4 * Write file data over NFS.
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
9 #include <linux/types.h>
10 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/file.h>
14 #include <linux/writeback.h>
15 #include <linux/swap.h>
16 #include <linux/migrate.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_mount.h>
21 #include <linux/nfs_page.h>
22 #include <linux/backing-dev.h>
24 #include <asm/uaccess.h>
26 #include "delegation.h"
33 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
35 #define MIN_POOL_WRITE (32)
36 #define MIN_POOL_COMMIT (4)
39 * Local function declarations
41 static void nfs_pageio_init_write(struct nfs_pageio_descriptor
*desc
,
42 struct inode
*inode
, int ioflags
);
43 static void nfs_redirty_request(struct nfs_page
*req
);
44 static const struct rpc_call_ops nfs_write_partial_ops
;
45 static const struct rpc_call_ops nfs_write_full_ops
;
46 static const struct rpc_call_ops nfs_commit_ops
;
48 static struct kmem_cache
*nfs_wdata_cachep
;
49 static mempool_t
*nfs_wdata_mempool
;
50 static mempool_t
*nfs_commit_mempool
;
52 struct nfs_write_data
*nfs_commitdata_alloc(void)
54 struct nfs_write_data
*p
= mempool_alloc(nfs_commit_mempool
, GFP_NOFS
);
57 memset(p
, 0, sizeof(*p
));
58 INIT_LIST_HEAD(&p
->pages
);
62 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc
);
64 void nfs_commit_free(struct nfs_write_data
*p
)
66 if (p
&& (p
->pagevec
!= &p
->page_array
[0]))
68 mempool_free(p
, nfs_commit_mempool
);
70 EXPORT_SYMBOL_GPL(nfs_commit_free
);
72 struct nfs_write_data
*nfs_writedata_alloc(unsigned int pagecount
)
74 struct nfs_write_data
*p
= mempool_alloc(nfs_wdata_mempool
, GFP_NOFS
);
77 memset(p
, 0, sizeof(*p
));
78 INIT_LIST_HEAD(&p
->pages
);
79 p
->npages
= pagecount
;
80 if (pagecount
<= ARRAY_SIZE(p
->page_array
))
81 p
->pagevec
= p
->page_array
;
83 p
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), GFP_NOFS
);
85 mempool_free(p
, nfs_wdata_mempool
);
93 void nfs_writedata_free(struct nfs_write_data
*p
)
95 if (p
&& (p
->pagevec
!= &p
->page_array
[0]))
97 mempool_free(p
, nfs_wdata_mempool
);
100 static void nfs_writedata_release(struct nfs_write_data
*wdata
)
102 put_lseg(wdata
->lseg
);
103 put_nfs_open_context(wdata
->args
.context
);
104 nfs_writedata_free(wdata
);
107 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
111 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
114 static struct nfs_page
*nfs_page_find_request_locked(struct page
*page
)
116 struct nfs_page
*req
= NULL
;
118 if (PagePrivate(page
)) {
119 req
= (struct nfs_page
*)page_private(page
);
121 kref_get(&req
->wb_kref
);
126 static struct nfs_page
*nfs_page_find_request(struct page
*page
)
128 struct inode
*inode
= page
->mapping
->host
;
129 struct nfs_page
*req
= NULL
;
131 spin_lock(&inode
->i_lock
);
132 req
= nfs_page_find_request_locked(page
);
133 spin_unlock(&inode
->i_lock
);
137 /* Adjust the file length if we're writing beyond the end */
138 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
140 struct inode
*inode
= page
->mapping
->host
;
144 spin_lock(&inode
->i_lock
);
145 i_size
= i_size_read(inode
);
146 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
147 if (i_size
> 0 && page
->index
< end_index
)
149 end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + ((loff_t
)offset
+count
);
152 i_size_write(inode
, end
);
153 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
155 spin_unlock(&inode
->i_lock
);
158 /* A writeback failed: mark the page as bad, and invalidate the page cache */
159 static void nfs_set_pageerror(struct page
*page
)
162 nfs_zap_mapping(page
->mapping
->host
, page
->mapping
);
165 /* We can set the PG_uptodate flag if we see that a write request
166 * covers the full page.
168 static void nfs_mark_uptodate(struct page
*page
, unsigned int base
, unsigned int count
)
170 if (PageUptodate(page
))
174 if (count
!= nfs_page_length(page
))
176 SetPageUptodate(page
);
179 static int wb_priority(struct writeback_control
*wbc
)
181 if (wbc
->for_reclaim
)
182 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
183 if (wbc
->for_kupdate
|| wbc
->for_background
)
184 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
185 return FLUSH_COND_STABLE
;
189 * NFS congestion control
192 int nfs_congestion_kb
;
194 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
195 #define NFS_CONGESTION_OFF_THRESH \
196 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
198 static int nfs_set_page_writeback(struct page
*page
)
200 int ret
= test_set_page_writeback(page
);
203 struct inode
*inode
= page
->mapping
->host
;
204 struct nfs_server
*nfss
= NFS_SERVER(inode
);
206 page_cache_get(page
);
207 if (atomic_long_inc_return(&nfss
->writeback
) >
208 NFS_CONGESTION_ON_THRESH
) {
209 set_bdi_congested(&nfss
->backing_dev_info
,
216 static void nfs_end_page_writeback(struct page
*page
)
218 struct inode
*inode
= page
->mapping
->host
;
219 struct nfs_server
*nfss
= NFS_SERVER(inode
);
221 end_page_writeback(page
);
222 page_cache_release(page
);
223 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
224 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
227 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
229 struct inode
*inode
= page
->mapping
->host
;
230 struct nfs_page
*req
;
233 spin_lock(&inode
->i_lock
);
235 req
= nfs_page_find_request_locked(page
);
238 if (nfs_set_page_tag_locked(req
))
240 /* Note: If we hold the page lock, as is the case in nfs_writepage,
241 * then the call to nfs_set_page_tag_locked() will always
242 * succeed provided that someone hasn't already marked the
243 * request as dirty (in which case we don't care).
245 spin_unlock(&inode
->i_lock
);
247 ret
= nfs_wait_on_request(req
);
250 nfs_release_request(req
);
253 spin_lock(&inode
->i_lock
);
255 spin_unlock(&inode
->i_lock
);
260 * Find an associated nfs write request, and prepare to flush it out
261 * May return an error if the user signalled nfs_wait_on_request().
263 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
264 struct page
*page
, bool nonblock
)
266 struct nfs_page
*req
;
269 req
= nfs_find_and_lock_request(page
, nonblock
);
276 ret
= nfs_set_page_writeback(page
);
278 BUG_ON(test_bit(PG_CLEAN
, &req
->wb_flags
));
280 if (!nfs_pageio_add_request(pgio
, req
)) {
281 nfs_redirty_request(req
);
282 ret
= pgio
->pg_error
;
288 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
290 struct inode
*inode
= page
->mapping
->host
;
293 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
294 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
296 nfs_pageio_cond_complete(pgio
, page
->index
);
297 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
298 if (ret
== -EAGAIN
) {
299 redirty_page_for_writepage(wbc
, page
);
306 * Write an mmapped page to the server.
308 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
310 struct nfs_pageio_descriptor pgio
;
313 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
));
314 err
= nfs_do_writepage(page
, wbc
, &pgio
);
315 nfs_pageio_complete(&pgio
);
318 if (pgio
.pg_error
< 0)
319 return pgio
.pg_error
;
323 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
327 ret
= nfs_writepage_locked(page
, wbc
);
332 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
336 ret
= nfs_do_writepage(page
, wbc
, data
);
341 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
343 struct inode
*inode
= mapping
->host
;
344 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
345 struct nfs_pageio_descriptor pgio
;
348 /* Stop dirtying of new pages while we sync */
349 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
350 nfs_wait_bit_killable
, TASK_KILLABLE
);
354 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
356 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
));
357 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
358 nfs_pageio_complete(&pgio
);
360 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
361 smp_mb__after_clear_bit();
362 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
375 * Insert a write request into an inode
377 static int nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
379 struct nfs_inode
*nfsi
= NFS_I(inode
);
382 error
= radix_tree_preload(GFP_NOFS
);
386 /* Lock the request! */
387 nfs_lock_request_dontget(req
);
389 spin_lock(&inode
->i_lock
);
390 error
= radix_tree_insert(&nfsi
->nfs_page_tree
, req
->wb_index
, req
);
392 if (!nfsi
->npages
&& nfs_have_delegation(inode
, FMODE_WRITE
))
394 set_bit(PG_MAPPED
, &req
->wb_flags
);
395 SetPagePrivate(req
->wb_page
);
396 set_page_private(req
->wb_page
, (unsigned long)req
);
398 kref_get(&req
->wb_kref
);
399 radix_tree_tag_set(&nfsi
->nfs_page_tree
, req
->wb_index
,
400 NFS_PAGE_TAG_LOCKED
);
401 spin_unlock(&inode
->i_lock
);
402 radix_tree_preload_end();
408 * Remove a write request from an inode
410 static void nfs_inode_remove_request(struct nfs_page
*req
)
412 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
413 struct nfs_inode
*nfsi
= NFS_I(inode
);
415 BUG_ON (!NFS_WBACK_BUSY(req
));
417 spin_lock(&inode
->i_lock
);
418 set_page_private(req
->wb_page
, 0);
419 ClearPagePrivate(req
->wb_page
);
420 clear_bit(PG_MAPPED
, &req
->wb_flags
);
421 radix_tree_delete(&nfsi
->nfs_page_tree
, req
->wb_index
);
423 spin_unlock(&inode
->i_lock
);
424 nfs_release_request(req
);
428 nfs_mark_request_dirty(struct nfs_page
*req
)
430 __set_page_dirty_nobuffers(req
->wb_page
);
433 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
435 * Add a request to the inode's commit list.
438 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
)
440 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
441 struct nfs_inode
*nfsi
= NFS_I(inode
);
443 spin_lock(&inode
->i_lock
);
444 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
445 radix_tree_tag_set(&nfsi
->nfs_page_tree
,
447 NFS_PAGE_TAG_COMMIT
);
449 spin_unlock(&inode
->i_lock
);
450 pnfs_mark_request_commit(req
, lseg
);
451 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
452 inc_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
453 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
457 nfs_clear_request_commit(struct nfs_page
*req
)
459 struct page
*page
= req
->wb_page
;
461 if (test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
)) {
462 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
463 dec_bdi_stat(page
->mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
470 int nfs_write_need_commit(struct nfs_write_data
*data
)
472 if (data
->verf
.committed
== NFS_DATA_SYNC
)
473 return data
->lseg
== NULL
;
475 return data
->verf
.committed
!= NFS_FILE_SYNC
;
479 int nfs_reschedule_unstable_write(struct nfs_page
*req
,
480 struct nfs_write_data
*data
)
482 if (test_and_clear_bit(PG_NEED_COMMIT
, &req
->wb_flags
)) {
483 nfs_mark_request_commit(req
, data
->lseg
);
486 if (test_and_clear_bit(PG_NEED_RESCHED
, &req
->wb_flags
)) {
487 nfs_mark_request_dirty(req
);
494 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
)
499 nfs_clear_request_commit(struct nfs_page
*req
)
505 int nfs_write_need_commit(struct nfs_write_data
*data
)
511 int nfs_reschedule_unstable_write(struct nfs_page
*req
,
512 struct nfs_write_data
*data
)
518 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
520 nfs_need_commit(struct nfs_inode
*nfsi
)
522 return radix_tree_tagged(&nfsi
->nfs_page_tree
, NFS_PAGE_TAG_COMMIT
);
526 * nfs_scan_commit - Scan an inode for commit requests
527 * @inode: NFS inode to scan
528 * @dst: destination list
529 * @idx_start: lower bound of page->index to scan.
530 * @npages: idx_start + npages sets the upper bound to scan.
532 * Moves requests from the inode's 'commit' request list.
533 * The requests are *not* checked to ensure that they form a contiguous set.
536 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
, pgoff_t idx_start
, unsigned int npages
)
538 struct nfs_inode
*nfsi
= NFS_I(inode
);
541 if (!nfs_need_commit(nfsi
))
544 spin_lock(&inode
->i_lock
);
545 ret
= nfs_scan_list(nfsi
, dst
, idx_start
, npages
, NFS_PAGE_TAG_COMMIT
);
547 nfsi
->ncommit
-= ret
;
548 spin_unlock(&inode
->i_lock
);
550 if (nfs_need_commit(NFS_I(inode
)))
551 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
556 static inline int nfs_need_commit(struct nfs_inode
*nfsi
)
561 static inline int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
, pgoff_t idx_start
, unsigned int npages
)
568 * Search for an existing write request, and attempt to update
569 * it to reflect a new dirty region on a given page.
571 * If the attempt fails, then the existing request is flushed out
574 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
579 struct nfs_page
*req
;
584 if (!PagePrivate(page
))
587 end
= offset
+ bytes
;
588 spin_lock(&inode
->i_lock
);
591 req
= nfs_page_find_request_locked(page
);
595 rqend
= req
->wb_offset
+ req
->wb_bytes
;
597 * Tell the caller to flush out the request if
598 * the offsets are non-contiguous.
599 * Note: nfs_flush_incompatible() will already
600 * have flushed out requests having wrong owners.
603 || end
< req
->wb_offset
)
606 if (nfs_set_page_tag_locked(req
))
609 /* The request is locked, so wait and then retry */
610 spin_unlock(&inode
->i_lock
);
611 error
= nfs_wait_on_request(req
);
612 nfs_release_request(req
);
615 spin_lock(&inode
->i_lock
);
618 if (nfs_clear_request_commit(req
) &&
619 radix_tree_tag_clear(&NFS_I(inode
)->nfs_page_tree
,
620 req
->wb_index
, NFS_PAGE_TAG_COMMIT
) != NULL
) {
621 NFS_I(inode
)->ncommit
--;
622 pnfs_clear_request_commit(req
);
625 /* Okay, the request matches. Update the region */
626 if (offset
< req
->wb_offset
) {
627 req
->wb_offset
= offset
;
628 req
->wb_pgbase
= offset
;
631 req
->wb_bytes
= end
- req
->wb_offset
;
633 req
->wb_bytes
= rqend
- req
->wb_offset
;
635 spin_unlock(&inode
->i_lock
);
638 spin_unlock(&inode
->i_lock
);
639 nfs_release_request(req
);
640 error
= nfs_wb_page(inode
, page
);
642 return ERR_PTR(error
);
646 * Try to update an existing write request, or create one if there is none.
648 * Note: Should always be called with the Page Lock held to prevent races
649 * if we have to add a new request. Also assumes that the caller has
650 * already called nfs_flush_incompatible() if necessary.
652 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
653 struct page
*page
, unsigned int offset
, unsigned int bytes
)
655 struct inode
*inode
= page
->mapping
->host
;
656 struct nfs_page
*req
;
659 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
662 req
= nfs_create_request(ctx
, inode
, page
, offset
, bytes
);
665 error
= nfs_inode_add_request(inode
, req
);
667 nfs_release_request(req
);
668 req
= ERR_PTR(error
);
674 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
675 unsigned int offset
, unsigned int count
)
677 struct nfs_page
*req
;
679 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
682 /* Update file length */
683 nfs_grow_file(page
, offset
, count
);
684 nfs_mark_uptodate(page
, req
->wb_pgbase
, req
->wb_bytes
);
685 nfs_mark_request_dirty(req
);
686 nfs_clear_page_tag_locked(req
);
690 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
692 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
693 struct nfs_page
*req
;
694 int do_flush
, status
;
696 * Look for a request corresponding to this page. If there
697 * is one, and it belongs to another file, we flush it out
698 * before we try to copy anything into the page. Do this
699 * due to the lack of an ACCESS-type call in NFSv2.
700 * Also do the same if we find a request from an existing
704 req
= nfs_page_find_request(page
);
707 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
||
708 req
->wb_lock_context
->lockowner
!= current
->files
||
709 req
->wb_lock_context
->pid
!= current
->tgid
;
710 nfs_release_request(req
);
713 status
= nfs_wb_page(page
->mapping
->host
, page
);
714 } while (status
== 0);
719 * If the page cache is marked as unsafe or invalid, then we can't rely on
720 * the PageUptodate() flag. In this case, we will need to turn off
721 * write optimisations that depend on the page contents being correct.
723 static int nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
725 return PageUptodate(page
) &&
726 !(NFS_I(inode
)->cache_validity
& (NFS_INO_REVAL_PAGECACHE
|NFS_INO_INVALID_DATA
));
730 * Update and possibly write a cached page of an NFS file.
732 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
733 * things with a page scheduled for an RPC call (e.g. invalidate it).
735 int nfs_updatepage(struct file
*file
, struct page
*page
,
736 unsigned int offset
, unsigned int count
)
738 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
739 struct inode
*inode
= page
->mapping
->host
;
742 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
744 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
745 file
->f_path
.dentry
->d_parent
->d_name
.name
,
746 file
->f_path
.dentry
->d_name
.name
, count
,
747 (long long)(page_offset(page
) + offset
));
749 /* If we're not using byte range locks, and we know the page
750 * is up to date, it may be more efficient to extend the write
751 * to cover the entire page in order to avoid fragmentation
754 if (nfs_write_pageuptodate(page
, inode
) &&
755 inode
->i_flock
== NULL
&&
756 !(file
->f_flags
& O_DSYNC
)) {
757 count
= max(count
+ offset
, nfs_page_length(page
));
761 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
763 nfs_set_pageerror(page
);
765 __set_page_dirty_nobuffers(page
);
767 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
768 status
, (long long)i_size_read(inode
));
772 static void nfs_writepage_release(struct nfs_page
*req
,
773 struct nfs_write_data
*data
)
775 struct page
*page
= req
->wb_page
;
777 if (PageError(req
->wb_page
) || !nfs_reschedule_unstable_write(req
, data
))
778 nfs_inode_remove_request(req
);
779 nfs_clear_page_tag_locked(req
);
780 nfs_end_page_writeback(page
);
783 static int flush_task_priority(int how
)
785 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
787 return RPC_PRIORITY_HIGH
;
789 return RPC_PRIORITY_LOW
;
791 return RPC_PRIORITY_NORMAL
;
794 int nfs_initiate_write(struct nfs_write_data
*data
,
795 struct rpc_clnt
*clnt
,
796 const struct rpc_call_ops
*call_ops
,
799 struct inode
*inode
= data
->inode
;
800 int priority
= flush_task_priority(how
);
801 struct rpc_task
*task
;
802 struct rpc_message msg
= {
803 .rpc_argp
= &data
->args
,
804 .rpc_resp
= &data
->res
,
805 .rpc_cred
= data
->cred
,
807 struct rpc_task_setup task_setup_data
= {
811 .callback_ops
= call_ops
,
812 .callback_data
= data
,
813 .workqueue
= nfsiod_workqueue
,
814 .flags
= RPC_TASK_ASYNC
,
815 .priority
= priority
,
819 /* Set up the initial task struct. */
820 NFS_PROTO(inode
)->write_setup(data
, &msg
);
822 dprintk("NFS: %5u initiated write call "
823 "(req %s/%lld, %u bytes @ offset %llu)\n",
826 (long long)NFS_FILEID(inode
),
828 (unsigned long long)data
->args
.offset
);
830 task
= rpc_run_task(&task_setup_data
);
835 if (how
& FLUSH_SYNC
) {
836 ret
= rpc_wait_for_completion_task(task
);
838 ret
= task
->tk_status
;
844 EXPORT_SYMBOL_GPL(nfs_initiate_write
);
847 * Set up the argument/result storage required for the RPC call.
849 static int nfs_write_rpcsetup(struct nfs_page
*req
,
850 struct nfs_write_data
*data
,
851 const struct rpc_call_ops
*call_ops
,
852 unsigned int count
, unsigned int offset
,
853 struct pnfs_layout_segment
*lseg
,
856 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
858 /* Set up the RPC argument and reply structs
859 * NB: take care not to mess about with data->commit et al. */
862 data
->inode
= inode
= req
->wb_context
->path
.dentry
->d_inode
;
863 data
->cred
= req
->wb_context
->cred
;
864 data
->lseg
= get_lseg(lseg
);
866 data
->args
.fh
= NFS_FH(inode
);
867 data
->args
.offset
= req_offset(req
) + offset
;
868 /* pnfs_set_layoutcommit needs this */
869 data
->mds_offset
= data
->args
.offset
;
870 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
871 data
->args
.pages
= data
->pagevec
;
872 data
->args
.count
= count
;
873 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
874 data
->args
.lock_context
= req
->wb_lock_context
;
875 data
->args
.stable
= NFS_UNSTABLE
;
876 if (how
& (FLUSH_STABLE
| FLUSH_COND_STABLE
)) {
877 data
->args
.stable
= NFS_DATA_SYNC
;
878 if (!nfs_need_commit(NFS_I(inode
)))
879 data
->args
.stable
= NFS_FILE_SYNC
;
882 data
->res
.fattr
= &data
->fattr
;
883 data
->res
.count
= count
;
884 data
->res
.verf
= &data
->verf
;
885 nfs_fattr_init(&data
->fattr
);
888 (pnfs_try_to_write_data(data
, call_ops
, how
) == PNFS_ATTEMPTED
))
891 return nfs_initiate_write(data
, NFS_CLIENT(inode
), call_ops
, how
);
894 /* If a nfs_flush_* function fails, it should remove reqs from @head and
895 * call this on each, which will prepare them to be retried on next
896 * writeback using standard nfs.
898 static void nfs_redirty_request(struct nfs_page
*req
)
900 struct page
*page
= req
->wb_page
;
902 nfs_mark_request_dirty(req
);
903 nfs_clear_page_tag_locked(req
);
904 nfs_end_page_writeback(page
);
908 * Generate multiple small requests to write out a single
909 * contiguous dirty area on one page.
911 static int nfs_flush_multi(struct nfs_pageio_descriptor
*desc
)
913 struct nfs_page
*req
= nfs_list_entry(desc
->pg_list
.next
);
914 struct page
*page
= req
->wb_page
;
915 struct nfs_write_data
*data
;
916 size_t wsize
= NFS_SERVER(desc
->pg_inode
)->wsize
, nbytes
;
920 struct pnfs_layout_segment
*lseg
;
923 nfs_list_remove_request(req
);
925 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
926 (desc
->pg_moreio
|| NFS_I(desc
->pg_inode
)->ncommit
||
927 desc
->pg_count
> wsize
))
928 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
931 nbytes
= desc
->pg_count
;
933 size_t len
= min(nbytes
, wsize
);
935 data
= nfs_writedata_alloc(1);
938 list_add(&data
->pages
, &list
);
941 } while (nbytes
!= 0);
942 atomic_set(&req
->wb_complete
, requests
);
944 BUG_ON(desc
->pg_lseg
);
945 lseg
= pnfs_update_layout(desc
->pg_inode
, req
->wb_context
,
946 req_offset(req
), desc
->pg_count
,
947 IOMODE_RW
, GFP_NOFS
);
948 ClearPageError(page
);
950 nbytes
= desc
->pg_count
;
954 data
= list_entry(list
.next
, struct nfs_write_data
, pages
);
955 list_del_init(&data
->pages
);
957 data
->pagevec
[0] = page
;
961 ret2
= nfs_write_rpcsetup(req
, data
, &nfs_write_partial_ops
,
962 wsize
, offset
, lseg
, desc
->pg_ioflags
);
967 } while (nbytes
!= 0);
970 desc
->pg_lseg
= NULL
;
974 while (!list_empty(&list
)) {
975 data
= list_entry(list
.next
, struct nfs_write_data
, pages
);
976 list_del(&data
->pages
);
977 nfs_writedata_free(data
);
979 nfs_redirty_request(req
);
984 * Create an RPC task for the given write request and kick it.
985 * The page must have been locked by the caller.
987 * It may happen that the page we're passed is not marked dirty.
988 * This is the case if nfs_updatepage detects a conflicting request
989 * that has been written but not committed.
991 static int nfs_flush_one(struct nfs_pageio_descriptor
*desc
)
993 struct nfs_page
*req
;
995 struct nfs_write_data
*data
;
996 struct list_head
*head
= &desc
->pg_list
;
997 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1000 data
= nfs_writedata_alloc(nfs_page_array_len(desc
->pg_base
,
1003 while (!list_empty(head
)) {
1004 req
= nfs_list_entry(head
->next
);
1005 nfs_list_remove_request(req
);
1006 nfs_redirty_request(req
);
1011 pages
= data
->pagevec
;
1012 while (!list_empty(head
)) {
1013 req
= nfs_list_entry(head
->next
);
1014 nfs_list_remove_request(req
);
1015 nfs_list_add_request(req
, &data
->pages
);
1016 ClearPageError(req
->wb_page
);
1017 *pages
++ = req
->wb_page
;
1019 req
= nfs_list_entry(data
->pages
.next
);
1020 if ((!lseg
) && list_is_singular(&data
->pages
))
1021 lseg
= pnfs_update_layout(desc
->pg_inode
, req
->wb_context
,
1022 req_offset(req
), desc
->pg_count
,
1023 IOMODE_RW
, GFP_NOFS
);
1025 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1026 (desc
->pg_moreio
|| NFS_I(desc
->pg_inode
)->ncommit
))
1027 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1029 /* Set up the argument struct */
1030 ret
= nfs_write_rpcsetup(req
, data
, &nfs_write_full_ops
, desc
->pg_count
, 0, lseg
, desc
->pg_ioflags
);
1032 put_lseg(lseg
); /* Cleans any gotten in ->pg_test */
1033 desc
->pg_lseg
= NULL
;
1037 static void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1038 struct inode
*inode
, int ioflags
)
1040 size_t wsize
= NFS_SERVER(inode
)->wsize
;
1042 if (wsize
< PAGE_CACHE_SIZE
)
1043 nfs_pageio_init(pgio
, inode
, nfs_flush_multi
, wsize
, ioflags
);
1045 nfs_pageio_init(pgio
, inode
, nfs_flush_one
, wsize
, ioflags
);
1049 * Handle a write reply that flushed part of a page.
1051 static void nfs_writeback_done_partial(struct rpc_task
*task
, void *calldata
)
1053 struct nfs_write_data
*data
= calldata
;
1055 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1057 data
->req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
1059 NFS_FILEID(data
->req
->wb_context
->path
.dentry
->d_inode
),
1060 data
->req
->wb_bytes
, (long long)req_offset(data
->req
));
1062 nfs_writeback_done(task
, data
);
1065 static void nfs_writeback_release_partial(void *calldata
)
1067 struct nfs_write_data
*data
= calldata
;
1068 struct nfs_page
*req
= data
->req
;
1069 struct page
*page
= req
->wb_page
;
1070 int status
= data
->task
.tk_status
;
1073 nfs_set_pageerror(page
);
1074 nfs_context_set_write_error(req
->wb_context
, status
);
1075 dprintk(", error = %d\n", status
);
1079 if (nfs_write_need_commit(data
)) {
1080 struct inode
*inode
= page
->mapping
->host
;
1082 spin_lock(&inode
->i_lock
);
1083 if (test_bit(PG_NEED_RESCHED
, &req
->wb_flags
)) {
1084 /* Do nothing we need to resend the writes */
1085 } else if (!test_and_set_bit(PG_NEED_COMMIT
, &req
->wb_flags
)) {
1086 memcpy(&req
->wb_verf
, &data
->verf
, sizeof(req
->wb_verf
));
1087 dprintk(" defer commit\n");
1088 } else if (memcmp(&req
->wb_verf
, &data
->verf
, sizeof(req
->wb_verf
))) {
1089 set_bit(PG_NEED_RESCHED
, &req
->wb_flags
);
1090 clear_bit(PG_NEED_COMMIT
, &req
->wb_flags
);
1091 dprintk(" server reboot detected\n");
1093 spin_unlock(&inode
->i_lock
);
1098 if (atomic_dec_and_test(&req
->wb_complete
))
1099 nfs_writepage_release(req
, data
);
1100 nfs_writedata_release(calldata
);
1103 #if defined(CONFIG_NFS_V4_1)
1104 void nfs_write_prepare(struct rpc_task
*task
, void *calldata
)
1106 struct nfs_write_data
*data
= calldata
;
1108 if (nfs4_setup_sequence(NFS_SERVER(data
->inode
),
1109 &data
->args
.seq_args
,
1110 &data
->res
.seq_res
, 1, task
))
1112 rpc_call_start(task
);
1114 #endif /* CONFIG_NFS_V4_1 */
1116 static const struct rpc_call_ops nfs_write_partial_ops
= {
1117 #if defined(CONFIG_NFS_V4_1)
1118 .rpc_call_prepare
= nfs_write_prepare
,
1119 #endif /* CONFIG_NFS_V4_1 */
1120 .rpc_call_done
= nfs_writeback_done_partial
,
1121 .rpc_release
= nfs_writeback_release_partial
,
1125 * Handle a write reply that flushes a whole page.
1127 * FIXME: There is an inherent race with invalidate_inode_pages and
1128 * writebacks since the page->count is kept > 1 for as long
1129 * as the page has a write request pending.
1131 static void nfs_writeback_done_full(struct rpc_task
*task
, void *calldata
)
1133 struct nfs_write_data
*data
= calldata
;
1135 nfs_writeback_done(task
, data
);
1138 static void nfs_writeback_release_full(void *calldata
)
1140 struct nfs_write_data
*data
= calldata
;
1141 int status
= data
->task
.tk_status
;
1143 /* Update attributes as result of writeback. */
1144 while (!list_empty(&data
->pages
)) {
1145 struct nfs_page
*req
= nfs_list_entry(data
->pages
.next
);
1146 struct page
*page
= req
->wb_page
;
1148 nfs_list_remove_request(req
);
1150 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1152 req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
1153 (long long)NFS_FILEID(req
->wb_context
->path
.dentry
->d_inode
),
1155 (long long)req_offset(req
));
1158 nfs_set_pageerror(page
);
1159 nfs_context_set_write_error(req
->wb_context
, status
);
1160 dprintk(", error = %d\n", status
);
1161 goto remove_request
;
1164 if (nfs_write_need_commit(data
)) {
1165 memcpy(&req
->wb_verf
, &data
->verf
, sizeof(req
->wb_verf
));
1166 nfs_mark_request_commit(req
, data
->lseg
);
1167 dprintk(" marked for commit\n");
1172 nfs_inode_remove_request(req
);
1174 nfs_clear_page_tag_locked(req
);
1175 nfs_end_page_writeback(page
);
1177 nfs_writedata_release(calldata
);
1180 static const struct rpc_call_ops nfs_write_full_ops
= {
1181 #if defined(CONFIG_NFS_V4_1)
1182 .rpc_call_prepare
= nfs_write_prepare
,
1183 #endif /* CONFIG_NFS_V4_1 */
1184 .rpc_call_done
= nfs_writeback_done_full
,
1185 .rpc_release
= nfs_writeback_release_full
,
1190 * This function is called when the WRITE call is complete.
1192 void nfs_writeback_done(struct rpc_task
*task
, struct nfs_write_data
*data
)
1194 struct nfs_writeargs
*argp
= &data
->args
;
1195 struct nfs_writeres
*resp
= &data
->res
;
1196 struct nfs_server
*server
= NFS_SERVER(data
->inode
);
1199 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1200 task
->tk_pid
, task
->tk_status
);
1203 * ->write_done will attempt to use post-op attributes to detect
1204 * conflicting writes by other clients. A strict interpretation
1205 * of close-to-open would allow us to continue caching even if
1206 * another writer had changed the file, but some applications
1207 * depend on tighter cache coherency when writing.
1209 status
= NFS_PROTO(data
->inode
)->write_done(task
, data
);
1212 nfs_add_stats(data
->inode
, NFSIOS_SERVERWRITTENBYTES
, resp
->count
);
1214 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1215 if (resp
->verf
->committed
< argp
->stable
&& task
->tk_status
>= 0) {
1216 /* We tried a write call, but the server did not
1217 * commit data to stable storage even though we
1219 * Note: There is a known bug in Tru64 < 5.0 in which
1220 * the server reports NFS_DATA_SYNC, but performs
1221 * NFS_FILE_SYNC. We therefore implement this checking
1222 * as a dprintk() in order to avoid filling syslog.
1224 static unsigned long complain
;
1226 /* Note this will print the MDS for a DS write */
1227 if (time_before(complain
, jiffies
)) {
1228 dprintk("NFS: faulty NFS server %s:"
1229 " (committed = %d) != (stable = %d)\n",
1230 server
->nfs_client
->cl_hostname
,
1231 resp
->verf
->committed
, argp
->stable
);
1232 complain
= jiffies
+ 300 * HZ
;
1236 /* Is this a short write? */
1237 if (task
->tk_status
>= 0 && resp
->count
< argp
->count
) {
1238 static unsigned long complain
;
1240 nfs_inc_stats(data
->inode
, NFSIOS_SHORTWRITE
);
1242 /* Has the server at least made some progress? */
1243 if (resp
->count
!= 0) {
1244 /* Was this an NFSv2 write or an NFSv3 stable write? */
1245 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1246 /* Resend from where the server left off */
1247 data
->mds_offset
+= resp
->count
;
1248 argp
->offset
+= resp
->count
;
1249 argp
->pgbase
+= resp
->count
;
1250 argp
->count
-= resp
->count
;
1252 /* Resend as a stable write in order to avoid
1253 * headaches in the case of a server crash.
1255 argp
->stable
= NFS_FILE_SYNC
;
1257 nfs_restart_rpc(task
, server
->nfs_client
);
1260 if (time_before(complain
, jiffies
)) {
1262 "NFS: Server wrote zero bytes, expected %u.\n",
1264 complain
= jiffies
+ 300 * HZ
;
1266 /* Can't do anything about it except throw an error. */
1267 task
->tk_status
= -EIO
;
1273 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1274 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1278 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1282 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1284 nfs_wait_bit_killable
,
1286 return (ret
< 0) ? ret
: 1;
1289 void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1291 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1292 smp_mb__after_clear_bit();
1293 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1295 EXPORT_SYMBOL_GPL(nfs_commit_clear_lock
);
1297 void nfs_commitdata_release(void *data
)
1299 struct nfs_write_data
*wdata
= data
;
1301 put_lseg(wdata
->lseg
);
1302 put_nfs_open_context(wdata
->args
.context
);
1303 nfs_commit_free(wdata
);
1305 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1307 int nfs_initiate_commit(struct nfs_write_data
*data
, struct rpc_clnt
*clnt
,
1308 const struct rpc_call_ops
*call_ops
,
1311 struct rpc_task
*task
;
1312 int priority
= flush_task_priority(how
);
1313 struct rpc_message msg
= {
1314 .rpc_argp
= &data
->args
,
1315 .rpc_resp
= &data
->res
,
1316 .rpc_cred
= data
->cred
,
1318 struct rpc_task_setup task_setup_data
= {
1319 .task
= &data
->task
,
1321 .rpc_message
= &msg
,
1322 .callback_ops
= call_ops
,
1323 .callback_data
= data
,
1324 .workqueue
= nfsiod_workqueue
,
1325 .flags
= RPC_TASK_ASYNC
,
1326 .priority
= priority
,
1328 /* Set up the initial task struct. */
1329 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1331 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1333 task
= rpc_run_task(&task_setup_data
);
1335 return PTR_ERR(task
);
1336 if (how
& FLUSH_SYNC
)
1337 rpc_wait_for_completion_task(task
);
1341 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1344 * Set up the argument/result storage required for the RPC call.
1346 void nfs_init_commit(struct nfs_write_data
*data
,
1347 struct list_head
*head
,
1348 struct pnfs_layout_segment
*lseg
)
1350 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1351 struct inode
*inode
= first
->wb_context
->path
.dentry
->d_inode
;
1353 /* Set up the RPC argument and reply structs
1354 * NB: take care not to mess about with data->commit et al. */
1356 list_splice_init(head
, &data
->pages
);
1358 data
->inode
= inode
;
1359 data
->cred
= first
->wb_context
->cred
;
1360 data
->lseg
= lseg
; /* reference transferred */
1361 data
->mds_ops
= &nfs_commit_ops
;
1363 data
->args
.fh
= NFS_FH(data
->inode
);
1364 /* Note: we always request a commit of the entire inode */
1365 data
->args
.offset
= 0;
1366 data
->args
.count
= 0;
1367 data
->args
.context
= get_nfs_open_context(first
->wb_context
);
1368 data
->res
.count
= 0;
1369 data
->res
.fattr
= &data
->fattr
;
1370 data
->res
.verf
= &data
->verf
;
1371 nfs_fattr_init(&data
->fattr
);
1373 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1375 void nfs_retry_commit(struct list_head
*page_list
,
1376 struct pnfs_layout_segment
*lseg
)
1378 struct nfs_page
*req
;
1380 while (!list_empty(page_list
)) {
1381 req
= nfs_list_entry(page_list
->next
);
1382 nfs_list_remove_request(req
);
1383 nfs_mark_request_commit(req
, lseg
);
1384 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1385 dec_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
,
1387 nfs_clear_page_tag_locked(req
);
1390 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1393 * Commit dirty pages
1396 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
)
1398 struct nfs_write_data
*data
;
1400 data
= nfs_commitdata_alloc();
1405 /* Set up the argument struct */
1406 nfs_init_commit(data
, head
, NULL
);
1407 return nfs_initiate_commit(data
, NFS_CLIENT(inode
), data
->mds_ops
, how
);
1409 nfs_retry_commit(head
, NULL
);
1410 nfs_commit_clear_lock(NFS_I(inode
));
1415 * COMMIT call returned
1417 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1419 struct nfs_write_data
*data
= calldata
;
1421 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1422 task
->tk_pid
, task
->tk_status
);
1424 /* Call the NFS version-specific code */
1425 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1428 void nfs_commit_release_pages(struct nfs_write_data
*data
)
1430 struct nfs_page
*req
;
1431 int status
= data
->task
.tk_status
;
1433 while (!list_empty(&data
->pages
)) {
1434 req
= nfs_list_entry(data
->pages
.next
);
1435 nfs_list_remove_request(req
);
1436 nfs_clear_request_commit(req
);
1438 dprintk("NFS: commit (%s/%lld %d@%lld)",
1439 req
->wb_context
->path
.dentry
->d_inode
->i_sb
->s_id
,
1440 (long long)NFS_FILEID(req
->wb_context
->path
.dentry
->d_inode
),
1442 (long long)req_offset(req
));
1444 nfs_context_set_write_error(req
->wb_context
, status
);
1445 nfs_inode_remove_request(req
);
1446 dprintk(", error = %d\n", status
);
1450 /* Okay, COMMIT succeeded, apparently. Check the verifier
1451 * returned by the server against all stored verfs. */
1452 if (!memcmp(req
->wb_verf
.verifier
, data
->verf
.verifier
, sizeof(data
->verf
.verifier
))) {
1453 /* We have a match */
1454 nfs_inode_remove_request(req
);
1458 /* We have a mismatch. Write the page again */
1459 dprintk(" mismatch\n");
1460 nfs_mark_request_dirty(req
);
1462 nfs_clear_page_tag_locked(req
);
1465 EXPORT_SYMBOL_GPL(nfs_commit_release_pages
);
1467 static void nfs_commit_release(void *calldata
)
1469 struct nfs_write_data
*data
= calldata
;
1471 nfs_commit_release_pages(data
);
1472 nfs_commit_clear_lock(NFS_I(data
->inode
));
1473 nfs_commitdata_release(calldata
);
1476 static const struct rpc_call_ops nfs_commit_ops
= {
1477 #if defined(CONFIG_NFS_V4_1)
1478 .rpc_call_prepare
= nfs_write_prepare
,
1479 #endif /* CONFIG_NFS_V4_1 */
1480 .rpc_call_done
= nfs_commit_done
,
1481 .rpc_release
= nfs_commit_release
,
1484 int nfs_commit_inode(struct inode
*inode
, int how
)
1487 int may_wait
= how
& FLUSH_SYNC
;
1490 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1492 goto out_mark_dirty
;
1493 res
= nfs_scan_commit(inode
, &head
, 0, 0);
1497 error
= pnfs_commit_list(inode
, &head
, how
);
1498 if (error
== PNFS_NOT_ATTEMPTED
)
1499 error
= nfs_commit_list(inode
, &head
, how
);
1503 goto out_mark_dirty
;
1504 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1506 nfs_wait_bit_killable
,
1511 nfs_commit_clear_lock(NFS_I(inode
));
1513 /* Note: If we exit without ensuring that the commit is complete,
1514 * we must mark the inode as dirty. Otherwise, future calls to
1515 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1516 * that the data is on the disk.
1519 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1523 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1525 struct nfs_inode
*nfsi
= NFS_I(inode
);
1526 int flags
= FLUSH_SYNC
;
1529 /* no commits means nothing needs to be done */
1533 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1534 /* Don't commit yet if this is a non-blocking flush and there
1535 * are a lot of outstanding writes for this mapping.
1537 if (nfsi
->ncommit
<= (nfsi
->npages
>> 1))
1538 goto out_mark_dirty
;
1540 /* don't wait for the COMMIT response */
1544 ret
= nfs_commit_inode(inode
, flags
);
1546 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1547 if (ret
< wbc
->nr_to_write
)
1548 wbc
->nr_to_write
-= ret
;
1550 wbc
->nr_to_write
= 0;
1555 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1559 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1565 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1569 ret
= nfs_commit_unstable_pages(inode
, wbc
);
1570 if (ret
>= 0 && test_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(inode
)->flags
)) {
1574 if (wbc
->sync_mode
== WB_SYNC_NONE
|| wbc
->nonblocking
||
1575 wbc
->for_background
)
1578 status
= pnfs_layoutcommit_inode(inode
, sync
);
1586 * flush the inode to disk.
1588 int nfs_wb_all(struct inode
*inode
)
1590 struct writeback_control wbc
= {
1591 .sync_mode
= WB_SYNC_ALL
,
1592 .nr_to_write
= LONG_MAX
,
1594 .range_end
= LLONG_MAX
,
1597 return sync_inode(inode
, &wbc
);
1600 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1602 struct nfs_page
*req
;
1605 BUG_ON(!PageLocked(page
));
1607 wait_on_page_writeback(page
);
1608 req
= nfs_page_find_request(page
);
1611 if (nfs_lock_request_dontget(req
)) {
1612 nfs_inode_remove_request(req
);
1614 * In case nfs_inode_remove_request has marked the
1615 * page as being dirty
1617 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1618 nfs_unlock_request(req
);
1621 ret
= nfs_wait_on_request(req
);
1622 nfs_release_request(req
);
1630 * Write back all requests on one page - we do this before reading it.
1632 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1634 loff_t range_start
= page_offset(page
);
1635 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1636 struct writeback_control wbc
= {
1637 .sync_mode
= WB_SYNC_ALL
,
1639 .range_start
= range_start
,
1640 .range_end
= range_end
,
1645 wait_on_page_writeback(page
);
1646 if (clear_page_dirty_for_io(page
)) {
1647 ret
= nfs_writepage_locked(page
, &wbc
);
1652 if (!PagePrivate(page
))
1654 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1663 #ifdef CONFIG_MIGRATION
1664 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1668 * If PagePrivate is set, then the page is currently associated with
1669 * an in-progress read or write request. Don't try to migrate it.
1671 * FIXME: we could do this in principle, but we'll need a way to ensure
1672 * that we can safely release the inode reference while holding
1675 if (PagePrivate(page
))
1678 nfs_fscache_release_page(page
, GFP_KERNEL
);
1680 return migrate_page(mapping
, newpage
, page
);
1684 int __init
nfs_init_writepagecache(void)
1686 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1687 sizeof(struct nfs_write_data
),
1688 0, SLAB_HWCACHE_ALIGN
,
1690 if (nfs_wdata_cachep
== NULL
)
1693 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1695 if (nfs_wdata_mempool
== NULL
)
1698 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1700 if (nfs_commit_mempool
== NULL
)
1704 * NFS congestion size, scale with available memory.
1716 * This allows larger machines to have larger/more transfers.
1717 * Limit the default to 256M
1719 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1720 if (nfs_congestion_kb
> 256*1024)
1721 nfs_congestion_kb
= 256*1024;
1726 void nfs_destroy_writepagecache(void)
1728 mempool_destroy(nfs_commit_mempool
);
1729 mempool_destroy(nfs_wdata_mempool
);
1730 kmem_cache_destroy(nfs_wdata_cachep
);