blob: 6e2b06e6ca794c1d467a4b2edeaf9b4469806413 [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{
Trond Myklebust93870d72010-05-13 12:51:03 -040043 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL);
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 {
Trond Myklebust93870d72010-05-13 12:51:03 -040053 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040054 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 Myklebust1ae88b22009-08-12 09:12:30 -040063void 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 Myklebust1ae88b22009-08-12 09:12:30 -040070static void nfs_readdata_release(struct nfs_read_data *rdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Trond Myklebust383ba712008-02-19 20:04:20 -050072 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
David Howellsf42b2932009-04-03 16:42:44 +0100114int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
115 struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
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{
David Howells7f8e05f2009-04-03 16:42:45 +0100142 struct inode *d_inode = req->wb_context->path.dentry->d_inode;
143
144 if (PageUptodate(req->wb_page))
145 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 unlock_page(req->wb_page);
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400150 req->wb_context->path.dentry->d_inode->i_sb->s_id,
151 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 req->wb_bytes,
153 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700154 nfs_clear_request(req);
155 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
158/*
159 * Set up the NFS read request struct
160 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400161static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500162 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 unsigned int count, unsigned int offset)
164{
Trond Myklebust84115e12007-07-14 15:39:59 -0400165 struct inode *inode = req->wb_context->path.dentry->d_inode;
166 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
Trond Myklebust07737692007-10-25 18:42:54 -0400167 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400168 struct rpc_message msg = {
169 .rpc_argp = &data->args,
170 .rpc_resp = &data->res,
171 .rpc_cred = req->wb_context->cred,
172 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400173 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -0400174 .task = &data->task,
Trond Myklebust84115e12007-07-14 15:39:59 -0400175 .rpc_client = NFS_CLIENT(inode),
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400176 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400177 .callback_ops = call_ops,
178 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500179 .workqueue = nfsiod_workqueue,
Trond Myklebust84115e12007-07-14 15:39:59 -0400180 .flags = RPC_TASK_ASYNC | swap_flags,
181 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 data->req = req;
Trond Myklebust84115e12007-07-14 15:39:59 -0400184 data->inode = inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400185 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 data->args.fh = NFS_FH(inode);
188 data->args.offset = req_offset(req) + offset;
189 data->args.pgbase = req->wb_pgbase + offset;
190 data->args.pages = data->pagevec;
191 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500192 data->args.context = get_nfs_open_context(req->wb_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 data->res.fattr = &data->fattr;
195 data->res.count = count;
196 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400197 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Trond Myklebustec06c092006-03-20 13:44:27 -0500199 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400200 NFS_PROTO(inode)->read_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Chuck Levera3f565b2007-01-31 12:14:01 -0500202 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 data->task.tk_pid,
204 inode->i_sb->s_id,
205 (long long)NFS_FILEID(inode),
206 count,
207 (unsigned long long)data->args.offset);
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400208
Trond Myklebust07737692007-10-25 18:42:54 -0400209 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400210 if (IS_ERR(task))
211 return PTR_ERR(task);
212 rpc_put_task(task);
213 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
216static void
217nfs_async_read_error(struct list_head *head)
218{
219 struct nfs_page *req;
220
221 while (!list_empty(head)) {
222 req = nfs_list_entry(head->next);
223 nfs_list_remove_request(req);
224 SetPageError(req->wb_page);
225 nfs_readpage_release(req);
226 }
227}
228
229/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 * Generate multiple requests to fill a single page.
231 *
232 * We optimize to reduce the number of read operations on the wire. If we
233 * detect that we're reading a page, or an area of a page, that is past the
234 * end of file, we do not generate NFS read operations but just clear the
235 * parts of the page that would have come back zero from the server anyway.
236 *
237 * We rely on the cached value of i_size to make this determination; another
238 * client can fill pages on the server past our cached end-of-file, but we
239 * won't see the new data until our attribute cache is updated. This is more
240 * or less conventional NFS client behavior.
241 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400242static 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 -0700243{
244 struct nfs_page *req = nfs_list_entry(head->next);
245 struct page *page = req->wb_page;
246 struct nfs_read_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700247 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
248 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400250 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 LIST_HEAD(list);
252
253 nfs_list_remove_request(req);
254
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400255 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700256 do {
257 size_t len = min(nbytes,rsize);
258
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400259 data = nfs_readdata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 if (!data)
261 goto out_bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 list_add(&data->pages, &list);
263 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700264 nbytes -= len;
265 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 atomic_set(&req->wb_complete, requests);
267
268 ClearPageError(page);
269 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400270 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 do {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400272 int ret2;
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 data = list_entry(list.next, struct nfs_read_data, pages);
275 list_del_init(&data->pages);
276
277 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400279 if (nbytes < rsize)
280 rsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400281 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400282 rsize, offset);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400283 if (ret == 0)
284 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400285 offset += rsize;
286 nbytes -= rsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 } while (nbytes != 0);
288
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400289 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291out_bad:
292 while (!list_empty(&list)) {
293 data = list_entry(list.next, struct nfs_read_data, pages);
294 list_del(&data->pages);
295 nfs_readdata_free(data);
296 }
297 SetPageError(page);
298 nfs_readpage_release(req);
299 return -ENOMEM;
300}
301
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400302static 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 -0700303{
304 struct nfs_page *req;
305 struct page **pages;
306 struct nfs_read_data *data;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400307 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400309 data = nfs_readdata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 if (!data)
311 goto out_bad;
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 while (!list_empty(head)) {
315 req = nfs_list_entry(head->next);
316 nfs_list_remove_request(req);
317 nfs_list_add_request(req, &data->pages);
318 ClearPageError(req->wb_page);
319 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321 req = nfs_list_entry(data->pages.next);
322
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400323 return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324out_bad:
325 nfs_async_read_error(head);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400326 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500330 * This is the callback from RPC telling us whether a reply was
331 * received or some error occurred (timeout or socket shutdown).
332 */
333int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
334{
335 int status;
336
Harvey Harrison3110ff82008-05-02 13:42:44 -0700337 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
Trond Myklebust0b671302006-11-14 16:12:23 -0500338 task->tk_status);
339
340 status = NFS_PROTO(data->inode)->read_done(task, data);
341 if (status != 0)
342 return status;
343
344 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
345
346 if (task->tk_status == -ESTALE) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200347 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
Trond Myklebust0b671302006-11-14 16:12:23 -0500348 nfs_mark_for_revalidate(data->inode);
349 }
Trond Myklebust0b671302006-11-14 16:12:23 -0500350 return 0;
351}
352
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400353static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
Trond Myklebust0b671302006-11-14 16:12:23 -0500354{
355 struct nfs_readargs *argp = &data->args;
356 struct nfs_readres *resp = &data->res;
357
358 if (resp->eof || resp->count == argp->count)
Trond Myklebustd61e6122009-12-05 19:32:19 -0500359 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500360
361 /* This is a short read! */
362 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
363 /* Has the server at least made some progress? */
364 if (resp->count == 0)
Trond Myklebustd61e6122009-12-05 19:32:19 -0500365 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500366
367 /* Yes, so retry the read at the end of the data */
368 argp->offset += resp->count;
369 argp->pgbase += resp->count;
370 argp->count -= resp->count;
Trond Myklebust0110ee12009-12-07 09:00:24 -0500371 nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client);
Trond Myklebust0b671302006-11-14 16:12:23 -0500372}
373
374/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * Handle a read reply that fills part of a page.
376 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500377static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Trond Myklebustec06c092006-03-20 13:44:27 -0500379 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Trond Myklebustec06c092006-03-20 13:44:27 -0500381 if (nfs_readpage_result(task, data) != 0)
382 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400383 if (task->tk_status < 0)
384 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500385
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400386 nfs_readpage_truncate_uninitialised_page(data);
387 nfs_readpage_retry(task, data);
388}
389
390static void nfs_readpage_release_partial(void *calldata)
391{
392 struct nfs_read_data *data = calldata;
393 struct nfs_page *req = data->req;
394 struct page *page = req->wb_page;
395 int status = data->task.tk_status;
396
397 if (status < 0)
Trond Myklebust0b671302006-11-14 16:12:23 -0500398 SetPageError(page);
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 if (atomic_dec_and_test(&req->wb_complete)) {
401 if (!PageError(page))
402 SetPageUptodate(page);
403 nfs_readpage_release(req);
404 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400405 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400408#if defined(CONFIG_NFS_V4_1)
409void nfs_read_prepare(struct rpc_task *task, void *calldata)
410{
411 struct nfs_read_data *data = calldata;
412
413 if (nfs4_setup_sequence(NFS_SERVER(data->inode)->nfs_client,
414 &data->args.seq_args, &data->res.seq_res,
415 0, task))
416 return;
417 rpc_call_start(task);
418}
419#endif /* CONFIG_NFS_V4_1 */
420
Trond Myklebustec06c092006-03-20 13:44:27 -0500421static const struct rpc_call_ops nfs_read_partial_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400422#if defined(CONFIG_NFS_V4_1)
423 .rpc_call_prepare = nfs_read_prepare,
424#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500425 .rpc_call_done = nfs_readpage_result_partial,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400426 .rpc_release = nfs_readpage_release_partial,
Trond Myklebustec06c092006-03-20 13:44:27 -0500427};
428
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400429static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
430{
431 unsigned int count = data->res.count;
432 unsigned int base = data->args.pgbase;
433 struct page **pages;
434
Trond Myklebust79558f32006-08-22 13:44:32 -0400435 if (data->res.eof)
436 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400437 if (unlikely(count == 0))
438 return;
439 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
440 base &= ~PAGE_CACHE_MASK;
441 count += base;
442 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
443 SetPageUptodate(*pages);
Trond Myklebust0b671302006-11-14 16:12:23 -0500444 if (count == 0)
445 return;
446 /* Was this a short read? */
447 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400448 SetPageUptodate(*pages);
449}
450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451/*
452 * This is the callback from RPC telling us whether a reply was
453 * received or some error occurred (timeout or socket shutdown).
454 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500455static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Trond Myklebustec06c092006-03-20 13:44:27 -0500457 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Trond Myklebust0b671302006-11-14 16:12:23 -0500459 if (nfs_readpage_result(task, data) != 0)
460 return;
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400461 if (task->tk_status < 0)
462 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400463 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500464 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400465 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500466 * to ensure that we call nfs_readpage_set_pages_uptodate()
467 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400468 */
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400469 nfs_readpage_truncate_uninitialised_page(data);
470 nfs_readpage_set_pages_uptodate(data);
471 nfs_readpage_retry(task, data);
472}
473
474static void nfs_readpage_release_full(void *calldata)
475{
476 struct nfs_read_data *data = calldata;
477
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 while (!list_empty(&data->pages)) {
479 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400481 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 nfs_readpage_release(req);
483 }
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400484 nfs_readdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Trond Myklebustec06c092006-03-20 13:44:27 -0500487static const struct rpc_call_ops nfs_read_full_ops = {
Andy Adamsonf11c88a2009-04-01 09:22:25 -0400488#if defined(CONFIG_NFS_V4_1)
489 .rpc_call_prepare = nfs_read_prepare,
490#endif /* CONFIG_NFS_V4_1 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500491 .rpc_call_done = nfs_readpage_result_full,
Trond Myklebustfdd1e742008-04-15 16:33:58 -0400492 .rpc_release = nfs_readpage_release_full,
Trond Myklebustec06c092006-03-20 13:44:27 -0500493};
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * Read a page over NFS.
497 * We read the page synchronously in the following case:
498 * - The error flag is set for this page. This happens only when a
499 * previous async read operation failed.
500 */
501int nfs_readpage(struct file *file, struct page *page)
502{
503 struct nfs_open_context *ctx;
504 struct inode *inode = page->mapping->host;
505 int error;
506
507 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
508 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500509 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
510 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 /*
513 * Try to flush any pending writes to the file..
514 *
515 * NOTE! Because we own the page lock, there cannot
516 * be any new pending writes generated at this point
517 * for this page (other pages can be written to).
518 */
519 error = nfs_wb_page(inode, page);
520 if (error)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400521 goto out_unlock;
522 if (PageUptodate(page))
523 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400525 error = -ESTALE;
526 if (NFS_STALE(inode))
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400527 goto out_unlock;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500530 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500531 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (ctx == NULL)
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400533 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400535 ctx = get_nfs_open_context(nfs_file_open_context(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
David Howells9a9fc1c2009-04-03 16:42:44 +0100537 if (!IS_SYNC(inode)) {
538 error = nfs_readpage_from_fscache(ctx, inode, page);
539 if (error == 0)
540 goto out;
541 }
542
Trond Myklebust8e0969f2006-12-13 15:23:44 -0500543 error = nfs_readpage_async(ctx, inode, page);
544
David Howells9a9fc1c2009-04-03 16:42:44 +0100545out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 put_nfs_open_context(ctx);
547 return error;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400548out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 unlock_page(page);
550 return error;
551}
552
553struct nfs_readdesc {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400554 struct nfs_pageio_descriptor *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 struct nfs_open_context *ctx;
556};
557
558static int
559readpage_async_filler(void *data, struct page *page)
560{
561 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
562 struct inode *inode = page->mapping->host;
563 struct nfs_page *new;
564 unsigned int len;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400565 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Trond Myklebust49a70f22006-12-05 00:35:38 -0500567 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (len == 0)
569 return nfs_return_empty_page(page);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 new = nfs_create_request(desc->ctx, inode, page, 0, len);
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400572 if (IS_ERR(new))
573 goto out_error;
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (len < PAGE_CACHE_SIZE)
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800576 zero_user_segment(page, len, PAGE_CACHE_SIZE);
Fred Isamanf8512ad2008-03-19 11:24:39 -0400577 if (!nfs_pageio_add_request(desc->pgio, new)) {
578 error = desc->pgio->pg_error;
579 goto out_unlock;
580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return 0;
Trond Myklebustde05a0c2007-05-20 13:05:05 -0400582out_error:
583 error = PTR_ERR(new);
584 SetPageError(page);
585out_unlock:
586 unlock_page(page);
587 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
589
590int nfs_readpages(struct file *filp, struct address_space *mapping,
591 struct list_head *pages, unsigned nr_pages)
592{
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400593 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 struct nfs_readdesc desc = {
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400595 .pgio = &pgio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 };
597 struct inode *inode = mapping->host;
598 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400599 size_t rsize = server->rsize;
600 unsigned long npages;
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400601 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
604 inode->i_sb->s_id,
605 (long long)NFS_FILEID(inode),
606 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500607 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400609 if (NFS_STALE(inode))
610 goto out;
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500613 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (desc.ctx == NULL)
615 return -EBADF;
616 } else
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400617 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
David Howells9a9fc1c2009-04-03 16:42:44 +0100618
619 /* attempt to read as many of the pages as possible from the cache
620 * - this returns -ENOBUFS immediately if the cookie is negative
621 */
622 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
623 pages, &nr_pages);
624 if (ret == 0)
625 goto read_complete; /* all pages were read */
626
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400627 if (rsize < PAGE_CACHE_SIZE)
628 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
629 else
630 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400633
634 nfs_pageio_complete(&pgio);
635 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
636 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
David Howells9a9fc1c2009-04-03 16:42:44 +0100637read_complete:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400639out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 return ret;
641}
642
David Howellsf7b422b2006-06-09 09:34:33 -0400643int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
645 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
646 sizeof(struct nfs_read_data),
647 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900648 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (nfs_rdata_cachep == NULL)
650 return -ENOMEM;
651
Matthew Dobson93d23412006-03-26 01:37:50 -0800652 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
653 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (nfs_rdata_mempool == NULL)
655 return -ENOMEM;
656
657 return 0;
658}
659
David Brownell266bee82006-06-27 12:59:15 -0700660void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
662 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700663 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}