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>
29 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
31 static int nfs_pagein_multi(struct inode
*, struct list_head
*, unsigned int, size_t, int);
32 static int nfs_pagein_one(struct inode
*, struct list_head
*, unsigned int, size_t, int);
33 static const struct rpc_call_ops nfs_read_partial_ops
;
34 static const struct rpc_call_ops nfs_read_full_ops
;
36 static struct kmem_cache
*nfs_rdata_cachep
;
37 static mempool_t
*nfs_rdata_mempool
;
39 #define MIN_POOL_READ (32)
41 struct nfs_read_data
*nfs_readdata_alloc(unsigned int pagecount
)
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 p
->res
.seq_res
.sr_slotid
= NFS4_MAX_SLOT_TABLE
;
50 if (pagecount
<= ARRAY_SIZE(p
->page_array
))
51 p
->pagevec
= p
->page_array
;
53 p
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), GFP_NOFS
);
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 new = nfs_create_request(ctx
, inode
, page
, 0, len
);
129 if (len
< PAGE_CACHE_SIZE
)
130 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
132 nfs_list_add_request(new, &one_request
);
133 if (NFS_SERVER(inode
)->rsize
< PAGE_CACHE_SIZE
)
134 nfs_pagein_multi(inode
, &one_request
, 1, len
, 0);
136 nfs_pagein_one(inode
, &one_request
, 1, len
, 0);
140 static void nfs_readpage_release(struct nfs_page
*req
)
142 struct inode
*d_inode
= req
->wb_context
->path
.dentry
->d_inode
;
144 if (PageUptodate(req
->wb_page
))
145 nfs_readpage_to_fscache(d_inode
, req
->wb_page
, 0);
147 unlock_page(req
->wb_page
);
149 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
150 req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
151 (long long)NFS_FILEID(req
->wb_context
->path
.dentry
->d_inode
),
153 (long long)req_offset(req
));
154 nfs_clear_request(req
);
155 nfs_release_request(req
);
159 * Set up the NFS read request struct
161 static int nfs_read_rpcsetup(struct nfs_page
*req
, struct nfs_read_data
*data
,
162 const struct rpc_call_ops
*call_ops
,
163 unsigned int count
, unsigned int offset
)
165 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
166 int swap_flags
= IS_SWAPFILE(inode
) ? NFS_RPC_SWAPFLAGS
: 0;
167 struct rpc_task
*task
;
168 struct rpc_message msg
= {
169 .rpc_argp
= &data
->args
,
170 .rpc_resp
= &data
->res
,
171 .rpc_cred
= req
->wb_context
->cred
,
173 struct rpc_task_setup task_setup_data
= {
175 .rpc_client
= NFS_CLIENT(inode
),
177 .callback_ops
= call_ops
,
178 .callback_data
= data
,
179 .workqueue
= nfsiod_workqueue
,
180 .flags
= RPC_TASK_ASYNC
| swap_flags
,
185 data
->cred
= msg
.rpc_cred
;
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
;
192 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
194 data
->res
.fattr
= &data
->fattr
;
195 data
->res
.count
= count
;
197 nfs_fattr_init(&data
->fattr
);
199 /* Set up the initial task struct. */
200 NFS_PROTO(inode
)->read_setup(data
, &msg
);
202 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
205 (long long)NFS_FILEID(inode
),
207 (unsigned long long)data
->args
.offset
);
209 task
= rpc_run_task(&task_setup_data
);
211 return PTR_ERR(task
);
217 nfs_async_read_error(struct list_head
*head
)
219 struct nfs_page
*req
;
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
);
230 * Generate multiple requests to fill a single page.
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.
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.
242 static int nfs_pagein_multi(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
244 struct nfs_page
*req
= nfs_list_entry(head
->next
);
245 struct page
*page
= req
->wb_page
;
246 struct nfs_read_data
*data
;
247 size_t rsize
= NFS_SERVER(inode
)->rsize
, nbytes
;
253 nfs_list_remove_request(req
);
257 size_t len
= min(nbytes
,rsize
);
259 data
= nfs_readdata_alloc(1);
262 list_add(&data
->pages
, &list
);
265 } while(nbytes
!= 0);
266 atomic_set(&req
->wb_complete
, requests
);
268 ClearPageError(page
);
274 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
275 list_del_init(&data
->pages
);
277 data
->pagevec
[0] = page
;
281 ret2
= nfs_read_rpcsetup(req
, data
, &nfs_read_partial_ops
,
287 } while (nbytes
!= 0);
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
);
298 nfs_readpage_release(req
);
302 static int nfs_pagein_one(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
304 struct nfs_page
*req
;
306 struct nfs_read_data
*data
;
309 data
= nfs_readdata_alloc(npages
);
313 pages
= data
->pagevec
;
314 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
;
321 req
= nfs_list_entry(data
->pages
.next
);
323 return nfs_read_rpcsetup(req
, data
, &nfs_read_full_ops
, count
, 0);
325 nfs_async_read_error(head
);
330 * This is the callback from RPC telling us whether a reply was
331 * received or some error occurred (timeout or socket shutdown).
333 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
337 dprintk("NFS: %s: %5u, (status %d)\n", __func__
, task
->tk_pid
,
340 status
= NFS_PROTO(data
->inode
)->read_done(task
, data
);
344 nfs_add_stats(data
->inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
346 if (task
->tk_status
== -ESTALE
) {
347 set_bit(NFS_INO_STALE
, &NFS_I(data
->inode
)->flags
);
348 nfs_mark_for_revalidate(data
->inode
);
353 static void nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
355 struct nfs_readargs
*argp
= &data
->args
;
356 struct nfs_readres
*resp
= &data
->res
;
358 if (resp
->eof
|| resp
->count
== argp
->count
)
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)
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
;
371 nfs4_restart_rpc(task
, NFS_SERVER(data
->inode
)->nfs_client
,
375 nfs4_sequence_free_slot(NFS_SERVER(data
->inode
)->nfs_client
,
382 * Handle a read reply that fills part of a page.
384 static void nfs_readpage_result_partial(struct rpc_task
*task
, void *calldata
)
386 struct nfs_read_data
*data
= calldata
;
388 if (nfs_readpage_result(task
, data
) != 0)
390 if (task
->tk_status
< 0)
393 nfs_readpage_truncate_uninitialised_page(data
);
394 nfs_readpage_retry(task
, data
);
397 static void nfs_readpage_release_partial(void *calldata
)
399 struct nfs_read_data
*data
= calldata
;
400 struct nfs_page
*req
= data
->req
;
401 struct page
*page
= req
->wb_page
;
402 int status
= data
->task
.tk_status
;
407 if (atomic_dec_and_test(&req
->wb_complete
)) {
408 if (!PageError(page
))
409 SetPageUptodate(page
);
410 nfs_readpage_release(req
);
412 nfs_readdata_release(calldata
);
415 #if defined(CONFIG_NFS_V4_1)
416 void nfs_read_prepare(struct rpc_task
*task
, void *calldata
)
418 struct nfs_read_data
*data
= calldata
;
420 if (nfs4_setup_sequence(NFS_SERVER(data
->inode
)->nfs_client
,
421 &data
->args
.seq_args
, &data
->res
.seq_res
,
424 rpc_call_start(task
);
426 #endif /* CONFIG_NFS_V4_1 */
428 static const struct rpc_call_ops nfs_read_partial_ops
= {
429 #if defined(CONFIG_NFS_V4_1)
430 .rpc_call_prepare
= nfs_read_prepare
,
431 #endif /* CONFIG_NFS_V4_1 */
432 .rpc_call_done
= nfs_readpage_result_partial
,
433 .rpc_release
= nfs_readpage_release_partial
,
436 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data
*data
)
438 unsigned int count
= data
->res
.count
;
439 unsigned int base
= data
->args
.pgbase
;
443 count
= data
->args
.count
;
444 if (unlikely(count
== 0))
446 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
447 base
&= ~PAGE_CACHE_MASK
;
449 for (;count
>= PAGE_CACHE_SIZE
; count
-= PAGE_CACHE_SIZE
, pages
++)
450 SetPageUptodate(*pages
);
453 /* Was this a short read? */
454 if (data
->res
.eof
|| data
->res
.count
== data
->args
.count
)
455 SetPageUptodate(*pages
);
459 * This is the callback from RPC telling us whether a reply was
460 * received or some error occurred (timeout or socket shutdown).
462 static void nfs_readpage_result_full(struct rpc_task
*task
, void *calldata
)
464 struct nfs_read_data
*data
= calldata
;
466 if (nfs_readpage_result(task
, data
) != 0)
468 if (task
->tk_status
< 0)
471 * Note: nfs_readpage_retry may change the values of
472 * data->args. In the multi-page case, we therefore need
473 * to ensure that we call nfs_readpage_set_pages_uptodate()
476 nfs_readpage_truncate_uninitialised_page(data
);
477 nfs_readpage_set_pages_uptodate(data
);
478 nfs_readpage_retry(task
, data
);
481 static void nfs_readpage_release_full(void *calldata
)
483 struct nfs_read_data
*data
= calldata
;
485 while (!list_empty(&data
->pages
)) {
486 struct nfs_page
*req
= nfs_list_entry(data
->pages
.next
);
488 nfs_list_remove_request(req
);
489 nfs_readpage_release(req
);
491 nfs_readdata_release(calldata
);
494 static const struct rpc_call_ops nfs_read_full_ops
= {
495 #if defined(CONFIG_NFS_V4_1)
496 .rpc_call_prepare
= nfs_read_prepare
,
497 #endif /* CONFIG_NFS_V4_1 */
498 .rpc_call_done
= nfs_readpage_result_full
,
499 .rpc_release
= nfs_readpage_release_full
,
503 * Read a page over NFS.
504 * We read the page synchronously in the following case:
505 * - The error flag is set for this page. This happens only when a
506 * previous async read operation failed.
508 int nfs_readpage(struct file
*file
, struct page
*page
)
510 struct nfs_open_context
*ctx
;
511 struct inode
*inode
= page
->mapping
->host
;
514 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
515 page
, PAGE_CACHE_SIZE
, page
->index
);
516 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
517 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
520 * Try to flush any pending writes to the file..
522 * NOTE! Because we own the page lock, there cannot
523 * be any new pending writes generated at this point
524 * for this page (other pages can be written to).
526 error
= nfs_wb_page(inode
, page
);
529 if (PageUptodate(page
))
533 if (NFS_STALE(inode
))
538 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
542 ctx
= get_nfs_open_context(nfs_file_open_context(file
));
544 if (!IS_SYNC(inode
)) {
545 error
= nfs_readpage_from_fscache(ctx
, inode
, page
);
550 error
= nfs_readpage_async(ctx
, inode
, page
);
553 put_nfs_open_context(ctx
);
560 struct nfs_readdesc
{
561 struct nfs_pageio_descriptor
*pgio
;
562 struct nfs_open_context
*ctx
;
566 readpage_async_filler(void *data
, struct page
*page
)
568 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
569 struct inode
*inode
= page
->mapping
->host
;
570 struct nfs_page
*new;
574 len
= nfs_page_length(page
);
576 return nfs_return_empty_page(page
);
578 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
582 if (len
< PAGE_CACHE_SIZE
)
583 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
584 if (!nfs_pageio_add_request(desc
->pgio
, new)) {
585 error
= desc
->pgio
->pg_error
;
590 error
= PTR_ERR(new);
597 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
598 struct list_head
*pages
, unsigned nr_pages
)
600 struct nfs_pageio_descriptor pgio
;
601 struct nfs_readdesc desc
= {
604 struct inode
*inode
= mapping
->host
;
605 struct nfs_server
*server
= NFS_SERVER(inode
);
606 size_t rsize
= server
->rsize
;
607 unsigned long npages
;
610 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
612 (long long)NFS_FILEID(inode
),
614 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
616 if (NFS_STALE(inode
))
620 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
621 if (desc
.ctx
== NULL
)
624 desc
.ctx
= get_nfs_open_context(nfs_file_open_context(filp
));
626 /* attempt to read as many of the pages as possible from the cache
627 * - this returns -ENOBUFS immediately if the cookie is negative
629 ret
= nfs_readpages_from_fscache(desc
.ctx
, inode
, mapping
,
632 goto read_complete
; /* all pages were read */
634 if (rsize
< PAGE_CACHE_SIZE
)
635 nfs_pageio_init(&pgio
, inode
, nfs_pagein_multi
, rsize
, 0);
637 nfs_pageio_init(&pgio
, inode
, nfs_pagein_one
, rsize
, 0);
639 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
641 nfs_pageio_complete(&pgio
);
642 npages
= (pgio
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
643 nfs_add_stats(inode
, NFSIOS_READPAGES
, npages
);
645 put_nfs_open_context(desc
.ctx
);
650 int __init
nfs_init_readpagecache(void)
652 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
653 sizeof(struct nfs_read_data
),
654 0, SLAB_HWCACHE_ALIGN
,
656 if (nfs_rdata_cachep
== NULL
)
659 nfs_rdata_mempool
= mempool_create_slab_pool(MIN_POOL_READ
,
661 if (nfs_rdata_mempool
== NULL
)
667 void nfs_destroy_readpagecache(void)
669 mempool_destroy(nfs_rdata_mempool
);
670 kmem_cache_destroy(nfs_rdata_cachep
);