlock;
498 if (PageUptodate(page))
499 goto out_unlock;
501 error = -ESTALE;
502 if (NFS_STALE(inode))
503 goto out_unlock;
505 if (file == NULL) {
506 error = -EBADF;
507 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
508 if (ctx == NULL)
509 goto out_unlock;
510 } else
511 ctx = get_nfs_open_context(nfs_file_open_context(file));
513 error = nfs_readpage_async(ctx, inode, page);
515 put_nfs_open_context(ctx);
516 return error;
517 out_unlock:
518 unlock_page(page);
519 return error;
522 struct nfs_readdesc {
523 struct nfs_pageio_descriptor *pgio;
524 struct nfs_open_context *ctx;
527 static int
528 readpage_async_filler(void *data, struct page *page)
530 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
531 struct inode *inode = page->mapping->host;
532 struct nfs_page *new;
533 unsigned int len;
534 int error;
536 error = nfs_wb_page(inode, page);
537 if (error)
538 goto out_unlock;
539 if (PageUptodate(page))
540 goto out_unlock;
542 len = nfs_page_length(page);
543 if (len == 0)
544 return nfs_return_empty_page(page);
546 new = nfs_create_request(desc->ctx, inode, page, 0, len);
547 if (IS_ERR(new))
548 goto out_error;
550 if (len < PAGE_CACHE_SIZE)
551 zero_user_segment(page, len, PAGE_CACHE_SIZE);
552 if (!nfs_pageio_add_request(desc->pgio, new)) {
553 error = desc->pgio->pg_error;
554 goto out_unlock;
556 return 0;
557 out_error:
558 error = PTR_ERR(new);
559 SetPageError(page);
560 out_unlock:
561 unlock_page(page);
562 return error;
565 int nfs_readpages(struct file *filp, struct address_space *mapping,
566 struct list_head *pages, unsigned nr_pages)
568 struct nfs_pageio_descriptor pgio;
569 struct nfs_readdesc desc = {
570 .pgio = &pgio,
572 struct inode *inode = mapping->host;
573 struct nfs_server *server = NFS_SERVER(inode);
574 size_t rsize = server->rsize;
575 unsigned long npages;
576 int ret = -ESTALE;
578 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
579 inode->i_sb->s_id,
580 (long long)NFS_FILEID(inode),
581 nr_pages);
582 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
584 if (NFS_STALE(inode))
585 goto out;
587 if (filp == NULL) {
588 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
589 if (desc.ctx == NULL)
590 return -EBADF;
591 } else
592 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
593 if (rsize < PAGE_CACHE_SIZE)
594 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
595 else
596 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
598 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
600 nfs_pageio_complete(&pgio);
601 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
602 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
603 put_nfs_open_context(desc.ctx);
604 out:
605 return ret;
608 int __init nfs_init_readpagecache(void)
610 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
611 sizeof(struct nfs_read_data),
612 0, SLAB_HWCACHE_ALIGN,
613 NULL);
614 if (nfs_rdata_cachep == NULL)
615 return -ENOMEM;
617 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
618 nfs_rdata_cachep);
619 if (nfs_rdata_mempool == NULL)
620 return -ENOMEM;
622 return 0;
625 void nfs_destroy_readpagecache(void)
627 mempool_destroy(nfs_rdata_mempool);
628 kmem_cache_destroy(nfs_rdata_cachep);