blob: 16f57e0af999a51dc166b181dd01c3e8ef72e79d [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 Myklebust5e4424a2008-02-25 21:53:49 -080061static void nfs_readdata_free(struct nfs_read_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050062{
63 if (p && (p->pagevec != &p->page_array[0]))
64 kfree(p->pagevec);
65 mempool_free(p, nfs_rdata_mempool);
66}
67
Trond Myklebust963d8fe2006-01-03 09:55:04 +010068void nfs_readdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
Trond Myklebust383ba712008-02-19 20:04:20 -050070 struct nfs_read_data *rdata = data;
71
72 put_nfs_open_context(rdata->args.context);
73 nfs_readdata_free(rdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074}
75
76static
Linus Torvalds1da177e2005-04-16 15:20:36 -070077int nfs_return_empty_page(struct page *page)
78{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080079 zero_user(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 SetPageUptodate(page);
81 unlock_page(page);
82 return 0;
83}
84
Trond Myklebust1de3fc12006-05-25 01:40:44 -040085static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
86{
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;
91
92 if (data->res.eof == 0 || remainder == 0)
93 return;
94 /*
95 * Note: "remainder" can never be negative, since we check for
96 * this in the XDR code.
97 */
98 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
99 base &= ~PAGE_CACHE_MASK;
100 pglen = PAGE_CACHE_SIZE - base;
Trond Myklebust79558f32006-08-22 13:44:32 -0400101 for (;;) {
102 if (remainder <= pglen) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800103 zero_user(*pages, base, remainder);
Trond Myklebust79558f32006-08-22 13:44:32 -0400104 break;
105 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800106 zero_user(*pages, base, pglen);
Trond Myklebust79558f32006-08-22 13:44:32 -0400107 pages++;
108 remainder -= pglen;
109 pglen = PAGE_CACHE_SIZE;
110 base = 0;
111 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400112}
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
115 struct page *page)
116{
117 LIST_HEAD(one_request);
118 struct nfs_page *new;
119 unsigned int len;
120
Trond Myklebust49a70f22006-12-05 00:35:38 -0500121 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 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);
128 }
129 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800130 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 nfs_list_add_request(new, &one_request);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400133 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400134 nfs_pagein_multi(inode, &one_request, 1, len, 0);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400135 else
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400136 nfs_pagein_one(inode, &one_request, 1, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 return 0;
138}
139
140static void nfs_readpage_release(struct nfs_page *req)
141{
142 unlock_page(req->wb_page);
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400145 req->wb_context->path.dentry->d_inode->i_sb->s_id,
146 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 req->wb_bytes,
148 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700149 nfs_clear_request(req);
150 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
153/*
154 * Set up the NFS read request struct
155 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400156static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500157 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 unsigned int count, unsigned int offset)
159{
Trond Myklebust84115e12007-07-14 15:39:59 -0400160 struct inode *inode = req->wb_context->path.dentry->d_inode;
161 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
Trond Myklebust07737692007-10-25 18:42:54 -0400162 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400163 struct rpc_message msg = {
164 .rpc_argp = &data->args,
165 .rpc_resp = &data->res,
166 .rpc_cred = req->wb_context->cred,
167 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400168 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -0400169 .task = &data->task,
Trond Myklebust84115e12007-07-14 15:39:59 -0400170 .rpc_client = NFS_CLIENT(inode),
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400171 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400172 .callback_ops = call_ops,
173 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500174 .workqueue = nfsiod_workqueue,
Trond Myklebust84115e12007-07-14 15:39:59 -0400175 .flags = RPC_TASK_ASYNC | swap_flags,
176 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 data->req = req;
Trond Myklebust84115e12007-07-14 15:39:59 -0400179 data->inode = inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400180 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
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;
Trond Myklebust383ba712008-02-19 20:04:20 -0500187 data->args.context = get_nfs_open_context(req->wb_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 data->res.fattr = &data->fattr;
190 data->res.count = count;
191 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400192 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Trond Myklebustec06c092006-03-20 13:44:27 -0500194 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400195 NFS_PROTO(inode)->read_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Chuck Levera3f565b2007-01-31 12:14:01 -0500197 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 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);
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400203
Trond Myklebust07737692007-10-25 18:42:54 -0400204 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400205 if (IS_ERR(task))
206 return PTR_ERR(task);
207 rpc_put_task(task);
208 return 0;
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;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400245 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 LIST_HEAD(list);
247
248 nfs_list_remove_request(req);
249
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400250 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700251 do {
252 size_t len = min(nbytes,rsize);
253
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400254 data = nfs_readdata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (!data)
256 goto out_bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 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 {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400267 int ret2;
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 data = list_entry(list.next, struct nfs_read_data, pages);
270 list_del_init(&data->pages);
271
272 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400274 if (nbytes < rsize)
275 rsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400276 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400277 rsize, offset);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400278 if (ret == 0)
279 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400280 offset += rsize;
281 nbytes -= rsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 } while (nbytes != 0);
283
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400284 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286out_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);
291 }
292 SetPageError(page);
293 nfs_readpage_release(req);
294 return -ENOMEM;
295}
296
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400297static 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 -0700298{
299 struct nfs_page *req;
300 struct page **pages;
301 struct nfs_read_data *data;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400302 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400304 data = nfs_readdata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 if (!data)
306 goto out_bad;
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 }
316 req = nfs_list_entry(data->pages.next);
317
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400318 return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319out_bad:
320 nfs_async_read_error(head);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400321 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500325 * This is the callback from RPC telling us whether a reply was
326 * received or some error occurred (timeout or socket shutdown).
327 */
328int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
329{
330 int status;
331
Chuck Levera3f565b2007-01-31 12:14:01 -0500332 dprintk("NFS: %s: %5u, (status %d)\n", __FUNCTION__, task->tk_pid,
Trond Myklebust0b671302006-11-14 16:12:23 -0500333 task->tk_status);
334
335 status = NFS_PROTO(data->inode)->read_done(task, data);
336 if (status != 0)
337 return status;
338
339 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
340
341 if (task->tk_status == -ESTALE) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200342 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
Trond Myklebust0b671302006-11-14 16:12:23 -0500343 nfs_mark_for_revalidate(data->inode);
344 }
Trond Myklebust0b671302006-11-14 16:12:23 -0500345 return 0;
346}
347
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400348static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
Trond Myklebust0b671302006-11-14 16:12:23 -0500349{
350 struct nfs_readargs *argp = &data->args;
351 struct nfs_readres *resp = &data->res;
352
353 if (resp->eof || resp->count == argp->count)
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400354 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500355
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)
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400360 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500361
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);
Trond Myklebust0b671302006-11-14 16:12:23 -0500367}
368
369/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * Handle a read reply that fills part of a page.
371 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500372static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Trond Myklebustec06c092006-03-20 13:44:27 -0500374 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Trond Myklebustec06c092006-03-20 13:44:27 -0500376 if (nfs_readpage_result(task, data) != 0)
377 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400378 if (task->tk_status < 0)
379 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500380
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400381 nfs_readpage_truncate_uninitialised_page(data);
382 nfs_readpage_retry(task, data);
383}
384
385static void nfs_readpage_release_partial(void *calldata)
386{
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;
391
392 if (status < 0)
Trond Myklebust0b671302006-11-14 16:12:23 -0500393 SetPageError(page);
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 if (atomic_dec_and_test(&req->wb_complete)) {
396 if (!PageError(page))
397 SetPageUptodate(page);
398 nfs_readpage_release(req);
399 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400400 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
Trond Myklebustec06c092006-03-20 13:44:27 -0500403static const struct rpc_call_ops nfs_read_partial_ops = {
404 .rpc_call_done = nfs_readpage_result_partial,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400405 .rpc_release = nfs_readpage_release_partial,
Trond Myklebustec06c092006-03-20 13:44:27 -0500406};
407
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400408static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
409{
410 unsigned int count = data->res.count;
411 unsigned int base = data->args.pgbase;
412 struct page **pages;
413
Trond Myklebust79558f32006-08-22 13:44:32 -0400414 if (data->res.eof)
415 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400416 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);
Trond Myklebust0b671302006-11-14 16:12:23 -0500423 if (count == 0)
424 return;
425 /* Was this a short read? */
426 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400427 SetPageUptodate(*pages);
428}
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430/*
431 * This is the callback from RPC telling us whether a reply was
432 * received or some error occurred (timeout or socket shutdown).
433 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500434static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
Trond Myklebustec06c092006-03-20 13:44:27 -0500436 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Trond Myklebust0b671302006-11-14 16:12:23 -0500438 if (nfs_readpage_result(task, data) != 0)
439 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400440 if (task->tk_status < 0)
441 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400442 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500443 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400444 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500445 * to ensure that we call nfs_readpage_set_pages_uptodate()
446 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400447 */
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400448 nfs_readpage_truncate_uninitialised_page(data);
449 nfs_readpage_set_pages_uptodate(data);
450 nfs_readpage_retry(task, data);
451}
452
453static void nfs_readpage_release_full(void *calldata)
454{
455 struct nfs_read_data *data = calldata;
456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 while (!list_empty(&data->pages)) {
458 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400460 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 nfs_readpage_release(req);
462 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400463 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
Trond Myklebustec06c092006-03-20 13:44:27 -0500466static const struct rpc_call_ops nfs_read_full_ops = {
467 .rpc_call_done = nfs_readpage_result_full,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400468 .rpc_release = nfs_readpage_release_full,
Trond Myklebustec06c092006-03-20 13:44:27 -0500469};
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 * 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.
476 */
477int nfs_readpage(struct file *file, struct page *page)
478{
479 struct nfs_open_context *ctx;
480 struct inode *inode = page->mapping->host;
481 int error;
482
483 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
484 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500485 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
486 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 /*
489 * Try to flush any pending writes to the file..
490 *
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).
494 */
495 error = nfs_wb_page(inode, page);
496 if (error)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400497 goto out_unlock;
498 if (PageUptodate(page))
499 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400501 error = -ESTALE;
502 if (NFS_STALE(inode))
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400503 goto out_unlock;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500506 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500507 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (ctx == NULL)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400509 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400511 ctx = get_nfs_open_context(nfs_file_open_context(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Trond Myklebust8e0969f2006-12-13 15:23:44 -0500513 error = nfs_readpage_async(ctx, inode, page);
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 put_nfs_open_context(ctx);
516 return error;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400517out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 unlock_page(page);
519 return error;
520}
521
522struct nfs_readdesc {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400523 struct nfs_pageio_descriptor *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 struct nfs_open_context *ctx;
525};
526
527static int
528readpage_async_filler(void *data, struct page *page)
529{
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;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400534 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400536 error = nfs_wb_page(inode, page);
537 if (error)
538 goto out_unlock;
539 if (PageUptodate(page))
540 goto out_unlock;
541
Trond Myklebust49a70f22006-12-05 00:35:38 -0500542 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 if (len == 0)
544 return nfs_return_empty_page(page);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 new = nfs_create_request(desc->ctx, inode, page, 0, len);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400547 if (IS_ERR(new))
548 goto out_error;
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800551 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Fred Isamanf8512ad2008-03-19 11:24:39 -0400552 if (!nfs_pageio_add_request(desc->pgio, new)) {
553 error = desc->pgio->pg_error;
554 goto out_unlock;
555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return 0;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400557out_error:
558 error = PTR_ERR(new);
559 SetPageError(page);
560out_unlock:
561 unlock_page(page);
562 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
564
565int nfs_readpages(struct file *filp, struct address_space *mapping,
566 struct list_head *pages, unsigned nr_pages)
567{
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400568 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 struct nfs_readdesc desc = {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400570 .pgio = &pgio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 };
572 struct inode *inode = mapping->host;
573 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400574 size_t rsize = server->rsize;
575 unsigned long npages;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400576 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
579 inode->i_sb->s_id,
580 (long long)NFS_FILEID(inode),
581 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500582 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400584 if (NFS_STALE(inode))
585 goto out;
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500588 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 if (desc.ctx == NULL)
590 return -EBADF;
591 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400592 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400593 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);
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400599
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400604out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return ret;
606}
607
David Howellsf7b422b2006-06-09 09:34:33 -0400608int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
610 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
611 sizeof(struct nfs_read_data),
612 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900613 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (nfs_rdata_cachep == NULL)
615 return -ENOMEM;
616
Matthew Dobson93d23412006-03-26 01:37:50 -0800617 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
618 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 if (nfs_rdata_mempool == NULL)
620 return -ENOMEM;
621
622 return 0;
623}
624
David Brownell266bee82006-06-27 12:59:15 -0700625void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626{
627 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700628 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}