blob: c2e49c397a27186ac9ba09ebd49710bb091935e8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#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
Chuck Lever91d5b472006-03-20 13:44:14 -050033#include "iostat.h"
34
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#define NFSDBG_FACILITY NFSDBG_PAGECACHE
36
37static int nfs_pagein_one(struct list_head *, struct inode *);
Trond Myklebustec06c092006-03-20 13:44:27 -050038static const struct rpc_call_ops nfs_read_partial_ops;
39static const struct rpc_call_ops nfs_read_full_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41static kmem_cache_t *nfs_rdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050042static mempool_t *nfs_rdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44#define MIN_POOL_READ (32)
45
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070046struct nfs_read_data *nfs_readdata_alloc(size_t len)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050047{
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070048 unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050049 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);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070054 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040055 if (pagecount <= ARRAY_SIZE(p->page_array))
56 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050057 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040058 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
59 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050060 mempool_free(p, nfs_rdata_mempool);
61 p = NULL;
62 }
63 }
64 }
65 return p;
66}
67
Adrian Bunke4e20512006-08-03 15:07:47 -040068static void nfs_readdata_free(struct nfs_read_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050069{
70 if (p && (p->pagevec != &p->page_array[0]))
71 kfree(p->pagevec);
72 mempool_free(p, nfs_rdata_mempool);
73}
74
Trond Myklebust963d8fe2006-01-03 09:55:04 +010075void nfs_readdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 nfs_readdata_free(data);
78}
79
80static
81unsigned 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
96static
97int 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
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400105static 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;
Trond Myklebust79558f32006-08-22 13:44:32 -0400121 for (;;) {
122 if (remainder <= pglen) {
123 memclear_highpage_flush(*pages, base, remainder);
124 break;
125 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400126 memclear_highpage_flush(*pages, base, pglen);
Trond Myklebust79558f32006-08-22 13:44:32 -0400127 pages++;
128 remainder -= pglen;
129 pglen = PAGE_CACHE_SIZE;
130 base = 0;
131 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400132}
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134/*
135 * Read a page synchronously.
136 */
137static 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
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700145 rdata = nfs_readdata_alloc(count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 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",
David Howells54ceac42006-08-22 20:06:13 -0400174 NFS_SERVER(inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 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;
Chuck Lever91d5b472006-03-20 13:44:14 -0500195 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, result);
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 /* 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);
Chuck Leverdc592502005-08-18 11:24:12 -0700203 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700205 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Trond Myklebust7a524112006-09-15 16:03:45 -0400207 if (rdata->res.eof || rdata->res.count == rdata->args.count) {
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400208 SetPageUptodate(page);
Trond Myklebust7a524112006-09-15 16:03:45 -0400209 if (rdata->res.eof && count != 0)
210 memclear_highpage_flush(page, rdata->args.pgbase, count);
211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 result = 0;
213
214io_error:
215 unlock_page(page);
216 nfs_readdata_free(rdata);
217 return result;
218}
219
220static 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 nfs_list_add_request(new, &one_request);
239 nfs_pagein_one(&one_request, inode);
240 return 0;
241}
242
243static void nfs_readpage_release(struct nfs_page *req)
244{
245 unlock_page(req->wb_page);
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 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));
Nick Wilson10d2c462005-09-22 21:44:28 -0700252 nfs_clear_request(req);
253 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
256/*
257 * Set up the NFS read request struct
258 */
259static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500260 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 unsigned int count, unsigned int offset)
262{
263 struct inode *inode;
Trond Myklebustec06c092006-03-20 13:44:27 -0500264 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
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;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400280 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Trond Myklebustec06c092006-03-20 13:44:27 -0500282 /* 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 NFS_PROTO(inode)->read_setup(data);
286
287 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
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
297static void
298nfs_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 */
313static 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 */
338static 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;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700343 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
344 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 int requests = 0;
346 LIST_HEAD(list);
347
348 nfs_list_remove_request(req);
349
350 nbytes = req->wb_bytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700351 do {
352 size_t len = min(nbytes,rsize);
353
354 data = nfs_readdata_alloc(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (!data)
356 goto out_bad;
357 INIT_LIST_HEAD(&data->pages);
358 list_add(&data->pages, &list);
359 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700360 nbytes -= len;
361 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373 if (nbytes > rsize) {
Trond Myklebustec06c092006-03-20 13:44:27 -0500374 nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
375 rsize, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 offset += rsize;
377 nbytes -= rsize;
378 } else {
Trond Myklebustec06c092006-03-20 13:44:27 -0500379 nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
380 nbytes, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 nbytes = 0;
382 }
383 nfs_execute_read(data);
384 } while (nbytes != 0);
385
386 return 0;
387
388out_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
399static 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
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700409 data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 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
Trond Myklebustec06c092006-03-20 13:44:27 -0500426 nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
428 nfs_execute_read(data);
429 return 0;
430out_bad:
431 nfs_async_read_error(head);
432 return -ENOMEM;
433}
434
435static int
436nfs_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 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500460static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Trond Myklebustec06c092006-03-20 13:44:27 -0500462 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 struct nfs_page *req = data->req;
464 struct page *page = req->wb_page;
465
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400466 if (likely(task->tk_status >= 0))
467 nfs_readpage_truncate_uninitialised_page(data);
468 else
469 SetPageError(page);
Trond Myklebustec06c092006-03-20 13:44:27 -0500470 if (nfs_readpage_result(task, data) != 0)
471 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 if (atomic_dec_and_test(&req->wb_complete)) {
473 if (!PageError(page))
474 SetPageUptodate(page);
475 nfs_readpage_release(req);
476 }
477}
478
Trond Myklebustec06c092006-03-20 13:44:27 -0500479static 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
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400484static 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
Trond Myklebust79558f32006-08-22 13:44:32 -0400490 if (data->res.eof)
491 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400492 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);
Trond Myklebust79558f32006-08-22 13:44:32 -0400499 if (count != 0)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400500 SetPageUptodate(*pages);
501}
502
503static 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);
Trond Myklebust79558f32006-08-22 13:44:32 -0400514 if (count != 0)
515 SetPageError(*pages);
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400516}
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518/*
519 * This is the callback from RPC telling us whether a reply was
520 * received or some error occurred (timeout or socket shutdown).
521 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500522static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Trond Myklebustec06c092006-03-20 13:44:27 -0500524 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400526 /*
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);
Trond Myklebustec06c092006-03-20 13:44:27 -0500537 if (nfs_readpage_result(task, data) != 0)
538 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 while (!list_empty(&data->pages)) {
540 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400542 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 nfs_readpage_release(req);
544 }
545}
546
Trond Myklebustec06c092006-03-20 13:44:27 -0500547static 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552/*
553 * This is the callback from RPC telling us whether a reply was
554 * received or some error occurred (timeout or socket shutdown).
555 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500556int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 struct nfs_readargs *argp = &data->args;
559 struct nfs_readres *resp = &data->res;
Trond Myklebustec06c092006-03-20 13:44:27 -0500560 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562 dprintk("NFS: %4d nfs_readpage_result, (status %d)\n",
Trond Myklebustec06c092006-03-20 13:44:27 -0500563 task->tk_pid, task->tk_status);
564
565 status = NFS_PROTO(data->inode)->read_done(task, data);
566 if (status != 0)
567 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
Chuck Lever91d5b472006-03-20 13:44:14 -0500569 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, resp->count);
570
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400571 if (task->tk_status < 0) {
572 if (task->tk_status == -ESTALE) {
573 set_bit(NFS_INO_STALE, &NFS_FLAGS(data->inode));
574 nfs_mark_for_revalidate(data->inode);
575 }
576 } else if (resp->count < argp->count && !resp->eof) {
577 /* This is a short read! */
Chuck Lever91d5b472006-03-20 13:44:14 -0500578 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 /* Has the server at least made some progress? */
580 if (resp->count != 0) {
581 /* Yes, so retry the read at the end of the data */
582 argp->offset += resp->count;
583 argp->pgbase += resp->count;
584 argp->count -= resp->count;
585 rpc_restart_call(task);
Trond Myklebustec06c092006-03-20 13:44:27 -0500586 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588 task->tk_status = -EIO;
589 }
Chuck Leverdc592502005-08-18 11:24:12 -0700590 spin_lock(&data->inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700591 NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700592 spin_unlock(&data->inode->i_lock);
Trond Myklebustec06c092006-03-20 13:44:27 -0500593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594}
595
596/*
597 * Read a page over NFS.
598 * We read the page synchronously in the following case:
599 * - The error flag is set for this page. This happens only when a
600 * previous async read operation failed.
601 */
602int nfs_readpage(struct file *file, struct page *page)
603{
604 struct nfs_open_context *ctx;
605 struct inode *inode = page->mapping->host;
606 int error;
607
608 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
609 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500610 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
611 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 /*
614 * Try to flush any pending writes to the file..
615 *
616 * NOTE! Because we own the page lock, there cannot
617 * be any new pending writes generated at this point
618 * for this page (other pages can be written to).
619 */
620 error = nfs_wb_page(inode, page);
621 if (error)
622 goto out_error;
623
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400624 error = -ESTALE;
625 if (NFS_STALE(inode))
626 goto out_error;
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (file == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500629 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (ctx == NULL)
631 return -EBADF;
632 } else
633 ctx = get_nfs_open_context((struct nfs_open_context *)
634 file->private_data);
635 if (!IS_SYNC(inode)) {
636 error = nfs_readpage_async(ctx, inode, page);
637 goto out;
638 }
639
640 error = nfs_readpage_sync(ctx, inode, page);
641 if (error < 0 && IS_SWAPFILE(inode))
642 printk("Aiee.. nfs swap-in of page failed!\n");
643out:
644 put_nfs_open_context(ctx);
645 return error;
646
647out_error:
648 unlock_page(page);
649 return error;
650}
651
652struct nfs_readdesc {
653 struct list_head *head;
654 struct nfs_open_context *ctx;
655};
656
657static int
658readpage_async_filler(void *data, struct page *page)
659{
660 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
661 struct inode *inode = page->mapping->host;
662 struct nfs_page *new;
663 unsigned int len;
664
665 nfs_wb_page(inode, page);
666 len = nfs_page_length(inode, page);
667 if (len == 0)
668 return nfs_return_empty_page(page);
669 new = nfs_create_request(desc->ctx, inode, page, 0, len);
670 if (IS_ERR(new)) {
671 SetPageError(page);
672 unlock_page(page);
673 return PTR_ERR(new);
674 }
675 if (len < PAGE_CACHE_SIZE)
676 memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 nfs_list_add_request(new, desc->head);
678 return 0;
679}
680
681int nfs_readpages(struct file *filp, struct address_space *mapping,
682 struct list_head *pages, unsigned nr_pages)
683{
684 LIST_HEAD(head);
685 struct nfs_readdesc desc = {
686 .head = &head,
687 };
688 struct inode *inode = mapping->host;
689 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400690 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
692 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
693 inode->i_sb->s_id,
694 (long long)NFS_FILEID(inode),
695 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500696 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400698 if (NFS_STALE(inode))
699 goto out;
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500702 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 if (desc.ctx == NULL)
704 return -EBADF;
705 } else
706 desc.ctx = get_nfs_open_context((struct nfs_open_context *)
707 filp->private_data);
708 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
709 if (!list_empty(&head)) {
710 int err = nfs_pagein_list(&head, server->rpages);
711 if (!ret)
Chuck Lever91d5b472006-03-20 13:44:14 -0500712 nfs_add_stats(inode, NFSIOS_READPAGES, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 ret = err;
714 }
715 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400716out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 return ret;
718}
719
David Howellsf7b422b2006-06-09 09:34:33 -0400720int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
723 sizeof(struct nfs_read_data),
724 0, SLAB_HWCACHE_ALIGN,
725 NULL, NULL);
726 if (nfs_rdata_cachep == NULL)
727 return -ENOMEM;
728
Matthew Dobson93d23412006-03-26 01:37:50 -0800729 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
730 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 if (nfs_rdata_mempool == NULL)
732 return -ENOMEM;
733
734 return 0;
735}
736
David Brownell266bee82006-06-27 12:59:15 -0700737void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700740 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}