6 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
7 * modified for async RPC by okir@monad.swb.de
10 #include <linux/time.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/fcntl.h>
14 #include <linux/stat.h>
16 #include <linux/slab.h>
17 #include <linux/pagemap.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_page.h>
21 #include <linux/smp_lock.h>
23 #include <asm/system.h>
28 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
30 static int nfs_pagein_multi(struct inode
*, struct list_head
*, size_t, int);
31 static int nfs_pagein_one(struct inode
*, struct list_head
*, size_t, int);
32 static const struct rpc_call_ops nfs_read_partial_ops
;
33 static const struct rpc_call_ops nfs_read_full_ops
;
35 static struct kmem_cache
*nfs_rdata_cachep
;
36 static mempool_t
*nfs_rdata_mempool
;
38 #define MIN_POOL_READ (32)
40 struct nfs_read_data
*nfs_readdata_alloc(size_t len
)
42 unsigned int pagecount
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
43 struct nfs_read_data
*p
= mempool_alloc(nfs_rdata_mempool
, GFP_NOFS
);
46 memset(p
, 0, sizeof(*p
));
47 INIT_LIST_HEAD(&p
->pages
);
48 p
->npages
= pagecount
;
49 if (pagecount
<= ARRAY_SIZE(p
->page_array
))
50 p
->pagevec
= p
->page_array
;
52 p
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), GFP_NOFS
);
54 mempool_free(p
, nfs_rdata_mempool
);
62 static void nfs_readdata_rcu_free(struct rcu_head
*head
)
64 struct nfs_read_data
*p
= container_of(head
, struct nfs_read_data
, task
.u
.tk_rcu
);
65 if (p
&& (p
->pagevec
!= &p
->page_array
[0]))
67 mempool_free(p
, nfs_rdata_mempool
);
70 static void nfs_readdata_free(struct nfs_read_data
*rdata
)
72 call_rcu_bh(&rdata
->task
.u
.tk_rcu
, nfs_readdata_rcu_free
);
75 void nfs_readdata_release(void *data
)
77 nfs_readdata_free(data
);
81 int nfs_return_empty_page(struct page
*page
)
83 memclear_highpage_flush(page
, 0, PAGE_CACHE_SIZE
);
84 SetPageUptodate(page
);
89 static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data
*data
)
91 unsigned int remainder
= data
->args
.count
- data
->res
.count
;
92 unsigned int base
= data
->args
.pgbase
+ data
->res
.count
;
96 if (data
->res
.eof
== 0 || remainder
== 0)
99 * Note: "remainder" can never be negative, since we check for
100 * this in the XDR code.
102 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
103 base
&= ~PAGE_CACHE_MASK
;
104 pglen
= PAGE_CACHE_SIZE
- base
;
106 if (remainder
<= pglen
) {
107 memclear_highpage_flush(*pages
, base
, remainder
);
110 memclear_highpage_flush(*pages
, base
, pglen
);
113 pglen
= PAGE_CACHE_SIZE
;
118 static int nfs_readpage_async(struct nfs_open_context
*ctx
, struct inode
*inode
,
121 LIST_HEAD(one_request
);
122 struct nfs_page
*new;
125 len
= nfs_page_length(page
);
127 return nfs_return_empty_page(page
);
128 new = nfs_create_request(ctx
, inode
, page
, 0, len
);
133 if (len
< PAGE_CACHE_SIZE
)
134 memclear_highpage_flush(page
, len
, PAGE_CACHE_SIZE
- len
);
136 nfs_list_add_request(new, &one_request
);
137 if (NFS_SERVER(inode
)->rsize
< PAGE_CACHE_SIZE
)
138 nfs_pagein_multi(inode
, &one_request
, len
, 0);
140 nfs_pagein_one(inode
, &one_request
, len
, 0);
144 static void nfs_readpage_release(struct nfs_page
*req
)
146 unlock_page(req
->wb_page
);
148 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
149 req
->wb_context
->dentry
->d_inode
->i_sb
->s_id
,
150 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
152 (long long)req_offset(req
));
153 nfs_clear_request(req
);
154 nfs_release_request(req
);
158 * Set up the NFS read request struct
160 static void nfs_read_rpcsetup(struct nfs_page
*req
, struct nfs_read_data
*data
,
161 const struct rpc_call_ops
*call_ops
,
162 unsigned int count
, unsigned int offset
)
168 data
->inode
= inode
= req
->wb_context
->dentry
->d_inode
;
169 data
->cred
= req
->wb_context
->cred
;
171 data
->args
.fh
= NFS_FH(inode
);
172 data
->args
.offset
= req_offset(req
) + offset
;
173 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
174 data
->args
.pages
= data
->pagevec
;
175 data
->args
.count
= count
;
176 data
->args
.context
= req
->wb_context
;
178 data
->res
.fattr
= &data
->fattr
;
179 data
->res
.count
= count
;
181 nfs_fattr_init(&data
->fattr
);
183 /* Set up the initial task struct. */
184 flags
= RPC_TASK_ASYNC
| (IS_SWAPFILE(inode
)? NFS_RPC_SWAPFLAGS
: 0);
185 rpc_init_task(&data
->task
, NFS_CLIENT(inode
), flags
, call_ops
, data
);
186 NFS_PROTO(inode
)->read_setup(data
);
188 data
->task
.tk_cookie
= (unsigned long)inode
;
190 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
193 (long long)NFS_FILEID(inode
),
195 (unsigned long long)data
->args
.offset
);
199 nfs_async_read_error(struct list_head
*head
)
201 struct nfs_page
*req
;
203 while (!list_empty(head
)) {
204 req
= nfs_list_entry(head
->next
);
205 nfs_list_remove_request(req
);
206 SetPageError(req
->wb_page
);
207 nfs_readpage_release(req
);
212 * Start an async read operation
214 static void nfs_execute_read(struct nfs_read_data
*data
)
216 struct rpc_clnt
*clnt
= NFS_CLIENT(data
->inode
);
219 rpc_clnt_sigmask(clnt
, &oldset
);
220 rpc_execute(&data
->task
);
221 rpc_clnt_sigunmask(clnt
, &oldset
);
225 * Generate multiple requests to fill a single page.
227 * We optimize to reduce the number of read operations on the wire. If we
228 * detect that we're reading a page, or an area of a page, that is past the
229 * end of file, we do not generate NFS read operations but just clear the
230 * parts of the page that would have come back zero from the server anyway.
232 * We rely on the cached value of i_size to make this determination; another
233 * client can fill pages on the server past our cached end-of-file, but we
234 * won't see the new data until our attribute cache is updated. This is more
235 * or less conventional NFS client behavior.
237 static int nfs_pagein_multi(struct inode
*inode
, struct list_head
*head
, size_t count
, int flags
)
239 struct nfs_page
*req
= nfs_list_entry(head
->next
);
240 struct page
*page
= req
->wb_page
;
241 struct nfs_read_data
*data
;
242 size_t rsize
= NFS_SERVER(inode
)->rsize
, nbytes
;
247 nfs_list_remove_request(req
);
251 size_t len
= min(nbytes
,rsize
);
253 data
= nfs_readdata_alloc(len
);
256 INIT_LIST_HEAD(&data
->pages
);
257 list_add(&data
->pages
, &list
);
260 } while(nbytes
!= 0);
261 atomic_set(&req
->wb_complete
, requests
);
263 ClearPageError(page
);
267 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
268 list_del_init(&data
->pages
);
270 data
->pagevec
[0] = page
;
274 nfs_read_rpcsetup(req
, data
, &nfs_read_partial_ops
,
278 nfs_execute_read(data
);
279 } while (nbytes
!= 0);
284 while (!list_empty(&list
)) {
285 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
286 list_del(&data
->pages
);
287 nfs_readdata_free(data
);
290 nfs_readpage_release(req
);
294 static int nfs_pagein_one(struct inode
*inode
, struct list_head
*head
, size_t count
, int flags
)
296 struct nfs_page
*req
;
298 struct nfs_read_data
*data
;
300 data
= nfs_readdata_alloc(count
);
304 INIT_LIST_HEAD(&data
->pages
);
305 pages
= data
->pagevec
;
306 while (!list_empty(head
)) {
307 req
= nfs_list_entry(head
->next
);
308 nfs_list_remove_request(req
);
309 nfs_list_add_request(req
, &data
->pages
);
310 ClearPageError(req
->wb_page
);
311 *pages
++ = req
->wb_page
;
313 req
= nfs_list_entry(data
->pages
.next
);
315 nfs_read_rpcsetup(req
, data
, &nfs_read_full_ops
, count
, 0);
317 nfs_execute_read(data
);
320 nfs_async_read_error(head
);
325 nfs_pagein_list(struct inode
*inode
, struct list_head
*head
, unsigned int rsize
)
327 struct nfs_pageio_descriptor desc
;
328 unsigned int pages
= 0;
331 if (rsize
< PAGE_CACHE_SIZE
)
332 nfs_pageio_init(&desc
, inode
, nfs_pagein_multi
, rsize
, 0);
334 nfs_pageio_init(&desc
, inode
, nfs_pagein_one
, rsize
, 0);
336 nfs_pageio_add_list(&desc
, head
);
337 nfs_pageio_complete(&desc
);
338 pages
+= (desc
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
340 nfs_async_read_error(head
);
347 * This is the callback from RPC telling us whether a reply was
348 * received or some error occurred (timeout or socket shutdown).
350 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
354 dprintk("NFS: %s: %5u, (status %d)\n", __FUNCTION__
, task
->tk_pid
,
357 status
= NFS_PROTO(data
->inode
)->read_done(task
, data
);
361 nfs_add_stats(data
->inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
363 if (task
->tk_status
== -ESTALE
) {
364 set_bit(NFS_INO_STALE
, &NFS_FLAGS(data
->inode
));
365 nfs_mark_for_revalidate(data
->inode
);
367 spin_lock(&data
->inode
->i_lock
);
368 NFS_I(data
->inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
369 spin_unlock(&data
->inode
->i_lock
);
373 static int nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
375 struct nfs_readargs
*argp
= &data
->args
;
376 struct nfs_readres
*resp
= &data
->res
;
378 if (resp
->eof
|| resp
->count
== argp
->count
)
381 /* This is a short read! */
382 nfs_inc_stats(data
->inode
, NFSIOS_SHORTREAD
);
383 /* Has the server at least made some progress? */
384 if (resp
->count
== 0)
387 /* Yes, so retry the read at the end of the data */
388 argp
->offset
+= resp
->count
;
389 argp
->pgbase
+= resp
->count
;
390 argp
->count
-= resp
->count
;
391 rpc_restart_call(task
);
396 * Handle a read reply that fills part of a page.
398 static void nfs_readpage_result_partial(struct rpc_task
*task
, void *calldata
)
400 struct nfs_read_data
*data
= calldata
;
401 struct nfs_page
*req
= data
->req
;
402 struct page
*page
= req
->wb_page
;
404 if (nfs_readpage_result(task
, data
) != 0)
407 if (likely(task
->tk_status
>= 0)) {
408 nfs_readpage_truncate_uninitialised_page(data
);
409 if (nfs_readpage_retry(task
, data
) != 0)
412 if (unlikely(task
->tk_status
< 0))
414 if (atomic_dec_and_test(&req
->wb_complete
)) {
415 if (!PageError(page
))
416 SetPageUptodate(page
);
417 nfs_readpage_release(req
);
421 static const struct rpc_call_ops nfs_read_partial_ops
= {
422 .rpc_call_done
= nfs_readpage_result_partial
,
423 .rpc_release
= nfs_readdata_release
,
426 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data
*data
)
428 unsigned int count
= data
->res
.count
;
429 unsigned int base
= data
->args
.pgbase
;
433 count
= data
->args
.count
;
434 if (unlikely(count
== 0))
436 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
437 base
&= ~PAGE_CACHE_MASK
;
439 for (;count
>= PAGE_CACHE_SIZE
; count
-= PAGE_CACHE_SIZE
, pages
++)
440 SetPageUptodate(*pages
);
443 /* Was this a short read? */
444 if (data
->res
.eof
|| data
->res
.count
== data
->args
.count
)
445 SetPageUptodate(*pages
);
449 * This is the callback from RPC telling us whether a reply was
450 * received or some error occurred (timeout or socket shutdown).
452 static void nfs_readpage_result_full(struct rpc_task
*task
, void *calldata
)
454 struct nfs_read_data
*data
= calldata
;
456 if (nfs_readpage_result(task
, data
) != 0)
459 * Note: nfs_readpage_retry may change the values of
460 * data->args. In the multi-page case, we therefore need
461 * to ensure that we call nfs_readpage_set_pages_uptodate()
464 if (likely(task
->tk_status
>= 0)) {
465 nfs_readpage_truncate_uninitialised_page(data
);
466 nfs_readpage_set_pages_uptodate(data
);
467 if (nfs_readpage_retry(task
, data
) != 0)
470 while (!list_empty(&data
->pages
)) {
471 struct nfs_page
*req
= nfs_list_entry(data
->pages
.next
);
473 nfs_list_remove_request(req
);
474 nfs_readpage_release(req
);
478 static const struct rpc_call_ops nfs_read_full_ops
= {
479 .rpc_call_done
= nfs_readpage_result_full
,
480 .rpc_release
= nfs_readdata_release
,
484 * Read a page over NFS.
485 * We read the page synchronously in the following case:
486 * - The error flag is set for this page. This happens only when a
487 * previous async read operation failed.
489 int nfs_readpage(struct file
*file
, struct page
*page
)
491 struct nfs_open_context
*ctx
;
492 struct inode
*inode
= page
->mapping
->host
;
495 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
496 page
, PAGE_CACHE_SIZE
, page
->index
);
497 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
498 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
501 * Try to flush any pending writes to the file..
503 * NOTE! Because we own the page lock, there cannot
504 * be any new pending writes generated at this point
505 * for this page (other pages can be written to).
507 error
= nfs_wb_page(inode
, page
);
512 if (NFS_STALE(inode
))
517 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
521 ctx
= get_nfs_open_context((struct nfs_open_context
*)
524 error
= nfs_readpage_async(ctx
, inode
, page
);
526 put_nfs_open_context(ctx
);
534 struct nfs_readdesc
{
535 struct list_head
*head
;
536 struct nfs_open_context
*ctx
;
540 readpage_async_filler(void *data
, struct page
*page
)
542 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
543 struct inode
*inode
= page
->mapping
->host
;
544 struct nfs_page
*new;
547 nfs_wb_page(inode
, page
);
548 len
= nfs_page_length(page
);
550 return nfs_return_empty_page(page
);
551 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
557 if (len
< PAGE_CACHE_SIZE
)
558 memclear_highpage_flush(page
, len
, PAGE_CACHE_SIZE
- len
);
559 nfs_list_add_request(new, desc
->head
);
563 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
564 struct list_head
*pages
, unsigned nr_pages
)
567 struct nfs_readdesc desc
= {
570 struct inode
*inode
= mapping
->host
;
571 struct nfs_server
*server
= NFS_SERVER(inode
);
574 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
576 (long long)NFS_FILEID(inode
),
578 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
580 if (NFS_STALE(inode
))
584 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
585 if (desc
.ctx
== NULL
)
588 desc
.ctx
= get_nfs_open_context((struct nfs_open_context
*)
590 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
591 if (!list_empty(&head
)) {
592 int err
= nfs_pagein_list(inode
, &head
, server
->rsize
);
594 nfs_add_stats(inode
, NFSIOS_READPAGES
, err
);
597 put_nfs_open_context(desc
.ctx
);
602 int __init
nfs_init_readpagecache(void)
604 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
605 sizeof(struct nfs_read_data
),
606 0, SLAB_HWCACHE_ALIGN
,
608 if (nfs_rdata_cachep
== NULL
)
611 nfs_rdata_mempool
= mempool_create_slab_pool(MIN_POOL_READ
,
613 if (nfs_rdata_mempool
== NULL
)
619 void nfs_destroy_readpagecache(void)
621 mempool_destroy(nfs_rdata_mempool
);
622 kmem_cache_destroy(nfs_rdata_cachep
);