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>
23 #include <linux/export.h>
25 #include <asm/uaccess.h>
27 #include "delegation.h"
34 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
36 #define MIN_POOL_WRITE (32)
37 #define MIN_POOL_COMMIT (4)
40 * Local function declarations
42 static void nfs_redirty_request(struct nfs_page
*req
);
43 static const struct rpc_call_ops nfs_write_common_ops
;
44 static const struct rpc_call_ops nfs_commit_ops
;
45 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
;
46 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
;
48 static struct kmem_cache
*nfs_wdata_cachep
;
49 static mempool_t
*nfs_wdata_mempool
;
50 static struct kmem_cache
*nfs_cdata_cachep
;
51 static mempool_t
*nfs_commit_mempool
;
53 struct nfs_commit_data
*nfs_commitdata_alloc(void)
55 struct nfs_commit_data
*p
= mempool_alloc(nfs_commit_mempool
, GFP_NOFS
);
58 memset(p
, 0, sizeof(*p
));
59 INIT_LIST_HEAD(&p
->pages
);
63 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc
);
65 void nfs_commit_free(struct nfs_commit_data
*p
)
67 mempool_free(p
, nfs_commit_mempool
);
69 EXPORT_SYMBOL_GPL(nfs_commit_free
);
71 struct nfs_write_header
*nfs_writehdr_alloc(void)
73 struct nfs_write_header
*p
= mempool_alloc(nfs_wdata_mempool
, GFP_NOFS
);
76 struct nfs_pgio_header
*hdr
= &p
->header
;
78 memset(p
, 0, sizeof(*p
));
79 INIT_LIST_HEAD(&hdr
->pages
);
80 INIT_LIST_HEAD(&hdr
->rpc_list
);
81 spin_lock_init(&hdr
->lock
);
82 atomic_set(&hdr
->refcnt
, 0);
87 static struct nfs_write_data
*nfs_writedata_alloc(struct nfs_pgio_header
*hdr
,
88 unsigned int pagecount
)
90 struct nfs_write_data
*data
, *prealloc
;
92 prealloc
= &container_of(hdr
, struct nfs_write_header
, header
)->rpc_data
;
93 if (prealloc
->header
== NULL
)
96 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
100 if (nfs_pgarray_set(&data
->pages
, pagecount
)) {
102 atomic_inc(&hdr
->refcnt
);
104 if (data
!= prealloc
)
112 void nfs_writehdr_free(struct nfs_pgio_header
*hdr
)
114 struct nfs_write_header
*whdr
= container_of(hdr
, struct nfs_write_header
, header
);
115 mempool_free(whdr
, nfs_wdata_mempool
);
118 void nfs_writedata_release(struct nfs_write_data
*wdata
)
120 struct nfs_pgio_header
*hdr
= wdata
->header
;
121 struct nfs_write_header
*write_header
= container_of(hdr
, struct nfs_write_header
, header
);
123 put_nfs_open_context(wdata
->args
.context
);
124 if (wdata
->pages
.pagevec
!= wdata
->pages
.page_array
)
125 kfree(wdata
->pages
.pagevec
);
126 if (wdata
!= &write_header
->rpc_data
)
129 wdata
->header
= NULL
;
130 if (atomic_dec_and_test(&hdr
->refcnt
))
131 hdr
->completion_ops
->completion(hdr
);
134 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
138 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
141 static struct nfs_page
*nfs_page_find_request_locked(struct page
*page
)
143 struct nfs_page
*req
= NULL
;
145 if (PagePrivate(page
)) {
146 req
= (struct nfs_page
*)page_private(page
);
148 kref_get(&req
->wb_kref
);
153 static struct nfs_page
*nfs_page_find_request(struct page
*page
)
155 struct inode
*inode
= page
->mapping
->host
;
156 struct nfs_page
*req
= NULL
;
158 spin_lock(&inode
->i_lock
);
159 req
= nfs_page_find_request_locked(page
);
160 spin_unlock(&inode
->i_lock
);
164 /* Adjust the file length if we're writing beyond the end */
165 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
167 struct inode
*inode
= page
->mapping
->host
;
171 spin_lock(&inode
->i_lock
);
172 i_size
= i_size_read(inode
);
173 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
174 if (i_size
> 0 && page
->index
< end_index
)
176 end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + ((loff_t
)offset
+count
);
179 i_size_write(inode
, end
);
180 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
182 spin_unlock(&inode
->i_lock
);
185 /* A writeback failed: mark the page as bad, and invalidate the page cache */
186 static void nfs_set_pageerror(struct page
*page
)
189 nfs_zap_mapping(page
->mapping
->host
, page
->mapping
);
192 /* We can set the PG_uptodate flag if we see that a write request
193 * covers the full page.
195 static void nfs_mark_uptodate(struct page
*page
, unsigned int base
, unsigned int count
)
197 if (PageUptodate(page
))
201 if (count
!= nfs_page_length(page
))
203 SetPageUptodate(page
);
206 static int wb_priority(struct writeback_control
*wbc
)
208 if (wbc
->for_reclaim
)
209 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
210 if (wbc
->for_kupdate
|| wbc
->for_background
)
211 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
212 return FLUSH_COND_STABLE
;
216 * NFS congestion control
219 int nfs_congestion_kb
;
221 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
222 #define NFS_CONGESTION_OFF_THRESH \
223 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
225 static int nfs_set_page_writeback(struct page
*page
)
227 int ret
= test_set_page_writeback(page
);
230 struct inode
*inode
= page
->mapping
->host
;
231 struct nfs_server
*nfss
= NFS_SERVER(inode
);
233 if (atomic_long_inc_return(&nfss
->writeback
) >
234 NFS_CONGESTION_ON_THRESH
) {
235 set_bdi_congested(&nfss
->backing_dev_info
,
242 static void nfs_end_page_writeback(struct page
*page
)
244 struct inode
*inode
= page
->mapping
->host
;
245 struct nfs_server
*nfss
= NFS_SERVER(inode
);
247 end_page_writeback(page
);
248 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
249 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
252 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
254 struct inode
*inode
= page
->mapping
->host
;
255 struct nfs_page
*req
;
258 spin_lock(&inode
->i_lock
);
260 req
= nfs_page_find_request_locked(page
);
263 if (nfs_lock_request(req
))
265 /* Note: If we hold the page lock, as is the case in nfs_writepage,
266 * then the call to nfs_lock_request() will always
267 * succeed provided that someone hasn't already marked the
268 * request as dirty (in which case we don't care).
270 spin_unlock(&inode
->i_lock
);
272 ret
= nfs_wait_on_request(req
);
275 nfs_release_request(req
);
278 spin_lock(&inode
->i_lock
);
280 spin_unlock(&inode
->i_lock
);
285 * Find an associated nfs write request, and prepare to flush it out
286 * May return an error if the user signalled nfs_wait_on_request().
288 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
289 struct page
*page
, bool nonblock
)
291 struct nfs_page
*req
;
294 req
= nfs_find_and_lock_request(page
, nonblock
);
301 ret
= nfs_set_page_writeback(page
);
303 BUG_ON(test_bit(PG_CLEAN
, &req
->wb_flags
));
305 if (!nfs_pageio_add_request(pgio
, req
)) {
306 nfs_redirty_request(req
);
307 ret
= pgio
->pg_error
;
313 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
315 struct inode
*inode
= page
->mapping
->host
;
318 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
319 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
321 nfs_pageio_cond_complete(pgio
, page
->index
);
322 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
323 if (ret
== -EAGAIN
) {
324 redirty_page_for_writepage(wbc
, page
);
331 * Write an mmapped page to the server.
333 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
335 struct nfs_pageio_descriptor pgio
;
338 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
),
339 &nfs_async_write_completion_ops
);
340 err
= nfs_do_writepage(page
, wbc
, &pgio
);
341 nfs_pageio_complete(&pgio
);
344 if (pgio
.pg_error
< 0)
345 return pgio
.pg_error
;
349 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
353 ret
= nfs_writepage_locked(page
, wbc
);
358 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
362 ret
= nfs_do_writepage(page
, wbc
, data
);
367 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
369 struct inode
*inode
= mapping
->host
;
370 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
371 struct nfs_pageio_descriptor pgio
;
374 /* Stop dirtying of new pages while we sync */
375 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
376 nfs_wait_bit_killable
, TASK_KILLABLE
);
380 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
382 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
),
383 &nfs_async_write_completion_ops
);
384 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
385 nfs_pageio_complete(&pgio
);
387 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
388 smp_mb__after_clear_bit();
389 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
402 * Insert a write request into an inode
404 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
406 struct nfs_inode
*nfsi
= NFS_I(inode
);
408 /* Lock the request! */
409 nfs_lock_request(req
);
411 spin_lock(&inode
->i_lock
);
412 if (!nfsi
->npages
&& nfs_have_delegation(inode
, FMODE_WRITE
))
414 set_bit(PG_MAPPED
, &req
->wb_flags
);
415 SetPagePrivate(req
->wb_page
);
416 set_page_private(req
->wb_page
, (unsigned long)req
);
418 kref_get(&req
->wb_kref
);
419 spin_unlock(&inode
->i_lock
);
423 * Remove a write request from an inode
425 static void nfs_inode_remove_request(struct nfs_page
*req
)
427 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
428 struct nfs_inode
*nfsi
= NFS_I(inode
);
430 BUG_ON (!NFS_WBACK_BUSY(req
));
432 spin_lock(&inode
->i_lock
);
433 set_page_private(req
->wb_page
, 0);
434 ClearPagePrivate(req
->wb_page
);
435 clear_bit(PG_MAPPED
, &req
->wb_flags
);
437 spin_unlock(&inode
->i_lock
);
438 nfs_release_request(req
);
442 nfs_mark_request_dirty(struct nfs_page
*req
)
444 __set_page_dirty_nobuffers(req
->wb_page
);
447 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
449 * nfs_request_add_commit_list - add request to a commit list
450 * @req: pointer to a struct nfs_page
451 * @dst: commit list head
452 * @cinfo: holds list lock and accounting info
454 * This sets the PG_CLEAN bit, updates the cinfo count of
455 * number of outstanding requests requiring a commit as well as
458 * The caller must _not_ hold the cinfo->lock, but must be
459 * holding the nfs_page lock.
462 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
463 struct nfs_commit_info
*cinfo
)
465 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
466 spin_lock(cinfo
->lock
);
467 nfs_list_add_request(req
, dst
);
468 cinfo
->mds
->ncommit
++;
469 spin_unlock(cinfo
->lock
);
471 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
472 inc_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
,
474 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
478 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
481 * nfs_request_remove_commit_list - Remove request from a commit list
482 * @req: pointer to a nfs_page
483 * @cinfo: holds list lock and accounting info
485 * This clears the PG_CLEAN bit, and updates the cinfo's count of
486 * number of outstanding requests requiring a commit
487 * It does not update the MM page stats.
489 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
492 nfs_request_remove_commit_list(struct nfs_page
*req
,
493 struct nfs_commit_info
*cinfo
)
495 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
497 nfs_list_remove_request(req
);
498 cinfo
->mds
->ncommit
--;
500 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
502 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
505 cinfo
->lock
= &inode
->i_lock
;
506 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
507 cinfo
->ds
= pnfs_get_ds_info(inode
);
509 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
512 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
514 struct nfs_direct_req
*dreq
)
517 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
519 nfs_init_cinfo_from_inode(cinfo
, inode
);
521 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
524 * Add a request to the inode's commit list.
527 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
528 struct nfs_commit_info
*cinfo
)
530 if (pnfs_mark_request_commit(req
, lseg
, cinfo
))
532 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
536 nfs_clear_page_commit(struct page
*page
)
538 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
539 dec_bdi_stat(page
->mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
543 nfs_clear_request_commit(struct nfs_page
*req
)
545 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
546 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
547 struct nfs_commit_info cinfo
;
549 nfs_init_cinfo_from_inode(&cinfo
, inode
);
550 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
551 spin_lock(cinfo
.lock
);
552 nfs_request_remove_commit_list(req
, &cinfo
);
553 spin_unlock(cinfo
.lock
);
555 nfs_clear_page_commit(req
->wb_page
);
560 int nfs_write_need_commit(struct nfs_write_data
*data
)
562 if (data
->verf
.committed
== NFS_DATA_SYNC
)
563 return data
->header
->lseg
== NULL
;
564 return data
->verf
.committed
!= NFS_FILE_SYNC
;
568 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
573 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
575 struct nfs_direct_req
*dreq
)
580 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
581 struct nfs_commit_info
*cinfo
)
586 nfs_clear_request_commit(struct nfs_page
*req
)
591 int nfs_write_need_commit(struct nfs_write_data
*data
)
598 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
600 struct nfs_commit_info cinfo
;
601 unsigned long bytes
= 0;
603 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
605 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
606 while (!list_empty(&hdr
->pages
)) {
607 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
609 bytes
+= req
->wb_bytes
;
610 nfs_list_remove_request(req
);
611 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
612 (hdr
->good_bytes
< bytes
)) {
613 nfs_set_pageerror(req
->wb_page
);
614 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
617 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
)) {
618 nfs_mark_request_dirty(req
);
621 if (test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
)) {
622 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
);
626 nfs_inode_remove_request(req
);
628 nfs_unlock_request(req
);
629 nfs_end_page_writeback(req
->wb_page
);
630 nfs_release_request(req
);
636 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
638 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
640 return cinfo
->mds
->ncommit
;
643 /* cinfo->lock held by caller */
645 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
646 struct nfs_commit_info
*cinfo
, int max
)
648 struct nfs_page
*req
, *tmp
;
651 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
652 if (!nfs_lock_request(req
))
654 kref_get(&req
->wb_kref
);
655 if (cond_resched_lock(cinfo
->lock
))
656 list_safe_reset_next(req
, tmp
, wb_list
);
657 nfs_request_remove_commit_list(req
, cinfo
);
658 nfs_list_add_request(req
, dst
);
660 if ((ret
== max
) && !cinfo
->dreq
)
667 * nfs_scan_commit - Scan an inode for commit requests
668 * @inode: NFS inode to scan
669 * @dst: mds destination list
670 * @cinfo: mds and ds lists of reqs ready to commit
672 * Moves requests from the inode's 'commit' request list.
673 * The requests are *not* checked to ensure that they form a contiguous set.
676 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
677 struct nfs_commit_info
*cinfo
)
681 spin_lock(cinfo
->lock
);
682 if (cinfo
->mds
->ncommit
> 0) {
683 const int max
= INT_MAX
;
685 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
687 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
689 spin_unlock(cinfo
->lock
);
694 static unsigned long nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
699 int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
700 struct nfs_commit_info
*cinfo
)
707 * Search for an existing write request, and attempt to update
708 * it to reflect a new dirty region on a given page.
710 * If the attempt fails, then the existing request is flushed out
713 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
718 struct nfs_page
*req
;
723 if (!PagePrivate(page
))
726 end
= offset
+ bytes
;
727 spin_lock(&inode
->i_lock
);
730 req
= nfs_page_find_request_locked(page
);
734 rqend
= req
->wb_offset
+ req
->wb_bytes
;
736 * Tell the caller to flush out the request if
737 * the offsets are non-contiguous.
738 * Note: nfs_flush_incompatible() will already
739 * have flushed out requests having wrong owners.
742 || end
< req
->wb_offset
)
745 if (nfs_lock_request(req
))
748 /* The request is locked, so wait and then retry */
749 spin_unlock(&inode
->i_lock
);
750 error
= nfs_wait_on_request(req
);
751 nfs_release_request(req
);
754 spin_lock(&inode
->i_lock
);
757 /* Okay, the request matches. Update the region */
758 if (offset
< req
->wb_offset
) {
759 req
->wb_offset
= offset
;
760 req
->wb_pgbase
= offset
;
763 req
->wb_bytes
= end
- req
->wb_offset
;
765 req
->wb_bytes
= rqend
- req
->wb_offset
;
767 spin_unlock(&inode
->i_lock
);
769 nfs_clear_request_commit(req
);
772 spin_unlock(&inode
->i_lock
);
773 nfs_release_request(req
);
774 error
= nfs_wb_page(inode
, page
);
776 return ERR_PTR(error
);
780 * Try to update an existing write request, or create one if there is none.
782 * Note: Should always be called with the Page Lock held to prevent races
783 * if we have to add a new request. Also assumes that the caller has
784 * already called nfs_flush_incompatible() if necessary.
786 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
787 struct page
*page
, unsigned int offset
, unsigned int bytes
)
789 struct inode
*inode
= page
->mapping
->host
;
790 struct nfs_page
*req
;
792 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
795 req
= nfs_create_request(ctx
, inode
, page
, offset
, bytes
);
798 nfs_inode_add_request(inode
, req
);
803 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
804 unsigned int offset
, unsigned int count
)
806 struct nfs_page
*req
;
808 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
811 /* Update file length */
812 nfs_grow_file(page
, offset
, count
);
813 nfs_mark_uptodate(page
, req
->wb_pgbase
, req
->wb_bytes
);
814 nfs_mark_request_dirty(req
);
815 nfs_unlock_and_release_request(req
);
819 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
821 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
822 struct nfs_page
*req
;
823 int do_flush
, status
;
825 * Look for a request corresponding to this page. If there
826 * is one, and it belongs to another file, we flush it out
827 * before we try to copy anything into the page. Do this
828 * due to the lack of an ACCESS-type call in NFSv2.
829 * Also do the same if we find a request from an existing
833 req
= nfs_page_find_request(page
);
836 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
||
837 req
->wb_lock_context
->lockowner
!= current
->files
||
838 req
->wb_lock_context
->pid
!= current
->tgid
;
839 nfs_release_request(req
);
842 status
= nfs_wb_page(page
->mapping
->host
, page
);
843 } while (status
== 0);
848 * If the page cache is marked as unsafe or invalid, then we can't rely on
849 * the PageUptodate() flag. In this case, we will need to turn off
850 * write optimisations that depend on the page contents being correct.
852 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
854 if (nfs_have_delegated_attributes(inode
))
856 if (NFS_I(inode
)->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
859 return PageUptodate(page
) != 0;
863 * Update and possibly write a cached page of an NFS file.
865 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
866 * things with a page scheduled for an RPC call (e.g. invalidate it).
868 int nfs_updatepage(struct file
*file
, struct page
*page
,
869 unsigned int offset
, unsigned int count
)
871 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
872 struct inode
*inode
= page
->mapping
->host
;
875 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
877 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
878 file
->f_path
.dentry
->d_parent
->d_name
.name
,
879 file
->f_path
.dentry
->d_name
.name
, count
,
880 (long long)(page_offset(page
) + offset
));
882 /* If we're not using byte range locks, and we know the page
883 * is up to date, it may be more efficient to extend the write
884 * to cover the entire page in order to avoid fragmentation
887 if (nfs_write_pageuptodate(page
, inode
) &&
888 inode
->i_flock
== NULL
&&
889 !(file
->f_flags
& O_DSYNC
)) {
890 count
= max(count
+ offset
, nfs_page_length(page
));
894 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
896 nfs_set_pageerror(page
);
898 __set_page_dirty_nobuffers(page
);
900 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
901 status
, (long long)i_size_read(inode
));
905 static int flush_task_priority(int how
)
907 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
909 return RPC_PRIORITY_HIGH
;
911 return RPC_PRIORITY_LOW
;
913 return RPC_PRIORITY_NORMAL
;
916 int nfs_initiate_write(struct rpc_clnt
*clnt
,
917 struct nfs_write_data
*data
,
918 const struct rpc_call_ops
*call_ops
,
921 struct inode
*inode
= data
->header
->inode
;
922 int priority
= flush_task_priority(how
);
923 struct rpc_task
*task
;
924 struct rpc_message msg
= {
925 .rpc_argp
= &data
->args
,
926 .rpc_resp
= &data
->res
,
927 .rpc_cred
= data
->header
->cred
,
929 struct rpc_task_setup task_setup_data
= {
933 .callback_ops
= call_ops
,
934 .callback_data
= data
,
935 .workqueue
= nfsiod_workqueue
,
936 .flags
= RPC_TASK_ASYNC
| flags
,
937 .priority
= priority
,
941 /* Set up the initial task struct. */
942 NFS_PROTO(inode
)->write_setup(data
, &msg
);
944 dprintk("NFS: %5u initiated write call "
945 "(req %s/%lld, %u bytes @ offset %llu)\n",
948 (long long)NFS_FILEID(inode
),
950 (unsigned long long)data
->args
.offset
);
952 task
= rpc_run_task(&task_setup_data
);
957 if (how
& FLUSH_SYNC
) {
958 ret
= rpc_wait_for_completion_task(task
);
960 ret
= task
->tk_status
;
966 EXPORT_SYMBOL_GPL(nfs_initiate_write
);
969 * Set up the argument/result storage required for the RPC call.
971 static void nfs_write_rpcsetup(struct nfs_write_data
*data
,
972 unsigned int count
, unsigned int offset
,
973 int how
, struct nfs_commit_info
*cinfo
)
975 struct nfs_page
*req
= data
->header
->req
;
977 /* Set up the RPC argument and reply structs
978 * NB: take care not to mess about with data->commit et al. */
980 data
->args
.fh
= NFS_FH(data
->header
->inode
);
981 data
->args
.offset
= req_offset(req
) + offset
;
982 /* pnfs_set_layoutcommit needs this */
983 data
->mds_offset
= data
->args
.offset
;
984 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
985 data
->args
.pages
= data
->pages
.pagevec
;
986 data
->args
.count
= count
;
987 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
988 data
->args
.lock_context
= req
->wb_lock_context
;
989 data
->args
.stable
= NFS_UNSTABLE
;
990 switch (how
& (FLUSH_STABLE
| FLUSH_COND_STABLE
)) {
993 case FLUSH_COND_STABLE
:
994 if (nfs_reqs_to_commit(cinfo
))
997 data
->args
.stable
= NFS_FILE_SYNC
;
1000 data
->res
.fattr
= &data
->fattr
;
1001 data
->res
.count
= count
;
1002 data
->res
.verf
= &data
->verf
;
1003 nfs_fattr_init(&data
->fattr
);
1006 static int nfs_do_write(struct nfs_write_data
*data
,
1007 const struct rpc_call_ops
*call_ops
,
1010 struct inode
*inode
= data
->header
->inode
;
1012 return nfs_initiate_write(NFS_CLIENT(inode
), data
, call_ops
, how
, 0);
1015 static int nfs_do_multiple_writes(struct list_head
*head
,
1016 const struct rpc_call_ops
*call_ops
,
1019 struct nfs_write_data
*data
;
1022 while (!list_empty(head
)) {
1025 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1026 list_del_init(&data
->list
);
1028 ret2
= nfs_do_write(data
, call_ops
, how
);
1035 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1036 * call this on each, which will prepare them to be retried on next
1037 * writeback using standard nfs.
1039 static void nfs_redirty_request(struct nfs_page
*req
)
1041 nfs_mark_request_dirty(req
);
1042 nfs_unlock_request(req
);
1043 nfs_end_page_writeback(req
->wb_page
);
1044 nfs_release_request(req
);
1047 static void nfs_async_write_error(struct list_head
*head
)
1049 struct nfs_page
*req
;
1051 while (!list_empty(head
)) {
1052 req
= nfs_list_entry(head
->next
);
1053 nfs_list_remove_request(req
);
1054 nfs_redirty_request(req
);
1058 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1059 .error_cleanup
= nfs_async_write_error
,
1060 .completion
= nfs_write_completion
,
1063 static void nfs_flush_error(struct nfs_pageio_descriptor
*desc
,
1064 struct nfs_pgio_header
*hdr
)
1066 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
1067 while (!list_empty(&hdr
->rpc_list
)) {
1068 struct nfs_write_data
*data
= list_first_entry(&hdr
->rpc_list
,
1069 struct nfs_write_data
, list
);
1070 list_del(&data
->list
);
1071 nfs_writedata_release(data
);
1073 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1077 * Generate multiple small requests to write out a single
1078 * contiguous dirty area on one page.
1080 static int nfs_flush_multi(struct nfs_pageio_descriptor
*desc
,
1081 struct nfs_pgio_header
*hdr
)
1083 struct nfs_page
*req
= hdr
->req
;
1084 struct page
*page
= req
->wb_page
;
1085 struct nfs_write_data
*data
;
1086 size_t wsize
= desc
->pg_bsize
, nbytes
;
1087 unsigned int offset
;
1089 struct nfs_commit_info cinfo
;
1091 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1093 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1094 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
) ||
1095 desc
->pg_count
> wsize
))
1096 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1100 nbytes
= desc
->pg_count
;
1102 size_t len
= min(nbytes
, wsize
);
1104 data
= nfs_writedata_alloc(hdr
, 1);
1106 nfs_flush_error(desc
, hdr
);
1109 data
->pages
.pagevec
[0] = page
;
1110 nfs_write_rpcsetup(data
, len
, offset
, desc
->pg_ioflags
, &cinfo
);
1111 list_add(&data
->list
, &hdr
->rpc_list
);
1115 } while (nbytes
!= 0);
1116 nfs_list_remove_request(req
);
1117 nfs_list_add_request(req
, &hdr
->pages
);
1118 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1123 * Create an RPC task for the given write request and kick it.
1124 * The page must have been locked by the caller.
1126 * It may happen that the page we're passed is not marked dirty.
1127 * This is the case if nfs_updatepage detects a conflicting request
1128 * that has been written but not committed.
1130 static int nfs_flush_one(struct nfs_pageio_descriptor
*desc
,
1131 struct nfs_pgio_header
*hdr
)
1133 struct nfs_page
*req
;
1134 struct page
**pages
;
1135 struct nfs_write_data
*data
;
1136 struct list_head
*head
= &desc
->pg_list
;
1137 struct nfs_commit_info cinfo
;
1139 data
= nfs_writedata_alloc(hdr
, nfs_page_array_len(desc
->pg_base
,
1142 nfs_flush_error(desc
, hdr
);
1146 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1147 pages
= data
->pages
.pagevec
;
1148 while (!list_empty(head
)) {
1149 req
= nfs_list_entry(head
->next
);
1150 nfs_list_remove_request(req
);
1151 nfs_list_add_request(req
, &hdr
->pages
);
1152 *pages
++ = req
->wb_page
;
1155 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1156 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
)))
1157 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1159 /* Set up the argument struct */
1160 nfs_write_rpcsetup(data
, desc
->pg_count
, 0, desc
->pg_ioflags
, &cinfo
);
1161 list_add(&data
->list
, &hdr
->rpc_list
);
1162 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1166 int nfs_generic_flush(struct nfs_pageio_descriptor
*desc
,
1167 struct nfs_pgio_header
*hdr
)
1169 if (desc
->pg_bsize
< PAGE_CACHE_SIZE
)
1170 return nfs_flush_multi(desc
, hdr
);
1171 return nfs_flush_one(desc
, hdr
);
1174 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1176 struct nfs_write_header
*whdr
;
1177 struct nfs_pgio_header
*hdr
;
1180 whdr
= nfs_writehdr_alloc();
1182 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1185 hdr
= &whdr
->header
;
1186 nfs_pgheader_init(desc
, hdr
, nfs_writehdr_free
);
1187 atomic_inc(&hdr
->refcnt
);
1188 ret
= nfs_generic_flush(desc
, hdr
);
1190 ret
= nfs_do_multiple_writes(&hdr
->rpc_list
,
1191 desc
->pg_rpc_callops
,
1193 if (atomic_dec_and_test(&hdr
->refcnt
))
1194 hdr
->completion_ops
->completion(hdr
);
1198 static const struct nfs_pageio_ops nfs_pageio_write_ops
= {
1199 .pg_test
= nfs_generic_pg_test
,
1200 .pg_doio
= nfs_generic_pg_writepages
,
1203 void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor
*pgio
,
1204 struct inode
*inode
, int ioflags
,
1205 const struct nfs_pgio_completion_ops
*compl_ops
)
1207 nfs_pageio_init(pgio
, inode
, &nfs_pageio_write_ops
, compl_ops
,
1208 NFS_SERVER(inode
)->wsize
, ioflags
);
1211 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1213 pgio
->pg_ops
= &nfs_pageio_write_ops
;
1214 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1216 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1218 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1219 struct inode
*inode
, int ioflags
,
1220 const struct nfs_pgio_completion_ops
*compl_ops
)
1222 if (!pnfs_pageio_init_write(pgio
, inode
, ioflags
, compl_ops
))
1223 nfs_pageio_init_write_mds(pgio
, inode
, ioflags
, compl_ops
);
1226 void nfs_write_prepare(struct rpc_task
*task
, void *calldata
)
1228 struct nfs_write_data
*data
= calldata
;
1229 NFS_PROTO(data
->header
->inode
)->write_rpc_prepare(task
, data
);
1232 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1234 struct nfs_commit_data
*data
= calldata
;
1236 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1240 * Handle a write reply that flushes a whole page.
1242 * FIXME: There is an inherent race with invalidate_inode_pages and
1243 * writebacks since the page->count is kept > 1 for as long
1244 * as the page has a write request pending.
1246 static void nfs_writeback_done_common(struct rpc_task
*task
, void *calldata
)
1248 struct nfs_write_data
*data
= calldata
;
1250 nfs_writeback_done(task
, data
);
1253 static void nfs_writeback_release_common(void *calldata
)
1255 struct nfs_write_data
*data
= calldata
;
1256 struct nfs_pgio_header
*hdr
= data
->header
;
1257 int status
= data
->task
.tk_status
;
1258 struct nfs_page
*req
= hdr
->req
;
1260 if ((status
>= 0) && nfs_write_need_commit(data
)) {
1261 spin_lock(&hdr
->lock
);
1262 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
))
1264 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
))
1265 memcpy(&req
->wb_verf
, &data
->verf
, sizeof(req
->wb_verf
));
1266 else if (memcmp(&req
->wb_verf
, &data
->verf
, sizeof(req
->wb_verf
)))
1267 set_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
);
1268 spin_unlock(&hdr
->lock
);
1270 nfs_writedata_release(data
);
1273 static const struct rpc_call_ops nfs_write_common_ops
= {
1274 .rpc_call_prepare
= nfs_write_prepare
,
1275 .rpc_call_done
= nfs_writeback_done_common
,
1276 .rpc_release
= nfs_writeback_release_common
,
1281 * This function is called when the WRITE call is complete.
1283 void nfs_writeback_done(struct rpc_task
*task
, struct nfs_write_data
*data
)
1285 struct nfs_writeargs
*argp
= &data
->args
;
1286 struct nfs_writeres
*resp
= &data
->res
;
1287 struct inode
*inode
= data
->header
->inode
;
1290 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1291 task
->tk_pid
, task
->tk_status
);
1294 * ->write_done will attempt to use post-op attributes to detect
1295 * conflicting writes by other clients. A strict interpretation
1296 * of close-to-open would allow us to continue caching even if
1297 * another writer had changed the file, but some applications
1298 * depend on tighter cache coherency when writing.
1300 status
= NFS_PROTO(inode
)->write_done(task
, data
);
1303 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, resp
->count
);
1305 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1306 if (resp
->verf
->committed
< argp
->stable
&& task
->tk_status
>= 0) {
1307 /* We tried a write call, but the server did not
1308 * commit data to stable storage even though we
1310 * Note: There is a known bug in Tru64 < 5.0 in which
1311 * the server reports NFS_DATA_SYNC, but performs
1312 * NFS_FILE_SYNC. We therefore implement this checking
1313 * as a dprintk() in order to avoid filling syslog.
1315 static unsigned long complain
;
1317 /* Note this will print the MDS for a DS write */
1318 if (time_before(complain
, jiffies
)) {
1319 dprintk("NFS: faulty NFS server %s:"
1320 " (committed = %d) != (stable = %d)\n",
1321 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1322 resp
->verf
->committed
, argp
->stable
);
1323 complain
= jiffies
+ 300 * HZ
;
1327 if (task
->tk_status
< 0)
1328 nfs_set_pgio_error(data
->header
, task
->tk_status
, argp
->offset
);
1329 else if (resp
->count
< argp
->count
) {
1330 static unsigned long complain
;
1332 /* This a short write! */
1333 nfs_inc_stats(inode
, NFSIOS_SHORTWRITE
);
1335 /* Has the server at least made some progress? */
1336 if (resp
->count
== 0) {
1337 if (time_before(complain
, jiffies
)) {
1339 "NFS: Server wrote zero bytes, expected %u.\n",
1341 complain
= jiffies
+ 300 * HZ
;
1343 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
1344 task
->tk_status
= -EIO
;
1347 /* Was this an NFSv2 write or an NFSv3 stable write? */
1348 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1349 /* Resend from where the server left off */
1350 data
->mds_offset
+= resp
->count
;
1351 argp
->offset
+= resp
->count
;
1352 argp
->pgbase
+= resp
->count
;
1353 argp
->count
-= resp
->count
;
1355 /* Resend as a stable write in order to avoid
1356 * headaches in the case of a server crash.
1358 argp
->stable
= NFS_FILE_SYNC
;
1360 rpc_restart_call_prepare(task
);
1365 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1366 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1370 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1374 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1376 nfs_wait_bit_killable
,
1378 return (ret
< 0) ? ret
: 1;
1381 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1383 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1384 smp_mb__after_clear_bit();
1385 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1388 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1390 put_nfs_open_context(data
->context
);
1391 nfs_commit_free(data
);
1393 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1395 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1396 const struct rpc_call_ops
*call_ops
,
1399 struct rpc_task
*task
;
1400 int priority
= flush_task_priority(how
);
1401 struct rpc_message msg
= {
1402 .rpc_argp
= &data
->args
,
1403 .rpc_resp
= &data
->res
,
1404 .rpc_cred
= data
->cred
,
1406 struct rpc_task_setup task_setup_data
= {
1407 .task
= &data
->task
,
1409 .rpc_message
= &msg
,
1410 .callback_ops
= call_ops
,
1411 .callback_data
= data
,
1412 .workqueue
= nfsiod_workqueue
,
1413 .flags
= RPC_TASK_ASYNC
| flags
,
1414 .priority
= priority
,
1416 /* Set up the initial task struct. */
1417 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1419 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1421 task
= rpc_run_task(&task_setup_data
);
1423 return PTR_ERR(task
);
1424 if (how
& FLUSH_SYNC
)
1425 rpc_wait_for_completion_task(task
);
1429 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1432 * Set up the argument/result storage required for the RPC call.
1434 void nfs_init_commit(struct nfs_commit_data
*data
,
1435 struct list_head
*head
,
1436 struct pnfs_layout_segment
*lseg
,
1437 struct nfs_commit_info
*cinfo
)
1439 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1440 struct inode
*inode
= first
->wb_context
->dentry
->d_inode
;
1442 /* Set up the RPC argument and reply structs
1443 * NB: take care not to mess about with data->commit et al. */
1445 list_splice_init(head
, &data
->pages
);
1447 data
->inode
= inode
;
1448 data
->cred
= first
->wb_context
->cred
;
1449 data
->lseg
= lseg
; /* reference transferred */
1450 data
->mds_ops
= &nfs_commit_ops
;
1451 data
->completion_ops
= cinfo
->completion_ops
;
1452 data
->dreq
= cinfo
->dreq
;
1454 data
->args
.fh
= NFS_FH(data
->inode
);
1455 /* Note: we always request a commit of the entire inode */
1456 data
->args
.offset
= 0;
1457 data
->args
.count
= 0;
1458 data
->context
= get_nfs_open_context(first
->wb_context
);
1459 data
->res
.fattr
= &data
->fattr
;
1460 data
->res
.verf
= &data
->verf
;
1461 nfs_fattr_init(&data
->fattr
);
1463 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1465 void nfs_retry_commit(struct list_head
*page_list
,
1466 struct pnfs_layout_segment
*lseg
,
1467 struct nfs_commit_info
*cinfo
)
1469 struct nfs_page
*req
;
1471 while (!list_empty(page_list
)) {
1472 req
= nfs_list_entry(page_list
->next
);
1473 nfs_list_remove_request(req
);
1474 nfs_mark_request_commit(req
, lseg
, cinfo
);
1476 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1477 dec_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
,
1480 nfs_unlock_and_release_request(req
);
1483 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1486 * Commit dirty pages
1489 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1490 struct nfs_commit_info
*cinfo
)
1492 struct nfs_commit_data
*data
;
1494 data
= nfs_commitdata_alloc();
1499 /* Set up the argument struct */
1500 nfs_init_commit(data
, head
, NULL
, cinfo
);
1501 atomic_inc(&cinfo
->mds
->rpcs_out
);
1502 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, data
->mds_ops
,
1505 nfs_retry_commit(head
, NULL
, cinfo
);
1506 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1511 * COMMIT call returned
1513 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1515 struct nfs_commit_data
*data
= calldata
;
1517 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1518 task
->tk_pid
, task
->tk_status
);
1520 /* Call the NFS version-specific code */
1521 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1524 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1526 struct nfs_page
*req
;
1527 int status
= data
->task
.tk_status
;
1528 struct nfs_commit_info cinfo
;
1530 while (!list_empty(&data
->pages
)) {
1531 req
= nfs_list_entry(data
->pages
.next
);
1532 nfs_list_remove_request(req
);
1533 nfs_clear_page_commit(req
->wb_page
);
1535 dprintk("NFS: commit (%s/%lld %d@%lld)",
1536 req
->wb_context
->dentry
->d_sb
->s_id
,
1537 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
1539 (long long)req_offset(req
));
1541 nfs_context_set_write_error(req
->wb_context
, status
);
1542 nfs_inode_remove_request(req
);
1543 dprintk(", error = %d\n", status
);
1547 /* Okay, COMMIT succeeded, apparently. Check the verifier
1548 * returned by the server against all stored verfs. */
1549 if (!memcmp(req
->wb_verf
.verifier
, data
->verf
.verifier
, sizeof(data
->verf
.verifier
))) {
1550 /* We have a match */
1551 nfs_inode_remove_request(req
);
1555 /* We have a mismatch. Write the page again */
1556 dprintk(" mismatch\n");
1557 nfs_mark_request_dirty(req
);
1559 nfs_unlock_and_release_request(req
);
1561 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1562 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1563 nfs_commit_clear_lock(NFS_I(data
->inode
));
1566 static void nfs_commit_release(void *calldata
)
1568 struct nfs_commit_data
*data
= calldata
;
1570 data
->completion_ops
->completion(data
);
1571 nfs_commitdata_release(calldata
);
1574 static const struct rpc_call_ops nfs_commit_ops
= {
1575 .rpc_call_prepare
= nfs_commit_prepare
,
1576 .rpc_call_done
= nfs_commit_done
,
1577 .rpc_release
= nfs_commit_release
,
1580 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1581 .completion
= nfs_commit_release_pages
,
1582 .error_cleanup
= nfs_commit_clear_lock
,
1585 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1586 int how
, struct nfs_commit_info
*cinfo
)
1590 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1591 if (status
== PNFS_NOT_ATTEMPTED
)
1592 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1596 int nfs_commit_inode(struct inode
*inode
, int how
)
1599 struct nfs_commit_info cinfo
;
1600 int may_wait
= how
& FLUSH_SYNC
;
1603 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1605 goto out_mark_dirty
;
1606 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1607 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1611 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1615 goto out_mark_dirty
;
1616 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1618 nfs_wait_bit_killable
,
1623 nfs_commit_clear_lock(NFS_I(inode
));
1625 /* Note: If we exit without ensuring that the commit is complete,
1626 * we must mark the inode as dirty. Otherwise, future calls to
1627 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1628 * that the data is on the disk.
1631 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1635 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1637 struct nfs_inode
*nfsi
= NFS_I(inode
);
1638 int flags
= FLUSH_SYNC
;
1641 /* no commits means nothing needs to be done */
1642 if (!nfsi
->commit_info
.ncommit
)
1645 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1646 /* Don't commit yet if this is a non-blocking flush and there
1647 * are a lot of outstanding writes for this mapping.
1649 if (nfsi
->commit_info
.ncommit
<= (nfsi
->npages
>> 1))
1650 goto out_mark_dirty
;
1652 /* don't wait for the COMMIT response */
1656 ret
= nfs_commit_inode(inode
, flags
);
1658 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1659 if (ret
< wbc
->nr_to_write
)
1660 wbc
->nr_to_write
-= ret
;
1662 wbc
->nr_to_write
= 0;
1667 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1671 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1677 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1681 ret
= nfs_commit_unstable_pages(inode
, wbc
);
1682 if (ret
>= 0 && test_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(inode
)->flags
)) {
1686 if (wbc
->sync_mode
== WB_SYNC_NONE
)
1689 status
= pnfs_layoutcommit_inode(inode
, sync
);
1697 * flush the inode to disk.
1699 int nfs_wb_all(struct inode
*inode
)
1701 struct writeback_control wbc
= {
1702 .sync_mode
= WB_SYNC_ALL
,
1703 .nr_to_write
= LONG_MAX
,
1705 .range_end
= LLONG_MAX
,
1708 return sync_inode(inode
, &wbc
);
1711 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1713 struct nfs_page
*req
;
1716 BUG_ON(!PageLocked(page
));
1718 wait_on_page_writeback(page
);
1719 req
= nfs_page_find_request(page
);
1722 if (nfs_lock_request(req
)) {
1723 nfs_clear_request_commit(req
);
1724 nfs_inode_remove_request(req
);
1726 * In case nfs_inode_remove_request has marked the
1727 * page as being dirty
1729 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1730 nfs_unlock_and_release_request(req
);
1733 ret
= nfs_wait_on_request(req
);
1734 nfs_release_request(req
);
1742 * Write back all requests on one page - we do this before reading it.
1744 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1746 loff_t range_start
= page_offset(page
);
1747 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1748 struct writeback_control wbc
= {
1749 .sync_mode
= WB_SYNC_ALL
,
1751 .range_start
= range_start
,
1752 .range_end
= range_end
,
1757 wait_on_page_writeback(page
);
1758 if (clear_page_dirty_for_io(page
)) {
1759 ret
= nfs_writepage_locked(page
, &wbc
);
1764 if (!PagePrivate(page
))
1766 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1775 #ifdef CONFIG_MIGRATION
1776 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1777 struct page
*page
, enum migrate_mode mode
)
1780 * If PagePrivate is set, then the page is currently associated with
1781 * an in-progress read or write request. Don't try to migrate it.
1783 * FIXME: we could do this in principle, but we'll need a way to ensure
1784 * that we can safely release the inode reference while holding
1787 if (PagePrivate(page
))
1790 nfs_fscache_release_page(page
, GFP_KERNEL
);
1792 return migrate_page(mapping
, newpage
, page
, mode
);
1796 int __init
nfs_init_writepagecache(void)
1798 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1799 sizeof(struct nfs_write_header
),
1800 0, SLAB_HWCACHE_ALIGN
,
1802 if (nfs_wdata_cachep
== NULL
)
1805 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1807 if (nfs_wdata_mempool
== NULL
)
1810 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1811 sizeof(struct nfs_commit_data
),
1812 0, SLAB_HWCACHE_ALIGN
,
1814 if (nfs_cdata_cachep
== NULL
)
1817 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1819 if (nfs_commit_mempool
== NULL
)
1823 * NFS congestion size, scale with available memory.
1835 * This allows larger machines to have larger/more transfers.
1836 * Limit the default to 256M
1838 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1839 if (nfs_congestion_kb
> 256*1024)
1840 nfs_congestion_kb
= 256*1024;
1845 void nfs_destroy_writepagecache(void)
1847 mempool_destroy(nfs_commit_mempool
);
1848 mempool_destroy(nfs_wdata_mempool
);
1849 kmem_cache_destroy(nfs_wdata_cachep
);