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/module.h>
28 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
30 static const struct nfs_pageio_ops nfs_pageio_read_ops
;
31 static const struct rpc_call_ops nfs_read_common_ops
;
32 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops
;
34 static struct kmem_cache
*nfs_rdata_cachep
;
36 struct nfs_read_header
*nfs_readhdr_alloc(void)
38 struct nfs_read_header
*rhdr
;
40 rhdr
= kmem_cache_zalloc(nfs_rdata_cachep
, GFP_KERNEL
);
42 struct nfs_pgio_header
*hdr
= &rhdr
->header
;
44 INIT_LIST_HEAD(&hdr
->pages
);
45 INIT_LIST_HEAD(&hdr
->rpc_list
);
46 spin_lock_init(&hdr
->lock
);
47 atomic_set(&hdr
->refcnt
, 0);
51 EXPORT_SYMBOL_GPL(nfs_readhdr_alloc
);
53 static struct nfs_read_data
*nfs_readdata_alloc(struct nfs_pgio_header
*hdr
,
54 unsigned int pagecount
)
56 struct nfs_read_data
*data
, *prealloc
;
58 prealloc
= &container_of(hdr
, struct nfs_read_header
, header
)->rpc_data
;
59 if (prealloc
->header
== NULL
)
62 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
66 if (nfs_pgarray_set(&data
->pages
, pagecount
)) {
68 atomic_inc(&hdr
->refcnt
);
78 void nfs_readhdr_free(struct nfs_pgio_header
*hdr
)
80 struct nfs_read_header
*rhdr
= container_of(hdr
, struct nfs_read_header
, header
);
82 kmem_cache_free(nfs_rdata_cachep
, rhdr
);
84 EXPORT_SYMBOL_GPL(nfs_readhdr_free
);
86 void nfs_readdata_release(struct nfs_read_data
*rdata
)
88 struct nfs_pgio_header
*hdr
= rdata
->header
;
89 struct nfs_read_header
*read_header
= container_of(hdr
, struct nfs_read_header
, header
);
91 put_nfs_open_context(rdata
->args
.context
);
92 if (rdata
->pages
.pagevec
!= rdata
->pages
.page_array
)
93 kfree(rdata
->pages
.pagevec
);
94 if (rdata
!= &read_header
->rpc_data
)
98 if (atomic_dec_and_test(&hdr
->refcnt
))
99 hdr
->completion_ops
->completion(hdr
);
101 EXPORT_SYMBOL_GPL(nfs_readdata_release
);
104 int nfs_return_empty_page(struct page
*page
)
106 zero_user(page
, 0, PAGE_CACHE_SIZE
);
107 SetPageUptodate(page
);
112 void nfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
,
114 const struct nfs_pgio_completion_ops
*compl_ops
)
116 nfs_pageio_init(pgio
, inode
, &nfs_pageio_read_ops
, compl_ops
,
117 NFS_SERVER(inode
)->rsize
, 0);
119 EXPORT_SYMBOL_GPL(nfs_pageio_init_read
);
121 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor
*pgio
)
123 pgio
->pg_ops
= &nfs_pageio_read_ops
;
124 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->rsize
;
126 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds
);
128 int nfs_readpage_async(struct nfs_open_context
*ctx
, struct inode
*inode
,
131 struct nfs_page
*new;
133 struct nfs_pageio_descriptor pgio
;
135 len
= nfs_page_length(page
);
137 return nfs_return_empty_page(page
);
138 new = nfs_create_request(ctx
, inode
, page
, 0, len
);
143 if (len
< PAGE_CACHE_SIZE
)
144 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
146 NFS_PROTO(inode
)->read_pageio_init(&pgio
, inode
, &nfs_async_read_completion_ops
);
147 nfs_pageio_add_request(&pgio
, new);
148 nfs_pageio_complete(&pgio
);
149 NFS_I(inode
)->read_io
+= pgio
.pg_bytes_written
;
153 static void nfs_readpage_release(struct nfs_page
*req
)
155 struct inode
*d_inode
= req
->wb_context
->dentry
->d_inode
;
157 if (PageUptodate(req
->wb_page
))
158 nfs_readpage_to_fscache(d_inode
, req
->wb_page
, 0);
160 unlock_page(req
->wb_page
);
162 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
163 req
->wb_context
->dentry
->d_inode
->i_sb
->s_id
,
164 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
166 (long long)req_offset(req
));
167 nfs_release_request(req
);
170 /* Note io was page aligned */
171 static void nfs_read_completion(struct nfs_pgio_header
*hdr
)
173 unsigned long bytes
= 0;
175 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
177 while (!list_empty(&hdr
->pages
)) {
178 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
179 struct page
*page
= req
->wb_page
;
181 if (test_bit(NFS_IOHDR_EOF
, &hdr
->flags
)) {
182 if (bytes
> hdr
->good_bytes
)
183 zero_user(page
, 0, PAGE_SIZE
);
184 else if (hdr
->good_bytes
- bytes
< PAGE_SIZE
)
185 zero_user_segment(page
,
186 hdr
->good_bytes
& ~PAGE_MASK
,
189 bytes
+= req
->wb_bytes
;
190 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
)) {
191 if (bytes
<= hdr
->good_bytes
)
192 SetPageUptodate(page
);
194 SetPageUptodate(page
);
195 nfs_list_remove_request(req
);
196 nfs_readpage_release(req
);
202 int nfs_initiate_read(struct rpc_clnt
*clnt
,
203 struct nfs_read_data
*data
,
204 const struct rpc_call_ops
*call_ops
, int flags
)
206 struct inode
*inode
= data
->header
->inode
;
207 int swap_flags
= IS_SWAPFILE(inode
) ? NFS_RPC_SWAPFLAGS
: 0;
208 struct rpc_task
*task
;
209 struct rpc_message msg
= {
210 .rpc_argp
= &data
->args
,
211 .rpc_resp
= &data
->res
,
212 .rpc_cred
= data
->header
->cred
,
214 struct rpc_task_setup task_setup_data
= {
218 .callback_ops
= call_ops
,
219 .callback_data
= data
,
220 .workqueue
= nfsiod_workqueue
,
221 .flags
= RPC_TASK_ASYNC
| swap_flags
| flags
,
224 /* Set up the initial task struct. */
225 NFS_PROTO(inode
)->read_setup(data
, &msg
);
227 dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
231 (long long)NFS_FILEID(inode
),
233 (unsigned long long)data
->args
.offset
);
235 task
= rpc_run_task(&task_setup_data
);
237 return PTR_ERR(task
);
241 EXPORT_SYMBOL_GPL(nfs_initiate_read
);
244 * Set up the NFS read request struct
246 static void nfs_read_rpcsetup(struct nfs_read_data
*data
,
247 unsigned int count
, unsigned int offset
)
249 struct nfs_page
*req
= data
->header
->req
;
251 data
->args
.fh
= NFS_FH(data
->header
->inode
);
252 data
->args
.offset
= req_offset(req
) + offset
;
253 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
254 data
->args
.pages
= data
->pages
.pagevec
;
255 data
->args
.count
= count
;
256 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
257 data
->args
.lock_context
= req
->wb_lock_context
;
259 data
->res
.fattr
= &data
->fattr
;
260 data
->res
.count
= count
;
262 nfs_fattr_init(&data
->fattr
);
265 static int nfs_do_read(struct nfs_read_data
*data
,
266 const struct rpc_call_ops
*call_ops
)
268 struct inode
*inode
= data
->header
->inode
;
270 return nfs_initiate_read(NFS_CLIENT(inode
), data
, call_ops
, 0);
274 nfs_do_multiple_reads(struct list_head
*head
,
275 const struct rpc_call_ops
*call_ops
)
277 struct nfs_read_data
*data
;
280 while (!list_empty(head
)) {
283 data
= list_first_entry(head
, struct nfs_read_data
, list
);
284 list_del_init(&data
->list
);
286 ret2
= nfs_do_read(data
, call_ops
);
294 nfs_async_read_error(struct list_head
*head
)
296 struct nfs_page
*req
;
298 while (!list_empty(head
)) {
299 req
= nfs_list_entry(head
->next
);
300 nfs_list_remove_request(req
);
301 nfs_readpage_release(req
);
305 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops
= {
306 .error_cleanup
= nfs_async_read_error
,
307 .completion
= nfs_read_completion
,
310 static void nfs_pagein_error(struct nfs_pageio_descriptor
*desc
,
311 struct nfs_pgio_header
*hdr
)
313 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
314 while (!list_empty(&hdr
->rpc_list
)) {
315 struct nfs_read_data
*data
= list_first_entry(&hdr
->rpc_list
,
316 struct nfs_read_data
, list
);
317 list_del(&data
->list
);
318 nfs_readdata_release(data
);
320 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
324 * Generate multiple requests to fill a single page.
326 * We optimize to reduce the number of read operations on the wire. If we
327 * detect that we're reading a page, or an area of a page, that is past the
328 * end of file, we do not generate NFS read operations but just clear the
329 * parts of the page that would have come back zero from the server anyway.
331 * We rely on the cached value of i_size to make this determination; another
332 * client can fill pages on the server past our cached end-of-file, but we
333 * won't see the new data until our attribute cache is updated. This is more
334 * or less conventional NFS client behavior.
336 static int nfs_pagein_multi(struct nfs_pageio_descriptor
*desc
,
337 struct nfs_pgio_header
*hdr
)
339 struct nfs_page
*req
= hdr
->req
;
340 struct page
*page
= req
->wb_page
;
341 struct nfs_read_data
*data
;
342 size_t rsize
= desc
->pg_bsize
, nbytes
;
346 nbytes
= desc
->pg_count
;
348 size_t len
= min(nbytes
,rsize
);
350 data
= nfs_readdata_alloc(hdr
, 1);
352 nfs_pagein_error(desc
, hdr
);
355 data
->pages
.pagevec
[0] = page
;
356 nfs_read_rpcsetup(data
, len
, offset
);
357 list_add(&data
->list
, &hdr
->rpc_list
);
360 } while (nbytes
!= 0);
362 nfs_list_remove_request(req
);
363 nfs_list_add_request(req
, &hdr
->pages
);
364 desc
->pg_rpc_callops
= &nfs_read_common_ops
;
368 static int nfs_pagein_one(struct nfs_pageio_descriptor
*desc
,
369 struct nfs_pgio_header
*hdr
)
371 struct nfs_page
*req
;
373 struct nfs_read_data
*data
;
374 struct list_head
*head
= &desc
->pg_list
;
376 data
= nfs_readdata_alloc(hdr
, nfs_page_array_len(desc
->pg_base
,
379 nfs_pagein_error(desc
, hdr
);
383 pages
= data
->pages
.pagevec
;
384 while (!list_empty(head
)) {
385 req
= nfs_list_entry(head
->next
);
386 nfs_list_remove_request(req
);
387 nfs_list_add_request(req
, &hdr
->pages
);
388 *pages
++ = req
->wb_page
;
391 nfs_read_rpcsetup(data
, desc
->pg_count
, 0);
392 list_add(&data
->list
, &hdr
->rpc_list
);
393 desc
->pg_rpc_callops
= &nfs_read_common_ops
;
397 int nfs_generic_pagein(struct nfs_pageio_descriptor
*desc
,
398 struct nfs_pgio_header
*hdr
)
400 if (desc
->pg_bsize
< PAGE_CACHE_SIZE
)
401 return nfs_pagein_multi(desc
, hdr
);
402 return nfs_pagein_one(desc
, hdr
);
404 EXPORT_SYMBOL_GPL(nfs_generic_pagein
);
406 static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
408 struct nfs_read_header
*rhdr
;
409 struct nfs_pgio_header
*hdr
;
412 rhdr
= nfs_readhdr_alloc();
414 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
418 nfs_pgheader_init(desc
, hdr
, nfs_readhdr_free
);
419 atomic_inc(&hdr
->refcnt
);
420 ret
= nfs_generic_pagein(desc
, hdr
);
422 ret
= nfs_do_multiple_reads(&hdr
->rpc_list
,
423 desc
->pg_rpc_callops
);
424 if (atomic_dec_and_test(&hdr
->refcnt
))
425 hdr
->completion_ops
->completion(hdr
);
429 static const struct nfs_pageio_ops nfs_pageio_read_ops
= {
430 .pg_test
= nfs_generic_pg_test
,
431 .pg_doio
= nfs_generic_pg_readpages
,
435 * This is the callback from RPC telling us whether a reply was
436 * received or some error occurred (timeout or socket shutdown).
438 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
440 struct inode
*inode
= data
->header
->inode
;
443 dprintk("NFS: %s: %5u, (status %d)\n", __func__
, task
->tk_pid
,
446 status
= NFS_PROTO(inode
)->read_done(task
, data
);
450 nfs_add_stats(inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
452 if (task
->tk_status
== -ESTALE
) {
453 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
454 nfs_mark_for_revalidate(inode
);
459 static void nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
461 struct nfs_readargs
*argp
= &data
->args
;
462 struct nfs_readres
*resp
= &data
->res
;
464 /* This is a short read! */
465 nfs_inc_stats(data
->header
->inode
, NFSIOS_SHORTREAD
);
466 /* Has the server at least made some progress? */
467 if (resp
->count
== 0) {
468 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
471 /* Yes, so retry the read at the end of the data */
472 data
->mds_offset
+= resp
->count
;
473 argp
->offset
+= resp
->count
;
474 argp
->pgbase
+= resp
->count
;
475 argp
->count
-= resp
->count
;
476 rpc_restart_call_prepare(task
);
479 static void nfs_readpage_result_common(struct rpc_task
*task
, void *calldata
)
481 struct nfs_read_data
*data
= calldata
;
482 struct nfs_pgio_header
*hdr
= data
->header
;
484 /* Note the only returns of nfs_readpage_result are 0 and -EAGAIN */
485 if (nfs_readpage_result(task
, data
) != 0)
487 if (task
->tk_status
< 0)
488 nfs_set_pgio_error(hdr
, task
->tk_status
, data
->args
.offset
);
489 else if (data
->res
.eof
) {
492 bound
= data
->args
.offset
+ data
->res
.count
;
493 spin_lock(&hdr
->lock
);
494 if (bound
< hdr
->io_start
+ hdr
->good_bytes
) {
495 set_bit(NFS_IOHDR_EOF
, &hdr
->flags
);
496 clear_bit(NFS_IOHDR_ERROR
, &hdr
->flags
);
497 hdr
->good_bytes
= bound
- hdr
->io_start
;
499 spin_unlock(&hdr
->lock
);
500 } else if (data
->res
.count
!= data
->args
.count
)
501 nfs_readpage_retry(task
, data
);
504 static void nfs_readpage_release_common(void *calldata
)
506 nfs_readdata_release(calldata
);
509 void nfs_read_prepare(struct rpc_task
*task
, void *calldata
)
511 struct nfs_read_data
*data
= calldata
;
512 NFS_PROTO(data
->header
->inode
)->read_rpc_prepare(task
, data
);
515 static const struct rpc_call_ops nfs_read_common_ops
= {
516 .rpc_call_prepare
= nfs_read_prepare
,
517 .rpc_call_done
= nfs_readpage_result_common
,
518 .rpc_release
= nfs_readpage_release_common
,
522 * Read a page over NFS.
523 * We read the page synchronously in the following case:
524 * - The error flag is set for this page. This happens only when a
525 * previous async read operation failed.
527 int nfs_readpage(struct file
*file
, struct page
*page
)
529 struct nfs_open_context
*ctx
;
530 struct inode
*inode
= page_file_mapping(page
)->host
;
533 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
534 page
, PAGE_CACHE_SIZE
, page_file_index(page
));
535 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
536 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
539 * Try to flush any pending writes to the file..
541 * NOTE! Because we own the page lock, there cannot
542 * be any new pending writes generated at this point
543 * for this page (other pages can be written to).
545 error
= nfs_wb_page(inode
, page
);
548 if (PageUptodate(page
))
552 if (NFS_STALE(inode
))
557 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
561 ctx
= get_nfs_open_context(nfs_file_open_context(file
));
563 if (!IS_SYNC(inode
)) {
564 error
= nfs_readpage_from_fscache(ctx
, inode
, page
);
569 error
= nfs_readpage_async(ctx
, inode
, page
);
572 put_nfs_open_context(ctx
);
579 struct nfs_readdesc
{
580 struct nfs_pageio_descriptor
*pgio
;
581 struct nfs_open_context
*ctx
;
585 readpage_async_filler(void *data
, struct page
*page
)
587 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
588 struct inode
*inode
= page_file_mapping(page
)->host
;
589 struct nfs_page
*new;
593 len
= nfs_page_length(page
);
595 return nfs_return_empty_page(page
);
597 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
601 if (len
< PAGE_CACHE_SIZE
)
602 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
603 if (!nfs_pageio_add_request(desc
->pgio
, new)) {
604 error
= desc
->pgio
->pg_error
;
609 error
= PTR_ERR(new);
615 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
616 struct list_head
*pages
, unsigned nr_pages
)
618 struct nfs_pageio_descriptor pgio
;
619 struct nfs_readdesc desc
= {
622 struct inode
*inode
= mapping
->host
;
623 unsigned long npages
;
626 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
628 (long long)NFS_FILEID(inode
),
630 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
632 if (NFS_STALE(inode
))
636 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
637 if (desc
.ctx
== NULL
)
640 desc
.ctx
= get_nfs_open_context(nfs_file_open_context(filp
));
642 /* attempt to read as many of the pages as possible from the cache
643 * - this returns -ENOBUFS immediately if the cookie is negative
645 ret
= nfs_readpages_from_fscache(desc
.ctx
, inode
, mapping
,
648 goto read_complete
; /* all pages were read */
650 NFS_PROTO(inode
)->read_pageio_init(&pgio
, inode
, &nfs_async_read_completion_ops
);
652 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
654 nfs_pageio_complete(&pgio
);
655 NFS_I(inode
)->read_io
+= pgio
.pg_bytes_written
;
656 npages
= (pgio
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
657 nfs_add_stats(inode
, NFSIOS_READPAGES
, npages
);
659 put_nfs_open_context(desc
.ctx
);
664 int __init
nfs_init_readpagecache(void)
666 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
667 sizeof(struct nfs_read_header
),
668 0, SLAB_HWCACHE_ALIGN
,
670 if (nfs_rdata_cachep
== NULL
)
676 void nfs_destroy_readpagecache(void)
678 kmem_cache_destroy(nfs_rdata_cachep
);