blob: 20a7f952e244c763065ce01f1b44aa34c3e2c25d [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>
Andy Adamson64419a92011-03-01 01:34:16 +000021#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23#include <asm/system.h>
Fred Isamanbae724e2011-03-01 01:34:15 +000024#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Andy Adamsonf11c88a2009-04-01 09:22:25 -040026#include "nfs4_fs.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050027#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050028#include "iostat.h"
David Howells9a9fc1c2009-04-03 16:42:44 +010029#include "fscache.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050030
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#define NFSDBG_FACILITY NFSDBG_PAGECACHE
32
Fred Isamanc76069b2011-03-03 15:13:48 +000033static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc);
34static int nfs_pagein_one(struct nfs_pageio_descriptor *desc);
Trond Myklebustec06c092006-03-20 13:44:27 -050035static const struct rpc_call_ops nfs_read_partial_ops;
36static const struct rpc_call_ops nfs_read_full_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Christoph Lametere18b8902006-12-06 20:33:20 -080038static struct kmem_cache *nfs_rdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050039static mempool_t *nfs_rdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define MIN_POOL_READ (32)
42
Trond Myklebust8d5658c2007-04-10 09:26:35 -040043struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050044{
Trond Myklebust93870d72010-05-13 12:51:03 -040045 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050046
47 if (p) {
48 memset(p, 0, sizeof(*p));
49 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070050 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040051 if (pagecount <= ARRAY_SIZE(p->page_array))
52 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050053 else {
Trond Myklebust93870d72010-05-13 12:51:03 -040054 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040055 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050056 mempool_free(p, nfs_rdata_mempool);
57 p = NULL;
58 }
59 }
60 }
61 return p;
62}
63
Trond Myklebust1ae88b22009-08-12 09:12:30 -040064void nfs_readdata_free(struct nfs_read_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050065{
66 if (p && (p->pagevec != &p->page_array[0]))
67 kfree(p->pagevec);
68 mempool_free(p, nfs_rdata_mempool);
69}
70
Trond Myklebust1ae88b22009-08-12 09:12:30 -040071static void nfs_readdata_release(struct nfs_read_data *rdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
Fred Isamanbae724e2011-03-01 01:34:15 +000073 put_lseg(rdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -050074 put_nfs_open_context(rdata->args.context);
75 nfs_readdata_free(rdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076}
77
78static
Linus Torvalds1da177e2005-04-16 15:20:36 -070079int nfs_return_empty_page(struct page *page)
80{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080081 zero_user(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 SetPageUptodate(page);
83 unlock_page(page);
84 return 0;
85}
86
Trond Myklebust1de3fc12006-05-25 01:40:44 -040087static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
88{
89 unsigned int remainder = data->args.count - data->res.count;
90 unsigned int base = data->args.pgbase + data->res.count;
91 unsigned int pglen;
92 struct page **pages;
93
94 if (data->res.eof == 0 || remainder == 0)
95 return;
96 /*
97 * Note: "remainder" can never be negative, since we check for
98 * this in the XDR code.
99 */
100 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
101 base &= ~PAGE_CACHE_MASK;
102 pglen = PAGE_CACHE_SIZE - base;
Trond Myklebust79558f32006-08-22 13:44:32 -0400103 for (;;) {
104 if (remainder <= pglen) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800105 zero_user(*pages, base, remainder);
Trond Myklebust79558f32006-08-22 13:44:32 -0400106 break;
107 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800108 zero_user(*pages, base, pglen);
Trond Myklebust79558f32006-08-22 13:44:32 -0400109 pages++;
110 remainder -= pglen;
111 pglen = PAGE_CACHE_SIZE;
112 base = 0;
113 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400114}
115
David Howellsf42b2932009-04-03 16:42:44 +0100116int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
117 struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 struct nfs_page *new;
120 unsigned int len;
Fred Isamanc76069b2011-03-03 15:13:48 +0000121 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Trond Myklebust49a70f22006-12-05 00:35:38 -0500123 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 if (len == 0)
125 return nfs_return_empty_page(page);
126 new = nfs_create_request(ctx, inode, page, 0, len);
127 if (IS_ERR(new)) {
128 unlock_page(page);
129 return PTR_ERR(new);
130 }
131 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800132 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Fred Isamanc76069b2011-03-03 15:13:48 +0000134 nfs_pageio_init(&pgio, inode, NULL, 0, 0);
135 nfs_list_add_request(new, &pgio.pg_list);
136 pgio.pg_count = len;
137
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400138 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
Fred Isamanc76069b2011-03-03 15:13:48 +0000139 nfs_pagein_multi(&pgio);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400140 else
Fred Isamanc76069b2011-03-03 15:13:48 +0000141 nfs_pagein_one(&pgio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 return 0;
143}
144
145static void nfs_readpage_release(struct nfs_page *req)
146{
David Howells7f8e05f2009-04-03 16:42:45 +0100147 struct inode *d_inode = req->wb_context->path.dentry->d_inode;
148
149 if (PageUptodate(req->wb_page))
150 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 unlock_page(req->wb_page);
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400155 req->wb_context->path.dentry->d_inode->i_sb->s_id,
156 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 req->wb_bytes,
158 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700159 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000162int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt,
Andy Adamson64419a92011-03-01 01:34:16 +0000163 const struct rpc_call_ops *call_ops)
164{
165 struct inode *inode = data->inode;
166 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
167 struct rpc_task *task;
168 struct rpc_message msg = {
169 .rpc_argp = &data->args,
170 .rpc_resp = &data->res,
171 .rpc_cred = data->cred,
172 };
173 struct rpc_task_setup task_setup_data = {
174 .task = &data->task,
175 .rpc_client = clnt,
176 .rpc_message = &msg,
177 .callback_ops = call_ops,
178 .callback_data = data,
179 .workqueue = nfsiod_workqueue,
180 .flags = RPC_TASK_ASYNC | swap_flags,
181 };
182
183 /* Set up the initial task struct. */
184 NFS_PROTO(inode)->read_setup(data, &msg);
185
186 dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
187 "offset %llu)\n",
188 data->task.tk_pid,
189 inode->i_sb->s_id,
190 (long long)NFS_FILEID(inode),
191 data->args.count,
192 (unsigned long long)data->args.offset);
193
194 task = rpc_run_task(&task_setup_data);
195 if (IS_ERR(task))
196 return PTR_ERR(task);
197 rpc_put_task(task);
198 return 0;
199}
Andy Adamsondc70d7b2011-03-01 01:34:19 +0000200EXPORT_SYMBOL_GPL(nfs_initiate_read);
Andy Adamson64419a92011-03-01 01:34:16 +0000201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/*
203 * Set up the NFS read request struct
204 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400205static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500206 const struct rpc_call_ops *call_ops,
Fred Isamanbae724e2011-03-01 01:34:15 +0000207 unsigned int count, unsigned int offset,
208 struct pnfs_layout_segment *lseg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
Trond Myklebust84115e12007-07-14 15:39:59 -0400210 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212 data->req = req;
Trond Myklebust84115e12007-07-14 15:39:59 -0400213 data->inode = inode;
Andy Adamson64419a92011-03-01 01:34:16 +0000214 data->cred = req->wb_context->cred;
Fred Isamanbae724e2011-03-01 01:34:15 +0000215 data->lseg = get_lseg(lseg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 data->args.fh = NFS_FH(inode);
218 data->args.offset = req_offset(req) + offset;
219 data->args.pgbase = req->wb_pgbase + offset;
220 data->args.pages = data->pagevec;
221 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500222 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400223 data->args.lock_context = req->wb_lock_context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
225 data->res.fattr = &data->fattr;
226 data->res.count = count;
227 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400228 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Andy Adamson64419a92011-03-01 01:34:16 +0000230 if (data->lseg &&
231 (pnfs_try_to_read_data(data, call_ops) == PNFS_ATTEMPTED))
232 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Andy Adamson64419a92011-03-01 01:34:16 +0000234 return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
237static void
238nfs_async_read_error(struct list_head *head)
239{
240 struct nfs_page *req;
241
242 while (!list_empty(head)) {
243 req = nfs_list_entry(head->next);
244 nfs_list_remove_request(req);
245 SetPageError(req->wb_page);
246 nfs_readpage_release(req);
247 }
248}
249
250/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 * Generate multiple requests to fill a single page.
252 *
253 * We optimize to reduce the number of read operations on the wire. If we
254 * detect that we're reading a page, or an area of a page, that is past the
255 * end of file, we do not generate NFS read operations but just clear the
256 * parts of the page that would have come back zero from the server anyway.
257 *
258 * We rely on the cached value of i_size to make this determination; another
259 * client can fill pages on the server past our cached end-of-file, but we
260 * won't see the new data until our attribute cache is updated. This is more
261 * or less conventional NFS client behavior.
262 */
Fred Isamanc76069b2011-03-03 15:13:48 +0000263static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
Fred Isamanc76069b2011-03-03 15:13:48 +0000265 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 struct page *page = req->wb_page;
267 struct nfs_read_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000268 size_t rsize = NFS_SERVER(desc->pg_inode)->rsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700269 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400271 int ret = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000272 struct pnfs_layout_segment *lseg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 LIST_HEAD(list);
274
275 nfs_list_remove_request(req);
276
Fred Isamanc76069b2011-03-03 15:13:48 +0000277 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700278 do {
279 size_t len = min(nbytes,rsize);
280
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400281 data = nfs_readdata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 if (!data)
283 goto out_bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 list_add(&data->pages, &list);
285 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700286 nbytes -= len;
287 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 atomic_set(&req->wb_complete, requests);
289
Fred Isamanc76069b2011-03-03 15:13:48 +0000290 BUG_ON(desc->pg_lseg != NULL);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300291 lseg = pnfs_update_layout(desc->pg_inode, req->wb_context,
292 req_offset(req), desc->pg_count,
293 IOMODE_READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 ClearPageError(page);
295 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000296 nbytes = desc->pg_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 do {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400298 int ret2;
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 data = list_entry(list.next, struct nfs_read_data, pages);
301 list_del_init(&data->pages);
302
303 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400305 if (nbytes < rsize)
306 rsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400307 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
Fred Isamanbae724e2011-03-01 01:34:15 +0000308 rsize, offset, lseg);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400309 if (ret == 0)
310 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400311 offset += rsize;
312 nbytes -= rsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 } while (nbytes != 0);
Fred Isamanbae724e2011-03-01 01:34:15 +0000314 put_lseg(lseg);
Fred Isaman36fe4322011-03-03 15:13:49 +0000315 desc->pg_lseg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400317 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319out_bad:
320 while (!list_empty(&list)) {
321 data = list_entry(list.next, struct nfs_read_data, pages);
322 list_del(&data->pages);
323 nfs_readdata_free(data);
324 }
325 SetPageError(page);
326 nfs_readpage_release(req);
327 return -ENOMEM;
328}
329
Fred Isamanc76069b2011-03-03 15:13:48 +0000330static int nfs_pagein_one(struct nfs_pageio_descriptor *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 struct nfs_page *req;
333 struct page **pages;
334 struct nfs_read_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000335 struct list_head *head = &desc->pg_list;
336 struct pnfs_layout_segment *lseg = desc->pg_lseg;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400337 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Fred Isamanc76069b2011-03-03 15:13:48 +0000339 data = nfs_readdata_alloc(nfs_page_array_len(desc->pg_base,
340 desc->pg_count));
Fred Isamanbae724e2011-03-01 01:34:15 +0000341 if (!data) {
342 nfs_async_read_error(head);
343 goto out;
344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 while (!list_empty(head)) {
348 req = nfs_list_entry(head->next);
349 nfs_list_remove_request(req);
350 nfs_list_add_request(req, &data->pages);
351 ClearPageError(req->wb_page);
352 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 }
354 req = nfs_list_entry(data->pages.next);
Fred Isamanbae724e2011-03-01 01:34:15 +0000355 if ((!lseg) && list_is_singular(&data->pages))
Benny Halevyfb3296e2011-05-22 19:47:26 +0300356 lseg = pnfs_update_layout(desc->pg_inode, req->wb_context,
357 req_offset(req), desc->pg_count,
358 IOMODE_READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Fred Isamanc76069b2011-03-03 15:13:48 +0000360 ret = nfs_read_rpcsetup(req, data, &nfs_read_full_ops, desc->pg_count,
361 0, lseg);
Fred Isamanbae724e2011-03-01 01:34:15 +0000362out:
363 put_lseg(lseg);
Fred Isaman36fe4322011-03-03 15:13:49 +0000364 desc->pg_lseg = NULL;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400365 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500369 * This is the callback from RPC telling us whether a reply was
370 * received or some error occurred (timeout or socket shutdown).
371 */
372int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
373{
374 int status;
375
Harvey Harrison3110ff82008-05-02 13:42:44 -0700376 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
Trond Myklebust0b671302006-11-14 16:12:23 -0500377 task->tk_status);
378
379 status = NFS_PROTO(data->inode)->read_done(task, data);
380 if (status != 0)
381 return status;
382
383 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
384
385 if (task->tk_status == -ESTALE) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200386 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
Trond Myklebust0b671302006-11-14 16:12:23 -0500387 nfs_mark_for_revalidate(data->inode);
388 }
Trond Myklebust0b671302006-11-14 16:12:23 -0500389 return 0;
390}
391
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400392static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
Trond Myklebust0b671302006-11-14 16:12:23 -0500393{
394 struct nfs_readargs *argp = &data->args;
395 struct nfs_readres *resp = &data->res;
396
397 if (resp->eof || resp->count == argp->count)
Trond Myklebustd61e6122009-12-05 19:32:19 -0500398 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500399
400 /* This is a short read! */
401 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
402 /* Has the server at least made some progress? */
403 if (resp->count == 0)
Trond Myklebustd61e6122009-12-05 19:32:19 -0500404 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500405
406 /* Yes, so retry the read at the end of the data */
Andy Adamsoncbdabc7f2011-03-01 01:34:20 +0000407 data->mds_offset += resp->count;
Trond Myklebust0b671302006-11-14 16:12:23 -0500408 argp->offset += resp->count;
409 argp->pgbase += resp->count;
410 argp->count -= resp->count;
Trond Myklebust0110ee12009-12-07 09:00:24 -0500411 nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client);
Trond Myklebust0b671302006-11-14 16:12:23 -0500412}
413
414/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 * Handle a read reply that fills part of a page.
416 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500417static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
Trond Myklebustec06c092006-03-20 13:44:27 -0500419 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Trond Myklebustec06c092006-03-20 13:44:27 -0500421 if (nfs_readpage_result(task, data) != 0)
422 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400423 if (task->tk_status < 0)
424 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500425
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400426 nfs_readpage_truncate_uninitialised_page(data);
427 nfs_readpage_retry(task, data);
428}
429
430static void nfs_readpage_release_partial(void *calldata)
431{
432 struct nfs_read_data *data = calldata;
433 struct nfs_page *req = data->req;
434 struct page *page = req->wb_page;
435 int status = data->task.tk_status;
436
437 if (status < 0)
Trond Myklebust0b671302006-11-14 16:12:23 -0500438 SetPageError(page);
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (atomic_dec_and_test(&req->wb_complete)) {
441 if (!PageError(page))
442 SetPageUptodate(page);
443 nfs_readpage_release(req);
444 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400445 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400448#if defined(CONFIG_NFS_V4_1)
449void nfs_read_prepare(struct rpc_task *task, void *calldata)
450{
451 struct nfs_read_data *data = calldata;
452
Trond Myklebust035168a2010-06-16 09:52:26 -0400453 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400454 &data->args.seq_args, &data->res.seq_res,
455 0, task))
456 return;
457 rpc_call_start(task);
458}
459#endif /* CONFIG_NFS_V4_1 */
460
Trond Myklebustec06c092006-03-20 13:44:27 -0500461static const struct rpc_call_ops nfs_read_partial_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400462#if defined(CONFIG_NFS_V4_1)
463 .rpc_call_prepare = nfs_read_prepare,
464#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500465 .rpc_call_done = nfs_readpage_result_partial,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400466 .rpc_release = nfs_readpage_release_partial,
Trond Myklebustec06c092006-03-20 13:44:27 -0500467};
468
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400469static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
470{
471 unsigned int count = data->res.count;
472 unsigned int base = data->args.pgbase;
473 struct page **pages;
474
Trond Myklebust79558f32006-08-22 13:44:32 -0400475 if (data->res.eof)
476 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400477 if (unlikely(count == 0))
478 return;
479 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
480 base &= ~PAGE_CACHE_MASK;
481 count += base;
482 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
483 SetPageUptodate(*pages);
Trond Myklebust0b671302006-11-14 16:12:23 -0500484 if (count == 0)
485 return;
486 /* Was this a short read? */
487 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400488 SetPageUptodate(*pages);
489}
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491/*
492 * This is the callback from RPC telling us whether a reply was
493 * received or some error occurred (timeout or socket shutdown).
494 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500495static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Trond Myklebustec06c092006-03-20 13:44:27 -0500497 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Trond Myklebust0b671302006-11-14 16:12:23 -0500499 if (nfs_readpage_result(task, data) != 0)
500 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400501 if (task->tk_status < 0)
502 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400503 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500504 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400505 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500506 * to ensure that we call nfs_readpage_set_pages_uptodate()
507 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400508 */
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400509 nfs_readpage_truncate_uninitialised_page(data);
510 nfs_readpage_set_pages_uptodate(data);
511 nfs_readpage_retry(task, data);
512}
513
514static void nfs_readpage_release_full(void *calldata)
515{
516 struct nfs_read_data *data = calldata;
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 while (!list_empty(&data->pages)) {
519 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400521 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 nfs_readpage_release(req);
523 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400524 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
Trond Myklebustec06c092006-03-20 13:44:27 -0500527static const struct rpc_call_ops nfs_read_full_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400528#if defined(CONFIG_NFS_V4_1)
529 .rpc_call_prepare = nfs_read_prepare,
530#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500531 .rpc_call_done = nfs_readpage_result_full,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400532 .rpc_release = nfs_readpage_release_full,
Trond Myklebustec06c092006-03-20 13:44:27 -0500533};
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * Read a page over NFS.
537 * We read the page synchronously in the following case:
538 * - The error flag is set for this page. This happens only when a
539 * previous async read operation failed.
540 */
541int nfs_readpage(struct file *file, struct page *page)
542{
543 struct nfs_open_context *ctx;
544 struct inode *inode = page->mapping->host;
545 int error;
546
547 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
548 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500549 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
550 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 /*
553 * Try to flush any pending writes to the file..
554 *
555 * NOTE! Because we own the page lock, there cannot
556 * be any new pending writes generated at this point
557 * for this page (other pages can be written to).
558 */
559 error = nfs_wb_page(inode, page);
560 if (error)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400561 goto out_unlock;
562 if (PageUptodate(page))
563 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400565 error = -ESTALE;
566 if (NFS_STALE(inode))
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400567 goto out_unlock;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500570 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500571 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (ctx == NULL)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400573 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400575 ctx = get_nfs_open_context(nfs_file_open_context(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
David Howells9a9fc1c2009-04-03 16:42:44 +0100577 if (!IS_SYNC(inode)) {
578 error = nfs_readpage_from_fscache(ctx, inode, page);
579 if (error == 0)
580 goto out;
581 }
582
Trond Myklebust8e0969f2006-12-13 15:23:44 -0500583 error = nfs_readpage_async(ctx, inode, page);
584
David Howells9a9fc1c2009-04-03 16:42:44 +0100585out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 put_nfs_open_context(ctx);
587 return error;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400588out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 unlock_page(page);
590 return error;
591}
592
593struct nfs_readdesc {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400594 struct nfs_pageio_descriptor *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 struct nfs_open_context *ctx;
596};
597
598static int
599readpage_async_filler(void *data, struct page *page)
600{
601 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
602 struct inode *inode = page->mapping->host;
603 struct nfs_page *new;
604 unsigned int len;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400605 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Trond Myklebust49a70f22006-12-05 00:35:38 -0500607 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 if (len == 0)
609 return nfs_return_empty_page(page);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 new = nfs_create_request(desc->ctx, inode, page, 0, len);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400612 if (IS_ERR(new))
613 goto out_error;
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800616 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Fred Isamanf8512ad2008-03-19 11:24:39 -0400617 if (!nfs_pageio_add_request(desc->pgio, new)) {
618 error = desc->pgio->pg_error;
619 goto out_unlock;
620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return 0;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400622out_error:
623 error = PTR_ERR(new);
624 SetPageError(page);
625out_unlock:
626 unlock_page(page);
627 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
630int nfs_readpages(struct file *filp, struct address_space *mapping,
631 struct list_head *pages, unsigned nr_pages)
632{
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400633 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 struct nfs_readdesc desc = {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400635 .pgio = &pgio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 };
637 struct inode *inode = mapping->host;
638 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400639 size_t rsize = server->rsize;
640 unsigned long npages;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400641 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
644 inode->i_sb->s_id,
645 (long long)NFS_FILEID(inode),
646 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500647 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400649 if (NFS_STALE(inode))
650 goto out;
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500653 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (desc.ctx == NULL)
655 return -EBADF;
656 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400657 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
David Howells9a9fc1c2009-04-03 16:42:44 +0100658
659 /* attempt to read as many of the pages as possible from the cache
660 * - this returns -ENOBUFS immediately if the cookie is negative
661 */
662 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
663 pages, &nr_pages);
664 if (ret == 0)
665 goto read_complete; /* all pages were read */
666
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400667 if (rsize < PAGE_CACHE_SIZE)
668 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
669 else
670 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400673
674 nfs_pageio_complete(&pgio);
675 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
676 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
David Howells9a9fc1c2009-04-03 16:42:44 +0100677read_complete:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400679out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 return ret;
681}
682
David Howellsf7b422b2006-06-09 09:34:33 -0400683int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
686 sizeof(struct nfs_read_data),
687 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900688 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (nfs_rdata_cachep == NULL)
690 return -ENOMEM;
691
Matthew Dobson93d23412006-03-26 01:37:50 -0800692 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
693 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (nfs_rdata_mempool == NULL)
695 return -ENOMEM;
696
697 return 0;
698}
699
David Brownell266bee82006-06-27 12:59:15 -0700700void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700703 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}