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
;
49 static void nfs_clear_request_commit(struct nfs_page
*req
);
50 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
52 static struct nfs_page
*
53 nfs_page_search_commits_for_head_request_locked(struct nfs_inode
*nfsi
,
56 static struct kmem_cache
*nfs_wdata_cachep
;
57 static mempool_t
*nfs_wdata_mempool
;
58 static struct kmem_cache
*nfs_cdata_cachep
;
59 static mempool_t
*nfs_commit_mempool
;
61 struct nfs_commit_data
*nfs_commitdata_alloc(void)
63 struct nfs_commit_data
*p
= mempool_alloc(nfs_commit_mempool
, GFP_NOIO
);
66 memset(p
, 0, sizeof(*p
));
67 INIT_LIST_HEAD(&p
->pages
);
71 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc
);
73 void nfs_commit_free(struct nfs_commit_data
*p
)
75 mempool_free(p
, nfs_commit_mempool
);
77 EXPORT_SYMBOL_GPL(nfs_commit_free
);
79 static struct nfs_pgio_header
*nfs_writehdr_alloc(void)
81 struct nfs_pgio_header
*p
= mempool_alloc(nfs_wdata_mempool
, GFP_NOIO
);
84 memset(p
, 0, sizeof(*p
));
88 static void nfs_writehdr_free(struct nfs_pgio_header
*hdr
)
90 mempool_free(hdr
, nfs_wdata_mempool
);
93 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
97 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
101 * nfs_page_find_head_request_locked - find head request associated with @page
103 * must be called while holding the inode lock.
105 * returns matching head request with reference held, or NULL if not found.
107 static struct nfs_page
*
108 nfs_page_find_head_request_locked(struct nfs_inode
*nfsi
, struct page
*page
)
110 struct nfs_page
*req
= NULL
;
112 if (PagePrivate(page
))
113 req
= (struct nfs_page
*)page_private(page
);
114 else if (unlikely(PageSwapCache(page
)))
115 req
= nfs_page_search_commits_for_head_request_locked(nfsi
,
119 WARN_ON_ONCE(req
->wb_head
!= req
);
120 kref_get(&req
->wb_kref
);
127 * nfs_page_find_head_request - find head request associated with @page
129 * returns matching head request with reference held, or NULL if not found.
131 static struct nfs_page
*nfs_page_find_head_request(struct page
*page
)
133 struct inode
*inode
= page_file_mapping(page
)->host
;
134 struct nfs_page
*req
= NULL
;
136 spin_lock(&inode
->i_lock
);
137 req
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
138 spin_unlock(&inode
->i_lock
);
142 /* Adjust the file length if we're writing beyond the end */
143 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
145 struct inode
*inode
= page_file_mapping(page
)->host
;
149 spin_lock(&inode
->i_lock
);
150 i_size
= i_size_read(inode
);
151 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
152 if (i_size
> 0 && page_file_index(page
) < end_index
)
154 end
= page_file_offset(page
) + ((loff_t
)offset
+count
);
157 i_size_write(inode
, end
);
158 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
160 spin_unlock(&inode
->i_lock
);
163 /* A writeback failed: mark the page as bad, and invalidate the page cache */
164 static void nfs_set_pageerror(struct page
*page
)
166 nfs_zap_mapping(page_file_mapping(page
)->host
, page_file_mapping(page
));
170 * nfs_page_group_search_locked
171 * @head - head request of page group
172 * @page_offset - offset into page
174 * Search page group with head @head to find a request that contains the
175 * page offset @page_offset.
177 * Returns a pointer to the first matching nfs request, or NULL if no
180 * Must be called with the page group lock held
182 static struct nfs_page
*
183 nfs_page_group_search_locked(struct nfs_page
*head
, unsigned int page_offset
)
185 struct nfs_page
*req
;
187 WARN_ON_ONCE(head
!= head
->wb_head
);
188 WARN_ON_ONCE(!test_bit(PG_HEADLOCK
, &head
->wb_head
->wb_flags
));
192 if (page_offset
>= req
->wb_pgbase
&&
193 page_offset
< (req
->wb_pgbase
+ req
->wb_bytes
))
196 req
= req
->wb_this_page
;
197 } while (req
!= head
);
203 * nfs_page_group_covers_page
204 * @head - head request of page group
206 * Return true if the page group with head @head covers the whole page,
207 * returns false otherwise
209 static bool nfs_page_group_covers_page(struct nfs_page
*req
)
211 struct nfs_page
*tmp
;
212 unsigned int pos
= 0;
213 unsigned int len
= nfs_page_length(req
->wb_page
);
215 nfs_page_group_lock(req
, false);
218 tmp
= nfs_page_group_search_locked(req
->wb_head
, pos
);
220 /* no way this should happen */
221 WARN_ON_ONCE(tmp
->wb_pgbase
!= pos
);
222 pos
+= tmp
->wb_bytes
- (pos
- tmp
->wb_pgbase
);
224 } while (tmp
&& pos
< len
);
226 nfs_page_group_unlock(req
);
227 WARN_ON_ONCE(pos
> len
);
231 /* We can set the PG_uptodate flag if we see that a write request
232 * covers the full page.
234 static void nfs_mark_uptodate(struct nfs_page
*req
)
236 if (PageUptodate(req
->wb_page
))
238 if (!nfs_page_group_covers_page(req
))
240 SetPageUptodate(req
->wb_page
);
243 static int wb_priority(struct writeback_control
*wbc
)
246 if (wbc
->for_reclaim
)
247 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
248 if (wbc
->sync_mode
== WB_SYNC_ALL
)
249 ret
= FLUSH_COND_STABLE
;
250 if (wbc
->for_kupdate
|| wbc
->for_background
)
256 * NFS congestion control
259 int nfs_congestion_kb
;
261 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
262 #define NFS_CONGESTION_OFF_THRESH \
263 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
265 static void nfs_set_page_writeback(struct page
*page
)
267 struct nfs_server
*nfss
= NFS_SERVER(page_file_mapping(page
)->host
);
268 int ret
= test_set_page_writeback(page
);
270 WARN_ON_ONCE(ret
!= 0);
272 if (atomic_long_inc_return(&nfss
->writeback
) >
273 NFS_CONGESTION_ON_THRESH
) {
274 set_bdi_congested(&nfss
->backing_dev_info
,
279 static void nfs_end_page_writeback(struct nfs_page
*req
)
281 struct inode
*inode
= page_file_mapping(req
->wb_page
)->host
;
282 struct nfs_server
*nfss
= NFS_SERVER(inode
);
284 if (!nfs_page_group_sync_on_bit(req
, PG_WB_END
))
287 end_page_writeback(req
->wb_page
);
288 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
289 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
293 /* nfs_page_group_clear_bits
294 * @req - an nfs request
295 * clears all page group related bits from @req
298 nfs_page_group_clear_bits(struct nfs_page
*req
)
300 clear_bit(PG_TEARDOWN
, &req
->wb_flags
);
301 clear_bit(PG_UNLOCKPAGE
, &req
->wb_flags
);
302 clear_bit(PG_UPTODATE
, &req
->wb_flags
);
303 clear_bit(PG_WB_END
, &req
->wb_flags
);
304 clear_bit(PG_REMOVE
, &req
->wb_flags
);
309 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
311 * this is a helper function for nfs_lock_and_join_requests
313 * @inode - inode associated with request page group, must be holding inode lock
314 * @head - head request of page group, must be holding head lock
315 * @req - request that couldn't lock and needs to wait on the req bit lock
316 * @nonblock - if true, don't actually wait
318 * NOTE: this must be called holding page_group bit lock and inode spin lock
319 * and BOTH will be released before returning.
321 * returns 0 on success, < 0 on error.
324 nfs_unroll_locks_and_wait(struct inode
*inode
, struct nfs_page
*head
,
325 struct nfs_page
*req
, bool nonblock
)
326 __releases(&inode
->i_lock
)
328 struct nfs_page
*tmp
;
331 /* relinquish all the locks successfully grabbed this run */
332 for (tmp
= head
; tmp
!= req
; tmp
= tmp
->wb_this_page
)
333 nfs_unlock_request(tmp
);
335 WARN_ON_ONCE(test_bit(PG_TEARDOWN
, &req
->wb_flags
));
337 /* grab a ref on the request that will be waited on */
338 kref_get(&req
->wb_kref
);
340 nfs_page_group_unlock(head
);
341 spin_unlock(&inode
->i_lock
);
343 /* release ref from nfs_page_find_head_request_locked */
344 nfs_release_request(head
);
347 ret
= nfs_wait_on_request(req
);
350 nfs_release_request(req
);
356 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
358 * @destroy_list - request list (using wb_this_page) terminated by @old_head
359 * @old_head - the old head of the list
361 * All subrequests must be locked and removed from all lists, so at this point
362 * they are only "active" in this function, and possibly in nfs_wait_on_request
363 * with a reference held by some other context.
366 nfs_destroy_unlinked_subrequests(struct nfs_page
*destroy_list
,
367 struct nfs_page
*old_head
)
369 while (destroy_list
) {
370 struct nfs_page
*subreq
= destroy_list
;
372 destroy_list
= (subreq
->wb_this_page
== old_head
) ?
373 NULL
: subreq
->wb_this_page
;
375 WARN_ON_ONCE(old_head
!= subreq
->wb_head
);
377 /* make sure old group is not used */
378 subreq
->wb_head
= subreq
;
379 subreq
->wb_this_page
= subreq
;
381 /* subreq is now totally disconnected from page group or any
382 * write / commit lists. last chance to wake any waiters */
383 nfs_unlock_request(subreq
);
385 if (!test_bit(PG_TEARDOWN
, &subreq
->wb_flags
)) {
386 /* release ref on old head request */
387 nfs_release_request(old_head
);
389 nfs_page_group_clear_bits(subreq
);
391 /* release the PG_INODE_REF reference */
392 if (test_and_clear_bit(PG_INODE_REF
, &subreq
->wb_flags
))
393 nfs_release_request(subreq
);
397 WARN_ON_ONCE(test_bit(PG_CLEAN
, &subreq
->wb_flags
));
398 /* zombie requests have already released the last
399 * reference and were waiting on the rest of the
400 * group to complete. Since it's no longer part of a
401 * group, simply free the request */
402 nfs_page_group_clear_bits(subreq
);
403 nfs_free_request(subreq
);
409 * nfs_lock_and_join_requests - join all subreqs to the head req and return
410 * a locked reference, cancelling any pending
411 * operations for this page.
413 * @page - the page used to lookup the "page group" of nfs_page structures
414 * @nonblock - if true, don't block waiting for request locks
416 * This function joins all sub requests to the head request by first
417 * locking all requests in the group, cancelling any pending operations
418 * and finally updating the head request to cover the whole range covered by
419 * the (former) group. All subrequests are removed from any write or commit
420 * lists, unlinked from the group and destroyed.
422 * Returns a locked, referenced pointer to the head request - which after
423 * this call is guaranteed to be the only request associated with the page.
424 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
425 * error was encountered.
427 static struct nfs_page
*
428 nfs_lock_and_join_requests(struct page
*page
, bool nonblock
)
430 struct inode
*inode
= page_file_mapping(page
)->host
;
431 struct nfs_page
*head
, *subreq
;
432 struct nfs_page
*destroy_list
= NULL
;
433 unsigned int total_bytes
;
439 WARN_ON_ONCE(destroy_list
);
441 spin_lock(&inode
->i_lock
);
444 * A reference is taken only on the head request which acts as a
445 * reference to the whole page group - the group will not be destroyed
446 * until the head reference is released.
448 head
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
451 spin_unlock(&inode
->i_lock
);
455 /* holding inode lock, so always make a non-blocking call to try the
457 ret
= nfs_page_group_lock(head
, true);
459 spin_unlock(&inode
->i_lock
);
461 if (!nonblock
&& ret
== -EAGAIN
) {
462 nfs_page_group_lock_wait(head
);
463 nfs_release_request(head
);
467 nfs_release_request(head
);
471 /* lock each request in the page group */
475 * Subrequests are always contiguous, non overlapping
476 * and in order - but may be repeated (mirrored writes).
478 if (subreq
->wb_offset
== (head
->wb_offset
+ total_bytes
)) {
479 /* keep track of how many bytes this group covers */
480 total_bytes
+= subreq
->wb_bytes
;
481 } else if (WARN_ON_ONCE(subreq
->wb_offset
< head
->wb_offset
||
482 ((subreq
->wb_offset
+ subreq
->wb_bytes
) >
483 (head
->wb_offset
+ total_bytes
)))) {
484 nfs_page_group_unlock(head
);
485 spin_unlock(&inode
->i_lock
);
486 return ERR_PTR(-EIO
);
489 if (!nfs_lock_request(subreq
)) {
490 /* releases page group bit lock and
491 * inode spin lock and all references */
492 ret
= nfs_unroll_locks_and_wait(inode
, head
,
501 subreq
= subreq
->wb_this_page
;
502 } while (subreq
!= head
);
504 /* Now that all requests are locked, make sure they aren't on any list.
505 * Commit list removal accounting is done after locks are dropped */
508 nfs_clear_request_commit(subreq
);
509 subreq
= subreq
->wb_this_page
;
510 } while (subreq
!= head
);
512 /* unlink subrequests from head, destroy them later */
513 if (head
->wb_this_page
!= head
) {
514 /* destroy list will be terminated by head */
515 destroy_list
= head
->wb_this_page
;
516 head
->wb_this_page
= head
;
518 /* change head request to cover whole range that
519 * the former page group covered */
520 head
->wb_bytes
= total_bytes
;
524 * prepare head request to be added to new pgio descriptor
526 nfs_page_group_clear_bits(head
);
529 * some part of the group was still on the inode list - otherwise
530 * the group wouldn't be involved in async write.
531 * grab a reference for the head request, iff it needs one.
533 if (!test_and_set_bit(PG_INODE_REF
, &head
->wb_flags
))
534 kref_get(&head
->wb_kref
);
536 nfs_page_group_unlock(head
);
538 /* drop lock to clean uprequests on destroy list */
539 spin_unlock(&inode
->i_lock
);
541 nfs_destroy_unlinked_subrequests(destroy_list
, head
);
543 /* still holds ref on head from nfs_page_find_head_request_locked
544 * and still has lock on head from lock loop */
549 * Find an associated nfs write request, and prepare to flush it out
550 * May return an error if the user signalled nfs_wait_on_request().
552 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
553 struct page
*page
, bool nonblock
)
555 struct nfs_page
*req
;
558 req
= nfs_lock_and_join_requests(page
, nonblock
);
565 nfs_set_page_writeback(page
);
566 WARN_ON_ONCE(test_bit(PG_CLEAN
, &req
->wb_flags
));
569 if (!nfs_pageio_add_request(pgio
, req
)) {
570 nfs_redirty_request(req
);
571 ret
= pgio
->pg_error
;
577 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
579 struct inode
*inode
= page_file_mapping(page
)->host
;
582 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
583 nfs_add_stats(inode
, NFSIOS_WRITEPAGES
, 1);
585 nfs_pageio_cond_complete(pgio
, page_file_index(page
));
586 ret
= nfs_page_async_flush(pgio
, page
, wbc
->sync_mode
== WB_SYNC_NONE
);
587 if (ret
== -EAGAIN
) {
588 redirty_page_for_writepage(wbc
, page
);
595 * Write an mmapped page to the server.
597 static int nfs_writepage_locked(struct page
*page
, struct writeback_control
*wbc
)
599 struct nfs_pageio_descriptor pgio
;
602 nfs_pageio_init_write(&pgio
, page
->mapping
->host
, wb_priority(wbc
),
603 false, &nfs_async_write_completion_ops
);
604 err
= nfs_do_writepage(page
, wbc
, &pgio
);
605 nfs_pageio_complete(&pgio
);
608 if (pgio
.pg_error
< 0)
609 return pgio
.pg_error
;
613 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
617 ret
= nfs_writepage_locked(page
, wbc
);
622 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
626 ret
= nfs_do_writepage(page
, wbc
, data
);
631 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
633 struct inode
*inode
= mapping
->host
;
634 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
635 struct nfs_pageio_descriptor pgio
;
638 /* Stop dirtying of new pages while we sync */
639 err
= wait_on_bit_lock_action(bitlock
, NFS_INO_FLUSHING
,
640 nfs_wait_bit_killable
, TASK_KILLABLE
);
644 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
646 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
), false,
647 &nfs_async_write_completion_ops
);
648 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
649 nfs_pageio_complete(&pgio
);
651 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
652 smp_mb__after_atomic();
653 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
666 * Insert a write request into an inode
668 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
670 struct nfs_inode
*nfsi
= NFS_I(inode
);
672 WARN_ON_ONCE(req
->wb_this_page
!= req
);
674 /* Lock the request! */
675 nfs_lock_request(req
);
677 spin_lock(&inode
->i_lock
);
678 if (!nfsi
->nrequests
&&
679 NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
682 * Swap-space should not get truncated. Hence no need to plug the race
683 * with invalidate/truncate.
685 if (likely(!PageSwapCache(req
->wb_page
))) {
686 set_bit(PG_MAPPED
, &req
->wb_flags
);
687 SetPagePrivate(req
->wb_page
);
688 set_page_private(req
->wb_page
, (unsigned long)req
);
691 /* this a head request for a page group - mark it as having an
692 * extra reference so sub groups can follow suit.
693 * This flag also informs pgio layer when to bump nrequests when
694 * adding subrequests. */
695 WARN_ON(test_and_set_bit(PG_INODE_REF
, &req
->wb_flags
));
696 kref_get(&req
->wb_kref
);
697 spin_unlock(&inode
->i_lock
);
701 * Remove a write request from an inode
703 static void nfs_inode_remove_request(struct nfs_page
*req
)
705 struct inode
*inode
= d_inode(req
->wb_context
->dentry
);
706 struct nfs_inode
*nfsi
= NFS_I(inode
);
707 struct nfs_page
*head
;
709 if (nfs_page_group_sync_on_bit(req
, PG_REMOVE
)) {
712 spin_lock(&inode
->i_lock
);
713 if (likely(!PageSwapCache(head
->wb_page
))) {
714 set_page_private(head
->wb_page
, 0);
715 ClearPagePrivate(head
->wb_page
);
716 smp_mb__after_atomic();
717 wake_up_page(head
->wb_page
, PG_private
);
718 clear_bit(PG_MAPPED
, &head
->wb_flags
);
721 spin_unlock(&inode
->i_lock
);
723 spin_lock(&inode
->i_lock
);
725 spin_unlock(&inode
->i_lock
);
728 if (test_and_clear_bit(PG_INODE_REF
, &req
->wb_flags
))
729 nfs_release_request(req
);
733 nfs_mark_request_dirty(struct nfs_page
*req
)
735 __set_page_dirty_nobuffers(req
->wb_page
);
739 * nfs_page_search_commits_for_head_request_locked
741 * Search through commit lists on @inode for the head request for @page.
742 * Must be called while holding the inode (which is cinfo) lock.
744 * Returns the head request if found, or NULL if not found.
746 static struct nfs_page
*
747 nfs_page_search_commits_for_head_request_locked(struct nfs_inode
*nfsi
,
750 struct nfs_page
*freq
, *t
;
751 struct nfs_commit_info cinfo
;
752 struct inode
*inode
= &nfsi
->vfs_inode
;
754 nfs_init_cinfo_from_inode(&cinfo
, inode
);
756 /* search through pnfs commit lists */
757 freq
= pnfs_search_commit_reqs(inode
, &cinfo
, page
);
759 return freq
->wb_head
;
761 /* Linearly search the commit list for the correct request */
762 list_for_each_entry_safe(freq
, t
, &cinfo
.mds
->list
, wb_list
) {
763 if (freq
->wb_page
== page
)
764 return freq
->wb_head
;
771 * nfs_request_add_commit_list - add request to a commit list
772 * @req: pointer to a struct nfs_page
773 * @dst: commit list head
774 * @cinfo: holds list lock and accounting info
776 * This sets the PG_CLEAN bit, updates the cinfo count of
777 * number of outstanding requests requiring a commit as well as
780 * The caller must _not_ hold the cinfo->lock, but must be
781 * holding the nfs_page lock.
784 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
785 struct nfs_commit_info
*cinfo
)
787 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
788 spin_lock(cinfo
->lock
);
789 nfs_list_add_request(req
, dst
);
790 cinfo
->mds
->ncommit
++;
791 spin_unlock(cinfo
->lock
);
793 nfs_mark_page_unstable(req
->wb_page
);
795 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
798 * nfs_request_remove_commit_list - Remove request from a commit list
799 * @req: pointer to a nfs_page
800 * @cinfo: holds list lock and accounting info
802 * This clears the PG_CLEAN bit, and updates the cinfo's count of
803 * number of outstanding requests requiring a commit
804 * It does not update the MM page stats.
806 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
809 nfs_request_remove_commit_list(struct nfs_page
*req
,
810 struct nfs_commit_info
*cinfo
)
812 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
814 nfs_list_remove_request(req
);
815 cinfo
->mds
->ncommit
--;
817 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
819 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
822 cinfo
->lock
= &inode
->i_lock
;
823 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
824 cinfo
->ds
= pnfs_get_ds_info(inode
);
826 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
829 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
831 struct nfs_direct_req
*dreq
)
834 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
836 nfs_init_cinfo_from_inode(cinfo
, inode
);
838 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
841 * Add a request to the inode's commit list.
844 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
845 struct nfs_commit_info
*cinfo
, u32 ds_commit_idx
)
847 if (pnfs_mark_request_commit(req
, lseg
, cinfo
, ds_commit_idx
))
849 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
853 nfs_clear_page_commit(struct page
*page
)
855 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
856 dec_wb_stat(&inode_to_bdi(page_file_mapping(page
)->host
)->wb
,
860 /* Called holding inode (/cinfo) lock */
862 nfs_clear_request_commit(struct nfs_page
*req
)
864 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
865 struct inode
*inode
= d_inode(req
->wb_context
->dentry
);
866 struct nfs_commit_info cinfo
;
868 nfs_init_cinfo_from_inode(&cinfo
, inode
);
869 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
870 nfs_request_remove_commit_list(req
, &cinfo
);
872 nfs_clear_page_commit(req
->wb_page
);
876 int nfs_write_need_commit(struct nfs_pgio_header
*hdr
)
878 if (hdr
->verf
.committed
== NFS_DATA_SYNC
)
879 return hdr
->lseg
== NULL
;
880 return hdr
->verf
.committed
!= NFS_FILE_SYNC
;
883 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
885 struct nfs_commit_info cinfo
;
886 unsigned long bytes
= 0;
888 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
890 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
891 while (!list_empty(&hdr
->pages
)) {
892 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
894 bytes
+= req
->wb_bytes
;
895 nfs_list_remove_request(req
);
896 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
897 (hdr
->good_bytes
< bytes
)) {
898 nfs_set_pageerror(req
->wb_page
);
899 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
902 if (nfs_write_need_commit(hdr
)) {
903 memcpy(&req
->wb_verf
, &hdr
->verf
.verifier
, sizeof(req
->wb_verf
));
904 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
,
905 hdr
->pgio_mirror_idx
);
909 nfs_inode_remove_request(req
);
911 nfs_unlock_request(req
);
912 nfs_end_page_writeback(req
);
913 nfs_release_request(req
);
920 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
922 return cinfo
->mds
->ncommit
;
925 /* cinfo->lock held by caller */
927 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
928 struct nfs_commit_info
*cinfo
, int max
)
930 struct nfs_page
*req
, *tmp
;
933 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
934 if (!nfs_lock_request(req
))
936 kref_get(&req
->wb_kref
);
937 if (cond_resched_lock(cinfo
->lock
))
938 list_safe_reset_next(req
, tmp
, wb_list
);
939 nfs_request_remove_commit_list(req
, cinfo
);
940 nfs_list_add_request(req
, dst
);
942 if ((ret
== max
) && !cinfo
->dreq
)
949 * nfs_scan_commit - Scan an inode for commit requests
950 * @inode: NFS inode to scan
951 * @dst: mds destination list
952 * @cinfo: mds and ds lists of reqs ready to commit
954 * Moves requests from the inode's 'commit' request list.
955 * The requests are *not* checked to ensure that they form a contiguous set.
958 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
959 struct nfs_commit_info
*cinfo
)
963 spin_lock(cinfo
->lock
);
964 if (cinfo
->mds
->ncommit
> 0) {
965 const int max
= INT_MAX
;
967 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
969 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
971 spin_unlock(cinfo
->lock
);
976 * Search for an existing write request, and attempt to update
977 * it to reflect a new dirty region on a given page.
979 * If the attempt fails, then the existing request is flushed out
982 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
987 struct nfs_page
*req
;
992 if (!PagePrivate(page
))
995 end
= offset
+ bytes
;
996 spin_lock(&inode
->i_lock
);
999 req
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
1003 /* should be handled by nfs_flush_incompatible */
1004 WARN_ON_ONCE(req
->wb_head
!= req
);
1005 WARN_ON_ONCE(req
->wb_this_page
!= req
);
1007 rqend
= req
->wb_offset
+ req
->wb_bytes
;
1009 * Tell the caller to flush out the request if
1010 * the offsets are non-contiguous.
1011 * Note: nfs_flush_incompatible() will already
1012 * have flushed out requests having wrong owners.
1015 || end
< req
->wb_offset
)
1018 if (nfs_lock_request(req
))
1021 /* The request is locked, so wait and then retry */
1022 spin_unlock(&inode
->i_lock
);
1023 error
= nfs_wait_on_request(req
);
1024 nfs_release_request(req
);
1027 spin_lock(&inode
->i_lock
);
1030 /* Okay, the request matches. Update the region */
1031 if (offset
< req
->wb_offset
) {
1032 req
->wb_offset
= offset
;
1033 req
->wb_pgbase
= offset
;
1036 req
->wb_bytes
= end
- req
->wb_offset
;
1038 req
->wb_bytes
= rqend
- req
->wb_offset
;
1041 nfs_clear_request_commit(req
);
1042 spin_unlock(&inode
->i_lock
);
1045 spin_unlock(&inode
->i_lock
);
1046 nfs_release_request(req
);
1047 error
= nfs_wb_page(inode
, page
);
1049 return ERR_PTR(error
);
1053 * Try to update an existing write request, or create one if there is none.
1055 * Note: Should always be called with the Page Lock held to prevent races
1056 * if we have to add a new request. Also assumes that the caller has
1057 * already called nfs_flush_incompatible() if necessary.
1059 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
1060 struct page
*page
, unsigned int offset
, unsigned int bytes
)
1062 struct inode
*inode
= page_file_mapping(page
)->host
;
1063 struct nfs_page
*req
;
1065 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
1068 req
= nfs_create_request(ctx
, page
, NULL
, offset
, bytes
);
1071 nfs_inode_add_request(inode
, req
);
1076 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
1077 unsigned int offset
, unsigned int count
)
1079 struct nfs_page
*req
;
1081 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
1083 return PTR_ERR(req
);
1084 /* Update file length */
1085 nfs_grow_file(page
, offset
, count
);
1086 nfs_mark_uptodate(req
);
1087 nfs_mark_request_dirty(req
);
1088 nfs_unlock_and_release_request(req
);
1092 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
1094 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
1095 struct nfs_lock_context
*l_ctx
;
1096 struct file_lock_context
*flctx
= file_inode(file
)->i_flctx
;
1097 struct nfs_page
*req
;
1098 int do_flush
, status
;
1100 * Look for a request corresponding to this page. If there
1101 * is one, and it belongs to another file, we flush it out
1102 * before we try to copy anything into the page. Do this
1103 * due to the lack of an ACCESS-type call in NFSv2.
1104 * Also do the same if we find a request from an existing
1108 req
= nfs_page_find_head_request(page
);
1111 l_ctx
= req
->wb_lock_context
;
1112 do_flush
= req
->wb_page
!= page
|| req
->wb_context
!= ctx
;
1113 /* for now, flush if more than 1 request in page_group */
1114 do_flush
|= req
->wb_this_page
!= req
;
1115 if (l_ctx
&& flctx
&&
1116 !(list_empty_careful(&flctx
->flc_posix
) &&
1117 list_empty_careful(&flctx
->flc_flock
))) {
1118 do_flush
|= l_ctx
->lockowner
.l_owner
!= current
->files
1119 || l_ctx
->lockowner
.l_pid
!= current
->tgid
;
1121 nfs_release_request(req
);
1124 status
= nfs_wb_page(page_file_mapping(page
)->host
, page
);
1125 } while (status
== 0);
1130 * Avoid buffered writes when a open context credential's key would
1133 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1135 * Return 0 and set a credential flag which triggers the inode to flush
1136 * and performs NFS_FILE_SYNC writes if the key will expired within
1137 * RPC_KEY_EXPIRE_TIMEO.
1140 nfs_key_timeout_notify(struct file
*filp
, struct inode
*inode
)
1142 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
1143 struct rpc_auth
*auth
= NFS_SERVER(inode
)->client
->cl_auth
;
1145 return rpcauth_key_timeout_notify(auth
, ctx
->cred
);
1149 * Test if the open context credential key is marked to expire soon.
1151 bool nfs_ctx_key_to_expire(struct nfs_open_context
*ctx
)
1153 return rpcauth_cred_key_to_expire(ctx
->cred
);
1157 * If the page cache is marked as unsafe or invalid, then we can't rely on
1158 * the PageUptodate() flag. In this case, we will need to turn off
1159 * write optimisations that depend on the page contents being correct.
1161 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
1163 struct nfs_inode
*nfsi
= NFS_I(inode
);
1165 if (nfs_have_delegated_attributes(inode
))
1167 if (nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
1170 if (test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
))
1173 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1175 return PageUptodate(page
) != 0;
1179 is_whole_file_wrlock(struct file_lock
*fl
)
1181 return fl
->fl_start
== 0 && fl
->fl_end
== OFFSET_MAX
&&
1182 fl
->fl_type
== F_WRLCK
;
1185 /* If we know the page is up to date, and we're not using byte range locks (or
1186 * if we have the whole file locked for writing), it may be more efficient to
1187 * extend the write to cover the entire page in order to avoid fragmentation
1190 * If the file is opened for synchronous writes then we can just skip the rest
1193 static int nfs_can_extend_write(struct file
*file
, struct page
*page
, struct inode
*inode
)
1196 struct file_lock_context
*flctx
= inode
->i_flctx
;
1197 struct file_lock
*fl
;
1199 if (file
->f_flags
& O_DSYNC
)
1201 if (!nfs_write_pageuptodate(page
, inode
))
1203 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
1205 if (!flctx
|| (list_empty_careful(&flctx
->flc_flock
) &&
1206 list_empty_careful(&flctx
->flc_posix
)))
1209 /* Check to see if there are whole file write locks */
1211 spin_lock(&flctx
->flc_lock
);
1212 if (!list_empty(&flctx
->flc_posix
)) {
1213 fl
= list_first_entry(&flctx
->flc_posix
, struct file_lock
,
1215 if (is_whole_file_wrlock(fl
))
1217 } else if (!list_empty(&flctx
->flc_flock
)) {
1218 fl
= list_first_entry(&flctx
->flc_flock
, struct file_lock
,
1220 if (fl
->fl_type
== F_WRLCK
)
1223 spin_unlock(&flctx
->flc_lock
);
1228 * Update and possibly write a cached page of an NFS file.
1230 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1231 * things with a page scheduled for an RPC call (e.g. invalidate it).
1233 int nfs_updatepage(struct file
*file
, struct page
*page
,
1234 unsigned int offset
, unsigned int count
)
1236 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
1237 struct inode
*inode
= page_file_mapping(page
)->host
;
1240 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
1242 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1243 file
, count
, (long long)(page_file_offset(page
) + offset
));
1245 if (nfs_can_extend_write(file
, page
, inode
)) {
1246 count
= max(count
+ offset
, nfs_page_length(page
));
1250 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
1252 nfs_set_pageerror(page
);
1254 __set_page_dirty_nobuffers(page
);
1256 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1257 status
, (long long)i_size_read(inode
));
1261 static int flush_task_priority(int how
)
1263 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
1265 return RPC_PRIORITY_HIGH
;
1267 return RPC_PRIORITY_LOW
;
1269 return RPC_PRIORITY_NORMAL
;
1272 static void nfs_initiate_write(struct nfs_pgio_header
*hdr
,
1273 struct rpc_message
*msg
,
1274 const struct nfs_rpc_ops
*rpc_ops
,
1275 struct rpc_task_setup
*task_setup_data
, int how
)
1277 int priority
= flush_task_priority(how
);
1279 task_setup_data
->priority
= priority
;
1280 rpc_ops
->write_setup(hdr
, msg
);
1282 nfs4_state_protect_write(NFS_SERVER(hdr
->inode
)->nfs_client
,
1283 &task_setup_data
->rpc_client
, msg
, hdr
);
1286 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1287 * call this on each, which will prepare them to be retried on next
1288 * writeback using standard nfs.
1290 static void nfs_redirty_request(struct nfs_page
*req
)
1292 nfs_mark_request_dirty(req
);
1293 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1294 nfs_unlock_request(req
);
1295 nfs_end_page_writeback(req
);
1296 nfs_release_request(req
);
1299 static void nfs_async_write_error(struct list_head
*head
)
1301 struct nfs_page
*req
;
1303 while (!list_empty(head
)) {
1304 req
= nfs_list_entry(head
->next
);
1305 nfs_list_remove_request(req
);
1306 nfs_redirty_request(req
);
1310 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1311 .error_cleanup
= nfs_async_write_error
,
1312 .completion
= nfs_write_completion
,
1315 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1316 struct inode
*inode
, int ioflags
, bool force_mds
,
1317 const struct nfs_pgio_completion_ops
*compl_ops
)
1319 struct nfs_server
*server
= NFS_SERVER(inode
);
1320 const struct nfs_pageio_ops
*pg_ops
= &nfs_pgio_rw_ops
;
1322 #ifdef CONFIG_NFS_V4_1
1323 if (server
->pnfs_curr_ld
&& !force_mds
)
1324 pg_ops
= server
->pnfs_curr_ld
->pg_write_ops
;
1326 nfs_pageio_init(pgio
, inode
, pg_ops
, compl_ops
, &nfs_rw_write_ops
,
1327 server
->wsize
, ioflags
);
1329 EXPORT_SYMBOL_GPL(nfs_pageio_init_write
);
1331 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1333 struct nfs_pgio_mirror
*mirror
;
1335 pgio
->pg_ops
= &nfs_pgio_rw_ops
;
1337 nfs_pageio_stop_mirroring(pgio
);
1339 mirror
= &pgio
->pg_mirrors
[0];
1340 mirror
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1342 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1345 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1347 struct nfs_commit_data
*data
= calldata
;
1349 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1353 * Special version of should_remove_suid() that ignores capabilities.
1355 static int nfs_should_remove_suid(const struct inode
*inode
)
1357 umode_t mode
= inode
->i_mode
;
1360 /* suid always must be killed */
1361 if (unlikely(mode
& S_ISUID
))
1362 kill
= ATTR_KILL_SUID
;
1365 * sgid without any exec bits is just a mandatory locking mark; leave
1366 * it alone. If some exec bits are set, it's a real sgid; kill it.
1368 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1369 kill
|= ATTR_KILL_SGID
;
1371 if (unlikely(kill
&& S_ISREG(mode
)))
1377 static void nfs_writeback_check_extend(struct nfs_pgio_header
*hdr
,
1378 struct nfs_fattr
*fattr
)
1380 struct nfs_pgio_args
*argp
= &hdr
->args
;
1381 struct nfs_pgio_res
*resp
= &hdr
->res
;
1382 u64 size
= argp
->offset
+ resp
->count
;
1384 if (!(fattr
->valid
& NFS_ATTR_FATTR_SIZE
))
1386 if (nfs_size_to_loff_t(fattr
->size
) < i_size_read(hdr
->inode
)) {
1387 fattr
->valid
&= ~NFS_ATTR_FATTR_SIZE
;
1390 if (size
!= fattr
->size
)
1392 /* Set attribute barrier */
1393 nfs_fattr_set_barrier(fattr
);
1394 /* ...and update size */
1395 fattr
->valid
|= NFS_ATTR_FATTR_SIZE
;
1398 void nfs_writeback_update_inode(struct nfs_pgio_header
*hdr
)
1400 struct nfs_fattr
*fattr
= &hdr
->fattr
;
1401 struct inode
*inode
= hdr
->inode
;
1403 spin_lock(&inode
->i_lock
);
1404 nfs_writeback_check_extend(hdr
, fattr
);
1405 nfs_post_op_update_inode_force_wcc_locked(inode
, fattr
);
1406 spin_unlock(&inode
->i_lock
);
1408 EXPORT_SYMBOL_GPL(nfs_writeback_update_inode
);
1411 * This function is called when the WRITE call is complete.
1413 static int nfs_writeback_done(struct rpc_task
*task
,
1414 struct nfs_pgio_header
*hdr
,
1415 struct inode
*inode
)
1420 * ->write_done will attempt to use post-op attributes to detect
1421 * conflicting writes by other clients. A strict interpretation
1422 * of close-to-open would allow us to continue caching even if
1423 * another writer had changed the file, but some applications
1424 * depend on tighter cache coherency when writing.
1426 status
= NFS_PROTO(inode
)->write_done(task
, hdr
);
1429 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, hdr
->res
.count
);
1431 if (hdr
->res
.verf
->committed
< hdr
->args
.stable
&&
1432 task
->tk_status
>= 0) {
1433 /* We tried a write call, but the server did not
1434 * commit data to stable storage even though we
1436 * Note: There is a known bug in Tru64 < 5.0 in which
1437 * the server reports NFS_DATA_SYNC, but performs
1438 * NFS_FILE_SYNC. We therefore implement this checking
1439 * as a dprintk() in order to avoid filling syslog.
1441 static unsigned long complain
;
1443 /* Note this will print the MDS for a DS write */
1444 if (time_before(complain
, jiffies
)) {
1445 dprintk("NFS: faulty NFS server %s:"
1446 " (committed = %d) != (stable = %d)\n",
1447 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1448 hdr
->res
.verf
->committed
, hdr
->args
.stable
);
1449 complain
= jiffies
+ 300 * HZ
;
1453 /* Deal with the suid/sgid bit corner case */
1454 if (nfs_should_remove_suid(inode
))
1455 nfs_mark_for_revalidate(inode
);
1460 * This function is called when the WRITE call is complete.
1462 static void nfs_writeback_result(struct rpc_task
*task
,
1463 struct nfs_pgio_header
*hdr
)
1465 struct nfs_pgio_args
*argp
= &hdr
->args
;
1466 struct nfs_pgio_res
*resp
= &hdr
->res
;
1468 if (resp
->count
< argp
->count
) {
1469 static unsigned long complain
;
1471 /* This a short write! */
1472 nfs_inc_stats(hdr
->inode
, NFSIOS_SHORTWRITE
);
1474 /* Has the server at least made some progress? */
1475 if (resp
->count
== 0) {
1476 if (time_before(complain
, jiffies
)) {
1478 "NFS: Server wrote zero bytes, expected %u.\n",
1480 complain
= jiffies
+ 300 * HZ
;
1482 nfs_set_pgio_error(hdr
, -EIO
, argp
->offset
);
1483 task
->tk_status
= -EIO
;
1486 /* Was this an NFSv2 write or an NFSv3 stable write? */
1487 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1488 /* Resend from where the server left off */
1489 hdr
->mds_offset
+= resp
->count
;
1490 argp
->offset
+= resp
->count
;
1491 argp
->pgbase
+= resp
->count
;
1492 argp
->count
-= resp
->count
;
1494 /* Resend as a stable write in order to avoid
1495 * headaches in the case of a server crash.
1497 argp
->stable
= NFS_FILE_SYNC
;
1499 rpc_restart_call_prepare(task
);
1504 static int nfs_commit_set_lock(struct nfs_inode
*nfsi
, int may_wait
)
1508 if (!test_and_set_bit(NFS_INO_COMMIT
, &nfsi
->flags
))
1512 ret
= out_of_line_wait_on_bit_lock(&nfsi
->flags
,
1514 nfs_wait_bit_killable
,
1516 return (ret
< 0) ? ret
: 1;
1519 static void nfs_commit_clear_lock(struct nfs_inode
*nfsi
)
1521 clear_bit(NFS_INO_COMMIT
, &nfsi
->flags
);
1522 smp_mb__after_atomic();
1523 wake_up_bit(&nfsi
->flags
, NFS_INO_COMMIT
);
1526 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1528 put_nfs_open_context(data
->context
);
1529 nfs_commit_free(data
);
1531 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1533 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1534 const struct nfs_rpc_ops
*nfs_ops
,
1535 const struct rpc_call_ops
*call_ops
,
1538 struct rpc_task
*task
;
1539 int priority
= flush_task_priority(how
);
1540 struct rpc_message msg
= {
1541 .rpc_argp
= &data
->args
,
1542 .rpc_resp
= &data
->res
,
1543 .rpc_cred
= data
->cred
,
1545 struct rpc_task_setup task_setup_data
= {
1546 .task
= &data
->task
,
1548 .rpc_message
= &msg
,
1549 .callback_ops
= call_ops
,
1550 .callback_data
= data
,
1551 .workqueue
= nfsiod_workqueue
,
1552 .flags
= RPC_TASK_ASYNC
| flags
,
1553 .priority
= priority
,
1555 /* Set up the initial task struct. */
1556 nfs_ops
->commit_setup(data
, &msg
);
1558 dprintk("NFS: initiated commit call\n");
1560 nfs4_state_protect(NFS_SERVER(data
->inode
)->nfs_client
,
1561 NFS_SP4_MACH_CRED_COMMIT
, &task_setup_data
.rpc_client
, &msg
);
1563 task
= rpc_run_task(&task_setup_data
);
1565 return PTR_ERR(task
);
1566 if (how
& FLUSH_SYNC
)
1567 rpc_wait_for_completion_task(task
);
1571 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1573 static loff_t
nfs_get_lwb(struct list_head
*head
)
1576 struct nfs_page
*req
;
1578 list_for_each_entry(req
, head
, wb_list
)
1579 if (lwb
< (req_offset(req
) + req
->wb_bytes
))
1580 lwb
= req_offset(req
) + req
->wb_bytes
;
1586 * Set up the argument/result storage required for the RPC call.
1588 void nfs_init_commit(struct nfs_commit_data
*data
,
1589 struct list_head
*head
,
1590 struct pnfs_layout_segment
*lseg
,
1591 struct nfs_commit_info
*cinfo
)
1593 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1594 struct inode
*inode
= d_inode(first
->wb_context
->dentry
);
1596 /* Set up the RPC argument and reply structs
1597 * NB: take care not to mess about with data->commit et al. */
1599 list_splice_init(head
, &data
->pages
);
1601 data
->inode
= inode
;
1602 data
->cred
= first
->wb_context
->cred
;
1603 data
->lseg
= lseg
; /* reference transferred */
1604 /* only set lwb for pnfs commit */
1606 data
->lwb
= nfs_get_lwb(&data
->pages
);
1607 data
->mds_ops
= &nfs_commit_ops
;
1608 data
->completion_ops
= cinfo
->completion_ops
;
1609 data
->dreq
= cinfo
->dreq
;
1611 data
->args
.fh
= NFS_FH(data
->inode
);
1612 /* Note: we always request a commit of the entire inode */
1613 data
->args
.offset
= 0;
1614 data
->args
.count
= 0;
1615 data
->context
= get_nfs_open_context(first
->wb_context
);
1616 data
->res
.fattr
= &data
->fattr
;
1617 data
->res
.verf
= &data
->verf
;
1618 nfs_fattr_init(&data
->fattr
);
1620 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1622 void nfs_retry_commit(struct list_head
*page_list
,
1623 struct pnfs_layout_segment
*lseg
,
1624 struct nfs_commit_info
*cinfo
,
1627 struct nfs_page
*req
;
1629 while (!list_empty(page_list
)) {
1630 req
= nfs_list_entry(page_list
->next
);
1631 nfs_list_remove_request(req
);
1632 nfs_mark_request_commit(req
, lseg
, cinfo
, ds_commit_idx
);
1634 nfs_clear_page_commit(req
->wb_page
);
1635 nfs_unlock_and_release_request(req
);
1638 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1641 * Commit dirty pages
1644 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1645 struct nfs_commit_info
*cinfo
)
1647 struct nfs_commit_data
*data
;
1649 data
= nfs_commitdata_alloc();
1654 /* Set up the argument struct */
1655 nfs_init_commit(data
, head
, NULL
, cinfo
);
1656 atomic_inc(&cinfo
->mds
->rpcs_out
);
1657 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, NFS_PROTO(inode
),
1658 data
->mds_ops
, how
, 0);
1660 nfs_retry_commit(head
, NULL
, cinfo
, 0);
1661 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1666 * COMMIT call returned
1668 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1670 struct nfs_commit_data
*data
= calldata
;
1672 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1673 task
->tk_pid
, task
->tk_status
);
1675 /* Call the NFS version-specific code */
1676 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1679 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1681 struct nfs_page
*req
;
1682 int status
= data
->task
.tk_status
;
1683 struct nfs_commit_info cinfo
;
1684 struct nfs_server
*nfss
;
1686 while (!list_empty(&data
->pages
)) {
1687 req
= nfs_list_entry(data
->pages
.next
);
1688 nfs_list_remove_request(req
);
1689 nfs_clear_page_commit(req
->wb_page
);
1691 dprintk("NFS: commit (%s/%llu %d@%lld)",
1692 req
->wb_context
->dentry
->d_sb
->s_id
,
1693 (unsigned long long)NFS_FILEID(d_inode(req
->wb_context
->dentry
)),
1695 (long long)req_offset(req
));
1697 nfs_context_set_write_error(req
->wb_context
, status
);
1698 nfs_inode_remove_request(req
);
1699 dprintk(", error = %d\n", status
);
1703 /* Okay, COMMIT succeeded, apparently. Check the verifier
1704 * returned by the server against all stored verfs. */
1705 if (!memcmp(&req
->wb_verf
, &data
->verf
.verifier
, sizeof(req
->wb_verf
))) {
1706 /* We have a match */
1707 nfs_inode_remove_request(req
);
1711 /* We have a mismatch. Write the page again */
1712 dprintk(" mismatch\n");
1713 nfs_mark_request_dirty(req
);
1714 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1716 nfs_unlock_and_release_request(req
);
1718 nfss
= NFS_SERVER(data
->inode
);
1719 if (atomic_long_read(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
1720 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
1722 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1723 if (atomic_dec_and_test(&cinfo
.mds
->rpcs_out
))
1724 nfs_commit_clear_lock(NFS_I(data
->inode
));
1727 static void nfs_commit_release(void *calldata
)
1729 struct nfs_commit_data
*data
= calldata
;
1731 data
->completion_ops
->completion(data
);
1732 nfs_commitdata_release(calldata
);
1735 static const struct rpc_call_ops nfs_commit_ops
= {
1736 .rpc_call_prepare
= nfs_commit_prepare
,
1737 .rpc_call_done
= nfs_commit_done
,
1738 .rpc_release
= nfs_commit_release
,
1741 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1742 .completion
= nfs_commit_release_pages
,
1743 .error_cleanup
= nfs_commit_clear_lock
,
1746 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1747 int how
, struct nfs_commit_info
*cinfo
)
1751 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1752 if (status
== PNFS_NOT_ATTEMPTED
)
1753 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1757 int nfs_commit_inode(struct inode
*inode
, int how
)
1760 struct nfs_commit_info cinfo
;
1761 int may_wait
= how
& FLUSH_SYNC
;
1764 res
= nfs_commit_set_lock(NFS_I(inode
), may_wait
);
1766 goto out_mark_dirty
;
1767 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1768 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1772 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1776 goto out_mark_dirty
;
1777 error
= wait_on_bit_action(&NFS_I(inode
)->flags
,
1779 nfs_wait_bit_killable
,
1784 nfs_commit_clear_lock(NFS_I(inode
));
1786 /* Note: If we exit without ensuring that the commit is complete,
1787 * we must mark the inode as dirty. Otherwise, future calls to
1788 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1789 * that the data is on the disk.
1792 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1796 static int nfs_commit_unstable_pages(struct inode
*inode
, struct writeback_control
*wbc
)
1798 struct nfs_inode
*nfsi
= NFS_I(inode
);
1799 int flags
= FLUSH_SYNC
;
1802 /* no commits means nothing needs to be done */
1803 if (!nfsi
->commit_info
.ncommit
)
1806 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1807 /* Don't commit yet if this is a non-blocking flush and there
1808 * are a lot of outstanding writes for this mapping.
1810 if (nfsi
->commit_info
.ncommit
<= (nfsi
->nrequests
>> 1))
1811 goto out_mark_dirty
;
1813 /* don't wait for the COMMIT response */
1817 ret
= nfs_commit_inode(inode
, flags
);
1819 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1820 if (ret
< wbc
->nr_to_write
)
1821 wbc
->nr_to_write
-= ret
;
1823 wbc
->nr_to_write
= 0;
1828 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1832 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1834 return nfs_commit_unstable_pages(inode
, wbc
);
1836 EXPORT_SYMBOL_GPL(nfs_write_inode
);
1839 * flush the inode to disk.
1841 int nfs_wb_all(struct inode
*inode
)
1845 trace_nfs_writeback_inode_enter(inode
);
1847 ret
= filemap_write_and_wait(inode
->i_mapping
);
1850 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1853 pnfs_sync_inode(inode
, true);
1857 trace_nfs_writeback_inode_exit(inode
, ret
);
1860 EXPORT_SYMBOL_GPL(nfs_wb_all
);
1862 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1864 struct nfs_page
*req
;
1867 wait_on_page_writeback(page
);
1869 /* blocking call to cancel all requests and join to a single (head)
1871 req
= nfs_lock_and_join_requests(page
, false);
1876 /* all requests from this page have been cancelled by
1877 * nfs_lock_and_join_requests, so just remove the head
1878 * request from the inode / page_private pointer and
1880 nfs_inode_remove_request(req
);
1881 nfs_unlock_and_release_request(req
);
1888 * Write back all requests on one page - we do this before reading it.
1890 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1892 loff_t range_start
= page_file_offset(page
);
1893 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1894 struct writeback_control wbc
= {
1895 .sync_mode
= WB_SYNC_ALL
,
1897 .range_start
= range_start
,
1898 .range_end
= range_end
,
1902 trace_nfs_writeback_page_enter(inode
);
1905 wait_on_page_writeback(page
);
1906 if (clear_page_dirty_for_io(page
)) {
1907 ret
= nfs_writepage_locked(page
, &wbc
);
1913 if (!PagePrivate(page
))
1915 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1920 trace_nfs_writeback_page_exit(inode
, ret
);
1924 #ifdef CONFIG_MIGRATION
1925 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1926 struct page
*page
, enum migrate_mode mode
)
1929 * If PagePrivate is set, then the page is currently associated with
1930 * an in-progress read or write request. Don't try to migrate it.
1932 * FIXME: we could do this in principle, but we'll need a way to ensure
1933 * that we can safely release the inode reference while holding
1936 if (PagePrivate(page
))
1939 if (!nfs_fscache_release_page(page
, GFP_KERNEL
))
1942 return migrate_page(mapping
, newpage
, page
, mode
);
1946 int __init
nfs_init_writepagecache(void)
1948 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1949 sizeof(struct nfs_pgio_header
),
1950 0, SLAB_HWCACHE_ALIGN
,
1952 if (nfs_wdata_cachep
== NULL
)
1955 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1957 if (nfs_wdata_mempool
== NULL
)
1958 goto out_destroy_write_cache
;
1960 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1961 sizeof(struct nfs_commit_data
),
1962 0, SLAB_HWCACHE_ALIGN
,
1964 if (nfs_cdata_cachep
== NULL
)
1965 goto out_destroy_write_mempool
;
1967 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
1969 if (nfs_commit_mempool
== NULL
)
1970 goto out_destroy_commit_cache
;
1973 * NFS congestion size, scale with available memory.
1985 * This allows larger machines to have larger/more transfers.
1986 * Limit the default to 256M
1988 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
1989 if (nfs_congestion_kb
> 256*1024)
1990 nfs_congestion_kb
= 256*1024;
1994 out_destroy_commit_cache
:
1995 kmem_cache_destroy(nfs_cdata_cachep
);
1996 out_destroy_write_mempool
:
1997 mempool_destroy(nfs_wdata_mempool
);
1998 out_destroy_write_cache
:
1999 kmem_cache_destroy(nfs_wdata_cachep
);
2003 void nfs_destroy_writepagecache(void)
2005 mempool_destroy(nfs_commit_mempool
);
2006 kmem_cache_destroy(nfs_cdata_cachep
);
2007 mempool_destroy(nfs_wdata_mempool
);
2008 kmem_cache_destroy(nfs_wdata_cachep
);
2011 static const struct nfs_rw_ops nfs_rw_write_ops
= {
2012 .rw_mode
= FMODE_WRITE
,
2013 .rw_alloc_header
= nfs_writehdr_alloc
,
2014 .rw_free_header
= nfs_writehdr_free
,
2015 .rw_done
= nfs_writeback_done
,
2016 .rw_result
= nfs_writeback_result
,
2017 .rw_initiate
= nfs_initiate_write
,