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"
36 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
38 #define MIN_POOL_WRITE (32)
39 #define MIN_POOL_COMMIT (4)
42 * Local function declarations
44 static void nfs_redirty_request(struct nfs_page
*req
);
45 static const struct rpc_call_ops nfs_commit_ops
;
46 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
;
47 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
;
48 static const struct nfs_rw_ops nfs_rw_write_ops
;
50 static struct kmem_cache
*nfs_wdata_cachep
;
51 static mempool_t
*nfs_wdata_mempool
;
52 static struct kmem_cache
*nfs_cdata_cachep
;
53 static mempool_t
*nfs_commit_mempool
;
55 struct nfs_commit_data
*nfs_commitdata_alloc(void)
57 struct nfs_commit_data
*p
= mempool_alloc(nfs_commit_mempool
, GFP_NOIO
);
60 memset(p
, 0, sizeof(*p
));
61 INIT_LIST_HEAD(&p
->pages
);
65 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc
);
67 void nfs_commit_free(struct nfs_commit_data
*p
)
69 mempool_free(p
, nfs_commit_mempool
);
71 EXPORT_SYMBOL_GPL(nfs_commit_free
);
73 static struct nfs_rw_header
*nfs_writehdr_alloc(void)
75 struct nfs_rw_header
*p
= mempool_alloc(nfs_wdata_mempool
, GFP_NOIO
);
78 memset(p
, 0, sizeof(*p
));
82 static void nfs_writehdr_free(struct nfs_rw_header
*whdr
)
84 mempool_free(whdr
, nfs_wdata_mempool
);
87 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
91 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
94 static struct nfs_page
*
95 nfs_page_find_request_locked(struct nfs_inode
*nfsi
, struct page
*page
)
97 struct nfs_page
*req
= NULL
;
99 if (PagePrivate(page
))
100 req
= (struct nfs_page
*)page_private(page
);
101 else if (unlikely(PageSwapCache(page
))) {
102 struct nfs_page
*freq
, *t
;
104 /* Linearly search the commit list for the correct req */
105 list_for_each_entry_safe(freq
, t
, &nfsi
->commit_info
.list
, wb_list
) {
106 if (freq
->wb_page
== page
) {
114 kref_get(&req
->wb_kref
);
119 static struct nfs_page
*nfs_page_find_request(struct page
*page
)
121 struct inode
*inode
= page_file_mapping(page
)->host
;
122 struct nfs_page
*req
= NULL
;
124 spin_lock(&inode
->i_lock
);
125 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
126 spin_unlock(&inode
->i_lock
);
130 /* Adjust the file length if we're writing beyond the end */
131 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
133 struct inode
*inode
= page_file_mapping(page
)->host
;
137 spin_lock(&inode
->i_lock
);
138 i_size
= i_size_read(inode
);
139 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
140 if (i_size
> 0 && page_file_index(page
) < end_index
)
142 end
= page_file_offset(page
) + ((loff_t
)offset
+count
);
145 i_size_write(inode
, end
);
146 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
148 spin_unlock(&inode
->i_lock
);
151 /* A writeback failed: mark the page as bad, and invalidate the page cache */
152 static void nfs_set_pageerror(struct page
*page
)
154 nfs_zap_mapping(page_file_mapping(page
)->host
, page_file_mapping(page
));
158 * nfs_page_group_search_locked
159 * @head - head request of page group
160 * @page_offset - offset into page
162 * Search page group with head @head to find a request that contains the
163 * page offset @page_offset.
165 * Returns a pointer to the first matching nfs request, or NULL if no
168 * Must be called with the page group lock held
170 static struct nfs_page
*
171 nfs_page_group_search_locked(struct nfs_page
*head
, unsigned int page_offset
)
173 struct nfs_page
*req
;
175 WARN_ON_ONCE(head
!= head
->wb_head
);
176 WARN_ON_ONCE(!test_bit(PG_HEADLOCK
, &head
->wb_head
->wb_flags
));
180 if (page_offset
>= req
->wb_pgbase
&&
181 page_offset
< (req
->wb_pgbase
+ req
->wb_bytes
))
184 req
= req
->wb_this_page
;
185 } while (req
!= head
);
191 * nfs_page_group_covers_page
192 * @head - head request of page group
194 * Return true if the page group with head @head covers the whole page,
195 * returns false otherwise
197 static bool nfs_page_group_covers_page(struct nfs_page
*req
)
199 struct nfs_page
*tmp
;
200 unsigned int pos
= 0;
201 unsigned int len
= nfs_page_length(req
->wb_page
);
203 nfs_page_group_lock(req
);
206 tmp
= nfs_page_group_search_locked(req
->wb_head
, pos
);
208 /* no way this should happen */
209 WARN_ON_ONCE(tmp
->wb_pgbase
!= pos
);
210 pos
+= tmp
->wb_bytes
- (pos
- tmp
->wb_pgbase
);
212 } while (tmp
&& pos
< len
);
214 nfs_page_group_unlock(req
);
215 WARN_ON_ONCE(pos
> len
);
219 /* We can set the PG_uptodate flag if we see that a write request
220 * covers the full page.
222 static void nfs_mark_uptodate(struct nfs_page
*req
)
224 if (PageUptodate(req
->wb_page
))
226 if (!nfs_page_group_covers_page(req
))
228 SetPageUptodate(req
->wb_page
);
231 static int wb_priority(struct writeback_control
*wbc
)
233 if (wbc
->for_reclaim
)
234 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
235 if (wbc
->for_kupdate
|| wbc
->for_background
)
236 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
237 return FLUSH_COND_STABLE
;
241 * NFS congestion control
244 int nfs_congestion_kb
;
246 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
247 #define NFS_CONGESTION_OFF_THRESH \
248 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
250 static void nfs_set_page_writeback(struct page
*page
)
252 struct nfs_server
*nfss
= NFS_SERVER(page_file_mapping(page
)->host
);
253 int ret
= test_set_page_writeback(page
);
255 WARN_ON_ONCE(ret
!= 0);
257 if (atomic_long_inc_return(&nfss
->writeback
) >
258 NFS_CONGESTION_ON_THRESH
) {
259 set_bdi_congested(&nfss
->backing_dev_info
,
264 static void nfs_end_page_writeback(struct nfs_page
*req
)
266 struct inode
*inode
= page_file_mapping(req
->wb_page
)->host
;
267 struct nfs_server
*nfss
= NFS_SERVER(inode
);
269 if (!nfs_page_group_sync_on_bit(req
, PG_WB_END
))
272 end_page_writeback(req
->wb_page
);
273 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
274 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
277 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
279 struct inode
*inode
= page_file_mapping(page
)->host
;
280 struct nfs_page
*req
;
283 spin_lock(&inode
->i_lock
);
285 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
288 if (nfs_lock_request(req
))
290 /* Note: If we hold the page lock, as is the case in nfs_writepage,
291 * then the call to nfs_lock_request() will always
292 * succeed provided that someone hasn't already marked the
293 * request as dirty (in which case we don't care).
295 spin_unlock(&inode
->i_lock
);
297 ret
= nfs_wait_on_request(req
);
300 nfs_release_request(req
);
303 spin_lock(&inode
->i_lock
);
305 spin_unlock(&inode
->i_lock
);
310 * Find an associated nfs write request, and prepare to flush it out
311 * May return an error if the user signalled nfs_wait_on_request().
313 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
314 struct page
*page
, bool nonblock
)
316 struct nfs_page
*req
;
319 req
= nfs_find_and_lock_request(page
, nonblock
);
326 nfs_set_page_writeback(page
);
327 WARN_ON_ONCE(test_bit(PG_CLEAN
, &req
->wb_flags
));
330 if (!nfs_pageio_add_request(pgio
, req
)) {
331 nfs_redirty_request(req
);
332 ret
= pgio
->pg_error
;
338 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
340 struct inode
*inode
= page_file_mapping(page
)->host
;
343 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
344 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
346 nfs_pageio_cond_complete(pgio
, page_file_index(page
));
347 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
348 if (ret
== -EAGAIN
) {
349 redirty_page_for_writepage(wbc
, page
);
356 * Write an mmapped page to the server.
358 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
360 struct nfs_pageio_descriptor pgio
;
363 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
),
364 false, &nfs_async_write_completion_ops
);
365 err
= nfs_do_writepage(page
, wbc
, &pgio
);
366 nfs_pageio_complete(&pgio
);
369 if (pgio
.pg_error
< 0)
370 return pgio
.pg_error
;
374 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
378 ret
= nfs_writepage_locked(page
, wbc
);
383 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
387 ret
= nfs_do_writepage(page
, wbc
, data
);
392 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
394 struct inode
*inode
= mapping
->host
;
395 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
396 struct nfs_pageio_descriptor pgio
;
399 /* Stop dirtying of new pages while we sync */
400 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
401 nfs_wait_bit_killable
, TASK_KILLABLE
);
405 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
407 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
), false,
408 &nfs_async_write_completion_ops
);
409 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
410 nfs_pageio_complete(&pgio
);
412 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
413 smp_mb__after_atomic();
414 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
427 * Insert a write request into an inode
429 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
431 struct nfs_inode
*nfsi
= NFS_I(inode
);
433 WARN_ON_ONCE(req
->wb_this_page
!= req
);
435 /* Lock the request! */
436 nfs_lock_request(req
);
438 spin_lock(&inode
->i_lock
);
439 if (!nfsi
->npages
&& NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
442 * Swap-space should not get truncated. Hence no need to plug the race
443 * with invalidate/truncate.
445 if (likely(!PageSwapCache(req
->wb_page
))) {
446 set_bit(PG_MAPPED
, &req
->wb_flags
);
447 SetPagePrivate(req
->wb_page
);
448 set_page_private(req
->wb_page
, (unsigned long)req
);
451 set_bit(PG_INODE_REF
, &req
->wb_flags
);
452 kref_get(&req
->wb_kref
);
453 spin_unlock(&inode
->i_lock
);
457 * Remove a write request from an inode
459 static void nfs_inode_remove_request(struct nfs_page
*req
)
461 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
462 struct nfs_inode
*nfsi
= NFS_I(inode
);
463 struct nfs_page
*head
;
465 if (nfs_page_group_sync_on_bit(req
, PG_REMOVE
)) {
468 spin_lock(&inode
->i_lock
);
469 if (likely(!PageSwapCache(head
->wb_page
))) {
470 set_page_private(head
->wb_page
, 0);
471 ClearPagePrivate(head
->wb_page
);
472 clear_bit(PG_MAPPED
, &head
->wb_flags
);
475 spin_unlock(&inode
->i_lock
);
477 nfs_release_request(req
);
481 nfs_mark_request_dirty(struct nfs_page
*req
)
483 __set_page_dirty_nobuffers(req
->wb_page
);
486 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
488 * nfs_request_add_commit_list - add request to a commit list
489 * @req: pointer to a struct nfs_page
490 * @dst: commit list head
491 * @cinfo: holds list lock and accounting info
493 * This sets the PG_CLEAN bit, updates the cinfo count of
494 * number of outstanding requests requiring a commit as well as
497 * The caller must _not_ hold the cinfo->lock, but must be
498 * holding the nfs_page lock.
501 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
502 struct nfs_commit_info
*cinfo
)
504 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
505 spin_lock(cinfo
->lock
);
506 nfs_list_add_request(req
, dst
);
507 cinfo
->mds
->ncommit
++;
508 spin_unlock(cinfo
->lock
);
510 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
511 inc_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
513 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
517 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
520 * nfs_request_remove_commit_list - Remove request from a commit list
521 * @req: pointer to a nfs_page
522 * @cinfo: holds list lock and accounting info
524 * This clears the PG_CLEAN bit, and updates the cinfo's count of
525 * number of outstanding requests requiring a commit
526 * It does not update the MM page stats.
528 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
531 nfs_request_remove_commit_list(struct nfs_page
*req
,
532 struct nfs_commit_info
*cinfo
)
534 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
536 nfs_list_remove_request(req
);
537 cinfo
->mds
->ncommit
--;
539 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
541 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
544 cinfo
->lock
= &inode
->i_lock
;
545 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
546 cinfo
->ds
= pnfs_get_ds_info(inode
);
548 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
551 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
553 struct nfs_direct_req
*dreq
)
556 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
558 nfs_init_cinfo_from_inode(cinfo
, inode
);
560 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
563 * Add a request to the inode's commit list.
566 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
567 struct nfs_commit_info
*cinfo
)
569 if (pnfs_mark_request_commit(req
, lseg
, cinfo
))
571 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
575 nfs_clear_page_commit(struct page
*page
)
577 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
578 dec_bdi_stat(page_file_mapping(page
)->backing_dev_info
, BDI_RECLAIMABLE
);
582 nfs_clear_request_commit(struct nfs_page
*req
)
584 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
585 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
586 struct nfs_commit_info cinfo
;
588 nfs_init_cinfo_from_inode(&cinfo
, inode
);
589 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
590 spin_lock(cinfo
.lock
);
591 nfs_request_remove_commit_list(req
, &cinfo
);
592 spin_unlock(cinfo
.lock
);
594 nfs_clear_page_commit(req
->wb_page
);
599 int nfs_write_need_commit(struct nfs_pgio_data
*data
)
601 if (data
->verf
.committed
== NFS_DATA_SYNC
)
602 return data
->header
->lseg
== NULL
;
603 return data
->verf
.committed
!= NFS_FILE_SYNC
;
607 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
612 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
614 struct nfs_direct_req
*dreq
)
619 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
620 struct nfs_commit_info
*cinfo
)
625 nfs_clear_request_commit(struct nfs_page
*req
)
630 int nfs_write_need_commit(struct nfs_pgio_data
*data
)
637 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
639 struct nfs_commit_info cinfo
;
640 unsigned long bytes
= 0;
643 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
645 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
646 while (!list_empty(&hdr
->pages
)) {
647 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
649 bytes
+= req
->wb_bytes
;
650 nfs_list_remove_request(req
);
651 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
652 (hdr
->good_bytes
< bytes
)) {
653 nfs_set_pageerror(req
->wb_page
);
654 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
657 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
)) {
658 nfs_mark_request_dirty(req
);
661 if (test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
)) {
662 memcpy(&req
->wb_verf
, &hdr
->verf
.verifier
, sizeof(req
->wb_verf
));
663 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
);
667 nfs_inode_remove_request(req
);
669 nfs_unlock_request(req
);
670 nfs_end_page_writeback(req
);
671 do_destroy
= !test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
);
672 nfs_release_request(req
);
678 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
680 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
682 return cinfo
->mds
->ncommit
;
685 /* cinfo->lock held by caller */
687 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
688 struct nfs_commit_info
*cinfo
, int max
)
690 struct nfs_page
*req
, *tmp
;
693 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
694 if (!nfs_lock_request(req
))
696 kref_get(&req
->wb_kref
);
697 if (cond_resched_lock(cinfo
->lock
))
698 list_safe_reset_next(req
, tmp
, wb_list
);
699 nfs_request_remove_commit_list(req
, cinfo
);
700 nfs_list_add_request(req
, dst
);
702 if ((ret
== max
) && !cinfo
->dreq
)
709 * nfs_scan_commit - Scan an inode for commit requests
710 * @inode: NFS inode to scan
711 * @dst: mds destination list
712 * @cinfo: mds and ds lists of reqs ready to commit
714 * Moves requests from the inode's 'commit' request list.
715 * The requests are *not* checked to ensure that they form a contiguous set.
718 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
719 struct nfs_commit_info
*cinfo
)
723 spin_lock(cinfo
->lock
);
724 if (cinfo
->mds
->ncommit
> 0) {
725 const int max
= INT_MAX
;
727 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
729 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
731 spin_unlock(cinfo
->lock
);
736 unsigned long nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
741 int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
742 struct nfs_commit_info
*cinfo
)
749 * Search for an existing write request, and attempt to update
750 * it to reflect a new dirty region on a given page.
752 * If the attempt fails, then the existing request is flushed out
755 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
760 struct nfs_page
*req
;
765 if (!PagePrivate(page
))
768 end
= offset
+ bytes
;
769 spin_lock(&inode
->i_lock
);
772 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
776 /* should be handled by nfs_flush_incompatible */
777 WARN_ON_ONCE(req
->wb_head
!= req
);
778 WARN_ON_ONCE(req
->wb_this_page
!= req
);
780 rqend
= req
->wb_offset
+ req
->wb_bytes
;
782 * Tell the caller to flush out the request if
783 * the offsets are non-contiguous.
784 * Note: nfs_flush_incompatible() will already
785 * have flushed out requests having wrong owners.
788 || end
< req
->wb_offset
)
791 if (nfs_lock_request(req
))
794 /* The request is locked, so wait and then retry */
795 spin_unlock(&inode
->i_lock
);
796 error
= nfs_wait_on_request(req
);
797 nfs_release_request(req
);
800 spin_lock(&inode
->i_lock
);
803 /* Okay, the request matches. Update the region */
804 if (offset
< req
->wb_offset
) {
805 req
->wb_offset
= offset
;
806 req
->wb_pgbase
= offset
;
809 req
->wb_bytes
= end
- req
->wb_offset
;
811 req
->wb_bytes
= rqend
- req
->wb_offset
;
813 spin_unlock(&inode
->i_lock
);
815 nfs_clear_request_commit(req
);
818 spin_unlock(&inode
->i_lock
);
819 nfs_release_request(req
);
820 error
= nfs_wb_page(inode
, page
);
822 return ERR_PTR(error
);
826 * Try to update an existing write request, or create one if there is none.
828 * Note: Should always be called with the Page Lock held to prevent races
829 * if we have to add a new request. Also assumes that the caller has
830 * already called nfs_flush_incompatible() if necessary.
832 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
833 struct page
*page
, unsigned int offset
, unsigned int bytes
)
835 struct inode
*inode
= page_file_mapping(page
)->host
;
836 struct nfs_page
*req
;
838 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
841 req
= nfs_create_request(ctx
, page
, NULL
, offset
, bytes
);
844 nfs_inode_add_request(inode
, req
);
849 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
850 unsigned int offset
, unsigned int count
)
852 struct nfs_page
*req
;
854 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
857 /* Update file length */
858 nfs_grow_file(page
, offset
, count
);
859 nfs_mark_uptodate(req
);
860 nfs_mark_request_dirty(req
);
861 nfs_unlock_and_release_request(req
);
865 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
867 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
868 struct nfs_lock_context
*l_ctx
;
869 struct nfs_page
*req
;
870 int do_flush
, status
;
872 * Look for a request corresponding to this page. If there
873 * is one, and it belongs to another file, we flush it out
874 * before we try to copy anything into the page. Do this
875 * due to the lack of an ACCESS-type call in NFSv2.
876 * Also do the same if we find a request from an existing
880 req
= nfs_page_find_request(page
);
883 l_ctx
= req
->wb_lock_context
;
884 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
;
885 /* for now, flush if more than 1 request in page_group */
886 do_flush
|= req
->wb_this_page
!= req
;
887 if (l_ctx
&& ctx
->dentry
->d_inode
->i_flock
!= NULL
) {
888 do_flush
|= l_ctx
->lockowner
.l_owner
!= current
->files
889 || l_ctx
->lockowner
.l_pid
!= current
->tgid
;
891 nfs_release_request(req
);
894 status
= nfs_wb_page(page_file_mapping(page
)->host
, page
);
895 } while (status
== 0);
900 * Avoid buffered writes when a open context credential's key would
903 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
905 * Return 0 and set a credential flag which triggers the inode to flush
906 * and performs NFS_FILE_SYNC writes if the key will expired within
907 * RPC_KEY_EXPIRE_TIMEO.
910 nfs_key_timeout_notify(struct file
*filp
, struct inode
*inode
)
912 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
913 struct rpc_auth
*auth
= NFS_SERVER(inode
)->client
->cl_auth
;
915 return rpcauth_key_timeout_notify(auth
, ctx
->cred
);
919 * Test if the open context credential key is marked to expire soon.
921 bool nfs_ctx_key_to_expire(struct nfs_open_context
*ctx
)
923 return rpcauth_cred_key_to_expire(ctx
->cred
);
927 * If the page cache is marked as unsafe or invalid, then we can't rely on
928 * the PageUptodate() flag. In this case, we will need to turn off
929 * write optimisations that depend on the page contents being correct.
931 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
933 struct nfs_inode
*nfsi
= NFS_I(inode
);
935 if (nfs_have_delegated_attributes(inode
))
937 if (nfsi
->cache_validity
& (NFS_INO_INVALID_DATA
|NFS_INO_REVAL_PAGECACHE
))
940 if (test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
))
943 return PageUptodate(page
) != 0;
946 /* If we know the page is up to date, and we're not using byte range locks (or
947 * if we have the whole file locked for writing), it may be more efficient to
948 * extend the write to cover the entire page in order to avoid fragmentation
951 * If the file is opened for synchronous writes then we can just skip the rest
954 static int nfs_can_extend_write(struct file
*file
, struct page
*page
, struct inode
*inode
)
956 if (file
->f_flags
& O_DSYNC
)
958 if (!nfs_write_pageuptodate(page
, inode
))
960 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
962 if (inode
->i_flock
== NULL
|| (inode
->i_flock
->fl_start
== 0 &&
963 inode
->i_flock
->fl_end
== OFFSET_MAX
&&
964 inode
->i_flock
->fl_type
!= F_RDLCK
))
970 * Update and possibly write a cached page of an NFS file.
972 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
973 * things with a page scheduled for an RPC call (e.g. invalidate it).
975 int nfs_updatepage(struct file
*file
, struct page
*page
,
976 unsigned int offset
, unsigned int count
)
978 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
979 struct inode
*inode
= page_file_mapping(page
)->host
;
982 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
984 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
985 file
, count
, (long long)(page_file_offset(page
) + offset
));
987 if (nfs_can_extend_write(file
, page
, inode
)) {
988 count
= max(count
+ offset
, nfs_page_length(page
));
992 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
994 nfs_set_pageerror(page
);
996 __set_page_dirty_nobuffers(page
);
998 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
999 status
, (long long)i_size_read(inode
));
1003 static int flush_task_priority(int how
)
1005 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
1007 return RPC_PRIORITY_HIGH
;
1009 return RPC_PRIORITY_LOW
;
1011 return RPC_PRIORITY_NORMAL
;
1014 static void nfs_initiate_write(struct nfs_pgio_data
*data
, struct rpc_message
*msg
,
1015 struct rpc_task_setup
*task_setup_data
, int how
)
1017 struct inode
*inode
= data
->header
->inode
;
1018 int priority
= flush_task_priority(how
);
1020 task_setup_data
->priority
= priority
;
1021 NFS_PROTO(inode
)->write_setup(data
, msg
);
1023 nfs4_state_protect_write(NFS_SERVER(inode
)->nfs_client
,
1024 &task_setup_data
->rpc_client
, msg
, data
);
1027 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1028 * call this on each, which will prepare them to be retried on next
1029 * writeback using standard nfs.
1031 static void nfs_redirty_request(struct nfs_page
*req
)
1033 nfs_mark_request_dirty(req
);
1034 nfs_unlock_request(req
);
1035 nfs_end_page_writeback(req
);
1036 nfs_release_request(req
);
1039 static void nfs_async_write_error(struct list_head
*head
)
1041 struct nfs_page
*req
;
1043 while (!list_empty(head
)) {
1044 req
= nfs_list_entry(head
->next
);
1045 nfs_list_remove_request(req
);
1046 nfs_redirty_request(req
);
1050 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1051 .error_cleanup
= nfs_async_write_error
,
1052 .completion
= nfs_write_completion
,
1055 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1056 struct inode
*inode
, int ioflags
, bool force_mds
,
1057 const struct nfs_pgio_completion_ops
*compl_ops
)
1059 struct nfs_server
*server
= NFS_SERVER(inode
);
1060 const struct nfs_pageio_ops
*pg_ops
= &nfs_pgio_rw_ops
;
1062 #ifdef CONFIG_NFS_V4_1
1063 if (server
->pnfs_curr_ld
&& !force_mds
)
1064 pg_ops
= server
->pnfs_curr_ld
->pg_write_ops
;
1066 nfs_pageio_init(pgio
, inode
, pg_ops
, compl_ops
, &nfs_rw_write_ops
,
1067 server
->wsize
, ioflags
);
1069 EXPORT_SYMBOL_GPL(nfs_pageio_init_write
);
1071 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1073 pgio
->pg_ops
= &nfs_pgio_rw_ops
;
1074 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1076 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1079 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1081 struct nfs_commit_data
*data
= calldata
;
1083 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1086 static void nfs_writeback_release_common(struct nfs_pgio_data
*data
)
1088 struct nfs_pgio_header
*hdr
= data
->header
;
1089 int status
= data
->task
.tk_status
;
1091 if ((status
>= 0) && nfs_write_need_commit(data
)) {
1092 spin_lock(&hdr
->lock
);
1093 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
))
1095 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
))
1096 memcpy(&hdr
->verf
, &data
->verf
, sizeof(hdr
->verf
));
1097 else if (memcmp(&hdr
->verf
, &data
->verf
, sizeof(hdr
->verf
)))
1098 set_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
);
1099 spin_unlock(&hdr
->lock
);
1104 * Special version of should_remove_suid() that ignores capabilities.
1106 static int nfs_should_remove_suid(const struct inode
*inode
)
1108 umode_t mode
= inode
->i_mode
;
1111 /* suid always must be killed */
1112 if (unlikely(mode
& S_ISUID
))
1113 kill
= ATTR_KILL_SUID
;
1116 * sgid without any exec bits is just a mandatory locking mark; leave
1117 * it alone. If some exec bits are set, it's a real sgid; kill it.
1119 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1120 kill
|= ATTR_KILL_SGID
;
1122 if (unlikely(kill
&& S_ISREG(mode
)))
1129 * This function is called when the WRITE call is complete.
1131 static int nfs_writeback_done(struct rpc_task
*task
, struct nfs_pgio_data
*data
,
1132 struct inode
*inode
)
1137 * ->write_done will attempt to use post-op attributes to detect
1138 * conflicting writes by other clients. A strict interpretation
1139 * of close-to-open would allow us to continue caching even if
1140 * another writer had changed the file, but some applications
1141 * depend on tighter cache coherency when writing.
1143 status
= NFS_PROTO(inode
)->write_done(task
, data
);
1146 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, data
->res
.count
);
1148 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1149 if (data
->res
.verf
->committed
< data
->args
.stable
&& task
->tk_status
>= 0) {
1150 /* We tried a write call, but the server did not
1151 * commit data to stable storage even though we
1153 * Note: There is a known bug in Tru64 < 5.0 in which
1154 * the server reports NFS_DATA_SYNC, but performs
1155 * NFS_FILE_SYNC. We therefore implement this checking
1156 * as a dprintk() in order to avoid filling syslog.
1158 static unsigned long complain
;
1160 /* Note this will print the MDS for a DS write */
1161 if (time_before(complain
, jiffies
)) {
1162 dprintk("NFS: faulty NFS server %s:"
1163 " (committed = %d) != (stable = %d)\n",
1164 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1165 data
->res
.verf
->committed
, data
->args
.stable
);
1166 complain
= jiffies
+ 300 * HZ
;
1171 /* Deal with the suid/sgid bit corner case */
1172 if (nfs_should_remove_suid(inode
))
1173 nfs_mark_for_revalidate(inode
);
1178 * This function is called when the WRITE call is complete.
1180 static void nfs_writeback_result(struct rpc_task
*task
, struct nfs_pgio_data
*data
)
1182 struct nfs_pgio_args
*argp
= &data
->args
;
1183 struct nfs_pgio_res
*resp
= &data
->res
;
1185 if (resp
->count
< argp
->count
) {
1186 static unsigned long complain
;
1188 /* This a short write! */
1189 nfs_inc_stats(data
->header
->inode
, NFSIOS_SHORTWRITE
);
1191 /* Has the server at least made some progress? */
1192 if (resp
->count
== 0) {
1193 if (time_before(complain
, jiffies
)) {
1195 "NFS: Server wrote zero bytes, expected %u.\n",
1197 complain
= jiffies
+ 300 * HZ
;
1199 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
1200 task
->tk_status
= -EIO
;
1203 /* Was this an NFSv2 write or an NFSv3 stable write? */
1204 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1205 /* Resend from where the server left off */
1206 data
->mds_offset
+= resp
->count
;
1207 argp
->offset
+= resp
->count
;
1208 argp
->pgbase
+= resp
->count
;
1209 argp
->count
-= resp
->count
;
1211 /* Resend as a stable write in order to avoid
1212 * headaches in the case of a server crash.
1214 argp
->stable
= NFS_FILE_SYNC
;
1216 rpc_restart_call_prepare(task
);
1221 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1222 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1226 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1230 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1232 nfs_wait_bit_killable
,
1234 return (ret
< 0) ? ret
: 1;
1237 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1239 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1240 smp_mb__after_atomic();
1241 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1244 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1246 put_nfs_open_context(data
->context
);
1247 nfs_commit_free(data
);
1249 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1251 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1252 const struct rpc_call_ops
*call_ops
,
1255 struct rpc_task
*task
;
1256 int priority
= flush_task_priority(how
);
1257 struct rpc_message msg
= {
1258 .rpc_argp
= &data
->args
,
1259 .rpc_resp
= &data
->res
,
1260 .rpc_cred
= data
->cred
,
1262 struct rpc_task_setup task_setup_data
= {
1263 .task
= &data
->task
,
1265 .rpc_message
= &msg
,
1266 .callback_ops
= call_ops
,
1267 .callback_data
= data
,
1268 .workqueue
= nfsiod_workqueue
,
1269 .flags
= RPC_TASK_ASYNC
| flags
,
1270 .priority
= priority
,
1272 /* Set up the initial task struct. */
1273 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1275 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1277 nfs4_state_protect(NFS_SERVER(data
->inode
)->nfs_client
,
1278 NFS_SP4_MACH_CRED_COMMIT
, &task_setup_data
.rpc_client
, &msg
);
1280 task
= rpc_run_task(&task_setup_data
);
1282 return PTR_ERR(task
);
1283 if (how
& FLUSH_SYNC
)
1284 rpc_wait_for_completion_task(task
);
1288 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1291 * Set up the argument/result storage required for the RPC call.
1293 void nfs_init_commit(struct nfs_commit_data
*data
,
1294 struct list_head
*head
,
1295 struct pnfs_layout_segment
*lseg
,
1296 struct nfs_commit_info
*cinfo
)
1298 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1299 struct inode
*inode
= first
->wb_context
->dentry
->d_inode
;
1301 /* Set up the RPC argument and reply structs
1302 * NB: take care not to mess about with data->commit et al. */
1304 list_splice_init(head
, &data
->pages
);
1306 data
->inode
= inode
;
1307 data
->cred
= first
->wb_context
->cred
;
1308 data
->lseg
= lseg
; /* reference transferred */
1309 data
->mds_ops
= &nfs_commit_ops
;
1310 data
->completion_ops
= cinfo
->completion_ops
;
1311 data
->dreq
= cinfo
->dreq
;
1313 data
->args
.fh
= NFS_FH(data
->inode
);
1314 /* Note: we always request a commit of the entire inode */
1315 data
->args
.offset
= 0;
1316 data
->args
.count
= 0;
1317 data
->context
= get_nfs_open_context(first
->wb_context
);
1318 data
->res
.fattr
= &data
->fattr
;
1319 data
->res
.verf
= &data
->verf
;
1320 nfs_fattr_init(&data
->fattr
);
1322 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1324 void nfs_retry_commit(struct list_head
*page_list
,
1325 struct pnfs_layout_segment
*lseg
,
1326 struct nfs_commit_info
*cinfo
)
1328 struct nfs_page
*req
;
1330 while (!list_empty(page_list
)) {
1331 req
= nfs_list_entry(page_list
->next
);
1332 nfs_list_remove_request(req
);
1333 nfs_mark_request_commit(req
, lseg
, cinfo
);
1335 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1336 dec_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
1339 nfs_unlock_and_release_request(req
);
1342 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1345 * Commit dirty pages
1348 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1349 struct nfs_commit_info
*cinfo
)
1351 struct nfs_commit_data
*data
;
1353 data
= nfs_commitdata_alloc();
1358 /* Set up the argument struct */
1359 nfs_init_commit(data
, head
, NULL
, cinfo
);
1360 atomic_inc(&cinfo
->mds
->rpcs_out
);
1361 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, data
->mds_ops
,
1364 nfs_retry_commit(head
, NULL
, cinfo
);
1365 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1370 * COMMIT call returned
1372 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1374 struct nfs_commit_data
*data
= calldata
;
1376 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1377 task
->tk_pid
, task
->tk_status
);
1379 /* Call the NFS version-specific code */
1380 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1383 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1385 struct nfs_page
*req
;
1386 int status
= data
->task
.tk_status
;
1387 struct nfs_commit_info cinfo
;
1389 while (!list_empty(&data
->pages
)) {
1390 req
= nfs_list_entry(data
->pages
.next
);
1391 nfs_list_remove_request(req
);
1392 nfs_clear_page_commit(req
->wb_page
);
1394 dprintk("NFS: commit (%s/%llu %d@%lld)",
1395 req
->wb_context
->dentry
->d_sb
->s_id
,
1396 (unsigned long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
1398 (long long)req_offset(req
));
1400 nfs_context_set_write_error(req
->wb_context
, status
);
1401 nfs_inode_remove_request(req
);
1402 dprintk(", error = %d\n", status
);
1406 /* Okay, COMMIT succeeded, apparently. Check the verifier
1407 * returned by the server against all stored verfs. */
1408 if (!memcmp(&req
->wb_verf
, &data
->verf
.verifier
, sizeof(req
->wb_verf
))) {
1409 /* We have a match */
1410 nfs_inode_remove_request(req
);
1414 /* We have a mismatch. Write the page again */
1415 dprintk(" mismatch\n");
1416 nfs_mark_request_dirty(req
);
1417 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1419 nfs_unlock_and_release_request(req
);
1421 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1422 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1423 nfs_commit_clear_lock(NFS_I(data
->inode
));
1426 static void nfs_commit_release(void *calldata
)
1428 struct nfs_commit_data
*data
= calldata
;
1430 data
->completion_ops
->completion(data
);
1431 nfs_commitdata_release(calldata
);
1434 static const struct rpc_call_ops nfs_commit_ops
= {
1435 .rpc_call_prepare
= nfs_commit_prepare
,
1436 .rpc_call_done
= nfs_commit_done
,
1437 .rpc_release
= nfs_commit_release
,
1440 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1441 .completion
= nfs_commit_release_pages
,
1442 .error_cleanup
= nfs_commit_clear_lock
,
1445 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1446 int how
, struct nfs_commit_info
*cinfo
)
1450 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1451 if (status
== PNFS_NOT_ATTEMPTED
)
1452 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1456 int nfs_commit_inode(struct inode
*inode
, int how
)
1459 struct nfs_commit_info cinfo
;
1460 int may_wait
= how
& FLUSH_SYNC
;
1463 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1465 goto out_mark_dirty
;
1466 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1467 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1471 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1475 goto out_mark_dirty
;
1476 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1478 nfs_wait_bit_killable
,
1483 nfs_commit_clear_lock(NFS_I(inode
));
1485 /* Note: If we exit without ensuring that the commit is complete,
1486 * we must mark the inode as dirty. Otherwise, future calls to
1487 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1488 * that the data is on the disk.
1491 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1495 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1497 struct nfs_inode
*nfsi
= NFS_I(inode
);
1498 int flags
= FLUSH_SYNC
;
1501 /* no commits means nothing needs to be done */
1502 if (!nfsi
->commit_info
.ncommit
)
1505 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1506 /* Don't commit yet if this is a non-blocking flush and there
1507 * are a lot of outstanding writes for this mapping.
1509 if (nfsi
->commit_info
.ncommit
<= (nfsi
->npages
>> 1))
1510 goto out_mark_dirty
;
1512 /* don't wait for the COMMIT response */
1516 ret
= nfs_commit_inode(inode
, flags
);
1518 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1519 if (ret
< wbc
->nr_to_write
)
1520 wbc
->nr_to_write
-= ret
;
1522 wbc
->nr_to_write
= 0;
1527 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1531 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1537 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1539 return nfs_commit_unstable_pages(inode
, wbc
);
1541 EXPORT_SYMBOL_GPL(nfs_write_inode
);
1544 * flush the inode to disk.
1546 int nfs_wb_all(struct inode
*inode
)
1548 struct writeback_control wbc
= {
1549 .sync_mode
= WB_SYNC_ALL
,
1550 .nr_to_write
= LONG_MAX
,
1552 .range_end
= LLONG_MAX
,
1556 trace_nfs_writeback_inode_enter(inode
);
1558 ret
= sync_inode(inode
, &wbc
);
1560 trace_nfs_writeback_inode_exit(inode
, ret
);
1563 EXPORT_SYMBOL_GPL(nfs_wb_all
);
1565 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1567 struct nfs_page
*req
;
1571 wait_on_page_writeback(page
);
1572 req
= nfs_page_find_request(page
);
1575 if (nfs_lock_request(req
)) {
1576 nfs_clear_request_commit(req
);
1577 nfs_inode_remove_request(req
);
1579 * In case nfs_inode_remove_request has marked the
1580 * page as being dirty
1582 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1583 nfs_unlock_and_release_request(req
);
1586 ret
= nfs_wait_on_request(req
);
1587 nfs_release_request(req
);
1595 * Write back all requests on one page - we do this before reading it.
1597 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1599 loff_t range_start
= page_file_offset(page
);
1600 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1601 struct writeback_control wbc
= {
1602 .sync_mode
= WB_SYNC_ALL
,
1604 .range_start
= range_start
,
1605 .range_end
= range_end
,
1609 trace_nfs_writeback_page_enter(inode
);
1612 wait_on_page_writeback(page
);
1613 if (clear_page_dirty_for_io(page
)) {
1614 ret
= nfs_writepage_locked(page
, &wbc
);
1620 if (!PagePrivate(page
))
1622 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1627 trace_nfs_writeback_page_exit(inode
, ret
);
1631 #ifdef CONFIG_MIGRATION
1632 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1633 struct page
*page
, enum migrate_mode mode
)
1636 * If PagePrivate is set, then the page is currently associated with
1637 * an in-progress read or write request. Don't try to migrate it.
1639 * FIXME: we could do this in principle, but we'll need a way to ensure
1640 * that we can safely release the inode reference while holding
1643 if (PagePrivate(page
))
1646 if (!nfs_fscache_release_page(page
, GFP_KERNEL
))
1649 return migrate_page(mapping
, newpage
, page
, mode
);
1653 int __init
nfs_init_writepagecache(void)
1655 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1656 sizeof(struct nfs_rw_header
),
1657 0, SLAB_HWCACHE_ALIGN
,
1659 if (nfs_wdata_cachep
== NULL
)
1662 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1664 if (nfs_wdata_mempool
== NULL
)
1665 goto out_destroy_write_cache
;
1667 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1668 sizeof(struct nfs_commit_data
),
1669 0, SLAB_HWCACHE_ALIGN
,
1671 if (nfs_cdata_cachep
== NULL
)
1672 goto out_destroy_write_mempool
;
1674 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1676 if (nfs_commit_mempool
== NULL
)
1677 goto out_destroy_commit_cache
;
1680 * NFS congestion size, scale with available memory.
1692 * This allows larger machines to have larger/more transfers.
1693 * Limit the default to 256M
1695 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1696 if (nfs_congestion_kb
> 256*1024)
1697 nfs_congestion_kb
= 256*1024;
1701 out_destroy_commit_cache
:
1702 kmem_cache_destroy(nfs_cdata_cachep
);
1703 out_destroy_write_mempool
:
1704 mempool_destroy(nfs_wdata_mempool
);
1705 out_destroy_write_cache
:
1706 kmem_cache_destroy(nfs_wdata_cachep
);
1710 void nfs_destroy_writepagecache(void)
1712 mempool_destroy(nfs_commit_mempool
);
1713 kmem_cache_destroy(nfs_cdata_cachep
);
1714 mempool_destroy(nfs_wdata_mempool
);
1715 kmem_cache_destroy(nfs_wdata_cachep
);
1718 static const struct nfs_rw_ops nfs_rw_write_ops
= {
1719 .rw_mode
= FMODE_WRITE
,
1720 .rw_alloc_header
= nfs_writehdr_alloc
,
1721 .rw_free_header
= nfs_writehdr_free
,
1722 .rw_release
= nfs_writeback_release_common
,
1723 .rw_done
= nfs_writeback_done
,
1724 .rw_result
= nfs_writeback_result
,
1725 .rw_initiate
= nfs_initiate_write
,