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
));
157 /* We can set the PG_uptodate flag if we see that a write request
158 * covers the full page.
160 static void nfs_mark_uptodate(struct page
*page
, unsigned int base
, unsigned int count
)
162 if (PageUptodate(page
))
166 if (count
!= nfs_page_length(page
))
168 SetPageUptodate(page
);
171 static int wb_priority(struct writeback_control
*wbc
)
173 if (wbc
->for_reclaim
)
174 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
175 if (wbc
->for_kupdate
|| wbc
->for_background
)
176 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
177 return FLUSH_COND_STABLE
;
181 * NFS congestion control
184 int nfs_congestion_kb
;
186 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
187 #define NFS_CONGESTION_OFF_THRESH \
188 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
190 static void nfs_set_page_writeback(struct page
*page
)
192 struct nfs_server
*nfss
= NFS_SERVER(page_file_mapping(page
)->host
);
193 int ret
= test_set_page_writeback(page
);
195 WARN_ON_ONCE(ret
!= 0);
197 if (atomic_long_inc_return(&nfss
->writeback
) >
198 NFS_CONGESTION_ON_THRESH
) {
199 set_bdi_congested(&nfss
->backing_dev_info
,
204 static void nfs_end_page_writeback(struct page
*page
)
206 struct inode
*inode
= page_file_mapping(page
)->host
;
207 struct nfs_server
*nfss
= NFS_SERVER(inode
);
209 end_page_writeback(page
);
210 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
211 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
214 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
216 struct inode
*inode
= page_file_mapping(page
)->host
;
217 struct nfs_page
*req
;
220 spin_lock(&inode
->i_lock
);
222 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
225 if (nfs_lock_request(req
))
227 /* Note: If we hold the page lock, as is the case in nfs_writepage,
228 * then the call to nfs_lock_request() will always
229 * succeed provided that someone hasn't already marked the
230 * request as dirty (in which case we don't care).
232 spin_unlock(&inode
->i_lock
);
234 ret
= nfs_wait_on_request(req
);
237 nfs_release_request(req
);
240 spin_lock(&inode
->i_lock
);
242 spin_unlock(&inode
->i_lock
);
247 * Find an associated nfs write request, and prepare to flush it out
248 * May return an error if the user signalled nfs_wait_on_request().
250 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
251 struct page
*page
, bool nonblock
)
253 struct nfs_page
*req
;
256 req
= nfs_find_and_lock_request(page
, nonblock
);
263 nfs_set_page_writeback(page
);
264 WARN_ON_ONCE(test_bit(PG_CLEAN
, &req
->wb_flags
));
267 if (!nfs_pageio_add_request(pgio
, req
)) {
268 nfs_redirty_request(req
);
269 ret
= pgio
->pg_error
;
275 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
277 struct inode
*inode
= page_file_mapping(page
)->host
;
280 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
281 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
283 nfs_pageio_cond_complete(pgio
, page_file_index(page
));
284 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
285 if (ret
== -EAGAIN
) {
286 redirty_page_for_writepage(wbc
, page
);
293 * Write an mmapped page to the server.
295 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
297 struct nfs_pageio_descriptor pgio
;
300 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
),
301 false, &nfs_async_write_completion_ops
);
302 err
= nfs_do_writepage(page
, wbc
, &pgio
);
303 nfs_pageio_complete(&pgio
);
306 if (pgio
.pg_error
< 0)
307 return pgio
.pg_error
;
311 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
315 ret
= nfs_writepage_locked(page
, wbc
);
320 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
324 ret
= nfs_do_writepage(page
, wbc
, data
);
329 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
331 struct inode
*inode
= mapping
->host
;
332 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
333 struct nfs_pageio_descriptor pgio
;
336 /* Stop dirtying of new pages while we sync */
337 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
338 nfs_wait_bit_killable
, TASK_KILLABLE
);
342 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
344 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
), false,
345 &nfs_async_write_completion_ops
);
346 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
347 nfs_pageio_complete(&pgio
);
349 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
350 smp_mb__after_clear_bit();
351 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
364 * Insert a write request into an inode
366 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
368 struct nfs_inode
*nfsi
= NFS_I(inode
);
370 /* Lock the request! */
371 nfs_lock_request(req
);
373 spin_lock(&inode
->i_lock
);
374 if (!nfsi
->npages
&& NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
377 * Swap-space should not get truncated. Hence no need to plug the race
378 * with invalidate/truncate.
380 if (likely(!PageSwapCache(req
->wb_page
))) {
381 set_bit(PG_MAPPED
, &req
->wb_flags
);
382 SetPagePrivate(req
->wb_page
);
383 set_page_private(req
->wb_page
, (unsigned long)req
);
386 kref_get(&req
->wb_kref
);
387 spin_unlock(&inode
->i_lock
);
391 * Remove a write request from an inode
393 static void nfs_inode_remove_request(struct nfs_page
*req
)
395 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
396 struct nfs_inode
*nfsi
= NFS_I(inode
);
398 spin_lock(&inode
->i_lock
);
399 if (likely(!PageSwapCache(req
->wb_page
))) {
400 set_page_private(req
->wb_page
, 0);
401 ClearPagePrivate(req
->wb_page
);
402 clear_bit(PG_MAPPED
, &req
->wb_flags
);
405 spin_unlock(&inode
->i_lock
);
406 nfs_release_request(req
);
410 nfs_mark_request_dirty(struct nfs_page
*req
)
412 __set_page_dirty_nobuffers(req
->wb_page
);
415 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
417 * nfs_request_add_commit_list - add request to a commit list
418 * @req: pointer to a struct nfs_page
419 * @dst: commit list head
420 * @cinfo: holds list lock and accounting info
422 * This sets the PG_CLEAN bit, updates the cinfo count of
423 * number of outstanding requests requiring a commit as well as
426 * The caller must _not_ hold the cinfo->lock, but must be
427 * holding the nfs_page lock.
430 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
431 struct nfs_commit_info
*cinfo
)
433 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
434 spin_lock(cinfo
->lock
);
435 nfs_list_add_request(req
, dst
);
436 cinfo
->mds
->ncommit
++;
437 spin_unlock(cinfo
->lock
);
439 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
440 inc_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
442 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
446 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
449 * nfs_request_remove_commit_list - Remove request from a commit list
450 * @req: pointer to a nfs_page
451 * @cinfo: holds list lock and accounting info
453 * This clears the PG_CLEAN bit, and updates the cinfo's count of
454 * number of outstanding requests requiring a commit
455 * It does not update the MM page stats.
457 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
460 nfs_request_remove_commit_list(struct nfs_page
*req
,
461 struct nfs_commit_info
*cinfo
)
463 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
465 nfs_list_remove_request(req
);
466 cinfo
->mds
->ncommit
--;
468 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
470 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
473 cinfo
->lock
= &inode
->i_lock
;
474 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
475 cinfo
->ds
= pnfs_get_ds_info(inode
);
477 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
480 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
482 struct nfs_direct_req
*dreq
)
485 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
487 nfs_init_cinfo_from_inode(cinfo
, inode
);
489 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
492 * Add a request to the inode's commit list.
495 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
496 struct nfs_commit_info
*cinfo
)
498 if (pnfs_mark_request_commit(req
, lseg
, cinfo
))
500 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
504 nfs_clear_page_commit(struct page
*page
)
506 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
507 dec_bdi_stat(page_file_mapping(page
)->backing_dev_info
, BDI_RECLAIMABLE
);
511 nfs_clear_request_commit(struct nfs_page
*req
)
513 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
514 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
515 struct nfs_commit_info cinfo
;
517 nfs_init_cinfo_from_inode(&cinfo
, inode
);
518 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
519 spin_lock(cinfo
.lock
);
520 nfs_request_remove_commit_list(req
, &cinfo
);
521 spin_unlock(cinfo
.lock
);
523 nfs_clear_page_commit(req
->wb_page
);
528 int nfs_write_need_commit(struct nfs_pgio_data
*data
)
530 if (data
->verf
.committed
== NFS_DATA_SYNC
)
531 return data
->header
->lseg
== NULL
;
532 return data
->verf
.committed
!= NFS_FILE_SYNC
;
536 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
541 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
543 struct nfs_direct_req
*dreq
)
548 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
549 struct nfs_commit_info
*cinfo
)
554 nfs_clear_request_commit(struct nfs_page
*req
)
559 int nfs_write_need_commit(struct nfs_pgio_data
*data
)
566 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
568 struct nfs_commit_info cinfo
;
569 unsigned long bytes
= 0;
571 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
573 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
574 while (!list_empty(&hdr
->pages
)) {
575 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
577 bytes
+= req
->wb_bytes
;
578 nfs_list_remove_request(req
);
579 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
580 (hdr
->good_bytes
< bytes
)) {
581 nfs_set_pageerror(req
->wb_page
);
582 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
585 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
)) {
586 nfs_mark_request_dirty(req
);
589 if (test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
)) {
590 memcpy(&req
->wb_verf
, &hdr
->verf
.verifier
, sizeof(req
->wb_verf
));
591 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
);
595 nfs_inode_remove_request(req
);
597 nfs_unlock_request(req
);
598 nfs_end_page_writeback(req
->wb_page
);
599 nfs_release_request(req
);
605 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
607 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
609 return cinfo
->mds
->ncommit
;
612 /* cinfo->lock held by caller */
614 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
615 struct nfs_commit_info
*cinfo
, int max
)
617 struct nfs_page
*req
, *tmp
;
620 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
621 if (!nfs_lock_request(req
))
623 kref_get(&req
->wb_kref
);
624 if (cond_resched_lock(cinfo
->lock
))
625 list_safe_reset_next(req
, tmp
, wb_list
);
626 nfs_request_remove_commit_list(req
, cinfo
);
627 nfs_list_add_request(req
, dst
);
629 if ((ret
== max
) && !cinfo
->dreq
)
636 * nfs_scan_commit - Scan an inode for commit requests
637 * @inode: NFS inode to scan
638 * @dst: mds destination list
639 * @cinfo: mds and ds lists of reqs ready to commit
641 * Moves requests from the inode's 'commit' request list.
642 * The requests are *not* checked to ensure that they form a contiguous set.
645 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
646 struct nfs_commit_info
*cinfo
)
650 spin_lock(cinfo
->lock
);
651 if (cinfo
->mds
->ncommit
> 0) {
652 const int max
= INT_MAX
;
654 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
656 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
658 spin_unlock(cinfo
->lock
);
663 unsigned long nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
668 int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
669 struct nfs_commit_info
*cinfo
)
676 * Search for an existing write request, and attempt to update
677 * it to reflect a new dirty region on a given page.
679 * If the attempt fails, then the existing request is flushed out
682 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
687 struct nfs_page
*req
;
692 if (!PagePrivate(page
))
695 end
= offset
+ bytes
;
696 spin_lock(&inode
->i_lock
);
699 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
703 rqend
= req
->wb_offset
+ req
->wb_bytes
;
705 * Tell the caller to flush out the request if
706 * the offsets are non-contiguous.
707 * Note: nfs_flush_incompatible() will already
708 * have flushed out requests having wrong owners.
711 || end
< req
->wb_offset
)
714 if (nfs_lock_request(req
))
717 /* The request is locked, so wait and then retry */
718 spin_unlock(&inode
->i_lock
);
719 error
= nfs_wait_on_request(req
);
720 nfs_release_request(req
);
723 spin_lock(&inode
->i_lock
);
726 /* Okay, the request matches. Update the region */
727 if (offset
< req
->wb_offset
) {
728 req
->wb_offset
= offset
;
729 req
->wb_pgbase
= offset
;
732 req
->wb_bytes
= end
- req
->wb_offset
;
734 req
->wb_bytes
= rqend
- req
->wb_offset
;
736 spin_unlock(&inode
->i_lock
);
738 nfs_clear_request_commit(req
);
741 spin_unlock(&inode
->i_lock
);
742 nfs_release_request(req
);
743 error
= nfs_wb_page(inode
, page
);
745 return ERR_PTR(error
);
749 * Try to update an existing write request, or create one if there is none.
751 * Note: Should always be called with the Page Lock held to prevent races
752 * if we have to add a new request. Also assumes that the caller has
753 * already called nfs_flush_incompatible() if necessary.
755 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
756 struct page
*page
, unsigned int offset
, unsigned int bytes
)
758 struct inode
*inode
= page_file_mapping(page
)->host
;
759 struct nfs_page
*req
;
761 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
764 req
= nfs_create_request(ctx
, inode
, page
, offset
, bytes
);
767 nfs_inode_add_request(inode
, req
);
772 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
773 unsigned int offset
, unsigned int count
)
775 struct nfs_page
*req
;
777 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
780 /* Update file length */
781 nfs_grow_file(page
, offset
, count
);
782 nfs_mark_uptodate(page
, req
->wb_pgbase
, req
->wb_bytes
);
783 nfs_mark_request_dirty(req
);
784 nfs_unlock_and_release_request(req
);
788 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
790 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
791 struct nfs_lock_context
*l_ctx
;
792 struct nfs_page
*req
;
793 int do_flush
, status
;
795 * Look for a request corresponding to this page. If there
796 * is one, and it belongs to another file, we flush it out
797 * before we try to copy anything into the page. Do this
798 * due to the lack of an ACCESS-type call in NFSv2.
799 * Also do the same if we find a request from an existing
803 req
= nfs_page_find_request(page
);
806 l_ctx
= req
->wb_lock_context
;
807 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
;
808 if (l_ctx
&& ctx
->dentry
->d_inode
->i_flock
!= NULL
) {
809 do_flush
|= l_ctx
->lockowner
.l_owner
!= current
->files
810 || l_ctx
->lockowner
.l_pid
!= current
->tgid
;
812 nfs_release_request(req
);
815 status
= nfs_wb_page(page_file_mapping(page
)->host
, page
);
816 } while (status
== 0);
821 * Avoid buffered writes when a open context credential's key would
824 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
826 * Return 0 and set a credential flag which triggers the inode to flush
827 * and performs NFS_FILE_SYNC writes if the key will expired within
828 * RPC_KEY_EXPIRE_TIMEO.
831 nfs_key_timeout_notify(struct file
*filp
, struct inode
*inode
)
833 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
834 struct rpc_auth
*auth
= NFS_SERVER(inode
)->client
->cl_auth
;
836 return rpcauth_key_timeout_notify(auth
, ctx
->cred
);
840 * Test if the open context credential key is marked to expire soon.
842 bool nfs_ctx_key_to_expire(struct nfs_open_context
*ctx
)
844 return rpcauth_cred_key_to_expire(ctx
->cred
);
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 struct nfs_inode
*nfsi
= NFS_I(inode
);
856 if (nfs_have_delegated_attributes(inode
))
858 if (nfsi
->cache_validity
& (NFS_INO_INVALID_DATA
|NFS_INO_REVAL_PAGECACHE
))
861 if (test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
))
864 return PageUptodate(page
) != 0;
867 /* If we know the page is up to date, and we're not using byte range locks (or
868 * if we have the whole file locked for writing), it may be more efficient to
869 * extend the write to cover the entire page in order to avoid fragmentation
872 * If the file is opened for synchronous writes then we can just skip the rest
875 static int nfs_can_extend_write(struct file
*file
, struct page
*page
, struct inode
*inode
)
877 if (file
->f_flags
& O_DSYNC
)
879 if (!nfs_write_pageuptodate(page
, inode
))
881 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
883 if (inode
->i_flock
== NULL
|| (inode
->i_flock
->fl_start
== 0 &&
884 inode
->i_flock
->fl_end
== OFFSET_MAX
&&
885 inode
->i_flock
->fl_type
!= F_RDLCK
))
891 * Update and possibly write a cached page of an NFS file.
893 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
894 * things with a page scheduled for an RPC call (e.g. invalidate it).
896 int nfs_updatepage(struct file
*file
, struct page
*page
,
897 unsigned int offset
, unsigned int count
)
899 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
900 struct inode
*inode
= page_file_mapping(page
)->host
;
903 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
905 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
906 file
, count
, (long long)(page_file_offset(page
) + offset
));
908 if (nfs_can_extend_write(file
, page
, inode
)) {
909 count
= max(count
+ offset
, nfs_page_length(page
));
913 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
915 nfs_set_pageerror(page
);
917 __set_page_dirty_nobuffers(page
);
919 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
920 status
, (long long)i_size_read(inode
));
924 static int flush_task_priority(int how
)
926 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
928 return RPC_PRIORITY_HIGH
;
930 return RPC_PRIORITY_LOW
;
932 return RPC_PRIORITY_NORMAL
;
935 static void nfs_initiate_write(struct nfs_pgio_data
*data
, struct rpc_message
*msg
,
936 struct rpc_task_setup
*task_setup_data
, int how
)
938 struct inode
*inode
= data
->header
->inode
;
939 int priority
= flush_task_priority(how
);
941 task_setup_data
->priority
= priority
;
942 NFS_PROTO(inode
)->write_setup(data
, msg
);
944 nfs4_state_protect_write(NFS_SERVER(inode
)->nfs_client
,
945 &task_setup_data
->rpc_client
, msg
, data
);
948 static int nfs_do_write(struct nfs_pgio_data
*data
,
949 const struct rpc_call_ops
*call_ops
,
952 struct inode
*inode
= data
->header
->inode
;
954 return nfs_initiate_pgio(NFS_CLIENT(inode
), data
, call_ops
, how
, 0);
957 static int nfs_do_multiple_writes(struct list_head
*head
,
958 const struct rpc_call_ops
*call_ops
,
961 struct nfs_pgio_data
*data
;
964 while (!list_empty(head
)) {
967 data
= list_first_entry(head
, struct nfs_pgio_data
, list
);
968 list_del_init(&data
->list
);
970 ret2
= nfs_do_write(data
, call_ops
, how
);
977 /* If a nfs_flush_* function fails, it should remove reqs from @head and
978 * call this on each, which will prepare them to be retried on next
979 * writeback using standard nfs.
981 static void nfs_redirty_request(struct nfs_page
*req
)
983 nfs_mark_request_dirty(req
);
984 nfs_unlock_request(req
);
985 nfs_end_page_writeback(req
->wb_page
);
986 nfs_release_request(req
);
989 static void nfs_async_write_error(struct list_head
*head
)
991 struct nfs_page
*req
;
993 while (!list_empty(head
)) {
994 req
= nfs_list_entry(head
->next
);
995 nfs_list_remove_request(req
);
996 nfs_redirty_request(req
);
1000 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1001 .error_cleanup
= nfs_async_write_error
,
1002 .completion
= nfs_write_completion
,
1005 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1007 struct nfs_rw_header
*whdr
;
1008 struct nfs_pgio_header
*hdr
;
1011 whdr
= nfs_rw_header_alloc(desc
->pg_rw_ops
);
1013 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1016 hdr
= &whdr
->header
;
1017 nfs_pgheader_init(desc
, hdr
, nfs_rw_header_free
);
1018 atomic_inc(&hdr
->refcnt
);
1019 ret
= nfs_generic_pgio(desc
, hdr
);
1021 ret
= nfs_do_multiple_writes(&hdr
->rpc_list
,
1022 desc
->pg_rpc_callops
,
1024 if (atomic_dec_and_test(&hdr
->refcnt
))
1025 hdr
->completion_ops
->completion(hdr
);
1029 static const struct nfs_pageio_ops nfs_pageio_write_ops
= {
1030 .pg_test
= nfs_generic_pg_test
,
1031 .pg_doio
= nfs_generic_pg_writepages
,
1034 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1035 struct inode
*inode
, int ioflags
, bool force_mds
,
1036 const struct nfs_pgio_completion_ops
*compl_ops
)
1038 struct nfs_server
*server
= NFS_SERVER(inode
);
1039 const struct nfs_pageio_ops
*pg_ops
= &nfs_pageio_write_ops
;
1041 #ifdef CONFIG_NFS_V4_1
1042 if (server
->pnfs_curr_ld
&& !force_mds
)
1043 pg_ops
= server
->pnfs_curr_ld
->pg_write_ops
;
1045 nfs_pageio_init(pgio
, inode
, pg_ops
, compl_ops
, &nfs_rw_write_ops
,
1046 server
->wsize
, ioflags
);
1048 EXPORT_SYMBOL_GPL(nfs_pageio_init_write
);
1050 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1052 pgio
->pg_ops
= &nfs_pageio_write_ops
;
1053 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1055 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1058 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1060 struct nfs_commit_data
*data
= calldata
;
1062 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1065 static void nfs_writeback_release_common(struct nfs_pgio_data
*data
)
1067 struct nfs_pgio_header
*hdr
= data
->header
;
1068 int status
= data
->task
.tk_status
;
1070 if ((status
>= 0) && nfs_write_need_commit(data
)) {
1071 spin_lock(&hdr
->lock
);
1072 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
))
1074 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
))
1075 memcpy(&hdr
->verf
, &data
->verf
, sizeof(hdr
->verf
));
1076 else if (memcmp(&hdr
->verf
, &data
->verf
, sizeof(hdr
->verf
)))
1077 set_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
);
1078 spin_unlock(&hdr
->lock
);
1083 * Special version of should_remove_suid() that ignores capabilities.
1085 static int nfs_should_remove_suid(const struct inode
*inode
)
1087 umode_t mode
= inode
->i_mode
;
1090 /* suid always must be killed */
1091 if (unlikely(mode
& S_ISUID
))
1092 kill
= ATTR_KILL_SUID
;
1095 * sgid without any exec bits is just a mandatory locking mark; leave
1096 * it alone. If some exec bits are set, it's a real sgid; kill it.
1098 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1099 kill
|= ATTR_KILL_SGID
;
1101 if (unlikely(kill
&& S_ISREG(mode
)))
1108 * This function is called when the WRITE call is complete.
1110 static int nfs_writeback_done(struct rpc_task
*task
, struct nfs_pgio_data
*data
,
1111 struct inode
*inode
)
1116 * ->write_done will attempt to use post-op attributes to detect
1117 * conflicting writes by other clients. A strict interpretation
1118 * of close-to-open would allow us to continue caching even if
1119 * another writer had changed the file, but some applications
1120 * depend on tighter cache coherency when writing.
1122 status
= NFS_PROTO(inode
)->write_done(task
, data
);
1125 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, data
->res
.count
);
1127 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1128 if (data
->res
.verf
->committed
< data
->args
.stable
&& task
->tk_status
>= 0) {
1129 /* We tried a write call, but the server did not
1130 * commit data to stable storage even though we
1132 * Note: There is a known bug in Tru64 < 5.0 in which
1133 * the server reports NFS_DATA_SYNC, but performs
1134 * NFS_FILE_SYNC. We therefore implement this checking
1135 * as a dprintk() in order to avoid filling syslog.
1137 static unsigned long complain
;
1139 /* Note this will print the MDS for a DS write */
1140 if (time_before(complain
, jiffies
)) {
1141 dprintk("NFS: faulty NFS server %s:"
1142 " (committed = %d) != (stable = %d)\n",
1143 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1144 data
->res
.verf
->committed
, data
->args
.stable
);
1145 complain
= jiffies
+ 300 * HZ
;
1150 /* Deal with the suid/sgid bit corner case */
1151 if (nfs_should_remove_suid(inode
))
1152 nfs_mark_for_revalidate(inode
);
1157 * This function is called when the WRITE call is complete.
1159 static void nfs_writeback_result(struct rpc_task
*task
, struct nfs_pgio_data
*data
)
1161 struct nfs_pgio_args
*argp
= &data
->args
;
1162 struct nfs_pgio_res
*resp
= &data
->res
;
1164 if (resp
->count
< argp
->count
) {
1165 static unsigned long complain
;
1167 /* This a short write! */
1168 nfs_inc_stats(data
->header
->inode
, NFSIOS_SHORTWRITE
);
1170 /* Has the server at least made some progress? */
1171 if (resp
->count
== 0) {
1172 if (time_before(complain
, jiffies
)) {
1174 "NFS: Server wrote zero bytes, expected %u.\n",
1176 complain
= jiffies
+ 300 * HZ
;
1178 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
1179 task
->tk_status
= -EIO
;
1182 /* Was this an NFSv2 write or an NFSv3 stable write? */
1183 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1184 /* Resend from where the server left off */
1185 data
->mds_offset
+= resp
->count
;
1186 argp
->offset
+= resp
->count
;
1187 argp
->pgbase
+= resp
->count
;
1188 argp
->count
-= resp
->count
;
1190 /* Resend as a stable write in order to avoid
1191 * headaches in the case of a server crash.
1193 argp
->stable
= NFS_FILE_SYNC
;
1195 rpc_restart_call_prepare(task
);
1200 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1201 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1205 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1209 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1211 nfs_wait_bit_killable
,
1213 return (ret
< 0) ? ret
: 1;
1216 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1218 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1219 smp_mb__after_clear_bit();
1220 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1223 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1225 put_nfs_open_context(data
->context
);
1226 nfs_commit_free(data
);
1228 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1230 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1231 const struct rpc_call_ops
*call_ops
,
1234 struct rpc_task
*task
;
1235 int priority
= flush_task_priority(how
);
1236 struct rpc_message msg
= {
1237 .rpc_argp
= &data
->args
,
1238 .rpc_resp
= &data
->res
,
1239 .rpc_cred
= data
->cred
,
1241 struct rpc_task_setup task_setup_data
= {
1242 .task
= &data
->task
,
1244 .rpc_message
= &msg
,
1245 .callback_ops
= call_ops
,
1246 .callback_data
= data
,
1247 .workqueue
= nfsiod_workqueue
,
1248 .flags
= RPC_TASK_ASYNC
| flags
,
1249 .priority
= priority
,
1251 /* Set up the initial task struct. */
1252 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1254 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1256 nfs4_state_protect(NFS_SERVER(data
->inode
)->nfs_client
,
1257 NFS_SP4_MACH_CRED_COMMIT
, &task_setup_data
.rpc_client
, &msg
);
1259 task
= rpc_run_task(&task_setup_data
);
1261 return PTR_ERR(task
);
1262 if (how
& FLUSH_SYNC
)
1263 rpc_wait_for_completion_task(task
);
1267 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1270 * Set up the argument/result storage required for the RPC call.
1272 void nfs_init_commit(struct nfs_commit_data
*data
,
1273 struct list_head
*head
,
1274 struct pnfs_layout_segment
*lseg
,
1275 struct nfs_commit_info
*cinfo
)
1277 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1278 struct inode
*inode
= first
->wb_context
->dentry
->d_inode
;
1280 /* Set up the RPC argument and reply structs
1281 * NB: take care not to mess about with data->commit et al. */
1283 list_splice_init(head
, &data
->pages
);
1285 data
->inode
= inode
;
1286 data
->cred
= first
->wb_context
->cred
;
1287 data
->lseg
= lseg
; /* reference transferred */
1288 data
->mds_ops
= &nfs_commit_ops
;
1289 data
->completion_ops
= cinfo
->completion_ops
;
1290 data
->dreq
= cinfo
->dreq
;
1292 data
->args
.fh
= NFS_FH(data
->inode
);
1293 /* Note: we always request a commit of the entire inode */
1294 data
->args
.offset
= 0;
1295 data
->args
.count
= 0;
1296 data
->context
= get_nfs_open_context(first
->wb_context
);
1297 data
->res
.fattr
= &data
->fattr
;
1298 data
->res
.verf
= &data
->verf
;
1299 nfs_fattr_init(&data
->fattr
);
1301 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1303 void nfs_retry_commit(struct list_head
*page_list
,
1304 struct pnfs_layout_segment
*lseg
,
1305 struct nfs_commit_info
*cinfo
)
1307 struct nfs_page
*req
;
1309 while (!list_empty(page_list
)) {
1310 req
= nfs_list_entry(page_list
->next
);
1311 nfs_list_remove_request(req
);
1312 nfs_mark_request_commit(req
, lseg
, cinfo
);
1314 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1315 dec_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
1318 nfs_unlock_and_release_request(req
);
1321 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1324 * Commit dirty pages
1327 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1328 struct nfs_commit_info
*cinfo
)
1330 struct nfs_commit_data
*data
;
1332 data
= nfs_commitdata_alloc();
1337 /* Set up the argument struct */
1338 nfs_init_commit(data
, head
, NULL
, cinfo
);
1339 atomic_inc(&cinfo
->mds
->rpcs_out
);
1340 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, data
->mds_ops
,
1343 nfs_retry_commit(head
, NULL
, cinfo
);
1344 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1349 * COMMIT call returned
1351 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1353 struct nfs_commit_data
*data
= calldata
;
1355 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1356 task
->tk_pid
, task
->tk_status
);
1358 /* Call the NFS version-specific code */
1359 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1362 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1364 struct nfs_page
*req
;
1365 int status
= data
->task
.tk_status
;
1366 struct nfs_commit_info cinfo
;
1368 while (!list_empty(&data
->pages
)) {
1369 req
= nfs_list_entry(data
->pages
.next
);
1370 nfs_list_remove_request(req
);
1371 nfs_clear_page_commit(req
->wb_page
);
1373 dprintk("NFS: commit (%s/%llu %d@%lld)",
1374 req
->wb_context
->dentry
->d_sb
->s_id
,
1375 (unsigned long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
1377 (long long)req_offset(req
));
1379 nfs_context_set_write_error(req
->wb_context
, status
);
1380 nfs_inode_remove_request(req
);
1381 dprintk(", error = %d\n", status
);
1385 /* Okay, COMMIT succeeded, apparently. Check the verifier
1386 * returned by the server against all stored verfs. */
1387 if (!memcmp(&req
->wb_verf
, &data
->verf
.verifier
, sizeof(req
->wb_verf
))) {
1388 /* We have a match */
1389 nfs_inode_remove_request(req
);
1393 /* We have a mismatch. Write the page again */
1394 dprintk(" mismatch\n");
1395 nfs_mark_request_dirty(req
);
1396 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1398 nfs_unlock_and_release_request(req
);
1400 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1401 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1402 nfs_commit_clear_lock(NFS_I(data
->inode
));
1405 static void nfs_commit_release(void *calldata
)
1407 struct nfs_commit_data
*data
= calldata
;
1409 data
->completion_ops
->completion(data
);
1410 nfs_commitdata_release(calldata
);
1413 static const struct rpc_call_ops nfs_commit_ops
= {
1414 .rpc_call_prepare
= nfs_commit_prepare
,
1415 .rpc_call_done
= nfs_commit_done
,
1416 .rpc_release
= nfs_commit_release
,
1419 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1420 .completion
= nfs_commit_release_pages
,
1421 .error_cleanup
= nfs_commit_clear_lock
,
1424 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1425 int how
, struct nfs_commit_info
*cinfo
)
1429 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1430 if (status
== PNFS_NOT_ATTEMPTED
)
1431 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1435 int nfs_commit_inode(struct inode
*inode
, int how
)
1438 struct nfs_commit_info cinfo
;
1439 int may_wait
= how
& FLUSH_SYNC
;
1442 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1444 goto out_mark_dirty
;
1445 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1446 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1450 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1454 goto out_mark_dirty
;
1455 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1457 nfs_wait_bit_killable
,
1462 nfs_commit_clear_lock(NFS_I(inode
));
1464 /* Note: If we exit without ensuring that the commit is complete,
1465 * we must mark the inode as dirty. Otherwise, future calls to
1466 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1467 * that the data is on the disk.
1470 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1474 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1476 struct nfs_inode
*nfsi
= NFS_I(inode
);
1477 int flags
= FLUSH_SYNC
;
1480 /* no commits means nothing needs to be done */
1481 if (!nfsi
->commit_info
.ncommit
)
1484 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1485 /* Don't commit yet if this is a non-blocking flush and there
1486 * are a lot of outstanding writes for this mapping.
1488 if (nfsi
->commit_info
.ncommit
<= (nfsi
->npages
>> 1))
1489 goto out_mark_dirty
;
1491 /* don't wait for the COMMIT response */
1495 ret
= nfs_commit_inode(inode
, flags
);
1497 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1498 if (ret
< wbc
->nr_to_write
)
1499 wbc
->nr_to_write
-= ret
;
1501 wbc
->nr_to_write
= 0;
1506 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1510 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1516 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1518 return nfs_commit_unstable_pages(inode
, wbc
);
1520 EXPORT_SYMBOL_GPL(nfs_write_inode
);
1523 * flush the inode to disk.
1525 int nfs_wb_all(struct inode
*inode
)
1527 struct writeback_control wbc
= {
1528 .sync_mode
= WB_SYNC_ALL
,
1529 .nr_to_write
= LONG_MAX
,
1531 .range_end
= LLONG_MAX
,
1535 trace_nfs_writeback_inode_enter(inode
);
1537 ret
= sync_inode(inode
, &wbc
);
1539 trace_nfs_writeback_inode_exit(inode
, ret
);
1542 EXPORT_SYMBOL_GPL(nfs_wb_all
);
1544 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1546 struct nfs_page
*req
;
1550 wait_on_page_writeback(page
);
1551 req
= nfs_page_find_request(page
);
1554 if (nfs_lock_request(req
)) {
1555 nfs_clear_request_commit(req
);
1556 nfs_inode_remove_request(req
);
1558 * In case nfs_inode_remove_request has marked the
1559 * page as being dirty
1561 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1562 nfs_unlock_and_release_request(req
);
1565 ret
= nfs_wait_on_request(req
);
1566 nfs_release_request(req
);
1574 * Write back all requests on one page - we do this before reading it.
1576 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1578 loff_t range_start
= page_file_offset(page
);
1579 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1580 struct writeback_control wbc
= {
1581 .sync_mode
= WB_SYNC_ALL
,
1583 .range_start
= range_start
,
1584 .range_end
= range_end
,
1588 trace_nfs_writeback_page_enter(inode
);
1591 wait_on_page_writeback(page
);
1592 if (clear_page_dirty_for_io(page
)) {
1593 ret
= nfs_writepage_locked(page
, &wbc
);
1599 if (!PagePrivate(page
))
1601 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1606 trace_nfs_writeback_page_exit(inode
, ret
);
1610 #ifdef CONFIG_MIGRATION
1611 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1612 struct page
*page
, enum migrate_mode mode
)
1615 * If PagePrivate is set, then the page is currently associated with
1616 * an in-progress read or write request. Don't try to migrate it.
1618 * FIXME: we could do this in principle, but we'll need a way to ensure
1619 * that we can safely release the inode reference while holding
1622 if (PagePrivate(page
))
1625 if (!nfs_fscache_release_page(page
, GFP_KERNEL
))
1628 return migrate_page(mapping
, newpage
, page
, mode
);
1632 int __init
nfs_init_writepagecache(void)
1634 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1635 sizeof(struct nfs_rw_header
),
1636 0, SLAB_HWCACHE_ALIGN
,
1638 if (nfs_wdata_cachep
== NULL
)
1641 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1643 if (nfs_wdata_mempool
== NULL
)
1644 goto out_destroy_write_cache
;
1646 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1647 sizeof(struct nfs_commit_data
),
1648 0, SLAB_HWCACHE_ALIGN
,
1650 if (nfs_cdata_cachep
== NULL
)
1651 goto out_destroy_write_mempool
;
1653 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1655 if (nfs_commit_mempool
== NULL
)
1656 goto out_destroy_commit_cache
;
1659 * NFS congestion size, scale with available memory.
1671 * This allows larger machines to have larger/more transfers.
1672 * Limit the default to 256M
1674 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1675 if (nfs_congestion_kb
> 256*1024)
1676 nfs_congestion_kb
= 256*1024;
1680 out_destroy_commit_cache
:
1681 kmem_cache_destroy(nfs_cdata_cachep
);
1682 out_destroy_write_mempool
:
1683 mempool_destroy(nfs_wdata_mempool
);
1684 out_destroy_write_cache
:
1685 kmem_cache_destroy(nfs_wdata_cachep
);
1689 void nfs_destroy_writepagecache(void)
1691 mempool_destroy(nfs_commit_mempool
);
1692 kmem_cache_destroy(nfs_cdata_cachep
);
1693 mempool_destroy(nfs_wdata_mempool
);
1694 kmem_cache_destroy(nfs_wdata_cachep
);
1697 static const struct nfs_rw_ops nfs_rw_write_ops
= {
1698 .rw_mode
= FMODE_WRITE
,
1699 .rw_alloc_header
= nfs_writehdr_alloc
,
1700 .rw_free_header
= nfs_writehdr_free
,
1701 .rw_release
= nfs_writeback_release_common
,
1702 .rw_done
= nfs_writeback_done
,
1703 .rw_result
= nfs_writeback_result
,
1704 .rw_initiate
= nfs_initiate_write
,