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_NOIO
);
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_NOIO
);
76 struct nfs_pgio_header
*hdr
= &p
->header
;
78 memset(p
, 0, sizeof(*p
));
79 INIT_LIST_HEAD(&hdr
->pages
);
80 INIT_LIST_HEAD(&hdr
->rpc_list
);
81 spin_lock_init(&hdr
->lock
);
82 atomic_set(&hdr
->refcnt
, 0);
87 EXPORT_SYMBOL_GPL(nfs_writehdr_alloc
);
89 static struct nfs_write_data
*nfs_writedata_alloc(struct nfs_pgio_header
*hdr
,
90 unsigned int pagecount
)
92 struct nfs_write_data
*data
, *prealloc
;
94 prealloc
= &container_of(hdr
, struct nfs_write_header
, header
)->rpc_data
;
95 if (prealloc
->header
== NULL
)
98 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
102 if (nfs_pgarray_set(&data
->pages
, pagecount
)) {
104 atomic_inc(&hdr
->refcnt
);
106 if (data
!= prealloc
)
114 void nfs_writehdr_free(struct nfs_pgio_header
*hdr
)
116 struct nfs_write_header
*whdr
= container_of(hdr
, struct nfs_write_header
, header
);
117 mempool_free(whdr
, nfs_wdata_mempool
);
119 EXPORT_SYMBOL_GPL(nfs_writehdr_free
);
121 void nfs_writedata_release(struct nfs_write_data
*wdata
)
123 struct nfs_pgio_header
*hdr
= wdata
->header
;
124 struct nfs_write_header
*write_header
= container_of(hdr
, struct nfs_write_header
, header
);
126 put_nfs_open_context(wdata
->args
.context
);
127 if (wdata
->pages
.pagevec
!= wdata
->pages
.page_array
)
128 kfree(wdata
->pages
.pagevec
);
129 if (wdata
== &write_header
->rpc_data
) {
130 wdata
->header
= NULL
;
133 if (atomic_dec_and_test(&hdr
->refcnt
))
134 hdr
->completion_ops
->completion(hdr
);
135 /* Note: we only free the rpc_task after callbacks are done.
136 * See the comment in rpc_free_task() for why
140 EXPORT_SYMBOL_GPL(nfs_writedata_release
);
142 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
146 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
149 static struct nfs_page
*
150 nfs_page_find_request_locked(struct nfs_inode
*nfsi
, struct page
*page
)
152 struct nfs_page
*req
= NULL
;
154 if (PagePrivate(page
))
155 req
= (struct nfs_page
*)page_private(page
);
156 else if (unlikely(PageSwapCache(page
))) {
157 struct nfs_page
*freq
, *t
;
159 /* Linearly search the commit list for the correct req */
160 list_for_each_entry_safe(freq
, t
, &nfsi
->commit_info
.list
, wb_list
) {
161 if (freq
->wb_page
== page
) {
169 kref_get(&req
->wb_kref
);
174 static struct nfs_page
*nfs_page_find_request(struct page
*page
)
176 struct inode
*inode
= page_file_mapping(page
)->host
;
177 struct nfs_page
*req
= NULL
;
179 spin_lock(&inode
->i_lock
);
180 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
181 spin_unlock(&inode
->i_lock
);
185 /* Adjust the file length if we're writing beyond the end */
186 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
188 struct inode
*inode
= page_file_mapping(page
)->host
;
192 spin_lock(&inode
->i_lock
);
193 i_size
= i_size_read(inode
);
194 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
195 if (i_size
> 0 && page_file_index(page
) < end_index
)
197 end
= page_file_offset(page
) + ((loff_t
)offset
+count
);
200 i_size_write(inode
, end
);
201 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
203 spin_unlock(&inode
->i_lock
);
206 /* A writeback failed: mark the page as bad, and invalidate the page cache */
207 static void nfs_set_pageerror(struct page
*page
)
209 nfs_zap_mapping(page_file_mapping(page
)->host
, page_file_mapping(page
));
212 /* We can set the PG_uptodate flag if we see that a write request
213 * covers the full page.
215 static void nfs_mark_uptodate(struct page
*page
, unsigned int base
, unsigned int count
)
217 if (PageUptodate(page
))
221 if (count
!= nfs_page_length(page
))
223 SetPageUptodate(page
);
226 static int wb_priority(struct writeback_control
*wbc
)
228 if (wbc
->for_reclaim
)
229 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
230 if (wbc
->for_kupdate
|| wbc
->for_background
)
231 return FLUSH_LOWPRI
| FLUSH_COND_STABLE
;
232 return FLUSH_COND_STABLE
;
236 * NFS congestion control
239 int nfs_congestion_kb
;
241 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
242 #define NFS_CONGESTION_OFF_THRESH \
243 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
245 static void nfs_set_page_writeback(struct page
*page
)
247 struct nfs_server
*nfss
= NFS_SERVER(page_file_mapping(page
)->host
);
248 int ret
= test_set_page_writeback(page
);
250 WARN_ON_ONCE(ret
!= 0);
252 if (atomic_long_inc_return(&nfss
->writeback
) >
253 NFS_CONGESTION_ON_THRESH
) {
254 set_bdi_congested(&nfss
->backing_dev_info
,
259 static void nfs_end_page_writeback(struct page
*page
)
261 struct inode
*inode
= page_file_mapping(page
)->host
;
262 struct nfs_server
*nfss
= NFS_SERVER(inode
);
264 end_page_writeback(page
);
265 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
266 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
269 static struct nfs_page
*nfs_find_and_lock_request(struct page
*page
, bool nonblock
)
271 struct inode
*inode
= page_file_mapping(page
)->host
;
272 struct nfs_page
*req
;
275 spin_lock(&inode
->i_lock
);
277 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
280 if (nfs_lock_request(req
))
282 /* Note: If we hold the page lock, as is the case in nfs_writepage,
283 * then the call to nfs_lock_request() will always
284 * succeed provided that someone hasn't already marked the
285 * request as dirty (in which case we don't care).
287 spin_unlock(&inode
->i_lock
);
289 ret
= nfs_wait_on_request(req
);
292 nfs_release_request(req
);
295 spin_lock(&inode
->i_lock
);
297 spin_unlock(&inode
->i_lock
);
302 * Find an associated nfs write request, and prepare to flush it out
303 * May return an error if the user signalled nfs_wait_on_request().
305 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
306 struct page
*page
, bool nonblock
)
308 struct nfs_page
*req
;
311 req
= nfs_find_and_lock_request(page
, nonblock
);
318 nfs_set_page_writeback(page
);
319 WARN_ON_ONCE(test_bit(PG_CLEAN
, &req
->wb_flags
));
322 if (!nfs_pageio_add_request(pgio
, req
)) {
323 nfs_redirty_request(req
);
324 ret
= pgio
->pg_error
;
330 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
332 struct inode
*inode
= page_file_mapping(page
)->host
;
335 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
336 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
338 nfs_pageio_cond_complete(pgio
, page_file_index(page
));
339 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
340 if (ret
== -EAGAIN
) {
341 redirty_page_for_writepage(wbc
, page
);
348 * Write an mmapped page to the server.
350 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
352 struct nfs_pageio_descriptor pgio
;
355 NFS_PROTO(page_file_mapping(page
)->host
)->write_pageio_init(&pgio
,
358 &nfs_async_write_completion_ops
);
359 err
= nfs_do_writepage(page
, wbc
, &pgio
);
360 nfs_pageio_complete(&pgio
);
363 if (pgio
.pg_error
< 0)
364 return pgio
.pg_error
;
368 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
372 ret
= nfs_writepage_locked(page
, wbc
);
377 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
381 ret
= nfs_do_writepage(page
, wbc
, data
);
386 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
388 struct inode
*inode
= mapping
->host
;
389 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
390 struct nfs_pageio_descriptor pgio
;
393 /* Stop dirtying of new pages while we sync */
394 err
= wait_on_bit_lock(bitlock
, NFS_INO_FLUSHING
,
395 nfs_wait_bit_killable
, TASK_KILLABLE
);
399 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
401 NFS_PROTO(inode
)->write_pageio_init(&pgio
, inode
, wb_priority(wbc
), &nfs_async_write_completion_ops
);
402 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
403 nfs_pageio_complete(&pgio
);
405 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
406 smp_mb__after_clear_bit();
407 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
420 * Insert a write request into an inode
422 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
424 struct nfs_inode
*nfsi
= NFS_I(inode
);
426 /* Lock the request! */
427 nfs_lock_request(req
);
429 spin_lock(&inode
->i_lock
);
430 if (!nfsi
->npages
&& NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
433 * Swap-space should not get truncated. Hence no need to plug the race
434 * with invalidate/truncate.
436 if (likely(!PageSwapCache(req
->wb_page
))) {
437 set_bit(PG_MAPPED
, &req
->wb_flags
);
438 SetPagePrivate(req
->wb_page
);
439 set_page_private(req
->wb_page
, (unsigned long)req
);
442 kref_get(&req
->wb_kref
);
443 spin_unlock(&inode
->i_lock
);
447 * Remove a write request from an inode
449 static void nfs_inode_remove_request(struct nfs_page
*req
)
451 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
452 struct nfs_inode
*nfsi
= NFS_I(inode
);
454 spin_lock(&inode
->i_lock
);
455 if (likely(!PageSwapCache(req
->wb_page
))) {
456 set_page_private(req
->wb_page
, 0);
457 ClearPagePrivate(req
->wb_page
);
458 clear_bit(PG_MAPPED
, &req
->wb_flags
);
461 spin_unlock(&inode
->i_lock
);
462 nfs_release_request(req
);
466 nfs_mark_request_dirty(struct nfs_page
*req
)
468 __set_page_dirty_nobuffers(req
->wb_page
);
471 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
473 * nfs_request_add_commit_list - add request to a commit list
474 * @req: pointer to a struct nfs_page
475 * @dst: commit list head
476 * @cinfo: holds list lock and accounting info
478 * This sets the PG_CLEAN bit, updates the cinfo count of
479 * number of outstanding requests requiring a commit as well as
482 * The caller must _not_ hold the cinfo->lock, but must be
483 * holding the nfs_page lock.
486 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
487 struct nfs_commit_info
*cinfo
)
489 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
490 spin_lock(cinfo
->lock
);
491 nfs_list_add_request(req
, dst
);
492 cinfo
->mds
->ncommit
++;
493 spin_unlock(cinfo
->lock
);
495 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
496 inc_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
498 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
502 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
505 * nfs_request_remove_commit_list - Remove request from a commit list
506 * @req: pointer to a nfs_page
507 * @cinfo: holds list lock and accounting info
509 * This clears the PG_CLEAN bit, and updates the cinfo's count of
510 * number of outstanding requests requiring a commit
511 * It does not update the MM page stats.
513 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
516 nfs_request_remove_commit_list(struct nfs_page
*req
,
517 struct nfs_commit_info
*cinfo
)
519 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
521 nfs_list_remove_request(req
);
522 cinfo
->mds
->ncommit
--;
524 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
526 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
529 cinfo
->lock
= &inode
->i_lock
;
530 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
531 cinfo
->ds
= pnfs_get_ds_info(inode
);
533 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
536 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
538 struct nfs_direct_req
*dreq
)
541 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
543 nfs_init_cinfo_from_inode(cinfo
, inode
);
545 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
548 * Add a request to the inode's commit list.
551 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
552 struct nfs_commit_info
*cinfo
)
554 if (pnfs_mark_request_commit(req
, lseg
, cinfo
))
556 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
560 nfs_clear_page_commit(struct page
*page
)
562 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
563 dec_bdi_stat(page_file_mapping(page
)->backing_dev_info
, BDI_RECLAIMABLE
);
567 nfs_clear_request_commit(struct nfs_page
*req
)
569 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
570 struct inode
*inode
= req
->wb_context
->dentry
->d_inode
;
571 struct nfs_commit_info cinfo
;
573 nfs_init_cinfo_from_inode(&cinfo
, inode
);
574 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
575 spin_lock(cinfo
.lock
);
576 nfs_request_remove_commit_list(req
, &cinfo
);
577 spin_unlock(cinfo
.lock
);
579 nfs_clear_page_commit(req
->wb_page
);
584 int nfs_write_need_commit(struct nfs_write_data
*data
)
586 if (data
->verf
.committed
== NFS_DATA_SYNC
)
587 return data
->header
->lseg
== NULL
;
588 return data
->verf
.committed
!= NFS_FILE_SYNC
;
592 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
597 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
599 struct nfs_direct_req
*dreq
)
604 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
605 struct nfs_commit_info
*cinfo
)
610 nfs_clear_request_commit(struct nfs_page
*req
)
615 int nfs_write_need_commit(struct nfs_write_data
*data
)
622 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
624 struct nfs_commit_info cinfo
;
625 unsigned long bytes
= 0;
627 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
629 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
630 while (!list_empty(&hdr
->pages
)) {
631 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
633 bytes
+= req
->wb_bytes
;
634 nfs_list_remove_request(req
);
635 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
636 (hdr
->good_bytes
< bytes
)) {
637 nfs_set_pageerror(req
->wb_page
);
638 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
641 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
)) {
642 nfs_mark_request_dirty(req
);
645 if (test_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
)) {
646 memcpy(&req
->wb_verf
, &hdr
->verf
->verifier
, sizeof(req
->wb_verf
));
647 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
);
651 nfs_inode_remove_request(req
);
653 nfs_unlock_request(req
);
654 nfs_end_page_writeback(req
->wb_page
);
655 nfs_release_request(req
);
661 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
663 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
665 return cinfo
->mds
->ncommit
;
668 /* cinfo->lock held by caller */
670 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
671 struct nfs_commit_info
*cinfo
, int max
)
673 struct nfs_page
*req
, *tmp
;
676 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
677 if (!nfs_lock_request(req
))
679 kref_get(&req
->wb_kref
);
680 if (cond_resched_lock(cinfo
->lock
))
681 list_safe_reset_next(req
, tmp
, wb_list
);
682 nfs_request_remove_commit_list(req
, cinfo
);
683 nfs_list_add_request(req
, dst
);
685 if ((ret
== max
) && !cinfo
->dreq
)
692 * nfs_scan_commit - Scan an inode for commit requests
693 * @inode: NFS inode to scan
694 * @dst: mds destination list
695 * @cinfo: mds and ds lists of reqs ready to commit
697 * Moves requests from the inode's 'commit' request list.
698 * The requests are *not* checked to ensure that they form a contiguous set.
701 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
702 struct nfs_commit_info
*cinfo
)
706 spin_lock(cinfo
->lock
);
707 if (cinfo
->mds
->ncommit
> 0) {
708 const int max
= INT_MAX
;
710 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
712 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
714 spin_unlock(cinfo
->lock
);
719 static unsigned long nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
724 int nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
725 struct nfs_commit_info
*cinfo
)
732 * Search for an existing write request, and attempt to update
733 * it to reflect a new dirty region on a given page.
735 * If the attempt fails, then the existing request is flushed out
738 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
743 struct nfs_page
*req
;
748 if (!PagePrivate(page
))
751 end
= offset
+ bytes
;
752 spin_lock(&inode
->i_lock
);
755 req
= nfs_page_find_request_locked(NFS_I(inode
), page
);
759 rqend
= req
->wb_offset
+ req
->wb_bytes
;
761 * Tell the caller to flush out the request if
762 * the offsets are non-contiguous.
763 * Note: nfs_flush_incompatible() will already
764 * have flushed out requests having wrong owners.
767 || end
< req
->wb_offset
)
770 if (nfs_lock_request(req
))
773 /* The request is locked, so wait and then retry */
774 spin_unlock(&inode
->i_lock
);
775 error
= nfs_wait_on_request(req
);
776 nfs_release_request(req
);
779 spin_lock(&inode
->i_lock
);
782 /* Okay, the request matches. Update the region */
783 if (offset
< req
->wb_offset
) {
784 req
->wb_offset
= offset
;
785 req
->wb_pgbase
= offset
;
788 req
->wb_bytes
= end
- req
->wb_offset
;
790 req
->wb_bytes
= rqend
- req
->wb_offset
;
792 spin_unlock(&inode
->i_lock
);
794 nfs_clear_request_commit(req
);
797 spin_unlock(&inode
->i_lock
);
798 nfs_release_request(req
);
799 error
= nfs_wb_page(inode
, page
);
801 return ERR_PTR(error
);
805 * Try to update an existing write request, or create one if there is none.
807 * Note: Should always be called with the Page Lock held to prevent races
808 * if we have to add a new request. Also assumes that the caller has
809 * already called nfs_flush_incompatible() if necessary.
811 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
812 struct page
*page
, unsigned int offset
, unsigned int bytes
)
814 struct inode
*inode
= page_file_mapping(page
)->host
;
815 struct nfs_page
*req
;
817 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
820 req
= nfs_create_request(ctx
, inode
, page
, offset
, bytes
);
823 nfs_inode_add_request(inode
, req
);
828 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
829 unsigned int offset
, unsigned int count
)
831 struct nfs_page
*req
;
833 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
836 /* Update file length */
837 nfs_grow_file(page
, offset
, count
);
838 nfs_mark_uptodate(page
, req
->wb_pgbase
, req
->wb_bytes
);
839 nfs_mark_request_dirty(req
);
840 nfs_unlock_and_release_request(req
);
844 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
846 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
847 struct nfs_lock_context
*l_ctx
;
848 struct nfs_page
*req
;
849 int do_flush
, status
;
851 * Look for a request corresponding to this page. If there
852 * is one, and it belongs to another file, we flush it out
853 * before we try to copy anything into the page. Do this
854 * due to the lack of an ACCESS-type call in NFSv2.
855 * Also do the same if we find a request from an existing
859 req
= nfs_page_find_request(page
);
862 l_ctx
= req
->wb_lock_context
;
863 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
;
865 do_flush
|= l_ctx
->lockowner
.l_owner
!= current
->files
866 || l_ctx
->lockowner
.l_pid
!= current
->tgid
;
868 nfs_release_request(req
);
871 status
= nfs_wb_page(page_file_mapping(page
)->host
, page
);
872 } while (status
== 0);
877 * If the page cache is marked as unsafe or invalid, then we can't rely on
878 * the PageUptodate() flag. In this case, we will need to turn off
879 * write optimisations that depend on the page contents being correct.
881 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
883 if (nfs_have_delegated_attributes(inode
))
885 if (NFS_I(inode
)->cache_validity
& (NFS_INO_INVALID_DATA
|NFS_INO_REVAL_PAGECACHE
))
888 return PageUptodate(page
) != 0;
891 /* If we know the page is up to date, and we're not using byte range locks (or
892 * if we have the whole file locked for writing), it may be more efficient to
893 * extend the write to cover the entire page in order to avoid fragmentation
896 * If the file is opened for synchronous writes or if we have a write delegation
897 * from the server then we can just skip the rest of the checks.
899 static int nfs_can_extend_write(struct file
*file
, struct page
*page
, struct inode
*inode
)
901 if (file
->f_flags
& O_DSYNC
)
903 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
905 if (nfs_write_pageuptodate(page
, inode
) && (inode
->i_flock
== NULL
||
906 (inode
->i_flock
->fl_start
== 0 &&
907 inode
->i_flock
->fl_end
== OFFSET_MAX
&&
908 inode
->i_flock
->fl_type
!= F_RDLCK
)))
914 * Update and possibly write a cached page of an NFS file.
916 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
917 * things with a page scheduled for an RPC call (e.g. invalidate it).
919 int nfs_updatepage(struct file
*file
, struct page
*page
,
920 unsigned int offset
, unsigned int count
)
922 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
923 struct inode
*inode
= page_file_mapping(page
)->host
;
926 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
928 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
929 file
->f_path
.dentry
->d_parent
->d_name
.name
,
930 file
->f_path
.dentry
->d_name
.name
, count
,
931 (long long)(page_file_offset(page
) + offset
));
933 if (nfs_can_extend_write(file
, page
, inode
)) {
934 count
= max(count
+ offset
, nfs_page_length(page
));
938 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
940 nfs_set_pageerror(page
);
942 __set_page_dirty_nobuffers(page
);
944 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
945 status
, (long long)i_size_read(inode
));
949 static int flush_task_priority(int how
)
951 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
953 return RPC_PRIORITY_HIGH
;
955 return RPC_PRIORITY_LOW
;
957 return RPC_PRIORITY_NORMAL
;
960 int nfs_initiate_write(struct rpc_clnt
*clnt
,
961 struct nfs_write_data
*data
,
962 const struct rpc_call_ops
*call_ops
,
965 struct inode
*inode
= data
->header
->inode
;
966 int priority
= flush_task_priority(how
);
967 struct rpc_task
*task
;
968 struct rpc_message msg
= {
969 .rpc_argp
= &data
->args
,
970 .rpc_resp
= &data
->res
,
971 .rpc_cred
= data
->header
->cred
,
973 struct rpc_task_setup task_setup_data
= {
977 .callback_ops
= call_ops
,
978 .callback_data
= data
,
979 .workqueue
= nfsiod_workqueue
,
980 .flags
= RPC_TASK_ASYNC
| flags
,
981 .priority
= priority
,
985 /* Set up the initial task struct. */
986 NFS_PROTO(inode
)->write_setup(data
, &msg
);
988 dprintk("NFS: %5u initiated write call "
989 "(req %s/%lld, %u bytes @ offset %llu)\n",
992 (long long)NFS_FILEID(inode
),
994 (unsigned long long)data
->args
.offset
);
996 task
= rpc_run_task(&task_setup_data
);
1001 if (how
& FLUSH_SYNC
) {
1002 ret
= rpc_wait_for_completion_task(task
);
1004 ret
= task
->tk_status
;
1010 EXPORT_SYMBOL_GPL(nfs_initiate_write
);
1013 * Set up the argument/result storage required for the RPC call.
1015 static void nfs_write_rpcsetup(struct nfs_write_data
*data
,
1016 unsigned int count
, unsigned int offset
,
1017 int how
, struct nfs_commit_info
*cinfo
)
1019 struct nfs_page
*req
= data
->header
->req
;
1021 /* Set up the RPC argument and reply structs
1022 * NB: take care not to mess about with data->commit et al. */
1024 data
->args
.fh
= NFS_FH(data
->header
->inode
);
1025 data
->args
.offset
= req_offset(req
) + offset
;
1026 /* pnfs_set_layoutcommit needs this */
1027 data
->mds_offset
= data
->args
.offset
;
1028 data
->args
.pgbase
= req
->wb_pgbase
+ offset
;
1029 data
->args
.pages
= data
->pages
.pagevec
;
1030 data
->args
.count
= count
;
1031 data
->args
.context
= get_nfs_open_context(req
->wb_context
);
1032 data
->args
.lock_context
= req
->wb_lock_context
;
1033 data
->args
.stable
= NFS_UNSTABLE
;
1034 switch (how
& (FLUSH_STABLE
| FLUSH_COND_STABLE
)) {
1037 case FLUSH_COND_STABLE
:
1038 if (nfs_reqs_to_commit(cinfo
))
1041 data
->args
.stable
= NFS_FILE_SYNC
;
1044 data
->res
.fattr
= &data
->fattr
;
1045 data
->res
.count
= count
;
1046 data
->res
.verf
= &data
->verf
;
1047 nfs_fattr_init(&data
->fattr
);
1050 static int nfs_do_write(struct nfs_write_data
*data
,
1051 const struct rpc_call_ops
*call_ops
,
1054 struct inode
*inode
= data
->header
->inode
;
1056 return nfs_initiate_write(NFS_CLIENT(inode
), data
, call_ops
, how
, 0);
1059 static int nfs_do_multiple_writes(struct list_head
*head
,
1060 const struct rpc_call_ops
*call_ops
,
1063 struct nfs_write_data
*data
;
1066 while (!list_empty(head
)) {
1069 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1070 list_del_init(&data
->list
);
1072 ret2
= nfs_do_write(data
, call_ops
, how
);
1079 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1080 * call this on each, which will prepare them to be retried on next
1081 * writeback using standard nfs.
1083 static void nfs_redirty_request(struct nfs_page
*req
)
1085 nfs_mark_request_dirty(req
);
1086 nfs_unlock_request(req
);
1087 nfs_end_page_writeback(req
->wb_page
);
1088 nfs_release_request(req
);
1091 static void nfs_async_write_error(struct list_head
*head
)
1093 struct nfs_page
*req
;
1095 while (!list_empty(head
)) {
1096 req
= nfs_list_entry(head
->next
);
1097 nfs_list_remove_request(req
);
1098 nfs_redirty_request(req
);
1102 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1103 .error_cleanup
= nfs_async_write_error
,
1104 .completion
= nfs_write_completion
,
1107 static void nfs_flush_error(struct nfs_pageio_descriptor
*desc
,
1108 struct nfs_pgio_header
*hdr
)
1110 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
1111 while (!list_empty(&hdr
->rpc_list
)) {
1112 struct nfs_write_data
*data
= list_first_entry(&hdr
->rpc_list
,
1113 struct nfs_write_data
, list
);
1114 list_del(&data
->list
);
1115 nfs_writedata_release(data
);
1117 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1121 * Generate multiple small requests to write out a single
1122 * contiguous dirty area on one page.
1124 static int nfs_flush_multi(struct nfs_pageio_descriptor
*desc
,
1125 struct nfs_pgio_header
*hdr
)
1127 struct nfs_page
*req
= hdr
->req
;
1128 struct page
*page
= req
->wb_page
;
1129 struct nfs_write_data
*data
;
1130 size_t wsize
= desc
->pg_bsize
, nbytes
;
1131 unsigned int offset
;
1133 struct nfs_commit_info cinfo
;
1135 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1137 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1138 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
) ||
1139 desc
->pg_count
> wsize
))
1140 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1144 nbytes
= desc
->pg_count
;
1146 size_t len
= min(nbytes
, wsize
);
1148 data
= nfs_writedata_alloc(hdr
, 1);
1150 nfs_flush_error(desc
, hdr
);
1153 data
->pages
.pagevec
[0] = page
;
1154 nfs_write_rpcsetup(data
, len
, offset
, desc
->pg_ioflags
, &cinfo
);
1155 list_add(&data
->list
, &hdr
->rpc_list
);
1159 } while (nbytes
!= 0);
1160 nfs_list_remove_request(req
);
1161 nfs_list_add_request(req
, &hdr
->pages
);
1162 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1167 * Create an RPC task for the given write request and kick it.
1168 * The page must have been locked by the caller.
1170 * It may happen that the page we're passed is not marked dirty.
1171 * This is the case if nfs_updatepage detects a conflicting request
1172 * that has been written but not committed.
1174 static int nfs_flush_one(struct nfs_pageio_descriptor
*desc
,
1175 struct nfs_pgio_header
*hdr
)
1177 struct nfs_page
*req
;
1178 struct page
**pages
;
1179 struct nfs_write_data
*data
;
1180 struct list_head
*head
= &desc
->pg_list
;
1181 struct nfs_commit_info cinfo
;
1183 data
= nfs_writedata_alloc(hdr
, nfs_page_array_len(desc
->pg_base
,
1186 nfs_flush_error(desc
, hdr
);
1190 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
1191 pages
= data
->pages
.pagevec
;
1192 while (!list_empty(head
)) {
1193 req
= nfs_list_entry(head
->next
);
1194 nfs_list_remove_request(req
);
1195 nfs_list_add_request(req
, &hdr
->pages
);
1196 *pages
++ = req
->wb_page
;
1199 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
1200 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
)))
1201 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
1203 /* Set up the argument struct */
1204 nfs_write_rpcsetup(data
, desc
->pg_count
, 0, desc
->pg_ioflags
, &cinfo
);
1205 list_add(&data
->list
, &hdr
->rpc_list
);
1206 desc
->pg_rpc_callops
= &nfs_write_common_ops
;
1210 int nfs_generic_flush(struct nfs_pageio_descriptor
*desc
,
1211 struct nfs_pgio_header
*hdr
)
1213 if (desc
->pg_bsize
< PAGE_CACHE_SIZE
)
1214 return nfs_flush_multi(desc
, hdr
);
1215 return nfs_flush_one(desc
, hdr
);
1217 EXPORT_SYMBOL_GPL(nfs_generic_flush
);
1219 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1221 struct nfs_write_header
*whdr
;
1222 struct nfs_pgio_header
*hdr
;
1225 whdr
= nfs_writehdr_alloc();
1227 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1230 hdr
= &whdr
->header
;
1231 nfs_pgheader_init(desc
, hdr
, nfs_writehdr_free
);
1232 atomic_inc(&hdr
->refcnt
);
1233 ret
= nfs_generic_flush(desc
, hdr
);
1235 ret
= nfs_do_multiple_writes(&hdr
->rpc_list
,
1236 desc
->pg_rpc_callops
,
1238 if (atomic_dec_and_test(&hdr
->refcnt
))
1239 hdr
->completion_ops
->completion(hdr
);
1243 static const struct nfs_pageio_ops nfs_pageio_write_ops
= {
1244 .pg_test
= nfs_generic_pg_test
,
1245 .pg_doio
= nfs_generic_pg_writepages
,
1248 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1249 struct inode
*inode
, int ioflags
,
1250 const struct nfs_pgio_completion_ops
*compl_ops
)
1252 nfs_pageio_init(pgio
, inode
, &nfs_pageio_write_ops
, compl_ops
,
1253 NFS_SERVER(inode
)->wsize
, ioflags
);
1255 EXPORT_SYMBOL_GPL(nfs_pageio_init_write
);
1257 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1259 pgio
->pg_ops
= &nfs_pageio_write_ops
;
1260 pgio
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1262 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1265 void nfs_write_prepare(struct rpc_task
*task
, void *calldata
)
1267 struct nfs_write_data
*data
= calldata
;
1268 NFS_PROTO(data
->header
->inode
)->write_rpc_prepare(task
, data
);
1269 if (unlikely(test_bit(NFS_CONTEXT_BAD
, &data
->args
.context
->flags
)))
1270 rpc_exit(task
, -EIO
);
1273 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1275 struct nfs_commit_data
*data
= calldata
;
1277 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1281 * Handle a write reply that flushes a whole page.
1283 * FIXME: There is an inherent race with invalidate_inode_pages and
1284 * writebacks since the page->count is kept > 1 for as long
1285 * as the page has a write request pending.
1287 static void nfs_writeback_done_common(struct rpc_task
*task
, void *calldata
)
1289 struct nfs_write_data
*data
= calldata
;
1291 nfs_writeback_done(task
, data
);
1294 static void nfs_writeback_release_common(void *calldata
)
1296 struct nfs_write_data
*data
= calldata
;
1297 struct nfs_pgio_header
*hdr
= data
->header
;
1298 int status
= data
->task
.tk_status
;
1300 if ((status
>= 0) && nfs_write_need_commit(data
)) {
1301 spin_lock(&hdr
->lock
);
1302 if (test_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
))
1304 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT
, &hdr
->flags
))
1305 memcpy(hdr
->verf
, &data
->verf
, sizeof(*hdr
->verf
));
1306 else if (memcmp(hdr
->verf
, &data
->verf
, sizeof(*hdr
->verf
)))
1307 set_bit(NFS_IOHDR_NEED_RESCHED
, &hdr
->flags
);
1308 spin_unlock(&hdr
->lock
);
1310 nfs_writedata_release(data
);
1313 static const struct rpc_call_ops nfs_write_common_ops
= {
1314 .rpc_call_prepare
= nfs_write_prepare
,
1315 .rpc_call_done
= nfs_writeback_done_common
,
1316 .rpc_release
= nfs_writeback_release_common
,
1321 * This function is called when the WRITE call is complete.
1323 void nfs_writeback_done(struct rpc_task
*task
, struct nfs_write_data
*data
)
1325 struct nfs_writeargs
*argp
= &data
->args
;
1326 struct nfs_writeres
*resp
= &data
->res
;
1327 struct inode
*inode
= data
->header
->inode
;
1330 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1331 task
->tk_pid
, task
->tk_status
);
1334 * ->write_done will attempt to use post-op attributes to detect
1335 * conflicting writes by other clients. A strict interpretation
1336 * of close-to-open would allow us to continue caching even if
1337 * another writer had changed the file, but some applications
1338 * depend on tighter cache coherency when writing.
1340 status
= NFS_PROTO(inode
)->write_done(task
, data
);
1343 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, resp
->count
);
1345 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1346 if (resp
->verf
->committed
< argp
->stable
&& task
->tk_status
>= 0) {
1347 /* We tried a write call, but the server did not
1348 * commit data to stable storage even though we
1350 * Note: There is a known bug in Tru64 < 5.0 in which
1351 * the server reports NFS_DATA_SYNC, but performs
1352 * NFS_FILE_SYNC. We therefore implement this checking
1353 * as a dprintk() in order to avoid filling syslog.
1355 static unsigned long complain
;
1357 /* Note this will print the MDS for a DS write */
1358 if (time_before(complain
, jiffies
)) {
1359 dprintk("NFS: faulty NFS server %s:"
1360 " (committed = %d) != (stable = %d)\n",
1361 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1362 resp
->verf
->committed
, argp
->stable
);
1363 complain
= jiffies
+ 300 * HZ
;
1367 if (task
->tk_status
< 0)
1368 nfs_set_pgio_error(data
->header
, task
->tk_status
, argp
->offset
);
1369 else if (resp
->count
< argp
->count
) {
1370 static unsigned long complain
;
1372 /* This a short write! */
1373 nfs_inc_stats(inode
, NFSIOS_SHORTWRITE
);
1375 /* Has the server at least made some progress? */
1376 if (resp
->count
== 0) {
1377 if (time_before(complain
, jiffies
)) {
1379 "NFS: Server wrote zero bytes, expected %u.\n",
1381 complain
= jiffies
+ 300 * HZ
;
1383 nfs_set_pgio_error(data
->header
, -EIO
, argp
->offset
);
1384 task
->tk_status
= -EIO
;
1387 /* Was this an NFSv2 write or an NFSv3 stable write? */
1388 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1389 /* Resend from where the server left off */
1390 data
->mds_offset
+= resp
->count
;
1391 argp
->offset
+= resp
->count
;
1392 argp
->pgbase
+= resp
->count
;
1393 argp
->count
-= resp
->count
;
1395 /* Resend as a stable write in order to avoid
1396 * headaches in the case of a server crash.
1398 argp
->stable
= NFS_FILE_SYNC
;
1400 rpc_restart_call_prepare(task
);
1405 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1406 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1410 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1414 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1416 nfs_wait_bit_killable
,
1418 return (ret
< 0) ? ret
: 1;
1421 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1423 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1424 smp_mb__after_clear_bit();
1425 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1428 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1430 put_nfs_open_context(data
->context
);
1431 nfs_commit_free(data
);
1433 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1435 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1436 const struct rpc_call_ops
*call_ops
,
1439 struct rpc_task
*task
;
1440 int priority
= flush_task_priority(how
);
1441 struct rpc_message msg
= {
1442 .rpc_argp
= &data
->args
,
1443 .rpc_resp
= &data
->res
,
1444 .rpc_cred
= data
->cred
,
1446 struct rpc_task_setup task_setup_data
= {
1447 .task
= &data
->task
,
1449 .rpc_message
= &msg
,
1450 .callback_ops
= call_ops
,
1451 .callback_data
= data
,
1452 .workqueue
= nfsiod_workqueue
,
1453 .flags
= RPC_TASK_ASYNC
| flags
,
1454 .priority
= priority
,
1456 /* Set up the initial task struct. */
1457 NFS_PROTO(data
->inode
)->commit_setup(data
, &msg
);
1459 dprintk("NFS: %5u initiated commit call\n", data
->task
.tk_pid
);
1461 task
= rpc_run_task(&task_setup_data
);
1463 return PTR_ERR(task
);
1464 if (how
& FLUSH_SYNC
)
1465 rpc_wait_for_completion_task(task
);
1469 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1472 * Set up the argument/result storage required for the RPC call.
1474 void nfs_init_commit(struct nfs_commit_data
*data
,
1475 struct list_head
*head
,
1476 struct pnfs_layout_segment
*lseg
,
1477 struct nfs_commit_info
*cinfo
)
1479 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1480 struct inode
*inode
= first
->wb_context
->dentry
->d_inode
;
1482 /* Set up the RPC argument and reply structs
1483 * NB: take care not to mess about with data->commit et al. */
1485 list_splice_init(head
, &data
->pages
);
1487 data
->inode
= inode
;
1488 data
->cred
= first
->wb_context
->cred
;
1489 data
->lseg
= lseg
; /* reference transferred */
1490 data
->mds_ops
= &nfs_commit_ops
;
1491 data
->completion_ops
= cinfo
->completion_ops
;
1492 data
->dreq
= cinfo
->dreq
;
1494 data
->args
.fh
= NFS_FH(data
->inode
);
1495 /* Note: we always request a commit of the entire inode */
1496 data
->args
.offset
= 0;
1497 data
->args
.count
= 0;
1498 data
->context
= get_nfs_open_context(first
->wb_context
);
1499 data
->res
.fattr
= &data
->fattr
;
1500 data
->res
.verf
= &data
->verf
;
1501 nfs_fattr_init(&data
->fattr
);
1503 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1505 void nfs_retry_commit(struct list_head
*page_list
,
1506 struct pnfs_layout_segment
*lseg
,
1507 struct nfs_commit_info
*cinfo
)
1509 struct nfs_page
*req
;
1511 while (!list_empty(page_list
)) {
1512 req
= nfs_list_entry(page_list
->next
);
1513 nfs_list_remove_request(req
);
1514 nfs_mark_request_commit(req
, lseg
, cinfo
);
1516 dec_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1517 dec_bdi_stat(page_file_mapping(req
->wb_page
)->backing_dev_info
,
1520 nfs_unlock_and_release_request(req
);
1523 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1526 * Commit dirty pages
1529 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1530 struct nfs_commit_info
*cinfo
)
1532 struct nfs_commit_data
*data
;
1534 data
= nfs_commitdata_alloc();
1539 /* Set up the argument struct */
1540 nfs_init_commit(data
, head
, NULL
, cinfo
);
1541 atomic_inc(&cinfo
->mds
->rpcs_out
);
1542 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, data
->mds_ops
,
1545 nfs_retry_commit(head
, NULL
, cinfo
);
1546 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1551 * COMMIT call returned
1553 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1555 struct nfs_commit_data
*data
= calldata
;
1557 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1558 task
->tk_pid
, task
->tk_status
);
1560 /* Call the NFS version-specific code */
1561 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1564 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1566 struct nfs_page
*req
;
1567 int status
= data
->task
.tk_status
;
1568 struct nfs_commit_info cinfo
;
1570 while (!list_empty(&data
->pages
)) {
1571 req
= nfs_list_entry(data
->pages
.next
);
1572 nfs_list_remove_request(req
);
1573 nfs_clear_page_commit(req
->wb_page
);
1575 dprintk("NFS: commit (%s/%lld %d@%lld)",
1576 req
->wb_context
->dentry
->d_sb
->s_id
,
1577 (long long)NFS_FILEID(req
->wb_context
->dentry
->d_inode
),
1579 (long long)req_offset(req
));
1581 nfs_context_set_write_error(req
->wb_context
, status
);
1582 nfs_inode_remove_request(req
);
1583 dprintk(", error = %d\n", status
);
1587 /* Okay, COMMIT succeeded, apparently. Check the verifier
1588 * returned by the server against all stored verfs. */
1589 if (!memcmp(&req
->wb_verf
, &data
->verf
.verifier
, sizeof(req
->wb_verf
))) {
1590 /* We have a match */
1591 nfs_inode_remove_request(req
);
1595 /* We have a mismatch. Write the page again */
1596 dprintk(" mismatch\n");
1597 nfs_mark_request_dirty(req
);
1598 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1600 nfs_unlock_and_release_request(req
);
1602 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1603 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1604 nfs_commit_clear_lock(NFS_I(data
->inode
));
1607 static void nfs_commit_release(void *calldata
)
1609 struct nfs_commit_data
*data
= calldata
;
1611 data
->completion_ops
->completion(data
);
1612 nfs_commitdata_release(calldata
);
1615 static const struct rpc_call_ops nfs_commit_ops
= {
1616 .rpc_call_prepare
= nfs_commit_prepare
,
1617 .rpc_call_done
= nfs_commit_done
,
1618 .rpc_release
= nfs_commit_release
,
1621 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1622 .completion
= nfs_commit_release_pages
,
1623 .error_cleanup
= nfs_commit_clear_lock
,
1626 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1627 int how
, struct nfs_commit_info
*cinfo
)
1631 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1632 if (status
== PNFS_NOT_ATTEMPTED
)
1633 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1637 int nfs_commit_inode(struct inode
*inode
, int how
)
1640 struct nfs_commit_info cinfo
;
1641 int may_wait
= how
& FLUSH_SYNC
;
1644 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1646 goto out_mark_dirty
;
1647 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1648 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1652 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1656 goto out_mark_dirty
;
1657 error
= wait_on_bit(&NFS_I(inode
)->flags
,
1659 nfs_wait_bit_killable
,
1664 nfs_commit_clear_lock(NFS_I(inode
));
1666 /* Note: If we exit without ensuring that the commit is complete,
1667 * we must mark the inode as dirty. Otherwise, future calls to
1668 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1669 * that the data is on the disk.
1672 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1676 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1678 struct nfs_inode
*nfsi
= NFS_I(inode
);
1679 int flags
= FLUSH_SYNC
;
1682 /* no commits means nothing needs to be done */
1683 if (!nfsi
->commit_info
.ncommit
)
1686 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1687 /* Don't commit yet if this is a non-blocking flush and there
1688 * are a lot of outstanding writes for this mapping.
1690 if (nfsi
->commit_info
.ncommit
<= (nfsi
->npages
>> 1))
1691 goto out_mark_dirty
;
1693 /* don't wait for the COMMIT response */
1697 ret
= nfs_commit_inode(inode
, flags
);
1699 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1700 if (ret
< wbc
->nr_to_write
)
1701 wbc
->nr_to_write
-= ret
;
1703 wbc
->nr_to_write
= 0;
1708 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1712 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1718 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1720 return nfs_commit_unstable_pages(inode
, wbc
);
1722 EXPORT_SYMBOL_GPL(nfs_write_inode
);
1725 * flush the inode to disk.
1727 int nfs_wb_all(struct inode
*inode
)
1729 struct writeback_control wbc
= {
1730 .sync_mode
= WB_SYNC_ALL
,
1731 .nr_to_write
= LONG_MAX
,
1733 .range_end
= LLONG_MAX
,
1736 return sync_inode(inode
, &wbc
);
1738 EXPORT_SYMBOL_GPL(nfs_wb_all
);
1740 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1742 struct nfs_page
*req
;
1746 wait_on_page_writeback(page
);
1747 req
= nfs_page_find_request(page
);
1750 if (nfs_lock_request(req
)) {
1751 nfs_clear_request_commit(req
);
1752 nfs_inode_remove_request(req
);
1754 * In case nfs_inode_remove_request has marked the
1755 * page as being dirty
1757 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
1758 nfs_unlock_and_release_request(req
);
1761 ret
= nfs_wait_on_request(req
);
1762 nfs_release_request(req
);
1770 * Write back all requests on one page - we do this before reading it.
1772 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1774 loff_t range_start
= page_file_offset(page
);
1775 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1776 struct writeback_control wbc
= {
1777 .sync_mode
= WB_SYNC_ALL
,
1779 .range_start
= range_start
,
1780 .range_end
= range_end
,
1785 wait_on_page_writeback(page
);
1786 if (clear_page_dirty_for_io(page
)) {
1787 ret
= nfs_writepage_locked(page
, &wbc
);
1792 if (!PagePrivate(page
))
1794 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1803 #ifdef CONFIG_MIGRATION
1804 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1805 struct page
*page
, enum migrate_mode mode
)
1808 * If PagePrivate is set, then the page is currently associated with
1809 * an in-progress read or write request. Don't try to migrate it.
1811 * FIXME: we could do this in principle, but we'll need a way to ensure
1812 * that we can safely release the inode reference while holding
1815 if (PagePrivate(page
))
1818 if (!nfs_fscache_release_page(page
, GFP_KERNEL
))
1821 return migrate_page(mapping
, newpage
, page
, mode
);
1825 int __init
nfs_init_writepagecache(void)
1827 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1828 sizeof(struct nfs_write_header
),
1829 0, SLAB_HWCACHE_ALIGN
,
1831 if (nfs_wdata_cachep
== NULL
)
1834 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1836 if (nfs_wdata_mempool
== NULL
)
1837 goto out_destroy_write_cache
;
1839 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1840 sizeof(struct nfs_commit_data
),
1841 0, SLAB_HWCACHE_ALIGN
,
1843 if (nfs_cdata_cachep
== NULL
)
1844 goto out_destroy_write_mempool
;
1846 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1848 if (nfs_commit_mempool
== NULL
)
1849 goto out_destroy_commit_cache
;
1852 * NFS congestion size, scale with available memory.
1864 * This allows larger machines to have larger/more transfers.
1865 * Limit the default to 256M
1867 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1868 if (nfs_congestion_kb
> 256*1024)
1869 nfs_congestion_kb
= 256*1024;
1873 out_destroy_commit_cache
:
1874 kmem_cache_destroy(nfs_cdata_cachep
);
1875 out_destroy_write_mempool
:
1876 mempool_destroy(nfs_wdata_mempool
);
1877 out_destroy_write_cache
:
1878 kmem_cache_destroy(nfs_wdata_cachep
);
1882 void nfs_destroy_writepagecache(void)
1884 mempool_destroy(nfs_commit_mempool
);
1885 kmem_cache_destroy(nfs_cdata_cachep
);
1886 mempool_destroy(nfs_wdata_mempool
);
1887 kmem_cache_destroy(nfs_wdata_cachep
);