blob: 8fd6dfbe1bc3b8c248595a85fc5d3f3b86779ee3 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 */
9
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#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>
22
23#include <asm/system.h>
24
Trond Myklebust49a70f22006-12-05 00:35:38 -050025#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050026#include "iostat.h"
27
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#define NFSDBG_FACILITY NFSDBG_PAGECACHE
29
Trond Myklebust8d5658c2007-04-10 09:26:35 -040030static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int);
31static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int);
Trond Myklebustec06c092006-03-20 13:44:27 -050032static const struct rpc_call_ops nfs_read_partial_ops;
33static const struct rpc_call_ops nfs_read_full_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Christoph Lametere18b8902006-12-06 20:33:20 -080035static struct kmem_cache *nfs_rdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050036static mempool_t *nfs_rdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#define MIN_POOL_READ (32)
39
Trond Myklebust8d5658c2007-04-10 09:26:35 -040040struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050041{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080042 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050043
44 if (p) {
45 memset(p, 0, sizeof(*p));
46 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070047 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040048 if (pagecount <= ARRAY_SIZE(p->page_array))
49 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050050 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040051 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
52 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050053 mempool_free(p, nfs_rdata_mempool);
54 p = NULL;
55 }
56 }
57 }
58 return p;
59}
60
Trond Myklebust8aca67f2006-11-13 16:23:44 -050061static void nfs_readdata_rcu_free(struct rcu_head *head)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050062{
Trond Myklebust8aca67f2006-11-13 16:23:44 -050063 struct nfs_read_data *p = container_of(head, struct nfs_read_data, task.u.tk_rcu);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050064 if (p && (p->pagevec != &p->page_array[0]))
65 kfree(p->pagevec);
66 mempool_free(p, nfs_rdata_mempool);
67}
68
Trond Myklebust8aca67f2006-11-13 16:23:44 -050069static void nfs_readdata_free(struct nfs_read_data *rdata)
70{
71 call_rcu_bh(&rdata->task.u.tk_rcu, nfs_readdata_rcu_free);
72}
73
Trond Myklebust963d8fe2006-01-03 09:55:04 +010074void nfs_readdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 nfs_readdata_free(data);
77}
78
79static
Linus Torvalds1da177e2005-04-16 15:20:36 -070080int nfs_return_empty_page(struct page *page)
81{
Nate Diller60945cb2007-05-10 22:55:08 -070082 zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 SetPageUptodate(page);
84 unlock_page(page);
85 return 0;
86}
87
Trond Myklebust1de3fc12006-05-25 01:40:44 -040088static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
89{
90 unsigned int remainder = data->args.count - data->res.count;
91 unsigned int base = data->args.pgbase + data->res.count;
92 unsigned int pglen;
93 struct page **pages;
94
95 if (data->res.eof == 0 || remainder == 0)
96 return;
97 /*
98 * Note: "remainder" can never be negative, since we check for
99 * this in the XDR code.
100 */
101 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
102 base &= ~PAGE_CACHE_MASK;
103 pglen = PAGE_CACHE_SIZE - base;
Trond Myklebust79558f32006-08-22 13:44:32 -0400104 for (;;) {
105 if (remainder <= pglen) {
Nate Diller60945cb2007-05-10 22:55:08 -0700106 zero_user_page(*pages, base, remainder, KM_USER0);
Trond Myklebust79558f32006-08-22 13:44:32 -0400107 break;
108 }
Nate Diller60945cb2007-05-10 22:55:08 -0700109 zero_user_page(*pages, base, pglen, KM_USER0);
Trond Myklebust79558f32006-08-22 13:44:32 -0400110 pages++;
111 remainder -= pglen;
112 pglen = PAGE_CACHE_SIZE;
113 base = 0;
114 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400115}
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
118 struct page *page)
119{
120 LIST_HEAD(one_request);
121 struct nfs_page *new;
122 unsigned int len;
123
Trond Myklebust49a70f22006-12-05 00:35:38 -0500124 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (len == 0)
126 return nfs_return_empty_page(page);
127 new = nfs_create_request(ctx, inode, page, 0, len);
128 if (IS_ERR(new)) {
129 unlock_page(page);
130 return PTR_ERR(new);
131 }
132 if (len < PAGE_CACHE_SIZE)
Nate Diller60945cb2007-05-10 22:55:08 -0700133 zero_user_page(page, len, PAGE_CACHE_SIZE - len, KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 nfs_list_add_request(new, &one_request);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400136 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400137 nfs_pagein_multi(inode, &one_request, 1, len, 0);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400138 else
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400139 nfs_pagein_one(inode, &one_request, 1, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 return 0;
141}
142
143static void nfs_readpage_release(struct nfs_page *req)
144{
145 unlock_page(req->wb_page);
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400148 req->wb_context->path.dentry->d_inode->i_sb->s_id,
149 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 req->wb_bytes,
151 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700152 nfs_clear_request(req);
153 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
156/*
157 * Set up the NFS read request struct
158 */
159static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500160 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 unsigned int count, unsigned int offset)
162{
Trond Myklebust84115e12007-07-14 15:39:59 -0400163 struct inode *inode = req->wb_context->path.dentry->d_inode;
164 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
Trond Myklebust07737692007-10-25 18:42:54 -0400165 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400166 struct rpc_message msg = {
167 .rpc_argp = &data->args,
168 .rpc_resp = &data->res,
169 .rpc_cred = req->wb_context->cred,
170 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400171 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -0400172 .task = &data->task,
Trond Myklebust84115e12007-07-14 15:39:59 -0400173 .rpc_client = NFS_CLIENT(inode),
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400174 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400175 .callback_ops = call_ops,
176 .callback_data = data,
177 .flags = RPC_TASK_ASYNC | swap_flags,
178 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 data->req = req;
Trond Myklebust84115e12007-07-14 15:39:59 -0400181 data->inode = inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400182 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 data->args.fh = NFS_FH(inode);
185 data->args.offset = req_offset(req) + offset;
186 data->args.pgbase = req->wb_pgbase + offset;
187 data->args.pages = data->pagevec;
188 data->args.count = count;
189 data->args.context = req->wb_context;
190
191 data->res.fattr = &data->fattr;
192 data->res.count = count;
193 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400194 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Trond Myklebustec06c092006-03-20 13:44:27 -0500196 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400197 NFS_PROTO(inode)->read_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Chuck Levera3f565b2007-01-31 12:14:01 -0500199 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 data->task.tk_pid,
201 inode->i_sb->s_id,
202 (long long)NFS_FILEID(inode),
203 count,
204 (unsigned long long)data->args.offset);
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400205
Trond Myklebust07737692007-10-25 18:42:54 -0400206 task = rpc_run_task(&task_setup_data);
207 if (!IS_ERR(task))
208 rpc_put_task(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
211static void
212nfs_async_read_error(struct list_head *head)
213{
214 struct nfs_page *req;
215
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);
221 }
222}
223
224/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 * Generate multiple requests to fill a single page.
226 *
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.
231 *
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.
236 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400237static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
239 struct nfs_page *req = nfs_list_entry(head->next);
240 struct page *page = req->wb_page;
241 struct nfs_read_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700242 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
243 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 int requests = 0;
245 LIST_HEAD(list);
246
247 nfs_list_remove_request(req);
248
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400249 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700250 do {
251 size_t len = min(nbytes,rsize);
252
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400253 data = nfs_readdata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (!data)
255 goto out_bad;
256 INIT_LIST_HEAD(&data->pages);
257 list_add(&data->pages, &list);
258 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700259 nbytes -= len;
260 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 atomic_set(&req->wb_complete, requests);
262
263 ClearPageError(page);
264 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400265 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 do {
267 data = list_entry(list.next, struct nfs_read_data, pages);
268 list_del_init(&data->pages);
269
270 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400272 if (nbytes < rsize)
273 rsize = nbytes;
274 nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
275 rsize, offset);
276 offset += rsize;
277 nbytes -= rsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 } while (nbytes != 0);
279
280 return 0;
281
282out_bad:
283 while (!list_empty(&list)) {
284 data = list_entry(list.next, struct nfs_read_data, pages);
285 list_del(&data->pages);
286 nfs_readdata_free(data);
287 }
288 SetPageError(page);
289 nfs_readpage_release(req);
290 return -ENOMEM;
291}
292
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400293static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 struct nfs_page *req;
296 struct page **pages;
297 struct nfs_read_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400299 data = nfs_readdata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 if (!data)
301 goto out_bad;
302
303 INIT_LIST_HEAD(&data->pages);
304 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 while (!list_empty(head)) {
306 req = nfs_list_entry(head->next);
307 nfs_list_remove_request(req);
308 nfs_list_add_request(req, &data->pages);
309 ClearPageError(req->wb_page);
310 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312 req = nfs_list_entry(data->pages.next);
313
Trond Myklebustec06c092006-03-20 13:44:27 -0500314 nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 return 0;
316out_bad:
317 nfs_async_read_error(head);
318 return -ENOMEM;
319}
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500322 * This is the callback from RPC telling us whether a reply was
323 * received or some error occurred (timeout or socket shutdown).
324 */
325int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
326{
327 int status;
328
Chuck Levera3f565b2007-01-31 12:14:01 -0500329 dprintk("NFS: %s: %5u, (status %d)\n", __FUNCTION__, task->tk_pid,
Trond Myklebust0b671302006-11-14 16:12:23 -0500330 task->tk_status);
331
332 status = NFS_PROTO(data->inode)->read_done(task, data);
333 if (status != 0)
334 return status;
335
336 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
337
338 if (task->tk_status == -ESTALE) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200339 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
Trond Myklebust0b671302006-11-14 16:12:23 -0500340 nfs_mark_for_revalidate(data->inode);
341 }
Trond Myklebust0b671302006-11-14 16:12:23 -0500342 return 0;
343}
344
345static int nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
346{
347 struct nfs_readargs *argp = &data->args;
348 struct nfs_readres *resp = &data->res;
349
350 if (resp->eof || resp->count == argp->count)
351 return 0;
352
353 /* This is a short read! */
354 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
355 /* Has the server at least made some progress? */
356 if (resp->count == 0)
357 return 0;
358
359 /* Yes, so retry the read at the end of the data */
360 argp->offset += resp->count;
361 argp->pgbase += resp->count;
362 argp->count -= resp->count;
363 rpc_restart_call(task);
364 return -EAGAIN;
365}
366
367/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 * Handle a read reply that fills part of a page.
369 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500370static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Trond Myklebustec06c092006-03-20 13:44:27 -0500372 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 struct nfs_page *req = data->req;
374 struct page *page = req->wb_page;
375
Trond Myklebustec06c092006-03-20 13:44:27 -0500376 if (nfs_readpage_result(task, data) != 0)
377 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500378
379 if (likely(task->tk_status >= 0)) {
380 nfs_readpage_truncate_uninitialised_page(data);
381 if (nfs_readpage_retry(task, data) != 0)
382 return;
383 }
384 if (unlikely(task->tk_status < 0))
385 SetPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 if (atomic_dec_and_test(&req->wb_complete)) {
387 if (!PageError(page))
388 SetPageUptodate(page);
389 nfs_readpage_release(req);
390 }
391}
392
Trond Myklebustec06c092006-03-20 13:44:27 -0500393static const struct rpc_call_ops nfs_read_partial_ops = {
394 .rpc_call_done = nfs_readpage_result_partial,
395 .rpc_release = nfs_readdata_release,
396};
397
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400398static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
399{
400 unsigned int count = data->res.count;
401 unsigned int base = data->args.pgbase;
402 struct page **pages;
403
Trond Myklebust79558f32006-08-22 13:44:32 -0400404 if (data->res.eof)
405 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400406 if (unlikely(count == 0))
407 return;
408 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
409 base &= ~PAGE_CACHE_MASK;
410 count += base;
411 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
412 SetPageUptodate(*pages);
Trond Myklebust0b671302006-11-14 16:12:23 -0500413 if (count == 0)
414 return;
415 /* Was this a short read? */
416 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400417 SetPageUptodate(*pages);
418}
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420/*
421 * This is the callback from RPC telling us whether a reply was
422 * received or some error occurred (timeout or socket shutdown).
423 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500424static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Trond Myklebustec06c092006-03-20 13:44:27 -0500426 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Trond Myklebust0b671302006-11-14 16:12:23 -0500428 if (nfs_readpage_result(task, data) != 0)
429 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400430 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500431 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400432 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500433 * to ensure that we call nfs_readpage_set_pages_uptodate()
434 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400435 */
436 if (likely(task->tk_status >= 0)) {
437 nfs_readpage_truncate_uninitialised_page(data);
438 nfs_readpage_set_pages_uptodate(data);
Trond Myklebust0b671302006-11-14 16:12:23 -0500439 if (nfs_readpage_retry(task, data) != 0)
440 return;
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 while (!list_empty(&data->pages)) {
443 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400445 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 nfs_readpage_release(req);
447 }
448}
449
Trond Myklebustec06c092006-03-20 13:44:27 -0500450static const struct rpc_call_ops nfs_read_full_ops = {
451 .rpc_call_done = nfs_readpage_result_full,
452 .rpc_release = nfs_readdata_release,
453};
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * Read a page over NFS.
457 * We read the page synchronously in the following case:
458 * - The error flag is set for this page. This happens only when a
459 * previous async read operation failed.
460 */
461int nfs_readpage(struct file *file, struct page *page)
462{
463 struct nfs_open_context *ctx;
464 struct inode *inode = page->mapping->host;
465 int error;
466
467 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
468 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500469 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
470 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 /*
473 * Try to flush any pending writes to the file..
474 *
475 * NOTE! Because we own the page lock, there cannot
476 * be any new pending writes generated at this point
477 * for this page (other pages can be written to).
478 */
479 error = nfs_wb_page(inode, page);
480 if (error)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400481 goto out_unlock;
482 if (PageUptodate(page))
483 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400485 error = -ESTALE;
486 if (NFS_STALE(inode))
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400487 goto out_unlock;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500490 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500491 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (ctx == NULL)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400493 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400495 ctx = get_nfs_open_context(nfs_file_open_context(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Trond Myklebust8e0969f2006-12-13 15:23:44 -0500497 error = nfs_readpage_async(ctx, inode, page);
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 put_nfs_open_context(ctx);
500 return error;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400501out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 unlock_page(page);
503 return error;
504}
505
506struct nfs_readdesc {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400507 struct nfs_pageio_descriptor *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct nfs_open_context *ctx;
509};
510
511static int
512readpage_async_filler(void *data, struct page *page)
513{
514 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
515 struct inode *inode = page->mapping->host;
516 struct nfs_page *new;
517 unsigned int len;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400518 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400520 error = nfs_wb_page(inode, page);
521 if (error)
522 goto out_unlock;
523 if (PageUptodate(page))
524 goto out_unlock;
525
Trond Myklebust49a70f22006-12-05 00:35:38 -0500526 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (len == 0)
528 return nfs_return_empty_page(page);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 new = nfs_create_request(desc->ctx, inode, page, 0, len);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400531 if (IS_ERR(new))
532 goto out_error;
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (len < PAGE_CACHE_SIZE)
Nate Diller60945cb2007-05-10 22:55:08 -0700535 zero_user_page(page, len, PAGE_CACHE_SIZE - len, KM_USER0);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400536 nfs_pageio_add_request(desc->pgio, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return 0;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400538out_error:
539 error = PTR_ERR(new);
540 SetPageError(page);
541out_unlock:
542 unlock_page(page);
543 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544}
545
546int nfs_readpages(struct file *filp, struct address_space *mapping,
547 struct list_head *pages, unsigned nr_pages)
548{
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400549 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 struct nfs_readdesc desc = {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400551 .pgio = &pgio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 };
553 struct inode *inode = mapping->host;
554 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400555 size_t rsize = server->rsize;
556 unsigned long npages;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400557 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
560 inode->i_sb->s_id,
561 (long long)NFS_FILEID(inode),
562 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500563 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400565 if (NFS_STALE(inode))
566 goto out;
567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500569 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (desc.ctx == NULL)
571 return -EBADF;
572 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400573 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400574 if (rsize < PAGE_CACHE_SIZE)
575 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
576 else
577 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400580
581 nfs_pageio_complete(&pgio);
582 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
583 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400585out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return ret;
587}
588
David Howellsf7b422b2006-06-09 09:34:33 -0400589int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
591 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
592 sizeof(struct nfs_read_data),
593 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900594 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (nfs_rdata_cachep == NULL)
596 return -ENOMEM;
597
Matthew Dobson93d23412006-03-26 01:37:50 -0800598 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
599 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (nfs_rdata_mempool == NULL)
601 return -ENOMEM;
602
603 return 0;
604}
605
David Brownell266bee82006-06-27 12:59:15 -0700606void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700609 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}