blob: a9c26521a9e2d8a13965e364dae9ac8392cecdb1 [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
8 *
9 * We do an ugly hack here in order to return proper error codes to the
10 * user program when a read request failed: since generic_file_read
11 * only checks the return value of inode->i_op->readpage() which is always 0
12 * for async RPC, we set the error bit of the page to 1 when an error occurs,
13 * and make nfs_readpage transmit requests synchronously when encountering this.
14 * This is only a small problem, though, since we now retry all operations
15 * within the RPC code when root squashing is suspected.
16 */
17
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/time.h>
19#include <linux/kernel.h>
20#include <linux/errno.h>
21#include <linux/fcntl.h>
22#include <linux/stat.h>
23#include <linux/mm.h>
24#include <linux/slab.h>
25#include <linux/pagemap.h>
26#include <linux/sunrpc/clnt.h>
27#include <linux/nfs_fs.h>
28#include <linux/nfs_page.h>
29#include <linux/smp_lock.h>
30
31#include <asm/system.h>
32
Trond Myklebust49a70f22006-12-05 00:35:38 -050033#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050034#include "iostat.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#define NFSDBG_FACILITY NFSDBG_PAGECACHE
37
38static int nfs_pagein_one(struct list_head *, struct inode *);
Trond Myklebustec06c092006-03-20 13:44:27 -050039static const struct rpc_call_ops nfs_read_partial_ops;
40static const struct rpc_call_ops nfs_read_full_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Christoph Lametere18b8902006-12-06 20:33:20 -080042static struct kmem_cache *nfs_rdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050043static mempool_t *nfs_rdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#define MIN_POOL_READ (32)
46
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070047struct nfs_read_data *nfs_readdata_alloc(size_t len)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050048{
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070049 unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080050 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050051
52 if (p) {
53 memset(p, 0, sizeof(*p));
54 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070055 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040056 if (pagecount <= ARRAY_SIZE(p->page_array))
57 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050058 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040059 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
60 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050061 mempool_free(p, nfs_rdata_mempool);
62 p = NULL;
63 }
64 }
65 }
66 return p;
67}
68
Trond Myklebust8aca67f2006-11-13 16:23:44 -050069static void nfs_readdata_rcu_free(struct rcu_head *head)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050070{
Trond Myklebust8aca67f2006-11-13 16:23:44 -050071 struct nfs_read_data *p = container_of(head, struct nfs_read_data, task.u.tk_rcu);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050072 if (p && (p->pagevec != &p->page_array[0]))
73 kfree(p->pagevec);
74 mempool_free(p, nfs_rdata_mempool);
75}
76
Trond Myklebust8aca67f2006-11-13 16:23:44 -050077static void nfs_readdata_free(struct nfs_read_data *rdata)
78{
79 call_rcu_bh(&rdata->task.u.tk_rcu, nfs_readdata_rcu_free);
80}
81
Trond Myklebust963d8fe2006-01-03 09:55:04 +010082void nfs_readdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 nfs_readdata_free(data);
85}
86
87static
Linus Torvalds1da177e2005-04-16 15:20:36 -070088int nfs_return_empty_page(struct page *page)
89{
90 memclear_highpage_flush(page, 0, PAGE_CACHE_SIZE);
91 SetPageUptodate(page);
92 unlock_page(page);
93 return 0;
94}
95
Trond Myklebust1de3fc12006-05-25 01:40:44 -040096static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
97{
98 unsigned int remainder = data->args.count - data->res.count;
99 unsigned int base = data->args.pgbase + data->res.count;
100 unsigned int pglen;
101 struct page **pages;
102
103 if (data->res.eof == 0 || remainder == 0)
104 return;
105 /*
106 * Note: "remainder" can never be negative, since we check for
107 * this in the XDR code.
108 */
109 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
110 base &= ~PAGE_CACHE_MASK;
111 pglen = PAGE_CACHE_SIZE - base;
Trond Myklebust79558f32006-08-22 13:44:32 -0400112 for (;;) {
113 if (remainder <= pglen) {
114 memclear_highpage_flush(*pages, base, remainder);
115 break;
116 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400117 memclear_highpage_flush(*pages, base, pglen);
Trond Myklebust79558f32006-08-22 13:44:32 -0400118 pages++;
119 remainder -= pglen;
120 pglen = PAGE_CACHE_SIZE;
121 base = 0;
122 }
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400123}
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125/*
126 * Read a page synchronously.
127 */
128static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode,
129 struct page *page)
130{
131 unsigned int rsize = NFS_SERVER(inode)->rsize;
132 unsigned int count = PAGE_CACHE_SIZE;
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500133 int result = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 struct nfs_read_data *rdata;
135
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700136 rdata = nfs_readdata_alloc(count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 if (!rdata)
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500138 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
140 memset(rdata, 0, sizeof(*rdata));
141 rdata->flags = (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0);
142 rdata->cred = ctx->cred;
143 rdata->inode = inode;
144 INIT_LIST_HEAD(&rdata->pages);
145 rdata->args.fh = NFS_FH(inode);
146 rdata->args.context = ctx;
147 rdata->args.pages = &page;
148 rdata->args.pgbase = 0UL;
149 rdata->args.count = rsize;
150 rdata->res.fattr = &rdata->fattr;
151
152 dprintk("NFS: nfs_readpage_sync(%p)\n", page);
153
154 /*
155 * This works now because the socket layer never tries to DMA
156 * into this buffer directly.
157 */
158 do {
159 if (count < rsize)
160 rdata->args.count = count;
161 rdata->res.count = rdata->args.count;
162 rdata->args.offset = page_offset(page) + rdata->args.pgbase;
163
164 dprintk("NFS: nfs_proc_read(%s, (%s/%Ld), %Lu, %u)\n",
David Howells54ceac42006-08-22 20:06:13 -0400165 NFS_SERVER(inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 inode->i_sb->s_id,
167 (long long)NFS_FILEID(inode),
168 (unsigned long long)rdata->args.pgbase,
169 rdata->args.count);
170
171 lock_kernel();
172 result = NFS_PROTO(inode)->read(rdata);
173 unlock_kernel();
174
175 /*
176 * Even if we had a partial success we can't mark the page
177 * cache valid.
178 */
179 if (result < 0) {
180 if (result == -EISDIR)
181 result = -EINVAL;
182 goto io_error;
183 }
184 count -= result;
185 rdata->args.pgbase += result;
Chuck Lever91d5b472006-03-20 13:44:14 -0500186 nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, result);
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 /* Note: result == 0 should only happen if we're caching
189 * a write that extends the file and punches a hole.
190 */
191 if (rdata->res.eof != 0 || result == 0)
192 break;
193 } while (count);
Chuck Leverdc592502005-08-18 11:24:12 -0700194 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700195 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700196 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Trond Myklebust7a524112006-09-15 16:03:45 -0400198 if (rdata->res.eof || rdata->res.count == rdata->args.count) {
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400199 SetPageUptodate(page);
Trond Myklebust7a524112006-09-15 16:03:45 -0400200 if (rdata->res.eof && count != 0)
201 memclear_highpage_flush(page, rdata->args.pgbase, count);
202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 result = 0;
204
205io_error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 nfs_readdata_free(rdata);
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500207out_unlock:
208 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 return result;
210}
211
212static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
213 struct page *page)
214{
215 LIST_HEAD(one_request);
216 struct nfs_page *new;
217 unsigned int len;
218
Trond Myklebust49a70f22006-12-05 00:35:38 -0500219 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (len == 0)
221 return nfs_return_empty_page(page);
222 new = nfs_create_request(ctx, inode, page, 0, len);
223 if (IS_ERR(new)) {
224 unlock_page(page);
225 return PTR_ERR(new);
226 }
227 if (len < PAGE_CACHE_SIZE)
228 memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 nfs_list_add_request(new, &one_request);
231 nfs_pagein_one(&one_request, inode);
232 return 0;
233}
234
235static void nfs_readpage_release(struct nfs_page *req)
236{
237 unlock_page(req->wb_page);
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
240 req->wb_context->dentry->d_inode->i_sb->s_id,
241 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
242 req->wb_bytes,
243 (long long)req_offset(req));
Nick Wilson10d2c462005-09-22 21:44:28 -0700244 nfs_clear_request(req);
245 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
248/*
249 * Set up the NFS read request struct
250 */
251static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
Trond Myklebustec06c092006-03-20 13:44:27 -0500252 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 unsigned int count, unsigned int offset)
254{
255 struct inode *inode;
Trond Myklebustec06c092006-03-20 13:44:27 -0500256 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 data->req = req;
259 data->inode = inode = req->wb_context->dentry->d_inode;
260 data->cred = req->wb_context->cred;
261
262 data->args.fh = NFS_FH(inode);
263 data->args.offset = req_offset(req) + offset;
264 data->args.pgbase = req->wb_pgbase + offset;
265 data->args.pages = data->pagevec;
266 data->args.count = count;
267 data->args.context = req->wb_context;
268
269 data->res.fattr = &data->fattr;
270 data->res.count = count;
271 data->res.eof = 0;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400272 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Trond Myklebustec06c092006-03-20 13:44:27 -0500274 /* Set up the initial task struct. */
275 flags = RPC_TASK_ASYNC | (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0);
276 rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 NFS_PROTO(inode)->read_setup(data);
278
279 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
282 data->task.tk_pid,
283 inode->i_sb->s_id,
284 (long long)NFS_FILEID(inode),
285 count,
286 (unsigned long long)data->args.offset);
287}
288
289static void
290nfs_async_read_error(struct list_head *head)
291{
292 struct nfs_page *req;
293
294 while (!list_empty(head)) {
295 req = nfs_list_entry(head->next);
296 nfs_list_remove_request(req);
297 SetPageError(req->wb_page);
298 nfs_readpage_release(req);
299 }
300}
301
302/*
303 * Start an async read operation
304 */
305static void nfs_execute_read(struct nfs_read_data *data)
306{
307 struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
308 sigset_t oldset;
309
310 rpc_clnt_sigmask(clnt, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 rpc_execute(&data->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 rpc_clnt_sigunmask(clnt, &oldset);
313}
314
315/*
316 * Generate multiple requests to fill a single page.
317 *
318 * We optimize to reduce the number of read operations on the wire. If we
319 * detect that we're reading a page, or an area of a page, that is past the
320 * end of file, we do not generate NFS read operations but just clear the
321 * parts of the page that would have come back zero from the server anyway.
322 *
323 * We rely on the cached value of i_size to make this determination; another
324 * client can fill pages on the server past our cached end-of-file, but we
325 * won't see the new data until our attribute cache is updated. This is more
326 * or less conventional NFS client behavior.
327 */
328static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
329{
330 struct nfs_page *req = nfs_list_entry(head->next);
331 struct page *page = req->wb_page;
332 struct nfs_read_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700333 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
334 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 int requests = 0;
336 LIST_HEAD(list);
337
338 nfs_list_remove_request(req);
339
340 nbytes = req->wb_bytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700341 do {
342 size_t len = min(nbytes,rsize);
343
344 data = nfs_readdata_alloc(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (!data)
346 goto out_bad;
347 INIT_LIST_HEAD(&data->pages);
348 list_add(&data->pages, &list);
349 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700350 nbytes -= len;
351 } while(nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 atomic_set(&req->wb_complete, requests);
353
354 ClearPageError(page);
355 offset = 0;
356 nbytes = req->wb_bytes;
357 do {
358 data = list_entry(list.next, struct nfs_read_data, pages);
359 list_del_init(&data->pages);
360
361 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 if (nbytes > rsize) {
Trond Myklebustec06c092006-03-20 13:44:27 -0500364 nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
365 rsize, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 offset += rsize;
367 nbytes -= rsize;
368 } else {
Trond Myklebustec06c092006-03-20 13:44:27 -0500369 nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
370 nbytes, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 nbytes = 0;
372 }
373 nfs_execute_read(data);
374 } while (nbytes != 0);
375
376 return 0;
377
378out_bad:
379 while (!list_empty(&list)) {
380 data = list_entry(list.next, struct nfs_read_data, pages);
381 list_del(&data->pages);
382 nfs_readdata_free(data);
383 }
384 SetPageError(page);
385 nfs_readpage_release(req);
386 return -ENOMEM;
387}
388
389static int nfs_pagein_one(struct list_head *head, struct inode *inode)
390{
391 struct nfs_page *req;
392 struct page **pages;
393 struct nfs_read_data *data;
394 unsigned int count;
395
396 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
397 return nfs_pagein_multi(head, inode);
398
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700399 data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 if (!data)
401 goto out_bad;
402
403 INIT_LIST_HEAD(&data->pages);
404 pages = data->pagevec;
405 count = 0;
406 while (!list_empty(head)) {
407 req = nfs_list_entry(head->next);
408 nfs_list_remove_request(req);
409 nfs_list_add_request(req, &data->pages);
410 ClearPageError(req->wb_page);
411 *pages++ = req->wb_page;
412 count += req->wb_bytes;
413 }
414 req = nfs_list_entry(data->pages.next);
415
Trond Myklebustec06c092006-03-20 13:44:27 -0500416 nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418 nfs_execute_read(data);
419 return 0;
420out_bad:
421 nfs_async_read_error(head);
422 return -ENOMEM;
423}
424
425static int
426nfs_pagein_list(struct list_head *head, int rpages)
427{
428 LIST_HEAD(one_request);
429 struct nfs_page *req;
430 int error = 0;
431 unsigned int pages = 0;
432
433 while (!list_empty(head)) {
434 pages += nfs_coalesce_requests(head, &one_request, rpages);
435 req = nfs_list_entry(one_request.next);
436 error = nfs_pagein_one(&one_request, req->wb_context->dentry->d_inode);
437 if (error < 0)
438 break;
439 }
440 if (error >= 0)
441 return pages;
442
443 nfs_async_read_error(head);
444 return error;
445}
446
447/*
Trond Myklebust0b671302006-11-14 16:12:23 -0500448 * This is the callback from RPC telling us whether a reply was
449 * received or some error occurred (timeout or socket shutdown).
450 */
451int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
452{
453 int status;
454
455 dprintk("%s: %4d, (status %d)\n", __FUNCTION__, task->tk_pid,
456 task->tk_status);
457
458 status = NFS_PROTO(data->inode)->read_done(task, data);
459 if (status != 0)
460 return status;
461
462 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
463
464 if (task->tk_status == -ESTALE) {
465 set_bit(NFS_INO_STALE, &NFS_FLAGS(data->inode));
466 nfs_mark_for_revalidate(data->inode);
467 }
468 spin_lock(&data->inode->i_lock);
469 NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME;
470 spin_unlock(&data->inode->i_lock);
471 return 0;
472}
473
474static int nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
475{
476 struct nfs_readargs *argp = &data->args;
477 struct nfs_readres *resp = &data->res;
478
479 if (resp->eof || resp->count == argp->count)
480 return 0;
481
482 /* This is a short read! */
483 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
484 /* Has the server at least made some progress? */
485 if (resp->count == 0)
486 return 0;
487
488 /* Yes, so retry the read at the end of the data */
489 argp->offset += resp->count;
490 argp->pgbase += resp->count;
491 argp->count -= resp->count;
492 rpc_restart_call(task);
493 return -EAGAIN;
494}
495
496/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 * Handle a read reply that fills part of a page.
498 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500499static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Trond Myklebustec06c092006-03-20 13:44:27 -0500501 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 struct nfs_page *req = data->req;
503 struct page *page = req->wb_page;
504
Trond Myklebustec06c092006-03-20 13:44:27 -0500505 if (nfs_readpage_result(task, data) != 0)
506 return;
Trond Myklebust0b671302006-11-14 16:12:23 -0500507
508 if (likely(task->tk_status >= 0)) {
509 nfs_readpage_truncate_uninitialised_page(data);
510 if (nfs_readpage_retry(task, data) != 0)
511 return;
512 }
513 if (unlikely(task->tk_status < 0))
514 SetPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (atomic_dec_and_test(&req->wb_complete)) {
516 if (!PageError(page))
517 SetPageUptodate(page);
518 nfs_readpage_release(req);
519 }
520}
521
Trond Myklebustec06c092006-03-20 13:44:27 -0500522static const struct rpc_call_ops nfs_read_partial_ops = {
523 .rpc_call_done = nfs_readpage_result_partial,
524 .rpc_release = nfs_readdata_release,
525};
526
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400527static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
528{
529 unsigned int count = data->res.count;
530 unsigned int base = data->args.pgbase;
531 struct page **pages;
532
Trond Myklebust79558f32006-08-22 13:44:32 -0400533 if (data->res.eof)
534 count = data->args.count;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400535 if (unlikely(count == 0))
536 return;
537 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
538 base &= ~PAGE_CACHE_MASK;
539 count += base;
540 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
541 SetPageUptodate(*pages);
Trond Myklebust0b671302006-11-14 16:12:23 -0500542 if (count == 0)
543 return;
544 /* Was this a short read? */
545 if (data->res.eof || data->res.count == data->args.count)
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400546 SetPageUptodate(*pages);
547}
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/*
550 * This is the callback from RPC telling us whether a reply was
551 * received or some error occurred (timeout or socket shutdown).
552 */
Trond Myklebustec06c092006-03-20 13:44:27 -0500553static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Trond Myklebustec06c092006-03-20 13:44:27 -0500555 struct nfs_read_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Trond Myklebust0b671302006-11-14 16:12:23 -0500557 if (nfs_readpage_result(task, data) != 0)
558 return;
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400559 /*
Trond Myklebust0b671302006-11-14 16:12:23 -0500560 * Note: nfs_readpage_retry may change the values of
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400561 * data->args. In the multi-page case, we therefore need
Trond Myklebust0b671302006-11-14 16:12:23 -0500562 * to ensure that we call nfs_readpage_set_pages_uptodate()
563 * first.
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400564 */
565 if (likely(task->tk_status >= 0)) {
566 nfs_readpage_truncate_uninitialised_page(data);
567 nfs_readpage_set_pages_uptodate(data);
Trond Myklebust0b671302006-11-14 16:12:23 -0500568 if (nfs_readpage_retry(task, data) != 0)
569 return;
570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 while (!list_empty(&data->pages)) {
572 struct nfs_page *req = nfs_list_entry(data->pages.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Trond Myklebust1de3fc12006-05-25 01:40:44 -0400574 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 nfs_readpage_release(req);
576 }
577}
578
Trond Myklebustec06c092006-03-20 13:44:27 -0500579static const struct rpc_call_ops nfs_read_full_ops = {
580 .rpc_call_done = nfs_readpage_result_full,
581 .rpc_release = nfs_readdata_release,
582};
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * Read a page over NFS.
586 * We read the page synchronously in the following case:
587 * - The error flag is set for this page. This happens only when a
588 * previous async read operation failed.
589 */
590int nfs_readpage(struct file *file, struct page *page)
591{
592 struct nfs_open_context *ctx;
593 struct inode *inode = page->mapping->host;
594 int error;
595
596 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
597 page, PAGE_CACHE_SIZE, page->index);
Chuck Lever91d5b472006-03-20 13:44:14 -0500598 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
599 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /*
602 * Try to flush any pending writes to the file..
603 *
604 * NOTE! Because we own the page lock, there cannot
605 * be any new pending writes generated at this point
606 * for this page (other pages can be written to).
607 */
608 error = nfs_wb_page(inode, page);
609 if (error)
610 goto out_error;
611
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400612 error = -ESTALE;
613 if (NFS_STALE(inode))
614 goto out_error;
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (file == NULL) {
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500617 error = -EBADF;
Trond Myklebustd5308382005-11-04 15:33:38 -0500618 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 if (ctx == NULL)
Trond Myklebustcf1308f2006-11-19 16:44:52 -0500620 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 } else
622 ctx = get_nfs_open_context((struct nfs_open_context *)
623 file->private_data);
624 if (!IS_SYNC(inode)) {
625 error = nfs_readpage_async(ctx, inode, page);
626 goto out;
627 }
628
629 error = nfs_readpage_sync(ctx, inode, page);
630 if (error < 0 && IS_SWAPFILE(inode))
631 printk("Aiee.. nfs swap-in of page failed!\n");
632out:
633 put_nfs_open_context(ctx);
634 return error;
635
636out_error:
637 unlock_page(page);
638 return error;
639}
640
641struct nfs_readdesc {
642 struct list_head *head;
643 struct nfs_open_context *ctx;
644};
645
646static int
647readpage_async_filler(void *data, struct page *page)
648{
649 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
650 struct inode *inode = page->mapping->host;
651 struct nfs_page *new;
652 unsigned int len;
653
654 nfs_wb_page(inode, page);
Trond Myklebust49a70f22006-12-05 00:35:38 -0500655 len = nfs_page_length(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 if (len == 0)
657 return nfs_return_empty_page(page);
658 new = nfs_create_request(desc->ctx, inode, page, 0, len);
659 if (IS_ERR(new)) {
660 SetPageError(page);
661 unlock_page(page);
662 return PTR_ERR(new);
663 }
664 if (len < PAGE_CACHE_SIZE)
665 memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 nfs_list_add_request(new, desc->head);
667 return 0;
668}
669
670int nfs_readpages(struct file *filp, struct address_space *mapping,
671 struct list_head *pages, unsigned nr_pages)
672{
673 LIST_HEAD(head);
674 struct nfs_readdesc desc = {
675 .head = &head,
676 };
677 struct inode *inode = mapping->host;
678 struct nfs_server *server = NFS_SERVER(inode);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400679 int ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
682 inode->i_sb->s_id,
683 (long long)NFS_FILEID(inode),
684 nr_pages);
Chuck Lever91d5b472006-03-20 13:44:14 -0500685 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400687 if (NFS_STALE(inode))
688 goto out;
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (filp == NULL) {
Trond Myklebustd5308382005-11-04 15:33:38 -0500691 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (desc.ctx == NULL)
693 return -EBADF;
694 } else
695 desc.ctx = get_nfs_open_context((struct nfs_open_context *)
696 filp->private_data);
697 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
698 if (!list_empty(&head)) {
699 int err = nfs_pagein_list(&head, server->rpages);
700 if (!ret)
Chuck Lever91d5b472006-03-20 13:44:14 -0500701 nfs_add_stats(inode, NFSIOS_READPAGES, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 ret = err;
703 }
704 put_nfs_open_context(desc.ctx);
Trond Myklebust5f004cf2006-09-14 14:03:14 -0400705out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return ret;
707}
708
David Howellsf7b422b2006-06-09 09:34:33 -0400709int __init nfs_init_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
712 sizeof(struct nfs_read_data),
713 0, SLAB_HWCACHE_ALIGN,
714 NULL, NULL);
715 if (nfs_rdata_cachep == NULL)
716 return -ENOMEM;
717
Matthew Dobson93d23412006-03-26 01:37:50 -0800718 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
719 nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (nfs_rdata_mempool == NULL)
721 return -ENOMEM;
722
723 return 0;
724}
725
David Brownell266bee82006-06-27 12:59:15 -0700726void nfs_destroy_readpagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 mempool_destroy(nfs_rdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700729 kmem_cache_destroy(nfs_rdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}