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);
52 static struct nfs_read_data
*nfs_readdata_alloc(struct nfs_pgio_header
*hdr
,
53 unsigned int pagecount
)
55 struct nfs_read_data
*data
, *prealloc
;
57 prealloc
= &container_of(hdr
, struct nfs_read_header
, header
)->rpc_data
;
58 if (prealloc
->header
== NULL
)
61 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
65 if (nfs_pgarray_set(&data
->pages
, pagecount
)) {
67 atomic_inc(&hdr
->refcnt
);
77 void nfs_readhdr_free(struct nfs_pgio_header
*hdr
)
79 struct nfs_read_header
*rhdr
= container_of(hdr
, struct nfs_read_header
, header
);
81 kmem_cache_free(nfs_rdata_cachep
, rhdr
);
84 void nfs_readdata_release(struct nfs_read_data
*rdata
)
86 struct nfs_pgio_header
*hdr
= rdata
->header
;
87 struct nfs_read_header
*read_header
= container_of(hdr
, struct nfs_read_header
, header
);
89 put_nfs_open_context(rdata
->args
.context
);
90 if (rdata
->pages
.pagevec
!= rdata
->pages
.page_array
)
91 kfree(rdata
->pages
.pagevec
);
92 if (rdata
!= &read_header
->rpc_data
)
96 if (atomic_dec_and_test(&hdr
->refcnt
))
97 hdr
->completion_ops
->completion(hdr
);
101 int nfs_return_empty_page(struct page
*page
)
103 zero_user(page
, 0, PAGE_CACHE_SIZE
);
104 SetPageUptodate(page
);
109 void nfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
,
111 const struct nfs_pgio_completion_ops
*compl_ops
)
113 nfs_pageio_init(pgio
, inode
, &nfs_pageio_read_ops
, compl_ops
,
114 NFS_SERVER(inode
)->rsize
, 0);
117 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor
*pgio
)
119 pgio
->pg_ops
= &nfs_pageio_read_ops
;
120 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->rsize
;
122 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds
);
124 int nfs_readpage_async(struct nfs_open_context
*ctx
, struct inode
*inode
,
127 struct nfs_page
*new;
129 struct nfs_pageio_descriptor pgio
;
131 len
= nfs_page_length(page
);
133 return nfs_return_empty_page(page
);
134 new = nfs_create_request(ctx
, inode
, page
, 0, len
);
139 if (len
< PAGE_CACHE_SIZE
)
140 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
142 NFS_PROTO(inode
)->read_pageio_init(&pgio
, inode
, &nfs_async_read_completion_ops
);
143 nfs_pageio_add_request(&pgio
, new);
144 nfs_pageio_complete(&pgio
);
145 NFS_I(inode
)->read_io
+= pgio
.pg_bytes_written
;
149 static void nfs_readpage_release(struct nfs_page
*req
)
151 struct inode
*d_inode
= req
->wb_context
->dentry
->d_inode
;
153 if (PageUptodate(req
->wb_page
))
154 nfs_readpage_to_fscache(d_inode
, req
->wb_page
, 0);
156 unlock_page(req
->wb_page
);
158 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
159 req
->wb_context
->dentry
->d_inode
->i_sb
->s_id
,
160 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
162 (long long)req_offset(req
));
163 nfs_release_request(req
);
166 /* Note io was page aligned */
167 static void nfs_read_completion(struct nfs_pgio_header
*hdr
)
169 unsigned long bytes
= 0;
171 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
173 while (!list_empty(&hdr
->pages
)) {
174 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
175 struct page
*page
= req
->wb_page
;
177 if (test_bit(NFS_IOHDR_EOF
, &hdr
->flags
)) {
178 if (bytes
> hdr
->good_bytes
)
179 zero_user(page
, 0, PAGE_SIZE
);
180 else if (hdr
->good_bytes
- bytes
< PAGE_SIZE
)
181 zero_user_segment(page
,
182 hdr
->good_bytes
& ~PAGE_MASK
,
185 bytes
+= req
->wb_bytes
;
186 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
)) {
187 if (bytes
<= hdr
->good_bytes
)
188 SetPageUptodate(page
);
190 SetPageUptodate(page
);
191 nfs_list_remove_request(req
);
192 nfs_readpage_release(req
);
198 int nfs_initiate_read(struct rpc_clnt
*clnt
,
199 struct nfs_read_data
*data
,
200 const struct rpc_call_ops
*call_ops
, int flags
)
202 struct inode
*inode
= data
->header
->inode
;
203 int swap_flags
= IS_SWAPFILE(inode
) ? NFS_RPC_SWAPFLAGS
: 0;
204 struct rpc_task
*task
;
205 struct rpc_message msg
= {
206 .rpc_argp
= &data
->args
,
207 .rpc_resp
= &data
->res
,
208 .rpc_cred
= data
->header
->cred
,
210 struct rpc_task_setup task_setup_data
= {
214 .callback_ops
= call_ops
,
215 .callback_data
= data
,
216 .workqueue
= nfsiod_workqueue
,
217 .flags
= RPC_TASK_ASYNC
| swap_flags
| flags
,
220 /* Set up the initial task struct. */
221 NFS_PROTO(inode
)->read_setup(data
, &msg
);
223 dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
227 (long long)NFS_FILEID(inode
),
229 (unsigned long long)data
->args
.offset
);
231 task
= rpc_run_task(&task_setup_data
);
233 return PTR_ERR(task
);
237 EXPORT_SYMBOL_GPL(nfs_initiate_read
);
240 * Set up the NFS read request struct
242 static void nfs_read_rpcsetup(struct nfs_read_data
*data
,
243 unsigned int count
, unsigned int offset
)
245 struct nfs_page
*req
= data
->header
->req
;
247 data
->args
.fh
= NFS_FH(data
->header
->inode
);
248 data
->args
.offset
= req_offset(req
) + offset
;
249 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
250 data
->args
.pages
= data
->pages
.pagevec
;
251 data
->args
.count
= count
;
252 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
253 data
->args
.lock_context
= req
->wb_lock_context
;
255 data
->res
.fattr
= &data
->fattr
;
256 data
->res
.count
= count
;
258 nfs_fattr_init(&data
->fattr
);
261 static int nfs_do_read(struct nfs_read_data
*data
,
262 const struct rpc_call_ops
*call_ops
)
264 struct inode
*inode
= data
->header
->inode
;
266 return nfs_initiate_read(NFS_CLIENT(inode
), data
, call_ops
, 0);
270 nfs_do_multiple_reads(struct list_head
*head
,
271 const struct rpc_call_ops
*call_ops
)
273 struct nfs_read_data
*data
;
276 while (!list_empty(head
)) {
279 data
= list_first_entry(head
, struct nfs_read_data
, list
);
280 list_del_init(&data
->list
);
282 ret2
= nfs_do_read(data
, call_ops
);
290 nfs_async_read_error(struct list_head
*head
)
292 struct nfs_page
*req
;
294 while (!list_empty(head
)) {
295 req
= nfs_list_entry(head
->next
);
296 nfs_list_remove_request(req
);
297 nfs_readpage_release(req
);
301 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops
= {
302 .error_cleanup
= nfs_async_read_error
,
303 .completion
= nfs_read_completion
,
306 static void nfs_pagein_error(struct nfs_pageio_descriptor
*desc
,
307 struct nfs_pgio_header
*hdr
)
309 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
310 while (!list_empty(&hdr
->rpc_list
)) {
311 struct nfs_read_data
*data
= list_first_entry(&hdr
->rpc_list
,
312 struct nfs_read_data
, list
);
313 list_del(&data
->list
);
314 nfs_readdata_release(data
);
316 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
320 * Generate multiple requests to fill a single page.
322 * We optimize to reduce the number of read operations on the wire. If we
323 * detect that we're reading a page, or an area of a page, that is past the
324 * end of file, we do not generate NFS read operations but just clear the
325 * parts of the page that would have come back zero from the server anyway.
327 * We rely on the cached value of i_size to make this determination; another
328 * client can fill pages on the server past our cached end-of-file, but we
329 * won't see the new data until our attribute cache is updated. This is more
330 * or less conventional NFS client behavior.
332 static int nfs_pagein_multi(struct nfs_pageio_descriptor
*desc
,
333 struct nfs_pgio_header
*hdr
)
335 struct nfs_page
*req
= hdr
->req
;
336 struct page
*page
= req
->wb_page
;
337 struct nfs_read_data
*data
;
338 size_t rsize
= desc
->pg_bsize
, nbytes
;
342 nbytes
= desc
->pg_count
;
344 size_t len
= min(nbytes
,rsize
);
346 data
= nfs_readdata_alloc(hdr
, 1);
348 nfs_pagein_error(desc
, hdr
);
351 data
->pages
.pagevec
[0] = page
;
352 nfs_read_rpcsetup(data
, len
, offset
);
353 list_add(&data
->list
, &hdr
->rpc_list
);
356 } while (nbytes
!= 0);
358 nfs_list_remove_request(req
);
359 nfs_list_add_request(req
, &hdr
->pages
);
360 desc
->pg_rpc_callops
= &nfs_read_common_ops
;
364 static int nfs_pagein_one(struct nfs_pageio_descriptor
*desc
,
365 struct nfs_pgio_header
*hdr
)
367 struct nfs_page
*req
;
369 struct nfs_read_data
*data
;
370 struct list_head
*head
= &desc
->pg_list
;
372 data
= nfs_readdata_alloc(hdr
, nfs_page_array_len(desc
->pg_base
,
375 nfs_pagein_error(desc
, hdr
);
379 pages
= data
->pages
.pagevec
;
380 while (!list_empty(head
)) {
381 req
= nfs_list_entry(head
->next
);
382 nfs_list_remove_request(req
);
383 nfs_list_add_request(req
, &hdr
->pages
);
384 *pages
++ = req
->wb_page
;
387 nfs_read_rpcsetup(data
, desc
->pg_count
, 0);
388 list_add(&data
->list
, &hdr
->rpc_list
);
389 desc
->pg_rpc_callops
= &nfs_read_common_ops
;
393 int nfs_generic_pagein(struct nfs_pageio_descriptor
*desc
,
394 struct nfs_pgio_header
*hdr
)
396 if (desc
->pg_bsize
< PAGE_CACHE_SIZE
)
397 return nfs_pagein_multi(desc
, hdr
);
398 return nfs_pagein_one(desc
, hdr
);
401 static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
403 struct nfs_read_header
*rhdr
;
404 struct nfs_pgio_header
*hdr
;
407 rhdr
= nfs_readhdr_alloc();
409 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
413 nfs_pgheader_init(desc
, hdr
, nfs_readhdr_free
);
414 atomic_inc(&hdr
->refcnt
);
415 ret
= nfs_generic_pagein(desc
, hdr
);
417 ret
= nfs_do_multiple_reads(&hdr
->rpc_list
,
418 desc
->pg_rpc_callops
);
419 if (atomic_dec_and_test(&hdr
->refcnt
))
420 hdr
->completion_ops
->completion(hdr
);
424 static const struct nfs_pageio_ops nfs_pageio_read_ops
= {
425 .pg_test
= nfs_generic_pg_test
,
426 .pg_doio
= nfs_generic_pg_readpages
,
430 * This is the callback from RPC telling us whether a reply was
431 * received or some error occurred (timeout or socket shutdown).
433 int nfs_readpage_result(struct rpc_task
*task
, struct nfs_read_data
*data
)
435 struct inode
*inode
= data
->header
->inode
;
438 dprintk("NFS: %s: %5u, (status %d)\n", __func__
, task
->tk_pid
,
441 status
= NFS_PROTO(inode
)->read_done(task
, data
);
445 nfs_add_stats(inode
, NFSIOS_SERVERREADBYTES
, data
->res
.count
);
447 if (task
->tk_status
== -ESTALE
) {
448 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
449 nfs_mark_for_revalidate(inode
);
454 static void nfs_readpage_retry(struct rpc_task
*task
, struct nfs_read_data
*data
)
456 struct nfs_readargs
*argp
= &data
->args
;
457 struct nfs_readres
*resp
= &data
->res
;
459 /* This is a short read! */
460 nfs_inc_stats(data
->header
->inode
, NFSIOS_SHORTREAD
);
461 /* Has the server at least made some progress? */
462 if (resp
->count
== 0) {
463 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
466 /* Yes, so retry the read at the end of the data */
467 data
->mds_offset
+= resp
->count
;
468 argp
->offset
+= resp
->count
;
469 argp
->pgbase
+= resp
->count
;
470 argp
->count
-= resp
->count
;
471 rpc_restart_call_prepare(task
);
474 static void nfs_readpage_result_common(struct rpc_task
*task
, void *calldata
)
476 struct nfs_read_data
*data
= calldata
;
477 struct nfs_pgio_header
*hdr
= data
->header
;
479 /* Note the only returns of nfs_readpage_result are 0 and -EAGAIN */
480 if (nfs_readpage_result(task
, data
) != 0)
482 if (task
->tk_status
< 0)
483 nfs_set_pgio_error(hdr
, task
->tk_status
, data
->args
.offset
);
484 else if (data
->res
.eof
) {
487 bound
= data
->args
.offset
+ data
->res
.count
;
488 spin_lock(&hdr
->lock
);
489 if (bound
< hdr
->io_start
+ hdr
->good_bytes
) {
490 set_bit(NFS_IOHDR_EOF
, &hdr
->flags
);
491 clear_bit(NFS_IOHDR_ERROR
, &hdr
->flags
);
492 hdr
->good_bytes
= bound
- hdr
->io_start
;
494 spin_unlock(&hdr
->lock
);
495 } else if (data
->res
.count
!= data
->args
.count
)
496 nfs_readpage_retry(task
, data
);
499 static void nfs_readpage_release_common(void *calldata
)
501 nfs_readdata_release(calldata
);
504 void nfs_read_prepare(struct rpc_task
*task
, void *calldata
)
506 struct nfs_read_data
*data
= calldata
;
507 NFS_PROTO(data
->header
->inode
)->read_rpc_prepare(task
, data
);
510 static const struct rpc_call_ops nfs_read_common_ops
= {
511 .rpc_call_prepare
= nfs_read_prepare
,
512 .rpc_call_done
= nfs_readpage_result_common
,
513 .rpc_release
= nfs_readpage_release_common
,
517 * Read a page over NFS.
518 * We read the page synchronously in the following case:
519 * - The error flag is set for this page. This happens only when a
520 * previous async read operation failed.
522 int nfs_readpage(struct file
*file
, struct page
*page
)
524 struct nfs_open_context
*ctx
;
525 struct inode
*inode
= page
->mapping
->host
;
528 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
529 page
, PAGE_CACHE_SIZE
, page
->index
);
530 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGE
);
531 nfs_add_stats(inode
, NFSIOS_READPAGES
, 1);
534 * Try to flush any pending writes to the file..
536 * NOTE! Because we own the page lock, there cannot
537 * be any new pending writes generated at this point
538 * for this page (other pages can be written to).
540 error
= nfs_wb_page(inode
, page
);
543 if (PageUptodate(page
))
547 if (NFS_STALE(inode
))
552 ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
556 ctx
= get_nfs_open_context(nfs_file_open_context(file
));
558 if (!IS_SYNC(inode
)) {
559 error
= nfs_readpage_from_fscache(ctx
, inode
, page
);
564 error
= nfs_readpage_async(ctx
, inode
, page
);
567 put_nfs_open_context(ctx
);
574 struct nfs_readdesc
{
575 struct nfs_pageio_descriptor
*pgio
;
576 struct nfs_open_context
*ctx
;
580 readpage_async_filler(void *data
, struct page
*page
)
582 struct nfs_readdesc
*desc
= (struct nfs_readdesc
*)data
;
583 struct inode
*inode
= page
->mapping
->host
;
584 struct nfs_page
*new;
588 len
= nfs_page_length(page
);
590 return nfs_return_empty_page(page
);
592 new = nfs_create_request(desc
->ctx
, inode
, page
, 0, len
);
596 if (len
< PAGE_CACHE_SIZE
)
597 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
598 if (!nfs_pageio_add_request(desc
->pgio
, new)) {
599 error
= desc
->pgio
->pg_error
;
604 error
= PTR_ERR(new);
610 int nfs_readpages(struct file
*filp
, struct address_space
*mapping
,
611 struct list_head
*pages
, unsigned nr_pages
)
613 struct nfs_pageio_descriptor pgio
;
614 struct nfs_readdesc desc
= {
617 struct inode
*inode
= mapping
->host
;
618 unsigned long npages
;
621 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
623 (long long)NFS_FILEID(inode
),
625 nfs_inc_stats(inode
, NFSIOS_VFSREADPAGES
);
627 if (NFS_STALE(inode
))
631 desc
.ctx
= nfs_find_open_context(inode
, NULL
, FMODE_READ
);
632 if (desc
.ctx
== NULL
)
635 desc
.ctx
= get_nfs_open_context(nfs_file_open_context(filp
));
637 /* attempt to read as many of the pages as possible from the cache
638 * - this returns -ENOBUFS immediately if the cookie is negative
640 ret
= nfs_readpages_from_fscache(desc
.ctx
, inode
, mapping
,
643 goto read_complete
; /* all pages were read */
645 NFS_PROTO(inode
)->read_pageio_init(&pgio
, inode
, &nfs_async_read_completion_ops
);
647 ret
= read_cache_pages(mapping
, pages
, readpage_async_filler
, &desc
);
649 nfs_pageio_complete(&pgio
);
650 NFS_I(inode
)->read_io
+= pgio
.pg_bytes_written
;
651 npages
= (pgio
.pg_bytes_written
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
652 nfs_add_stats(inode
, NFSIOS_READPAGES
, npages
);
654 put_nfs_open_context(desc
.ctx
);
659 int __init
nfs_init_readpagecache(void)
661 nfs_rdata_cachep
= kmem_cache_create("nfs_read_data",
662 sizeof(struct nfs_read_header
),
663 0, SLAB_HWCACHE_ALIGN
,
665 if (nfs_rdata_cachep
== NULL
)
671 void nfs_destroy_readpagecache(void)
673 kmem_cache_destroy(nfs_rdata_cachep
);