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>
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 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_free(struct nfs_read_data
*p
)
64 if (p
&& (p
->pagevec
!= &p
->page_array
[0]))
66 mempool_free(p
, nfs_rdata_mempool
);
69 void nfs_readdata_release(void *data
)
71 struct nfs_read_data
*rdata
= data
;
73 put_nfs_open_context(rdata
->args
.context
);
74 nfs_readdata_free(rdata
);
78 int nfs_return_empty_page(struct page
*page
)
80 zero_user(page
, 0, PAGE_CACHE_SIZE
);
81 SetPageUptodate(page
);
86 static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data
*data
)
88 unsigned int remainder
= data
->args
.count
- data
->res
.count
;
89 unsigned int base
= data
->args
.pgbase
+ data
->res
.count
;
93 if (data
->res
.eof
== 0 || remainder
== 0)
96 * Note: "remainder" can never be negative, since we check for
97 * this in the XDR code.
99 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
100 base
&= ~PAGE_CACHE_MASK
;
101 pglen
= PAGE_CACHE_SIZE
- base
;
103 if (remainder
<= pglen
) {
104 zero_user(*pages
, base
, remainder
);
107 zero_user(*pages
, base
, pglen
);
110 pglen
= PAGE_CACHE_SIZE
;
115 int nfs_readpage_async(struct nfs_open_context
*ctx
, struct inode
*inode
,
118 LIST_HEAD(one_request
);
119 struct nfs_page
*new;
122 len
= nfs_page_length(page
);
124 return nfs_return_empty_page(page
);
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 unlock_page(req
->wb_page
);
145 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
146 req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
147 (long long)NFS_FILEID(req
->wb_context
->path
.dentry
->d_inode
),
149 (long long)req_offset(req
));
150 nfs_clear_request(req
);
151 nfs_release_request(req
);
155 * Set up the NFS read request struct
157 static int nfs_read_rpcsetup(struct nfs_page
*req
, struct nfs_read_data
*data
,
158 const struct rpc_call_ops
*call_ops
,
159 unsigned int count
, unsigned int offset
)
161 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
162 int swap_flags
= IS_SWAPFILE(inode
) ? NFS_RPC_SWAPFLAGS
: 0;
163 struct rpc_task
*task
;
164 struct rpc_message msg
= {
165 .rpc_argp
= &data
->args
,
166 .rpc_resp
= &data
->res
,
167 .rpc_cred
= req
->wb_context
->cred
,
169 struct rpc_task_setup task_setup_data
= {
171 .rpc_client
= NFS_CLIENT(inode
),
173 .callback_ops
= call_ops
,
174 .callback_data
= data
,
175 .workqueue
= nfsiod_workqueue
,
176 .flags
= RPC_TASK_ASYNC
| swap_flags
,
181 data
->cred
= msg
.rpc_cred
;
183 data
->args
.fh
= NFS_FH(inode
);
184 data
->args
.offset
= req_offset(req
) + offset
;
185 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
186 data
->args
.pages
= data
->pagevec
;
187 data
->args
.count
= count
;
188 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
190 data
->res
.fattr
= &data
->fattr
;
191 data
->res
.count
= count
;
193 nfs_fattr_init(&data
->fattr
);
195 /* Set up the initial task struct. */
196 NFS_PROTO(inode
)->read_setup(data
, &msg
);
198 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
201 (long long)NFS_FILEID(inode
),
203 (unsigned long long)data
->args
.offset
);
205 task
= rpc_run_task(&task_setup_data
);
207 return PTR_ERR(task
);
213 nfs_async_read_error(struct list_head
*head
)
215 struct nfs_page
*req
;
217 while (!list_empty(head
)) {
218 req
= nfs_list_entry(head
->next
);
219 nfs_list_remove_request(req
);
220 SetPageError(req
->wb_page
);
221 nfs_readpage_release(req
);
226 * Generate multiple requests to fill a single page.
228 * We optimize to reduce the number of read operations on the wire. If we
229 * detect that we're reading a page, or an area of a page, that is past the
230 * end of file, we do not generate NFS read operations but just clear the
231 * parts of the page that would have come back zero from the server anyway.
233 * We rely on the cached value of i_size to make this determination; another
234 * client can fill pages on the server past our cached end-of-file, but we
235 * won't see the new data until our attribute cache is updated. This is more
236 * or less conventional NFS client behavior.
238 static int nfs_pagein_multi(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
240 struct nfs_page
*req
= nfs_list_entry(head
->next
);
241 struct page
*page
= req
->wb_page
;
242 struct nfs_read_data
*data
;
243 size_t rsize
= NFS_SERVER(inode
)->rsize
, nbytes
;
249 nfs_list_remove_request(req
);
253 size_t len
= min(nbytes
,rsize
);
255 data
= nfs_readdata_alloc(1);
258 list_add(&data
->pages
, &list
);
261 } while(nbytes
!= 0);
262 atomic_set(&req
->wb_complete
, requests
);
264 ClearPageError(page
);
270 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
271 list_del_init(&data
->pages
);
273 data
->pagevec
[0] = page
;
277 ret2
= nfs_read_rpcsetup(req
, data
, &nfs_read_partial_ops
,
283 } while (nbytes
!= 0);
288 while (!list_empty(&list
)) {
289 data
= list_entry(list
.next
, struct nfs_read_data
, pages
);
290 list_del(&data
->pages
);
291 nfs_readdata_free(data
);
294 nfs_readpage_release(req
);
298 static int nfs_pagein_one(struct inode
*inode
, struct list_head
*head
, unsigned int npages
, size_t count
, int flags
)
300 struct nfs_page
*req
;
302 struct nfs_read_data
*data
;
305 data
= nfs_readdata_alloc(npages
);
309 pages
= data
->pagevec
;
310 while (!list_empty(head
)) {
311 req
= nfs_list_entry(head
->next
);
312 nfs_list_remove_request(req
);
313 nfs_list_add_request(req
, &data
->pages
);
314 ClearPageError(req
->wb_page
);
315 *pages
++ = req
->wb_page
;
317 req
= nfs_list_entry(data
->pages
.next
);
319 return nfs_read_rpcsetup(req
, data
, &nfs_read_full_ops
, count
, 0);
321 nfs_async_read_error(head
);
326 * This is the callback from RPC telling us whether a reply was
327 * received or some error occurred (timeout or socket shutdown).
329 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
333 dprintk("NFS: %s: %5u, (status %d)\n", __func__
, task
->tk_pid
,
336 status
= NFS_PROTO(data
->inode
)->read_done(task
, data
);
340 nfs_add_stats(data
->inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
342 if (task
->tk_status
== -ESTALE
) {
343 set_bit(NFS_INO_STALE
, &NFS_I(data
->inode
)->flags
);
344 nfs_mark_for_revalidate(data
->inode
);
349 static void nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
351 struct nfs_readargs
*argp
= &data
->args
;
352 struct nfs_readres
*resp
= &data
->res
;
354 if (resp
->eof
|| resp
->count
== argp
->count
)
357 /* This is a short read! */
358 nfs_inc_stats(data
->inode
, NFSIOS_SHORTREAD
);
359 /* Has the server at least made some progress? */
360 if (resp
->count
== 0)
363 /* Yes, so retry the read at the end of the data */
364 argp
->offset
+= resp
->count
;
365 argp
->pgbase
+= resp
->count
;
366 argp
->count
-= resp
->count
;
367 rpc_restart_call(task
);
371 * Handle a read reply that fills part of a page.
373 static void nfs_readpage_result_partial(struct rpc_task
*task
, void *calldata
)
375 struct nfs_read_data
*data
= calldata
;
377 if (nfs_readpage_result(task
, data
) != 0)
379 if (task
->tk_status
< 0)
382 nfs_readpage_truncate_uninitialised_page(data
);
383 nfs_readpage_retry(task
, data
);
386 static void nfs_readpage_release_partial(void *calldata
)
388 struct nfs_read_data
*data
= calldata
;
389 struct nfs_page
*req
= data
->req
;
390 struct page
*page
= req
->wb_page
;
391 int status
= data
->task
.tk_status
;
396 if (atomic_dec_and_test(&req
->wb_complete
)) {
397 if (!PageError(page
))
398 SetPageUptodate(page
);
399 nfs_readpage_release(req
);
401 nfs_readdata_release(calldata
);
404 static const struct rpc_call_ops nfs_read_partial_ops
= {
405 .rpc_call_done
= nfs_readpage_result_partial
,
406 .rpc_release
= nfs_readpage_release_partial
,
409 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data
*data
)
411 unsigned int count
= data
->res
.count
;
412 unsigned int base
= data
->args
.pgbase
;
416 count
= data
->args
.count
;
417 if (unlikely(count
== 0))
419 pages
= &data
->args
.pages
[base
>> PAGE_CACHE_SHIFT
];
420 base
&= ~PAGE_CACHE_MASK
;
422 for (;count
>= PAGE_CACHE_SIZE
; count
-= PAGE_CACHE_SIZE
, pages
++)
423 SetPageUptodate(*pages
);
426 /* Was this a short read? */
427 if (data
->res
.eof
|| data
->res
.count
== data
->args
.count
)
428 SetPageUptodate(*pages
);
432 * This is the callback from RPC telling us whether a reply was
433 * received or some error occurred (timeout or socket shutdown).
435 static void nfs_readpage_result_full(struct rpc_task
*task
, void *calldata
)
437 struct nfs_read_data
*data
= calldata
;
439 if (nfs_readpage_result(task
, data
) != 0)
441 if (task
->tk_status
< 0)
444 * Note: nfs_readpage_retry may change the values of
445 * data->args. In the multi-page case, we therefore need
446 * to ensure that we call nfs_readpage_set_pages_uptodate()
449 nfs_readpage_truncate_uninitialised_page(data
);
450 nfs_readpage_set_pages_uptodate(data
);
451 nfs_readpage_retry(task
, data
);
454 static void nfs_readpage_release_full(void *calldata
)
456 struct nfs_read_data
*data
= calldata
;
458 while (!list_empty(&data
->pages
)) {
459 struct nfs_page
*req
= nfs_list_entry(data
->pages
.next
);
461 nfs_list_remove_request(req
);
462 nfs_readpage_release(req
);
464 nfs_readdata_release(calldata
);
467 static const struct rpc_call_ops nfs_read_full_ops
= {
468 .rpc_call_done
= nfs_readpage_result_full
,
469 .rpc_release
= nfs_readpage_release_full
,
473 * Read a page over NFS.
474 * We read the page synchronously in the following case:
475 * - The error flag is set for this page. This happens only when a
476 * previous async read operation failed.
478 int nfs_readpage(struct file
*file
, struct page
*page
)
480 struct nfs_open_context
*ctx
;
481 struct inode
*inode
= page
->mapping
->host
;
484 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
485 page
, PAGE_CACHE_SIZE
, page
->index
);
486 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
487 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
490 * Try to flush any pending writes to the file..
492 * NOTE! Because we own the page lock, there cannot
493 * be any new pending writes generated at this point
494 * for this page (other pages can be written to).
496 error
= nfs_wb_page(inode
, page
);
499 if (PageUptodate(page
))
503 if (NFS_STALE(inode
))
508 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
512 ctx
= get_nfs_open_context(nfs_file_open_context(file
));
514 if (!IS_SYNC(inode
)) {
515 error
= nfs_readpage_from_fscache(ctx
, inode
, page
);
520 error
= nfs_readpage_async(ctx
, inode
, page
);
523 put_nfs_open_context(ctx
);
530 struct nfs_readdesc
{
531 struct nfs_pageio_descriptor
*pgio
;
532 struct nfs_open_context
*ctx
;
536 readpage_async_filler(void *data
, struct page
*page
)
538 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
539 struct inode
*inode
= page
->mapping
->host
;
540 struct nfs_page
*new;
544 len
= nfs_page_length(page
);
546 return nfs_return_empty_page(page
);
548 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
552 if (len
< PAGE_CACHE_SIZE
)
553 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
554 if (!nfs_pageio_add_request(desc
->pgio
, new)) {
555 error
= desc
->pgio
->pg_error
;
560 error
= PTR_ERR(new);
567 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
568 struct list_head
*pages
, unsigned nr_pages
)
570 struct nfs_pageio_descriptor pgio
;
571 struct nfs_readdesc desc
= {
574 struct inode
*inode
= mapping
->host
;
575 struct nfs_server
*server
= NFS_SERVER(inode
);
576 size_t rsize
= server
->rsize
;
577 unsigned long npages
;
580 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
582 (long long)NFS_FILEID(inode
),
584 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
586 if (NFS_STALE(inode
))
590 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
591 if (desc
.ctx
== NULL
)
594 desc
.ctx
= get_nfs_open_context(nfs_file_open_context(filp
));
596 /* attempt to read as many of the pages as possible from the cache
597 * - this returns -ENOBUFS immediately if the cookie is negative
599 ret
= nfs_readpages_from_fscache(desc
.ctx
, inode
, mapping
,
602 goto read_complete
; /* all pages were read */
604 if (rsize
< PAGE_CACHE_SIZE
)
605 nfs_pageio_init(&pgio
, inode
, nfs_pagein_multi
, rsize
, 0);
607 nfs_pageio_init(&pgio
, inode
, nfs_pagein_one
, rsize
, 0);
609 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
611 nfs_pageio_complete(&pgio
);
612 npages
= (pgio
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
613 nfs_add_stats(inode
, NFSIOS_READPAGES
, npages
);
615 put_nfs_open_context(desc
.ctx
);
620 int __init
nfs_init_readpagecache(void)
622 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
623 sizeof(struct nfs_read_data
),
624 0, SLAB_HWCACHE_ALIGN
,
626 if (nfs_rdata_cachep
== NULL
)
629 nfs_rdata_mempool
= mempool_create_slab_pool(MIN_POOL_READ
,
631 if (nfs_rdata_mempool
== NULL
)
637 void nfs_destroy_readpagecache(void)
639 mempool_destroy(nfs_rdata_mempool
);
640 kmem_cache_destroy(nfs_rdata_cachep
);