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);
88 static struct nfs_write_data
*nfs_writedata_alloc(struct nfs_pgio_header
*hdr
,
89 unsigned int pagecount
)
91 struct nfs_write_data
*data
, *prealloc
;
93 prealloc
= &container_of(hdr
, struct nfs_write_header
, header
)->rpc_data
;
94 if (prealloc
->header
== NULL
)
97 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
101 if (nfs_pgarray_set(&data
->pages
, pagecount
)) {
103 atomic_inc(&hdr
->refcnt
);
105 if (data
!= prealloc
)
113 void nfs_writehdr_free(struct nfs_pgio_header
*hdr
)
115 struct nfs_write_header
*whdr
= container_of(hdr
, struct nfs_write_header
, header
);
116 mempool_free(whdr
, nfs_wdata_mempool
);
119 void nfs_writedata_release(struct nfs_write_data
*wdata
)
121 struct nfs_pgio_header
*hdr
= wdata
->header
;
122 struct nfs_write_header
*write_header
= container_of(hdr
, struct nfs_write_header
, header
);
124 put_nfs_open_context(wdata
->args
.context
);
125 if (wdata
->pages
.pagevec
!= wdata
->pages
.page_array
)
126 kfree(wdata
->pages
.pagevec
);
127 if (wdata
!= &write_header
->rpc_data
)
130 wdata
->header
= NULL
;
131 if (atomic_dec_and_test(&hdr
->refcnt
))
132 hdr
->completion_ops
->completion(hdr
);
135 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
139 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
142 static struct nfs_page
*nfs_page_find_request_locked(struct page
*page
)
144 struct nfs_page
*req
= NULL
;
146 if (PagePrivate(page
)) {
147 req
= (struct nfs_page
*)page_private(page
);
149 kref_get(&req
->wb_kref
);
154 static struct nfs_page
*nfs_page_find_request(struct page
*page
)
156 struct inode
*inode
= page
->mapping
->host
;
157 struct nfs_page
*req
= NULL
;
159 spin_lock(&inode
->i_lock
);
160 req
= nfs_page_find_request_locked(page
);
161 spin_unlock(&inode
->i_lock
);
165 /* Adjust the file length if we're writing beyond the end */
166 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
168 struct inode
*inode
= page
->mapping
->host
;
172 spin_lock(&inode
->i_lock
);
173 i_size
= i_size_read(inode
);
174 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
175 if (i_size
> 0 && page
->index
< end_index
)
177 end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + ((loff_t
)offset
+count
);
180 i_size_write(inode
, end
);
181 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
183 spin_unlock(&inode
->i_lock
);
186 /* A writeback failed: mark the page as bad, and invalidate the page cache */
187 static void nfs_set_pageerror(struct page
*page
)
190 nfs_zap_mapping(page
->mapping
->host
, page
->mapping
);
193 /* We can set the PG_uptodate flag if we see that a write request
194 * covers the full page.
196 static void nfs_mark_uptodate(struct page
*page
, unsigned int base
, unsigned int count
)
198 if (PageUptodate(page
))
202 if (count
!= nfs_page_length(page
))
204 SetPageUptodate(page
);
207 static int wb_priority(struct writeback_control
*wbc
)
209 if (wbc
->for_reclaim
)
210 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
211 if (wbc
->for_kupdate
|| wbc
->for_background
)
212 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
213 return FLUSH_COND_STABLE
;
217 * NFS congestion control
220 int nfs_congestion_kb
;
222 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
223 #define NFS_CONGESTION_OFF_THRESH \
224 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
226 static int nfs_set_page_writeback(struct page
*page
)
228 int ret
= test_set_page_writeback(page
);
231 struct inode
*inode
= page
->mapping
->host
;
232 struct nfs_server
*nfss
= NFS_SERVER(inode
);
234 if (atomic_long_inc_return(&nfss
->writeback
) >
235 NFS_CONGESTION_ON_THRESH
) {
236 set_bdi_congested(&nfss
->backing_dev_info
,
243 static void nfs_end_page_writeback(struct page
*page
)
245 struct inode
*inode
= page
->mapping
->host
;
246 struct nfs_server
*nfss
= NFS_SERVER(inode
);
248 end_page_writeback(page
);
249 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
250 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
253 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
255 struct inode
*inode
= page
->mapping
->host
;
256 struct nfs_page
*req
;
259 spin_lock(&inode
->i_lock
);
261 req
= nfs_page_find_request_locked(page
);
264 if (nfs_lock_request(req
))
266 /* Note: If we hold the page lock, as is the case in nfs_writepage,
267 * then the call to nfs_lock_request() will always
268 * succeed provided that someone hasn't already marked the
269 * request as dirty (in which case we don't care).
271 spin_unlock(&inode
->i_lock
);
273 ret
= nfs_wait_on_request(req
);
276 nfs_release_request(req
);
279 spin_lock(&inode
->i_lock
);
281 spin_unlock(&inode
->i_lock
);
286 * Find an associated nfs write request, and prepare to flush it out
287 * May return an error if the user signalled nfs_wait_on_request().
289 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
290 struct page
*page
, bool nonblock
)
292 struct nfs_page
*req
;
295 req
= nfs_find_and_lock_request(page
, nonblock
);
302 ret
= nfs_set_page_writeback(page
);
304 BUG_ON(test_bit(PG_CLEAN
, &req
->wb_flags
));
306 if (!nfs_pageio_add_request(pgio
, req
)) {
307 nfs_redirty_request(req
);
308 ret
= pgio
->pg_error
;
314 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
316 struct inode
*inode
= page
->mapping
->host
;
319 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
320 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
322 nfs_pageio_cond_complete(pgio
, page
->index
);
323 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
324 if (ret
== -EAGAIN
) {
325 redirty_page_for_writepage(wbc
, page
);
332 * Write an mmapped page to the server.
334 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
336 struct nfs_pageio_descriptor pgio
;
339 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
),
340 &nfs_async_write_completion_ops
);
341 err
= nfs_do_writepage(page
, wbc
, &pgio
);
342 nfs_pageio_complete(&pgio
);
345 if (pgio
.pg_error
< 0)
346 return pgio
.pg_error
;
350 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
354 ret
= nfs_writepage_locked(page
, wbc
);
359 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
363 ret
= nfs_do_writepage(page
, wbc
, data
);
368 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
370 struct inode
*inode
= mapping
->host
;
371 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
372 struct nfs_pageio_descriptor pgio
;
375 /* Stop dirtying of new pages while we sync */
376 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
377 nfs_wait_bit_killable
, TASK_KILLABLE
);
381 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
383 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
),
384 &nfs_async_write_completion_ops
);
385 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
386 nfs_pageio_complete(&pgio
);
388 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
389 smp_mb__after_clear_bit();
390 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
403 * Insert a write request into an inode
405 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
407 struct nfs_inode
*nfsi
= NFS_I(inode
);
409 /* Lock the request! */
410 nfs_lock_request(req
);
412 spin_lock(&inode
->i_lock
);
413 if (!nfsi
->npages
&& nfs_have_delegation(inode
, FMODE_WRITE
))
415 set_bit(PG_MAPPED
, &req
->wb_flags
);
416 SetPagePrivate(req
->wb_page
);
417 set_page_private(req
->wb_page
, (unsigned long)req
);
419 kref_get(&req
->wb_kref
);
420 spin_unlock(&inode
->i_lock
);
424 * Remove a write request from an inode
426 static void nfs_inode_remove_request(struct nfs_page
*req
)
428 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
429 struct nfs_inode
*nfsi
= NFS_I(inode
);
431 BUG_ON (!NFS_WBACK_BUSY(req
));
433 spin_lock(&inode
->i_lock
);
434 set_page_private(req
->wb_page
, 0);
435 ClearPagePrivate(req
->wb_page
);
436 clear_bit(PG_MAPPED
, &req
->wb_flags
);
438 spin_unlock(&inode
->i_lock
);
439 nfs_release_request(req
);
443 nfs_mark_request_dirty(struct nfs_page
*req
)
445 __set_page_dirty_nobuffers(req
->wb_page
);
448 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
450 * nfs_request_add_commit_list - add request to a commit list
451 * @req: pointer to a struct nfs_page
452 * @dst: commit list head
453 * @cinfo: holds list lock and accounting info
455 * This sets the PG_CLEAN bit, updates the cinfo count of
456 * number of outstanding requests requiring a commit as well as
459 * The caller must _not_ hold the cinfo->lock, but must be
460 * holding the nfs_page lock.
463 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
464 struct nfs_commit_info
*cinfo
)
466 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
467 spin_lock(cinfo
->lock
);
468 nfs_list_add_request(req
, dst
);
469 cinfo
->mds
->ncommit
++;
470 spin_unlock(cinfo
->lock
);
472 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
473 inc_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
,
475 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
479 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
482 * nfs_request_remove_commit_list - Remove request from a commit list
483 * @req: pointer to a nfs_page
484 * @cinfo: holds list lock and accounting info
486 * This clears the PG_CLEAN bit, and updates the cinfo's count of
487 * number of outstanding requests requiring a commit
488 * It does not update the MM page stats.
490 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
493 nfs_request_remove_commit_list(struct nfs_page
*req
,
494 struct nfs_commit_info
*cinfo
)
496 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
498 nfs_list_remove_request(req
);
499 cinfo
->mds
->ncommit
--;
501 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
503 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
506 cinfo
->lock
= &inode
->i_lock
;
507 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
508 cinfo
->ds
= pnfs_get_ds_info(inode
);
510 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
513 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
515 struct nfs_direct_req
*dreq
)
518 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
520 nfs_init_cinfo_from_inode(cinfo
, inode
);
522 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
525 * Add a request to the inode's commit list.
528 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
529 struct nfs_commit_info
*cinfo
)
531 if (pnfs_mark_request_commit(req
, lseg
, cinfo
))
533 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
537 nfs_clear_page_commit(struct page
*page
)
539 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
540 dec_bdi_stat(page
->mapping
->backing_dev_info
, BDI_RECLAIMABLE
);
544 nfs_clear_request_commit(struct nfs_page
*req
)
546 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
547 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
548 struct nfs_commit_info cinfo
;
550 nfs_init_cinfo_from_inode(&cinfo
, inode
);
551 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
552 spin_lock(cinfo
.lock
);
553 nfs_request_remove_commit_list(req
, &cinfo
);
554 spin_unlock(cinfo
.lock
);
556 nfs_clear_page_commit(req
->wb_page
);
561 int nfs_write_need_commit(struct nfs_write_data
*data
)
563 if (data
->verf
.committed
== NFS_DATA_SYNC
)
564 return data
->header
->lseg
== NULL
;
565 return data
->verf
.committed
!= NFS_FILE_SYNC
;
569 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
574 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
576 struct nfs_direct_req
*dreq
)
581 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
582 struct nfs_commit_info
*cinfo
)
587 nfs_clear_request_commit(struct nfs_page
*req
)
592 int nfs_write_need_commit(struct nfs_write_data
*data
)
599 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
601 struct nfs_commit_info cinfo
;
602 unsigned long bytes
= 0;
604 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
606 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
607 while (!list_empty(&hdr
->pages
)) {
608 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
610 bytes
+= req
->wb_bytes
;
611 nfs_list_remove_request(req
);
612 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
613 (hdr
->good_bytes
< bytes
)) {
614 nfs_set_pageerror(req
->wb_page
);
615 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
618 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
)) {
619 nfs_mark_request_dirty(req
);
622 if (test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
)) {
623 memcpy(&req
->wb_verf
, hdr
->verf
, sizeof(req
->wb_verf
));
624 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
);
628 nfs_inode_remove_request(req
);
630 nfs_unlock_request(req
);
631 nfs_end_page_writeback(req
->wb_page
);
632 nfs_release_request(req
);
638 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
640 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
642 return cinfo
->mds
->ncommit
;
645 /* cinfo->lock held by caller */
647 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
648 struct nfs_commit_info
*cinfo
, int max
)
650 struct nfs_page
*req
, *tmp
;
653 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
654 if (!nfs_lock_request(req
))
656 kref_get(&req
->wb_kref
);
657 if (cond_resched_lock(cinfo
->lock
))
658 list_safe_reset_next(req
, tmp
, wb_list
);
659 nfs_request_remove_commit_list(req
, cinfo
);
660 nfs_list_add_request(req
, dst
);
662 if ((ret
== max
) && !cinfo
->dreq
)
669 * nfs_scan_commit - Scan an inode for commit requests
670 * @inode: NFS inode to scan
671 * @dst: mds destination list
672 * @cinfo: mds and ds lists of reqs ready to commit
674 * Moves requests from the inode's 'commit' request list.
675 * The requests are *not* checked to ensure that they form a contiguous set.
678 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
679 struct nfs_commit_info
*cinfo
)
683 spin_lock(cinfo
->lock
);
684 if (cinfo
->mds
->ncommit
> 0) {
685 const int max
= INT_MAX
;
687 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
689 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
691 spin_unlock(cinfo
->lock
);
696 static unsigned long nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
701 int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
702 struct nfs_commit_info
*cinfo
)
709 * Search for an existing write request, and attempt to update
710 * it to reflect a new dirty region on a given page.
712 * If the attempt fails, then the existing request is flushed out
715 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
720 struct nfs_page
*req
;
725 if (!PagePrivate(page
))
728 end
= offset
+ bytes
;
729 spin_lock(&inode
->i_lock
);
732 req
= nfs_page_find_request_locked(page
);
736 rqend
= req
->wb_offset
+ req
->wb_bytes
;
738 * Tell the caller to flush out the request if
739 * the offsets are non-contiguous.
740 * Note: nfs_flush_incompatible() will already
741 * have flushed out requests having wrong owners.
744 || end
< req
->wb_offset
)
747 if (nfs_lock_request(req
))
750 /* The request is locked, so wait and then retry */
751 spin_unlock(&inode
->i_lock
);
752 error
= nfs_wait_on_request(req
);
753 nfs_release_request(req
);
756 spin_lock(&inode
->i_lock
);
759 /* Okay, the request matches. Update the region */
760 if (offset
< req
->wb_offset
) {
761 req
->wb_offset
= offset
;
762 req
->wb_pgbase
= offset
;
765 req
->wb_bytes
= end
- req
->wb_offset
;
767 req
->wb_bytes
= rqend
- req
->wb_offset
;
769 spin_unlock(&inode
->i_lock
);
771 nfs_clear_request_commit(req
);
774 spin_unlock(&inode
->i_lock
);
775 nfs_release_request(req
);
776 error
= nfs_wb_page(inode
, page
);
778 return ERR_PTR(error
);
782 * Try to update an existing write request, or create one if there is none.
784 * Note: Should always be called with the Page Lock held to prevent races
785 * if we have to add a new request. Also assumes that the caller has
786 * already called nfs_flush_incompatible() if necessary.
788 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
789 struct page
*page
, unsigned int offset
, unsigned int bytes
)
791 struct inode
*inode
= page
->mapping
->host
;
792 struct nfs_page
*req
;
794 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
797 req
= nfs_create_request(ctx
, inode
, page
, offset
, bytes
);
800 nfs_inode_add_request(inode
, req
);
805 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
806 unsigned int offset
, unsigned int count
)
808 struct nfs_page
*req
;
810 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
813 /* Update file length */
814 nfs_grow_file(page
, offset
, count
);
815 nfs_mark_uptodate(page
, req
->wb_pgbase
, req
->wb_bytes
);
816 nfs_mark_request_dirty(req
);
817 nfs_unlock_and_release_request(req
);
821 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
823 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
824 struct nfs_page
*req
;
825 int do_flush
, status
;
827 * Look for a request corresponding to this page. If there
828 * is one, and it belongs to another file, we flush it out
829 * before we try to copy anything into the page. Do this
830 * due to the lack of an ACCESS-type call in NFSv2.
831 * Also do the same if we find a request from an existing
835 req
= nfs_page_find_request(page
);
838 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
||
839 req
->wb_lock_context
->lockowner
!= current
->files
||
840 req
->wb_lock_context
->pid
!= current
->tgid
;
841 nfs_release_request(req
);
844 status
= nfs_wb_page(page
->mapping
->host
, page
);
845 } while (status
== 0);
850 * If the page cache is marked as unsafe or invalid, then we can't rely on
851 * the PageUptodate() flag. In this case, we will need to turn off
852 * write optimisations that depend on the page contents being correct.
854 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
856 if (nfs_have_delegated_attributes(inode
))
858 if (NFS_I(inode
)->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
861 return PageUptodate(page
) != 0;
865 * Update and possibly write a cached page of an NFS file.
867 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
868 * things with a page scheduled for an RPC call (e.g. invalidate it).
870 int nfs_updatepage(struct file
*file
, struct page
*page
,
871 unsigned int offset
, unsigned int count
)
873 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
874 struct inode
*inode
= page
->mapping
->host
;
877 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
879 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
880 file
->f_path
.dentry
->d_parent
->d_name
.name
,
881 file
->f_path
.dentry
->d_name
.name
, count
,
882 (long long)(page_offset(page
) + offset
));
884 /* If we're not using byte range locks, and we know the page
885 * is up to date, it may be more efficient to extend the write
886 * to cover the entire page in order to avoid fragmentation
889 if (nfs_write_pageuptodate(page
, inode
) &&
890 inode
->i_flock
== NULL
&&
891 !(file
->f_flags
& O_DSYNC
)) {
892 count
= max(count
+ offset
, nfs_page_length(page
));
896 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
898 nfs_set_pageerror(page
);
900 __set_page_dirty_nobuffers(page
);
902 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
903 status
, (long long)i_size_read(inode
));
907 static int flush_task_priority(int how
)
909 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
911 return RPC_PRIORITY_HIGH
;
913 return RPC_PRIORITY_LOW
;
915 return RPC_PRIORITY_NORMAL
;
918 int nfs_initiate_write(struct rpc_clnt
*clnt
,
919 struct nfs_write_data
*data
,
920 const struct rpc_call_ops
*call_ops
,
923 struct inode
*inode
= data
->header
->inode
;
924 int priority
= flush_task_priority(how
);
925 struct rpc_task
*task
;
926 struct rpc_message msg
= {
927 .rpc_argp
= &data
->args
,
928 .rpc_resp
= &data
->res
,
929 .rpc_cred
= data
->header
->cred
,
931 struct rpc_task_setup task_setup_data
= {
935 .callback_ops
= call_ops
,
936 .callback_data
= data
,
937 .workqueue
= nfsiod_workqueue
,
938 .flags
= RPC_TASK_ASYNC
| flags
,
939 .priority
= priority
,
943 /* Set up the initial task struct. */
944 NFS_PROTO(inode
)->write_setup(data
, &msg
);
946 dprintk("NFS: %5u initiated write call "
947 "(req %s/%lld, %u bytes @ offset %llu)\n",
950 (long long)NFS_FILEID(inode
),
952 (unsigned long long)data
->args
.offset
);
954 task
= rpc_run_task(&task_setup_data
);
959 if (how
& FLUSH_SYNC
) {
960 ret
= rpc_wait_for_completion_task(task
);
962 ret
= task
->tk_status
;
968 EXPORT_SYMBOL_GPL(nfs_initiate_write
);
971 * Set up the argument/result storage required for the RPC call.
973 static void nfs_write_rpcsetup(struct nfs_write_data
*data
,
974 unsigned int count
, unsigned int offset
,
975 int how
, struct nfs_commit_info
*cinfo
)
977 struct nfs_page
*req
= data
->header
->req
;
979 /* Set up the RPC argument and reply structs
980 * NB: take care not to mess about with data->commit et al. */
982 data
->args
.fh
= NFS_FH(data
->header
->inode
);
983 data
->args
.offset
= req_offset(req
) + offset
;
984 /* pnfs_set_layoutcommit needs this */
985 data
->mds_offset
= data
->args
.offset
;
986 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
987 data
->args
.pages
= data
->pages
.pagevec
;
988 data
->args
.count
= count
;
989 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
990 data
->args
.lock_context
= req
->wb_lock_context
;
991 data
->args
.stable
= NFS_UNSTABLE
;
992 switch (how
& (FLUSH_STABLE
| FLUSH_COND_STABLE
)) {
995 case FLUSH_COND_STABLE
:
996 if (nfs_reqs_to_commit(cinfo
))
999 data
->args
.stable
= NFS_FILE_SYNC
;
1002 data
->res
.fattr
= &data
->fattr
;
1003 data
->res
.count
= count
;
1004 data
->res
.verf
= &data
->verf
;
1005 nfs_fattr_init(&data
->fattr
);
1008 static int nfs_do_write(struct nfs_write_data
*data
,
1009 const struct rpc_call_ops
*call_ops
,
1012 struct inode
*inode
= data
->header
->inode
;
1014 return nfs_initiate_write(NFS_CLIENT(inode
), data
, call_ops
, how
, 0);
1017 static int nfs_do_multiple_writes(struct list_head
*head
,
1018 const struct rpc_call_ops
*call_ops
,
1021 struct nfs_write_data
*data
;
1024 while (!list_empty(head
)) {
1027 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1028 list_del_init(&data
->list
);
1030 ret2
= nfs_do_write(data
, call_ops
, how
);
1037 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1038 * call this on each, which will prepare them to be retried on next
1039 * writeback using standard nfs.
1041 static void nfs_redirty_request(struct nfs_page
*req
)
1043 nfs_mark_request_dirty(req
);
1044 nfs_unlock_request(req
);
1045 nfs_end_page_writeback(req
->wb_page
);
1046 nfs_release_request(req
);
1049 static void nfs_async_write_error(struct list_head
*head
)
1051 struct nfs_page
*req
;
1053 while (!list_empty(head
)) {
1054 req
= nfs_list_entry(head
->next
);
1055 nfs_list_remove_request(req
);
1056 nfs_redirty_request(req
);
1060 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1061 .error_cleanup
= nfs_async_write_error
,
1062 .completion
= nfs_write_completion
,
1065 static void nfs_flush_error(struct nfs_pageio_descriptor
*desc
,
1066 struct nfs_pgio_header
*hdr
)
1068 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
1069 while (!list_empty(&hdr
->rpc_list
)) {
1070 struct nfs_write_data
*data
= list_first_entry(&hdr
->rpc_list
,
1071 struct nfs_write_data
, list
);
1072 list_del(&data
->list
);
1073 nfs_writedata_release(data
);
1075 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1079 * Generate multiple small requests to write out a single
1080 * contiguous dirty area on one page.
1082 static int nfs_flush_multi(struct nfs_pageio_descriptor
*desc
,
1083 struct nfs_pgio_header
*hdr
)
1085 struct nfs_page
*req
= hdr
->req
;
1086 struct page
*page
= req
->wb_page
;
1087 struct nfs_write_data
*data
;
1088 size_t wsize
= desc
->pg_bsize
, nbytes
;
1089 unsigned int offset
;
1091 struct nfs_commit_info cinfo
;
1093 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1095 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1096 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
) ||
1097 desc
->pg_count
> wsize
))
1098 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1102 nbytes
= desc
->pg_count
;
1104 size_t len
= min(nbytes
, wsize
);
1106 data
= nfs_writedata_alloc(hdr
, 1);
1108 nfs_flush_error(desc
, hdr
);
1111 data
->pages
.pagevec
[0] = page
;
1112 nfs_write_rpcsetup(data
, len
, offset
, desc
->pg_ioflags
, &cinfo
);
1113 list_add(&data
->list
, &hdr
->rpc_list
);
1117 } while (nbytes
!= 0);
1118 nfs_list_remove_request(req
);
1119 nfs_list_add_request(req
, &hdr
->pages
);
1120 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1125 * Create an RPC task for the given write request and kick it.
1126 * The page must have been locked by the caller.
1128 * It may happen that the page we're passed is not marked dirty.
1129 * This is the case if nfs_updatepage detects a conflicting request
1130 * that has been written but not committed.
1132 static int nfs_flush_one(struct nfs_pageio_descriptor
*desc
,
1133 struct nfs_pgio_header
*hdr
)
1135 struct nfs_page
*req
;
1136 struct page
**pages
;
1137 struct nfs_write_data
*data
;
1138 struct list_head
*head
= &desc
->pg_list
;
1139 struct nfs_commit_info cinfo
;
1141 data
= nfs_writedata_alloc(hdr
, nfs_page_array_len(desc
->pg_base
,
1144 nfs_flush_error(desc
, hdr
);
1148 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1149 pages
= data
->pages
.pagevec
;
1150 while (!list_empty(head
)) {
1151 req
= nfs_list_entry(head
->next
);
1152 nfs_list_remove_request(req
);
1153 nfs_list_add_request(req
, &hdr
->pages
);
1154 *pages
++ = req
->wb_page
;
1157 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1158 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
)))
1159 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1161 /* Set up the argument struct */
1162 nfs_write_rpcsetup(data
, desc
->pg_count
, 0, desc
->pg_ioflags
, &cinfo
);
1163 list_add(&data
->list
, &hdr
->rpc_list
);
1164 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1168 int nfs_generic_flush(struct nfs_pageio_descriptor
*desc
,
1169 struct nfs_pgio_header
*hdr
)
1171 if (desc
->pg_bsize
< PAGE_CACHE_SIZE
)
1172 return nfs_flush_multi(desc
, hdr
);
1173 return nfs_flush_one(desc
, hdr
);
1176 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1178 struct nfs_write_header
*whdr
;
1179 struct nfs_pgio_header
*hdr
;
1182 whdr
= nfs_writehdr_alloc();
1184 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1187 hdr
= &whdr
->header
;
1188 nfs_pgheader_init(desc
, hdr
, nfs_writehdr_free
);
1189 atomic_inc(&hdr
->refcnt
);
1190 ret
= nfs_generic_flush(desc
, hdr
);
1192 ret
= nfs_do_multiple_writes(&hdr
->rpc_list
,
1193 desc
->pg_rpc_callops
,
1195 if (atomic_dec_and_test(&hdr
->refcnt
))
1196 hdr
->completion_ops
->completion(hdr
);
1200 static const struct nfs_pageio_ops nfs_pageio_write_ops
= {
1201 .pg_test
= nfs_generic_pg_test
,
1202 .pg_doio
= nfs_generic_pg_writepages
,
1205 void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor
*pgio
,
1206 struct inode
*inode
, int ioflags
,
1207 const struct nfs_pgio_completion_ops
*compl_ops
)
1209 nfs_pageio_init(pgio
, inode
, &nfs_pageio_write_ops
, compl_ops
,
1210 NFS_SERVER(inode
)->wsize
, ioflags
);
1213 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1215 pgio
->pg_ops
= &nfs_pageio_write_ops
;
1216 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1218 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1220 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1221 struct inode
*inode
, int ioflags
,
1222 const struct nfs_pgio_completion_ops
*compl_ops
)
1224 if (!pnfs_pageio_init_write(pgio
, inode
, ioflags
, compl_ops
))
1225 nfs_pageio_init_write_mds(pgio
, inode
, ioflags
, compl_ops
);
1228 void nfs_write_prepare(struct rpc_task
*task
, void *calldata
)
1230 struct nfs_write_data
*data
= calldata
;
1231 NFS_PROTO(data
->header
->inode
)->write_rpc_prepare(task
, data
);
1234 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1236 struct nfs_commit_data
*data
= calldata
;
1238 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1242 * Handle a write reply that flushes a whole page.
1244 * FIXME: There is an inherent race with invalidate_inode_pages and
1245 * writebacks since the page->count is kept > 1 for as long
1246 * as the page has a write request pending.
1248 static void nfs_writeback_done_common(struct rpc_task
*task
, void *calldata
)
1250 struct nfs_write_data
*data
= calldata
;
1252 nfs_writeback_done(task
, data
);
1255 static void nfs_writeback_release_common(void *calldata
)
1257 struct nfs_write_data
*data
= calldata
;
1258 struct nfs_pgio_header
*hdr
= data
->header
;
1259 int status
= data
->task
.tk_status
;
1261 if ((status
>= 0) && nfs_write_need_commit(data
)) {
1262 spin_lock(&hdr
->lock
);
1263 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
))
1265 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
))
1266 memcpy(hdr
->verf
, &data
->verf
, sizeof(*hdr
->verf
));
1267 else if (memcmp(hdr
->verf
, &data
->verf
, sizeof(*hdr
->verf
)))
1268 set_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
);
1269 spin_unlock(&hdr
->lock
);
1271 nfs_writedata_release(data
);
1274 static const struct rpc_call_ops nfs_write_common_ops
= {
1275 .rpc_call_prepare
= nfs_write_prepare
,
1276 .rpc_call_done
= nfs_writeback_done_common
,
1277 .rpc_release
= nfs_writeback_release_common
,
1282 * This function is called when the WRITE call is complete.
1284 void nfs_writeback_done(struct rpc_task
*task
, struct nfs_write_data
*data
)
1286 struct nfs_writeargs
*argp
= &data
->args
;
1287 struct nfs_writeres
*resp
= &data
->res
;
1288 struct inode
*inode
= data
->header
->inode
;
1291 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1292 task
->tk_pid
, task
->tk_status
);
1295 * ->write_done will attempt to use post-op attributes to detect
1296 * conflicting writes by other clients. A strict interpretation
1297 * of close-to-open would allow us to continue caching even if
1298 * another writer had changed the file, but some applications
1299 * depend on tighter cache coherency when writing.
1301 status
= NFS_PROTO(inode
)->write_done(task
, data
);
1304 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, resp
->count
);
1306 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1307 if (resp
->verf
->committed
< argp
->stable
&& task
->tk_status
>= 0) {
1308 /* We tried a write call, but the server did not
1309 * commit data to stable storage even though we
1311 * Note: There is a known bug in Tru64 < 5.0 in which
1312 * the server reports NFS_DATA_SYNC, but performs
1313 * NFS_FILE_SYNC. We therefore implement this checking
1314 * as a dprintk() in order to avoid filling syslog.
1316 static unsigned long complain
;
1318 /* Note this will print the MDS for a DS write */
1319 if (time_before(complain
, jiffies
)) {
1320 dprintk("NFS: faulty NFS server %s:"
1321 " (committed = %d) != (stable = %d)\n",
1322 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1323 resp
->verf
->committed
, argp
->stable
);
1324 complain
= jiffies
+ 300 * HZ
;
1328 if (task
->tk_status
< 0)
1329 nfs_set_pgio_error(data
->header
, task
->tk_status
, argp
->offset
);
1330 else if (resp
->count
< argp
->count
) {
1331 static unsigned long complain
;
1333 /* This a short write! */
1334 nfs_inc_stats(inode
, NFSIOS_SHORTWRITE
);
1336 /* Has the server at least made some progress? */
1337 if (resp
->count
== 0) {
1338 if (time_before(complain
, jiffies
)) {
1340 "NFS: Server wrote zero bytes, expected %u.\n",
1342 complain
= jiffies
+ 300 * HZ
;
1344 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
1345 task
->tk_status
= -EIO
;
1348 /* Was this an NFSv2 write or an NFSv3 stable write? */
1349 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1350 /* Resend from where the server left off */
1351 data
->mds_offset
+= resp
->count
;
1352 argp
->offset
+= resp
->count
;
1353 argp
->pgbase
+= resp
->count
;
1354 argp
->count
-= resp
->count
;
1356 /* Resend as a stable write in order to avoid
1357 * headaches in the case of a server crash.
1359 argp
->stable
= NFS_FILE_SYNC
;
1361 rpc_restart_call_prepare(task
);
1366 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1367 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1371 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1375 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1377 nfs_wait_bit_killable
,
1379 return (ret
< 0) ? ret
: 1;
1382 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1384 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1385 smp_mb__after_clear_bit();
1386 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1389 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1391 put_nfs_open_context(data
->context
);
1392 nfs_commit_free(data
);
1394 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1396 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1397 const struct rpc_call_ops
*call_ops
,
1400 struct rpc_task
*task
;
1401 int priority
= flush_task_priority(how
);
1402 struct rpc_message msg
= {
1403 .rpc_argp
= &data
->args
,
1404 .rpc_resp
= &data
->res
,
1405 .rpc_cred
= data
->cred
,
1407 struct rpc_task_setup task_setup_data
= {
1408 .task
= &data
->task
,
1410 .rpc_message
= &msg
,
1411 .callback_ops
= call_ops
,
1412 .callback_data
= data
,
1413 .workqueue
= nfsiod_workqueue
,
1414 .flags
= RPC_TASK_ASYNC
| flags
,
1415 .priority
= priority
,
1417 /* Set up the initial task struct. */
1418 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1420 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1422 task
= rpc_run_task(&task_setup_data
);
1424 return PTR_ERR(task
);
1425 if (how
& FLUSH_SYNC
)
1426 rpc_wait_for_completion_task(task
);
1430 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1433 * Set up the argument/result storage required for the RPC call.
1435 void nfs_init_commit(struct nfs_commit_data
*data
,
1436 struct list_head
*head
,
1437 struct pnfs_layout_segment
*lseg
,
1438 struct nfs_commit_info
*cinfo
)
1440 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1441 struct inode
*inode
= first
->wb_context
->dentry
->d_inode
;
1443 /* Set up the RPC argument and reply structs
1444 * NB: take care not to mess about with data->commit et al. */
1446 list_splice_init(head
, &data
->pages
);
1448 data
->inode
= inode
;
1449 data
->cred
= first
->wb_context
->cred
;
1450 data
->lseg
= lseg
; /* reference transferred */
1451 data
->mds_ops
= &nfs_commit_ops
;
1452 data
->completion_ops
= cinfo
->completion_ops
;
1453 data
->dreq
= cinfo
->dreq
;
1455 data
->args
.fh
= NFS_FH(data
->inode
);
1456 /* Note: we always request a commit of the entire inode */
1457 data
->args
.offset
= 0;
1458 data
->args
.count
= 0;
1459 data
->context
= get_nfs_open_context(first
->wb_context
);
1460 data
->res
.fattr
= &data
->fattr
;
1461 data
->res
.verf
= &data
->verf
;
1462 nfs_fattr_init(&data
->fattr
);
1464 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1466 void nfs_retry_commit(struct list_head
*page_list
,
1467 struct pnfs_layout_segment
*lseg
,
1468 struct nfs_commit_info
*cinfo
)
1470 struct nfs_page
*req
;
1472 while (!list_empty(page_list
)) {
1473 req
= nfs_list_entry(page_list
->next
);
1474 nfs_list_remove_request(req
);
1475 nfs_mark_request_commit(req
, lseg
, cinfo
);
1477 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1478 dec_bdi_stat(req
->wb_page
->mapping
->backing_dev_info
,
1481 nfs_unlock_and_release_request(req
);
1484 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1487 * Commit dirty pages
1490 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1491 struct nfs_commit_info
*cinfo
)
1493 struct nfs_commit_data
*data
;
1495 data
= nfs_commitdata_alloc();
1500 /* Set up the argument struct */
1501 nfs_init_commit(data
, head
, NULL
, cinfo
);
1502 atomic_inc(&cinfo
->mds
->rpcs_out
);
1503 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, data
->mds_ops
,
1506 nfs_retry_commit(head
, NULL
, cinfo
);
1507 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1512 * COMMIT call returned
1514 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1516 struct nfs_commit_data
*data
= calldata
;
1518 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1519 task
->tk_pid
, task
->tk_status
);
1521 /* Call the NFS version-specific code */
1522 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1525 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1527 struct nfs_page
*req
;
1528 int status
= data
->task
.tk_status
;
1529 struct nfs_commit_info cinfo
;
1531 while (!list_empty(&data
->pages
)) {
1532 req
= nfs_list_entry(data
->pages
.next
);
1533 nfs_list_remove_request(req
);
1534 nfs_clear_page_commit(req
->wb_page
);
1536 dprintk("NFS: commit (%s/%lld %d@%lld)",
1537 req
->wb_context
->dentry
->d_sb
->s_id
,
1538 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
1540 (long long)req_offset(req
));
1542 nfs_context_set_write_error(req
->wb_context
, status
);
1543 nfs_inode_remove_request(req
);
1544 dprintk(", error = %d\n", status
);
1548 /* Okay, COMMIT succeeded, apparently. Check the verifier
1549 * returned by the server against all stored verfs. */
1550 if (!memcmp(req
->wb_verf
.verifier
, data
->verf
.verifier
, sizeof(data
->verf
.verifier
))) {
1551 /* We have a match */
1552 nfs_inode_remove_request(req
);
1556 /* We have a mismatch. Write the page again */
1557 dprintk(" mismatch\n");
1558 nfs_mark_request_dirty(req
);
1560 nfs_unlock_and_release_request(req
);
1562 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1563 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1564 nfs_commit_clear_lock(NFS_I(data
->inode
));
1567 static void nfs_commit_release(void *calldata
)
1569 struct nfs_commit_data
*data
= calldata
;
1571 data
->completion_ops
->completion(data
);
1572 nfs_commitdata_release(calldata
);
1575 static const struct rpc_call_ops nfs_commit_ops
= {
1576 .rpc_call_prepare
= nfs_commit_prepare
,
1577 .rpc_call_done
= nfs_commit_done
,
1578 .rpc_release
= nfs_commit_release
,
1581 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1582 .completion
= nfs_commit_release_pages
,
1583 .error_cleanup
= nfs_commit_clear_lock
,
1586 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1587 int how
, struct nfs_commit_info
*cinfo
)
1591 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1592 if (status
== PNFS_NOT_ATTEMPTED
)
1593 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1597 int nfs_commit_inode(struct inode
*inode
, int how
)
1600 struct nfs_commit_info cinfo
;
1601 int may_wait
= how
& FLUSH_SYNC
;
1604 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1606 goto out_mark_dirty
;
1607 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1608 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1612 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1616 goto out_mark_dirty
;
1617 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1619 nfs_wait_bit_killable
,
1624 nfs_commit_clear_lock(NFS_I(inode
));
1626 /* Note: If we exit without ensuring that the commit is complete,
1627 * we must mark the inode as dirty. Otherwise, future calls to
1628 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1629 * that the data is on the disk.
1632 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1636 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1638 struct nfs_inode
*nfsi
= NFS_I(inode
);
1639 int flags
= FLUSH_SYNC
;
1642 /* no commits means nothing needs to be done */
1643 if (!nfsi
->commit_info
.ncommit
)
1646 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1647 /* Don't commit yet if this is a non-blocking flush and there
1648 * are a lot of outstanding writes for this mapping.
1650 if (nfsi
->commit_info
.ncommit
<= (nfsi
->npages
>> 1))
1651 goto out_mark_dirty
;
1653 /* don't wait for the COMMIT response */
1657 ret
= nfs_commit_inode(inode
, flags
);
1659 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1660 if (ret
< wbc
->nr_to_write
)
1661 wbc
->nr_to_write
-= ret
;
1663 wbc
->nr_to_write
= 0;
1668 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1672 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1678 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1682 ret
= nfs_commit_unstable_pages(inode
, wbc
);
1683 if (ret
>= 0 && test_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(inode
)->flags
)) {
1687 if (wbc
->sync_mode
== WB_SYNC_NONE
)
1690 status
= pnfs_layoutcommit_inode(inode
, sync
);
1698 * flush the inode to disk.
1700 int nfs_wb_all(struct inode
*inode
)
1702 struct writeback_control wbc
= {
1703 .sync_mode
= WB_SYNC_ALL
,
1704 .nr_to_write
= LONG_MAX
,
1706 .range_end
= LLONG_MAX
,
1709 return sync_inode(inode
, &wbc
);
1712 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1714 struct nfs_page
*req
;
1717 BUG_ON(!PageLocked(page
));
1719 wait_on_page_writeback(page
);
1720 req
= nfs_page_find_request(page
);
1723 if (nfs_lock_request(req
)) {
1724 nfs_clear_request_commit(req
);
1725 nfs_inode_remove_request(req
);
1727 * In case nfs_inode_remove_request has marked the
1728 * page as being dirty
1730 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1731 nfs_unlock_and_release_request(req
);
1734 ret
= nfs_wait_on_request(req
);
1735 nfs_release_request(req
);
1743 * Write back all requests on one page - we do this before reading it.
1745 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1747 loff_t range_start
= page_offset(page
);
1748 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1749 struct writeback_control wbc
= {
1750 .sync_mode
= WB_SYNC_ALL
,
1752 .range_start
= range_start
,
1753 .range_end
= range_end
,
1758 wait_on_page_writeback(page
);
1759 if (clear_page_dirty_for_io(page
)) {
1760 ret
= nfs_writepage_locked(page
, &wbc
);
1765 if (!PagePrivate(page
))
1767 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1776 #ifdef CONFIG_MIGRATION
1777 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1778 struct page
*page
, enum migrate_mode mode
)
1781 * If PagePrivate is set, then the page is currently associated with
1782 * an in-progress read or write request. Don't try to migrate it.
1784 * FIXME: we could do this in principle, but we'll need a way to ensure
1785 * that we can safely release the inode reference while holding
1788 if (PagePrivate(page
))
1791 nfs_fscache_release_page(page
, GFP_KERNEL
);
1793 return migrate_page(mapping
, newpage
, page
, mode
);
1797 int __init
nfs_init_writepagecache(void)
1799 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1800 sizeof(struct nfs_write_header
),
1801 0, SLAB_HWCACHE_ALIGN
,
1803 if (nfs_wdata_cachep
== NULL
)
1806 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1808 if (nfs_wdata_mempool
== NULL
)
1811 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1812 sizeof(struct nfs_commit_data
),
1813 0, SLAB_HWCACHE_ALIGN
,
1815 if (nfs_cdata_cachep
== NULL
)
1818 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1820 if (nfs_commit_mempool
== NULL
)
1824 * NFS congestion size, scale with available memory.
1836 * This allows larger machines to have larger/more transfers.
1837 * Limit the default to 256M
1839 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1840 if (nfs_congestion_kb
> 256*1024)
1841 nfs_congestion_kb
= 256*1024;
1846 void nfs_destroy_writepagecache(void)
1848 mempool_destroy(nfs_commit_mempool
);
1849 mempool_destroy(nfs_wdata_mempool
);
1850 kmem_cache_destroy(nfs_wdata_cachep
);