blob: 73ea5e8d66ceebbfab4079c4ab91b62809093ed7 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <asm/system.h>
23
Andy Adamsonf11c88a2009-04-01 09:22:25 -040024#include "nfs4_fs.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050025#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050026#include "iostat.h"
David Howells9a9fc1c2009-04-03 16:42:44 +010027#include "fscache.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050028
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#define NFSDBG_FACILITY NFSDBG_PAGECACHE
30
Trond Myklebust8d5658c2007-04-10 09:26:35 -040031static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int);
32static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int);
Trond Myklebustec06c092006-03-20 13:44:27 -050033static const struct rpc_call_ops nfs_read_partial_ops;
34static const struct rpc_call_ops nfs_read_full_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Christoph Lametere18b8902006-12-06 20:33:20 -080036static struct kmem_cache *nfs_rdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050037static mempool_t *nfs_rdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#define MIN_POOL_READ (32)
40
Trond Myklebust8d5658c2007-04-10 09:26:35 -040041struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050042{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080043 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050044
45 if (p) {
46 memset(p, 0, sizeof(*p));
47 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070048 p->npages = pagecount;
Andy Adamson5f7dbd52009-04-01 09:22:05 -040049 p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040050 if (pagecount <= ARRAY_SIZE(p->page_array))
51 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050052 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040053 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
54 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050055 mempool_free(p, nfs_rdata_mempool);
56 p = NULL;
57 }
58 }
59 }
60 return p;
61}
62
Trond Myklebust5e4424a2008-02-25 21:53:49 -080063static void nfs_readdata_free(struct nfs_read_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050064{
65 if (p && (p->pagevec != &p->page_array[0]))
66 kfree(p->pagevec);
67 mempool_free(p, nfs_rdata_mempool);
68}
69
Trond Myklebust963d8fe2006-01-03 09:55:04 +010070void nfs_readdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Trond Myklebust383ba712008-02-19 20:04:20 -050072 struct nfs_read_data *rdata = data;
73
74 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{
119 LIST_HEAD(one_request);
120 struct nfs_page *new;
121 unsigned int len;
122
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 nfs_list_add_request(new, &one_request);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400135 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400136 nfs_pagein_multi(inode, &one_request, 1, len, 0);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400137 else
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400138 nfs_pagein_one(inode, &one_request, 1, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return 0;
140}
141
142static void nfs_readpage_release(struct nfs_page *req)
143{
David Howells7f8e05f2009-04-03 16:42:45 +0100144 struct inode *d_inode = req->wb_context->path.dentry->d_inode;
145
146 if (PageUptodate(req->wb_page))
147 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 unlock_page(req->wb_page);
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400152 req->wb_context->path.dentry->d_inode->i_sb->s_id,
153 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 req->wb_bytes,
155 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700156 nfs_clear_request(req);
157 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
160/*
161 * Set up the NFS read request struct
162 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400163static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500164 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 unsigned int count, unsigned int offset)
166{
Trond Myklebust84115e12007-07-14 15:39:59 -0400167 struct inode *inode = req->wb_context->path.dentry->d_inode;
168 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
Trond Myklebust07737692007-10-25 18:42:54 -0400169 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400170 struct rpc_message msg = {
171 .rpc_argp = &data->args,
172 .rpc_resp = &data->res,
173 .rpc_cred = req->wb_context->cred,
174 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400175 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -0400176 .task = &data->task,
Trond Myklebust84115e12007-07-14 15:39:59 -0400177 .rpc_client = NFS_CLIENT(inode),
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400178 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400179 .callback_ops = call_ops,
180 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500181 .workqueue = nfsiod_workqueue,
Trond Myklebust84115e12007-07-14 15:39:59 -0400182 .flags = RPC_TASK_ASYNC | swap_flags,
183 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
185 data->req = req;
Trond Myklebust84115e12007-07-14 15:39:59 -0400186 data->inode = inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400187 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 data->args.fh = NFS_FH(inode);
190 data->args.offset = req_offset(req) + offset;
191 data->args.pgbase = req->wb_pgbase + offset;
192 data->args.pages = data->pagevec;
193 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500194 data->args.context = get_nfs_open_context(req->wb_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 data->res.fattr = &data->fattr;
197 data->res.count = count;
198 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400199 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Trond Myklebustec06c092006-03-20 13:44:27 -0500201 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400202 NFS_PROTO(inode)->read_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Chuck Levera3f565b2007-01-31 12:14:01 -0500204 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 data->task.tk_pid,
206 inode->i_sb->s_id,
207 (long long)NFS_FILEID(inode),
208 count,
209 (unsigned long long)data->args.offset);
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400210
Trond Myklebust07737692007-10-25 18:42:54 -0400211 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400212 if (IS_ERR(task))
213 return PTR_ERR(task);
214 rpc_put_task(task);
215 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
218static void
219nfs_async_read_error(struct list_head *head)
220{
221 struct nfs_page *req;
222
223 while (!list_empty(head)) {
224 req = nfs_list_entry(head->next);
225 nfs_list_remove_request(req);
226 SetPageError(req->wb_page);
227 nfs_readpage_release(req);
228 }
229}
230
231/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 * Generate multiple requests to fill a single page.
233 *
234 * We optimize to reduce the number of read operations on the wire. If we
235 * detect that we're reading a page, or an area of a page, that is past the
236 * end of file, we do not generate NFS read operations but just clear the
237 * parts of the page that would have come back zero from the server anyway.
238 *
239 * We rely on the cached value of i_size to make this determination; another
240 * client can fill pages on the server past our cached end-of-file, but we
241 * won't see the new data until our attribute cache is updated. This is more
242 * or less conventional NFS client behavior.
243 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400244static 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 -0700245{
246 struct nfs_page *req = nfs_list_entry(head->next);
247 struct page *page = req->wb_page;
248 struct nfs_read_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700249 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
250 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400252 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 LIST_HEAD(list);
254
255 nfs_list_remove_request(req);
256
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400257 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700258 do {
259 size_t len = min(nbytes,rsize);
260
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400261 data = nfs_readdata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 if (!data)
263 goto out_bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 list_add(&data->pages, &list);
265 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700266 nbytes -= len;
267 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 atomic_set(&req->wb_complete, requests);
269
270 ClearPageError(page);
271 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400272 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 do {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400274 int ret2;
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 data = list_entry(list.next, struct nfs_read_data, pages);
277 list_del_init(&data->pages);
278
279 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400281 if (nbytes < rsize)
282 rsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400283 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400284 rsize, offset);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400285 if (ret == 0)
286 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400287 offset += rsize;
288 nbytes -= rsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 } while (nbytes != 0);
290
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400291 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293out_bad:
294 while (!list_empty(&list)) {
295 data = list_entry(list.next, struct nfs_read_data, pages);
296 list_del(&data->pages);
297 nfs_readdata_free(data);
298 }
299 SetPageError(page);
300 nfs_readpage_release(req);
301 return -ENOMEM;
302}
303
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400304static 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 -0700305{
306 struct nfs_page *req;
307 struct page **pages;
308 struct nfs_read_data *data;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400309 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400311 data = nfs_readdata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 if (!data)
313 goto out_bad;
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 while (!list_empty(head)) {
317 req = nfs_list_entry(head->next);
318 nfs_list_remove_request(req);
319 nfs_list_add_request(req, &data->pages);
320 ClearPageError(req->wb_page);
321 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 }
323 req = nfs_list_entry(data->pages.next);
324
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400325 return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326out_bad:
327 nfs_async_read_error(head);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400328 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500332 * This is the callback from RPC telling us whether a reply was
333 * received or some error occurred (timeout or socket shutdown).
334 */
335int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
336{
337 int status;
338
Harvey Harrison3110ff82008-05-02 13:42:44 -0700339 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
Trond Myklebust0b671302006-11-14 16:12:23 -0500340 task->tk_status);
341
342 status = NFS_PROTO(data->inode)->read_done(task, data);
343 if (status != 0)
344 return status;
345
346 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
347
348 if (task->tk_status == -ESTALE) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200349 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
Trond Myklebust0b671302006-11-14 16:12:23 -0500350 nfs_mark_for_revalidate(data->inode);
351 }
Trond Myklebust0b671302006-11-14 16:12:23 -0500352 return 0;
353}
354
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400355static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
Trond Myklebust0b671302006-11-14 16:12:23 -0500356{
357 struct nfs_readargs *argp = &data->args;
358 struct nfs_readres *resp = &data->res;
359
360 if (resp->eof || resp->count == argp->count)
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400361 goto out;
Trond Myklebust0b671302006-11-14 16:12:23 -0500362
363 /* This is a short read! */
364 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
365 /* Has the server at least made some progress? */
366 if (resp->count == 0)
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400367 goto out;
Trond Myklebust0b671302006-11-14 16:12:23 -0500368
369 /* Yes, so retry the read at the end of the data */
370 argp->offset += resp->count;
371 argp->pgbase += resp->count;
372 argp->count -= resp->count;
Andy Adamsoneedc0202009-04-01 09:22:41 -0400373 nfs4_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client);
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400374 return;
375out:
376 nfs4_sequence_free_slot(NFS_SERVER(data->inode)->nfs_client,
377 &data->res.seq_res);
378 return;
379
Trond Myklebust0b671302006-11-14 16:12:23 -0500380}
381
382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * Handle a read reply that fills part of a page.
384 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500385static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Trond Myklebustec06c092006-03-20 13:44:27 -0500387 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Trond Myklebustec06c092006-03-20 13:44:27 -0500389 if (nfs_readpage_result(task, data) != 0)
390 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400391 if (task->tk_status < 0)
392 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500393
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400394 nfs_readpage_truncate_uninitialised_page(data);
395 nfs_readpage_retry(task, data);
396}
397
398static void nfs_readpage_release_partial(void *calldata)
399{
400 struct nfs_read_data *data = calldata;
401 struct nfs_page *req = data->req;
402 struct page *page = req->wb_page;
403 int status = data->task.tk_status;
404
405 if (status < 0)
Trond Myklebust0b671302006-11-14 16:12:23 -0500406 SetPageError(page);
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (atomic_dec_and_test(&req->wb_complete)) {
409 if (!PageError(page))
410 SetPageUptodate(page);
411 nfs_readpage_release(req);
412 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400413 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400416#if defined(CONFIG_NFS_V4_1)
417void nfs_read_prepare(struct rpc_task *task, void *calldata)
418{
419 struct nfs_read_data *data = calldata;
420
421 if (nfs4_setup_sequence(NFS_SERVER(data->inode)->nfs_client,
422 &data->args.seq_args, &data->res.seq_res,
423 0, task))
424 return;
425 rpc_call_start(task);
426}
427#endif /* CONFIG_NFS_V4_1 */
428
Trond Myklebustec06c092006-03-20 13:44:27 -0500429static const struct rpc_call_ops nfs_read_partial_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400430#if defined(CONFIG_NFS_V4_1)
431 .rpc_call_prepare = nfs_read_prepare,
432#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500433 .rpc_call_done = nfs_readpage_result_partial,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400434 .rpc_release = nfs_readpage_release_partial,
Trond Myklebustec06c092006-03-20 13:44:27 -0500435};
436
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400437static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
438{
439 unsigned int count = data->res.count;
440 unsigned int base = data->args.pgbase;
441 struct page **pages;
442
Trond Myklebust79558f32006-08-22 13:44:32 -0400443 if (data->res.eof)
444 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400445 if (unlikely(count == 0))
446 return;
447 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
448 base &= ~PAGE_CACHE_MASK;
449 count += base;
450 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
451 SetPageUptodate(*pages);
Trond Myklebust0b671302006-11-14 16:12:23 -0500452 if (count == 0)
453 return;
454 /* Was this a short read? */
455 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400456 SetPageUptodate(*pages);
457}
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459/*
460 * This is the callback from RPC telling us whether a reply was
461 * received or some error occurred (timeout or socket shutdown).
462 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500463static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Trond Myklebustec06c092006-03-20 13:44:27 -0500465 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Trond Myklebust0b671302006-11-14 16:12:23 -0500467 if (nfs_readpage_result(task, data) != 0)
468 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400469 if (task->tk_status < 0)
470 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400471 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500472 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400473 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500474 * to ensure that we call nfs_readpage_set_pages_uptodate()
475 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400476 */
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400477 nfs_readpage_truncate_uninitialised_page(data);
478 nfs_readpage_set_pages_uptodate(data);
479 nfs_readpage_retry(task, data);
480}
481
482static void nfs_readpage_release_full(void *calldata)
483{
484 struct nfs_read_data *data = calldata;
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 while (!list_empty(&data->pages)) {
487 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400489 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 nfs_readpage_release(req);
491 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400492 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
494
Trond Myklebustec06c092006-03-20 13:44:27 -0500495static const struct rpc_call_ops nfs_read_full_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400496#if defined(CONFIG_NFS_V4_1)
497 .rpc_call_prepare = nfs_read_prepare,
498#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500499 .rpc_call_done = nfs_readpage_result_full,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400500 .rpc_release = nfs_readpage_release_full,
Trond Myklebustec06c092006-03-20 13:44:27 -0500501};
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 * Read a page over NFS.
505 * We read the page synchronously in the following case:
506 * - The error flag is set for this page. This happens only when a
507 * previous async read operation failed.
508 */
509int nfs_readpage(struct file *file, struct page *page)
510{
511 struct nfs_open_context *ctx;
512 struct inode *inode = page->mapping->host;
513 int error;
514
515 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
516 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500517 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
518 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 /*
521 * Try to flush any pending writes to the file..
522 *
523 * NOTE! Because we own the page lock, there cannot
524 * be any new pending writes generated at this point
525 * for this page (other pages can be written to).
526 */
527 error = nfs_wb_page(inode, page);
528 if (error)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400529 goto out_unlock;
530 if (PageUptodate(page))
531 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400533 error = -ESTALE;
534 if (NFS_STALE(inode))
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400535 goto out_unlock;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500538 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500539 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (ctx == NULL)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400541 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400543 ctx = get_nfs_open_context(nfs_file_open_context(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
David Howells9a9fc1c2009-04-03 16:42:44 +0100545 if (!IS_SYNC(inode)) {
546 error = nfs_readpage_from_fscache(ctx, inode, page);
547 if (error == 0)
548 goto out;
549 }
550
Trond Myklebust8e0969f2006-12-13 15:23:44 -0500551 error = nfs_readpage_async(ctx, inode, page);
552
David Howells9a9fc1c2009-04-03 16:42:44 +0100553out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 put_nfs_open_context(ctx);
555 return error;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400556out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 unlock_page(page);
558 return error;
559}
560
561struct nfs_readdesc {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400562 struct nfs_pageio_descriptor *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 struct nfs_open_context *ctx;
564};
565
566static int
567readpage_async_filler(void *data, struct page *page)
568{
569 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
570 struct inode *inode = page->mapping->host;
571 struct nfs_page *new;
572 unsigned int len;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400573 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Trond Myklebust49a70f22006-12-05 00:35:38 -0500575 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (len == 0)
577 return nfs_return_empty_page(page);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 new = nfs_create_request(desc->ctx, inode, page, 0, len);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400580 if (IS_ERR(new))
581 goto out_error;
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800584 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Fred Isamanf8512ad2008-03-19 11:24:39 -0400585 if (!nfs_pageio_add_request(desc->pgio, new)) {
586 error = desc->pgio->pg_error;
587 goto out_unlock;
588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return 0;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400590out_error:
591 error = PTR_ERR(new);
592 SetPageError(page);
593out_unlock:
594 unlock_page(page);
595 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
598int nfs_readpages(struct file *filp, struct address_space *mapping,
599 struct list_head *pages, unsigned nr_pages)
600{
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400601 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 struct nfs_readdesc desc = {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400603 .pgio = &pgio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 };
605 struct inode *inode = mapping->host;
606 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400607 size_t rsize = server->rsize;
608 unsigned long npages;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400609 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
612 inode->i_sb->s_id,
613 (long long)NFS_FILEID(inode),
614 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500615 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400617 if (NFS_STALE(inode))
618 goto out;
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500621 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (desc.ctx == NULL)
623 return -EBADF;
624 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400625 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
David Howells9a9fc1c2009-04-03 16:42:44 +0100626
627 /* attempt to read as many of the pages as possible from the cache
628 * - this returns -ENOBUFS immediately if the cookie is negative
629 */
630 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
631 pages, &nr_pages);
632 if (ret == 0)
633 goto read_complete; /* all pages were read */
634
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400635 if (rsize < PAGE_CACHE_SIZE)
636 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
637 else
638 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400641
642 nfs_pageio_complete(&pgio);
643 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
644 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
David Howells9a9fc1c2009-04-03 16:42:44 +0100645read_complete:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400647out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 return ret;
649}
650
David Howellsf7b422b2006-06-09 09:34:33 -0400651int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
654 sizeof(struct nfs_read_data),
655 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900656 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 if (nfs_rdata_cachep == NULL)
658 return -ENOMEM;
659
Matthew Dobson93d23412006-03-26 01:37:50 -0800660 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
661 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (nfs_rdata_mempool == NULL)
663 return -ENOMEM;
664
665 return 0;
666}
667
David Brownell266bee82006-06-27 12:59:15 -0700668void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700671 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}