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>
22 #include <asm/system.h>
30 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
32 static int nfs_pagein_multi(struct inode
*, struct list_head
*, unsigned int, size_t, int);
33 static int nfs_pagein_one(struct inode
*, struct list_head
*, unsigned int, size_t, int);
34 static const struct rpc_call_ops nfs_read_partial_ops
;
35 static const struct rpc_call_ops nfs_read_full_ops
;
37 static struct kmem_cache
*nfs_rdata_cachep
;
38 static mempool_t
*nfs_rdata_mempool
;
40 #define MIN_POOL_READ (32)
42 struct nfs_read_data
*nfs_readdata_alloc(unsigned int pagecount
)
44 struct nfs_read_data
*p
= mempool_alloc(nfs_rdata_mempool
, GFP_KERNEL
);
47 memset(p
, 0, sizeof(*p
));
48 INIT_LIST_HEAD(&p
->pages
);
49 p
->npages
= pagecount
;
50 if (pagecount
<= ARRAY_SIZE(p
->page_array
))
51 p
->pagevec
= p
->page_array
;
53 p
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), GFP_KERNEL
);
55 mempool_free(p
, nfs_rdata_mempool
);
63 void nfs_readdata_free(struct nfs_read_data
*p
)
65 if (p
&& (p
->pagevec
!= &p
->page_array
[0]))
67 mempool_free(p
, nfs_rdata_mempool
);
70 static void nfs_readdata_release(struct nfs_read_data
*rdata
)
72 put_nfs_open_context(rdata
->args
.context
);
73 nfs_readdata_free(rdata
);
77 int nfs_return_empty_page(struct page
*page
)
79 zero_user(page
, 0, PAGE_CACHE_SIZE
);
80 SetPageUptodate(page
);
85 static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data
*data
)
87 unsigned int remainder
= data
->args
.count
- data
->res
.count
;
88 unsigned int base
= data
->args
.pgbase
+ data
->res
.count
;
92 if (data
->res
.eof
== 0 || remainder
== 0)
95 * Note: "remainder" can never be negative, since we check for
96 * this in the XDR code.
98 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
99 base
&= ~PAGE_CACHE_MASK
;
100 pglen
= PAGE_CACHE_SIZE
- base
;
102 if (remainder
<= pglen
) {
103 zero_user(*pages
, base
, remainder
);
106 zero_user(*pages
, base
, pglen
);
109 pglen
= PAGE_CACHE_SIZE
;
114 int nfs_readpage_async(struct nfs_open_context
*ctx
, struct inode
*inode
,
117 LIST_HEAD(one_request
);
118 struct nfs_page
*new;
121 len
= nfs_page_length(page
);
123 return nfs_return_empty_page(page
);
124 pnfs_update_layout(inode
, ctx
, IOMODE_READ
);
125 new = nfs_create_request(ctx
, inode
, page
, 0, len
);
130 if (len
< PAGE_CACHE_SIZE
)
131 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
133 nfs_list_add_request(new, &one_request
);
134 if (NFS_SERVER(inode
)->rsize
< PAGE_CACHE_SIZE
)
135 nfs_pagein_multi(inode
, &one_request
, 1, len
, 0);
137 nfs_pagein_one(inode
, &one_request
, 1, len
, 0);
141 static void nfs_readpage_release(struct nfs_page
*req
)
143 struct inode
*d_inode
= req
->wb_context
->path
.dentry
->d_inode
;
145 if (PageUptodate(req
->wb_page
))
146 nfs_readpage_to_fscache(d_inode
, req
->wb_page
, 0);
148 unlock_page(req
->wb_page
);
150 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
151 req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
152 (long long)NFS_FILEID(req
->wb_context
->path
.dentry
->d_inode
),
154 (long long)req_offset(req
));
155 nfs_clear_request(req
);
156 nfs_release_request(req
);
160 * Set up the NFS read request struct
162 static int nfs_read_rpcsetup(struct nfs_page
*req
, struct nfs_read_data
*data
,
163 const struct rpc_call_ops
*call_ops
,
164 unsigned int count
, unsigned int offset
)
166 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
167 int swap_flags
= IS_SWAPFILE(inode
) ? NFS_RPC_SWAPFLAGS
: 0;
168 struct rpc_task
*task
;
169 struct rpc_message msg
= {
170 .rpc_argp
= &data
->args
,
171 .rpc_resp
= &data
->res
,
172 .rpc_cred
= req
->wb_context
->cred
,
174 struct rpc_task_setup task_setup_data
= {
176 .rpc_client
= NFS_CLIENT(inode
),
178 .callback_ops
= call_ops
,
179 .callback_data
= data
,
180 .workqueue
= nfsiod_workqueue
,
181 .flags
= RPC_TASK_ASYNC
| swap_flags
,
186 data
->cred
= msg
.rpc_cred
;
188 data
->args
.fh
= NFS_FH(inode
);
189 data
->args
.offset
= req_offset(req
) + offset
;
190 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
191 data
->args
.pages
= data
->pagevec
;
192 data
->args
.count
= count
;
193 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
194 data
->args
.lock_context
= req
->wb_lock_context
;
196 data
->res
.fattr
= &data
->fattr
;
197 data
->res
.count
= count
;
199 nfs_fattr_init(&data
->fattr
);
201 /* Set up the initial task struct. */
202 NFS_PROTO(inode
)->read_setup(data
, &msg
);
204 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
207 (long long)NFS_FILEID(inode
),
209 (unsigned long long)data
->args
.offset
);
211 task
= rpc_run_task(&task_setup_data
);
213 return PTR_ERR(task
);
219 nfs_async_read_error(struct list_head
*head
)
221 struct nfs_page
*req
;
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
);
232 * Generate multiple requests to fill a single page.
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.
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.
244 static int nfs_pagein_multi(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
246 struct nfs_page
*req
= nfs_list_entry(head
->next
);
247 struct page
*page
= req
->wb_page
;
248 struct nfs_read_data
*data
;
249 size_t rsize
= NFS_SERVER(inode
)->rsize
, nbytes
;
255 nfs_list_remove_request(req
);
259 size_t len
= min(nbytes
,rsize
);
261 data
= nfs_readdata_alloc(1);
264 list_add(&data
->pages
, &list
);
267 } while(nbytes
!= 0);
268 atomic_set(&req
->wb_complete
, requests
);
270 ClearPageError(page
);
276 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
277 list_del_init(&data
->pages
);
279 data
->pagevec
[0] = page
;
283 ret2
= nfs_read_rpcsetup(req
, data
, &nfs_read_partial_ops
,
289 } while (nbytes
!= 0);
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
);
300 nfs_readpage_release(req
);
304 static int nfs_pagein_one(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
306 struct nfs_page
*req
;
308 struct nfs_read_data
*data
;
311 data
= nfs_readdata_alloc(npages
);
315 pages
= data
->pagevec
;
316 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
;
323 req
= nfs_list_entry(data
->pages
.next
);
325 return nfs_read_rpcsetup(req
, data
, &nfs_read_full_ops
, count
, 0);
327 nfs_async_read_error(head
);
332 * This is the callback from RPC telling us whether a reply was
333 * received or some error occurred (timeout or socket shutdown).
335 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
339 dprintk("NFS: %s: %5u, (status %d)\n", __func__
, task
->tk_pid
,
342 status
= NFS_PROTO(data
->inode
)->read_done(task
, data
);
346 nfs_add_stats(data
->inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
348 if (task
->tk_status
== -ESTALE
) {
349 set_bit(NFS_INO_STALE
, &NFS_I(data
->inode
)->flags
);
350 nfs_mark_for_revalidate(data
->inode
);
355 static void nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
357 struct nfs_readargs
*argp
= &data
->args
;
358 struct nfs_readres
*resp
= &data
->res
;
360 if (resp
->eof
|| resp
->count
== argp
->count
)
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)
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
;
373 nfs_restart_rpc(task
, NFS_SERVER(data
->inode
)->nfs_client
);
377 * Handle a read reply that fills part of a page.
379 static void nfs_readpage_result_partial(struct rpc_task
*task
, void *calldata
)
381 struct nfs_read_data
*data
= calldata
;
383 if (nfs_readpage_result(task
, data
) != 0)
385 if (task
->tk_status
< 0)
388 nfs_readpage_truncate_uninitialised_page(data
);
389 nfs_readpage_retry(task
, data
);
392 static void nfs_readpage_release_partial(void *calldata
)
394 struct nfs_read_data
*data
= calldata
;
395 struct nfs_page
*req
= data
->req
;
396 struct page
*page
= req
->wb_page
;
397 int status
= data
->task
.tk_status
;
402 if (atomic_dec_and_test(&req
->wb_complete
)) {
403 if (!PageError(page
))
404 SetPageUptodate(page
);
405 nfs_readpage_release(req
);
407 nfs_readdata_release(calldata
);
410 #if defined(CONFIG_NFS_V4_1)
411 void nfs_read_prepare(struct rpc_task
*task
, void *calldata
)
413 struct nfs_read_data
*data
= calldata
;
415 if (nfs4_setup_sequence(NFS_SERVER(data
->inode
),
416 &data
->args
.seq_args
, &data
->res
.seq_res
,
419 rpc_call_start(task
);
421 #endif /* CONFIG_NFS_V4_1 */
423 static const struct rpc_call_ops nfs_read_partial_ops
= {
424 #if defined(CONFIG_NFS_V4_1)
425 .rpc_call_prepare
= nfs_read_prepare
,
426 #endif /* CONFIG_NFS_V4_1 */
427 .rpc_call_done
= nfs_readpage_result_partial
,
428 .rpc_release
= nfs_readpage_release_partial
,
431 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data
*data
)
433 unsigned int count
= data
->res
.count
;
434 unsigned int base
= data
->args
.pgbase
;
438 count
= data
->args
.count
;
439 if (unlikely(count
== 0))
441 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
442 base
&= ~PAGE_CACHE_MASK
;
444 for (;count
>= PAGE_CACHE_SIZE
; count
-= PAGE_CACHE_SIZE
, pages
++)
445 SetPageUptodate(*pages
);
448 /* Was this a short read? */
449 if (data
->res
.eof
|| data
->res
.count
== data
->args
.count
)
450 SetPageUptodate(*pages
);
454 * This is the callback from RPC telling us whether a reply was
455 * received or some error occurred (timeout or socket shutdown).
457 static void nfs_readpage_result_full(struct rpc_task
*task
, void *calldata
)
459 struct nfs_read_data
*data
= calldata
;
461 if (nfs_readpage_result(task
, data
) != 0)
463 if (task
->tk_status
< 0)
466 * Note: nfs_readpage_retry may change the values of
467 * data->args. In the multi-page case, we therefore need
468 * to ensure that we call nfs_readpage_set_pages_uptodate()
471 nfs_readpage_truncate_uninitialised_page(data
);
472 nfs_readpage_set_pages_uptodate(data
);
473 nfs_readpage_retry(task
, data
);
476 static void nfs_readpage_release_full(void *calldata
)
478 struct nfs_read_data
*data
= calldata
;
480 while (!list_empty(&data
->pages
)) {
481 struct nfs_page
*req
= nfs_list_entry(data
->pages
.next
);
483 nfs_list_remove_request(req
);
484 nfs_readpage_release(req
);
486 nfs_readdata_release(calldata
);
489 static const struct rpc_call_ops nfs_read_full_ops
= {
490 #if defined(CONFIG_NFS_V4_1)
491 .rpc_call_prepare
= nfs_read_prepare
,
492 #endif /* CONFIG_NFS_V4_1 */
493 .rpc_call_done
= nfs_readpage_result_full
,
494 .rpc_release
= nfs_readpage_release_full
,
498 * Read a page over NFS.
499 * We read the page synchronously in the following case:
500 * - The error flag is set for this page. This happens only when a
501 * previous async read operation failed.
503 int nfs_readpage(struct file
*file
, struct page
*page
)
505 struct nfs_open_context
*ctx
;
506 struct inode
*inode
= page
->mapping
->host
;
509 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
510 page
, PAGE_CACHE_SIZE
, page
->index
);
511 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
512 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
515 * Try to flush any pending writes to the file..
517 * NOTE! Because we own the page lock, there cannot
518 * be any new pending writes generated at this point
519 * for this page (other pages can be written to).
521 error
= nfs_wb_page(inode
, page
);
524 if (PageUptodate(page
))
528 if (NFS_STALE(inode
))
533 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
537 ctx
= get_nfs_open_context(nfs_file_open_context(file
));
539 if (!IS_SYNC(inode
)) {
540 error
= nfs_readpage_from_fscache(ctx
, inode
, page
);
545 error
= nfs_readpage_async(ctx
, inode
, page
);
548 put_nfs_open_context(ctx
);
555 struct nfs_readdesc
{
556 struct nfs_pageio_descriptor
*pgio
;
557 struct nfs_open_context
*ctx
;
561 readpage_async_filler(void *data
, struct page
*page
)
563 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
564 struct inode
*inode
= page
->mapping
->host
;
565 struct nfs_page
*new;
569 len
= nfs_page_length(page
);
571 return nfs_return_empty_page(page
);
573 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
577 if (len
< PAGE_CACHE_SIZE
)
578 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
579 if (!nfs_pageio_add_request(desc
->pgio
, new)) {
580 error
= desc
->pgio
->pg_error
;
585 error
= PTR_ERR(new);
592 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
593 struct list_head
*pages
, unsigned nr_pages
)
595 struct nfs_pageio_descriptor pgio
;
596 struct nfs_readdesc desc
= {
599 struct inode
*inode
= mapping
->host
;
600 struct nfs_server
*server
= NFS_SERVER(inode
);
601 size_t rsize
= server
->rsize
;
602 unsigned long npages
;
605 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
607 (long long)NFS_FILEID(inode
),
609 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
611 if (NFS_STALE(inode
))
615 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
616 if (desc
.ctx
== NULL
)
619 desc
.ctx
= get_nfs_open_context(nfs_file_open_context(filp
));
621 /* attempt to read as many of the pages as possible from the cache
622 * - this returns -ENOBUFS immediately if the cookie is negative
624 ret
= nfs_readpages_from_fscache(desc
.ctx
, inode
, mapping
,
627 goto read_complete
; /* all pages were read */
629 pnfs_update_layout(inode
, desc
.ctx
, IOMODE_READ
);
630 if (rsize
< PAGE_CACHE_SIZE
)
631 nfs_pageio_init(&pgio
, inode
, nfs_pagein_multi
, rsize
, 0);
633 nfs_pageio_init(&pgio
, inode
, nfs_pagein_one
, rsize
, 0);
635 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
637 nfs_pageio_complete(&pgio
);
638 npages
= (pgio
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
639 nfs_add_stats(inode
, NFSIOS_READPAGES
, npages
);
641 put_nfs_open_context(desc
.ctx
);
646 int __init
nfs_init_readpagecache(void)
648 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
649 sizeof(struct nfs_read_data
),
650 0, SLAB_HWCACHE_ALIGN
,
652 if (nfs_rdata_cachep
== NULL
)
655 nfs_rdata_mempool
= mempool_create_slab_pool(MIN_POOL_READ
,
657 if (nfs_rdata_mempool
== NULL
)
663 void nfs_destroy_readpagecache(void)
665 mempool_destroy(nfs_rdata_mempool
);
666 kmem_cache_destroy(nfs_rdata_cachep
);