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>
24 #include <linux/freezer.h>
25 #include <linux/wait.h>
27 #include <asm/uaccess.h>
29 #include "delegation.h"
38 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
40 #define MIN_POOL_WRITE (32)
41 #define MIN_POOL_COMMIT (4)
44 * Local function declarations
46 static void nfs_redirty_request(struct nfs_page
*req
);
47 static const struct rpc_call_ops nfs_commit_ops
;
48 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
;
49 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
;
50 static const struct nfs_rw_ops nfs_rw_write_ops
;
51 static void nfs_clear_request_commit(struct nfs_page
*req
);
52 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
54 static struct nfs_page
*
55 nfs_page_search_commits_for_head_request_locked(struct nfs_inode
*nfsi
,
58 static struct kmem_cache
*nfs_wdata_cachep
;
59 static mempool_t
*nfs_wdata_mempool
;
60 static struct kmem_cache
*nfs_cdata_cachep
;
61 static mempool_t
*nfs_commit_mempool
;
63 struct nfs_commit_data
*nfs_commitdata_alloc(void)
65 struct nfs_commit_data
*p
= mempool_alloc(nfs_commit_mempool
, GFP_NOIO
);
68 memset(p
, 0, sizeof(*p
));
69 INIT_LIST_HEAD(&p
->pages
);
73 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc
);
75 void nfs_commit_free(struct nfs_commit_data
*p
)
77 mempool_free(p
, nfs_commit_mempool
);
79 EXPORT_SYMBOL_GPL(nfs_commit_free
);
81 static struct nfs_pgio_header
*nfs_writehdr_alloc(void)
83 struct nfs_pgio_header
*p
= mempool_alloc(nfs_wdata_mempool
, GFP_NOIO
);
86 memset(p
, 0, sizeof(*p
));
90 static void nfs_writehdr_free(struct nfs_pgio_header
*hdr
)
92 mempool_free(hdr
, nfs_wdata_mempool
);
95 static void nfs_context_set_write_error(struct nfs_open_context
*ctx
, int error
)
99 set_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
103 * nfs_page_find_head_request_locked - find head request associated with @page
105 * must be called while holding the inode lock.
107 * returns matching head request with reference held, or NULL if not found.
109 static struct nfs_page
*
110 nfs_page_find_head_request_locked(struct nfs_inode
*nfsi
, struct page
*page
)
112 struct nfs_page
*req
= NULL
;
114 if (PagePrivate(page
))
115 req
= (struct nfs_page
*)page_private(page
);
116 else if (unlikely(PageSwapCache(page
)))
117 req
= nfs_page_search_commits_for_head_request_locked(nfsi
,
121 WARN_ON_ONCE(req
->wb_head
!= req
);
122 kref_get(&req
->wb_kref
);
129 * nfs_page_find_head_request - find head request associated with @page
131 * returns matching head request with reference held, or NULL if not found.
133 static struct nfs_page
*nfs_page_find_head_request(struct page
*page
)
135 struct inode
*inode
= page_file_mapping(page
)->host
;
136 struct nfs_page
*req
= NULL
;
138 spin_lock(&inode
->i_lock
);
139 req
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
140 spin_unlock(&inode
->i_lock
);
144 /* Adjust the file length if we're writing beyond the end */
145 static void nfs_grow_file(struct page
*page
, unsigned int offset
, unsigned int count
)
147 struct inode
*inode
= page_file_mapping(page
)->host
;
151 spin_lock(&inode
->i_lock
);
152 i_size
= i_size_read(inode
);
153 end_index
= (i_size
- 1) >> PAGE_CACHE_SHIFT
;
154 if (i_size
> 0 && page_file_index(page
) < end_index
)
156 end
= page_file_offset(page
) + ((loff_t
)offset
+count
);
159 i_size_write(inode
, end
);
160 nfs_inc_stats(inode
, NFSIOS_EXTENDWRITE
);
162 spin_unlock(&inode
->i_lock
);
165 /* A writeback failed: mark the page as bad, and invalidate the page cache */
166 static void nfs_set_pageerror(struct page
*page
)
168 nfs_zap_mapping(page_file_mapping(page
)->host
, page_file_mapping(page
));
172 * nfs_page_group_search_locked
173 * @head - head request of page group
174 * @page_offset - offset into page
176 * Search page group with head @head to find a request that contains the
177 * page offset @page_offset.
179 * Returns a pointer to the first matching nfs request, or NULL if no
182 * Must be called with the page group lock held
184 static struct nfs_page
*
185 nfs_page_group_search_locked(struct nfs_page
*head
, unsigned int page_offset
)
187 struct nfs_page
*req
;
189 WARN_ON_ONCE(head
!= head
->wb_head
);
190 WARN_ON_ONCE(!test_bit(PG_HEADLOCK
, &head
->wb_head
->wb_flags
));
194 if (page_offset
>= req
->wb_pgbase
&&
195 page_offset
< (req
->wb_pgbase
+ req
->wb_bytes
))
198 req
= req
->wb_this_page
;
199 } while (req
!= head
);
205 * nfs_page_group_covers_page
206 * @head - head request of page group
208 * Return true if the page group with head @head covers the whole page,
209 * returns false otherwise
211 static bool nfs_page_group_covers_page(struct nfs_page
*req
)
213 struct nfs_page
*tmp
;
214 unsigned int pos
= 0;
215 unsigned int len
= nfs_page_length(req
->wb_page
);
217 nfs_page_group_lock(req
, false);
220 tmp
= nfs_page_group_search_locked(req
->wb_head
, pos
);
222 /* no way this should happen */
223 WARN_ON_ONCE(tmp
->wb_pgbase
!= pos
);
224 pos
+= tmp
->wb_bytes
- (pos
- tmp
->wb_pgbase
);
226 } while (tmp
&& pos
< len
);
228 nfs_page_group_unlock(req
);
229 WARN_ON_ONCE(pos
> len
);
233 /* We can set the PG_uptodate flag if we see that a write request
234 * covers the full page.
236 static void nfs_mark_uptodate(struct nfs_page
*req
)
238 if (PageUptodate(req
->wb_page
))
240 if (!nfs_page_group_covers_page(req
))
242 SetPageUptodate(req
->wb_page
);
245 static int wb_priority(struct writeback_control
*wbc
)
248 if (wbc
->for_reclaim
)
249 return FLUSH_HIGHPRI
| FLUSH_STABLE
;
250 if (wbc
->sync_mode
== WB_SYNC_ALL
)
251 ret
= FLUSH_COND_STABLE
;
252 if (wbc
->for_kupdate
|| wbc
->for_background
)
258 * NFS congestion control
261 int nfs_congestion_kb
;
263 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
264 #define NFS_CONGESTION_OFF_THRESH \
265 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
267 static void nfs_set_page_writeback(struct page
*page
)
269 struct nfs_server
*nfss
= NFS_SERVER(page_file_mapping(page
)->host
);
270 int ret
= test_set_page_writeback(page
);
272 WARN_ON_ONCE(ret
!= 0);
274 if (atomic_long_inc_return(&nfss
->writeback
) >
275 NFS_CONGESTION_ON_THRESH
) {
276 set_bdi_congested(&nfss
->backing_dev_info
,
281 static void nfs_end_page_writeback(struct nfs_page
*req
)
283 struct inode
*inode
= page_file_mapping(req
->wb_page
)->host
;
284 struct nfs_server
*nfss
= NFS_SERVER(inode
);
286 if (!nfs_page_group_sync_on_bit(req
, PG_WB_END
))
289 end_page_writeback(req
->wb_page
);
290 if (atomic_long_dec_return(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
291 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
295 /* nfs_page_group_clear_bits
296 * @req - an nfs request
297 * clears all page group related bits from @req
300 nfs_page_group_clear_bits(struct nfs_page
*req
)
302 clear_bit(PG_TEARDOWN
, &req
->wb_flags
);
303 clear_bit(PG_UNLOCKPAGE
, &req
->wb_flags
);
304 clear_bit(PG_UPTODATE
, &req
->wb_flags
);
305 clear_bit(PG_WB_END
, &req
->wb_flags
);
306 clear_bit(PG_REMOVE
, &req
->wb_flags
);
311 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
313 * this is a helper function for nfs_lock_and_join_requests
315 * @inode - inode associated with request page group, must be holding inode lock
316 * @head - head request of page group, must be holding head lock
317 * @req - request that couldn't lock and needs to wait on the req bit lock
318 * @nonblock - if true, don't actually wait
320 * NOTE: this must be called holding page_group bit lock and inode spin lock
321 * and BOTH will be released before returning.
323 * returns 0 on success, < 0 on error.
326 nfs_unroll_locks_and_wait(struct inode
*inode
, struct nfs_page
*head
,
327 struct nfs_page
*req
, bool nonblock
)
328 __releases(&inode
->i_lock
)
330 struct nfs_page
*tmp
;
333 /* relinquish all the locks successfully grabbed this run */
334 for (tmp
= head
; tmp
!= req
; tmp
= tmp
->wb_this_page
)
335 nfs_unlock_request(tmp
);
337 WARN_ON_ONCE(test_bit(PG_TEARDOWN
, &req
->wb_flags
));
339 /* grab a ref on the request that will be waited on */
340 kref_get(&req
->wb_kref
);
342 nfs_page_group_unlock(head
);
343 spin_unlock(&inode
->i_lock
);
345 /* release ref from nfs_page_find_head_request_locked */
346 nfs_release_request(head
);
349 ret
= nfs_wait_on_request(req
);
352 nfs_release_request(req
);
358 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
360 * @destroy_list - request list (using wb_this_page) terminated by @old_head
361 * @old_head - the old head of the list
363 * All subrequests must be locked and removed from all lists, so at this point
364 * they are only "active" in this function, and possibly in nfs_wait_on_request
365 * with a reference held by some other context.
368 nfs_destroy_unlinked_subrequests(struct nfs_page
*destroy_list
,
369 struct nfs_page
*old_head
)
371 while (destroy_list
) {
372 struct nfs_page
*subreq
= destroy_list
;
374 destroy_list
= (subreq
->wb_this_page
== old_head
) ?
375 NULL
: subreq
->wb_this_page
;
377 WARN_ON_ONCE(old_head
!= subreq
->wb_head
);
379 /* make sure old group is not used */
380 subreq
->wb_head
= subreq
;
381 subreq
->wb_this_page
= subreq
;
383 /* subreq is now totally disconnected from page group or any
384 * write / commit lists. last chance to wake any waiters */
385 nfs_unlock_request(subreq
);
387 if (!test_bit(PG_TEARDOWN
, &subreq
->wb_flags
)) {
388 /* release ref on old head request */
389 nfs_release_request(old_head
);
391 nfs_page_group_clear_bits(subreq
);
393 /* release the PG_INODE_REF reference */
394 if (test_and_clear_bit(PG_INODE_REF
, &subreq
->wb_flags
))
395 nfs_release_request(subreq
);
399 WARN_ON_ONCE(test_bit(PG_CLEAN
, &subreq
->wb_flags
));
400 /* zombie requests have already released the last
401 * reference and were waiting on the rest of the
402 * group to complete. Since it's no longer part of a
403 * group, simply free the request */
404 nfs_page_group_clear_bits(subreq
);
405 nfs_free_request(subreq
);
411 * nfs_lock_and_join_requests - join all subreqs to the head req and return
412 * a locked reference, cancelling any pending
413 * operations for this page.
415 * @page - the page used to lookup the "page group" of nfs_page structures
416 * @nonblock - if true, don't block waiting for request locks
418 * This function joins all sub requests to the head request by first
419 * locking all requests in the group, cancelling any pending operations
420 * and finally updating the head request to cover the whole range covered by
421 * the (former) group. All subrequests are removed from any write or commit
422 * lists, unlinked from the group and destroyed.
424 * Returns a locked, referenced pointer to the head request - which after
425 * this call is guaranteed to be the only request associated with the page.
426 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
427 * error was encountered.
429 static struct nfs_page
*
430 nfs_lock_and_join_requests(struct page
*page
, bool nonblock
)
432 struct inode
*inode
= page_file_mapping(page
)->host
;
433 struct nfs_page
*head
, *subreq
;
434 struct nfs_page
*destroy_list
= NULL
;
435 unsigned int total_bytes
;
441 WARN_ON_ONCE(destroy_list
);
443 spin_lock(&inode
->i_lock
);
446 * A reference is taken only on the head request which acts as a
447 * reference to the whole page group - the group will not be destroyed
448 * until the head reference is released.
450 head
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
453 spin_unlock(&inode
->i_lock
);
457 /* holding inode lock, so always make a non-blocking call to try the
459 ret
= nfs_page_group_lock(head
, true);
461 spin_unlock(&inode
->i_lock
);
463 if (!nonblock
&& ret
== -EAGAIN
) {
464 nfs_page_group_lock_wait(head
);
465 nfs_release_request(head
);
469 nfs_release_request(head
);
473 /* lock each request in the page group */
477 * Subrequests are always contiguous, non overlapping
478 * and in order - but may be repeated (mirrored writes).
480 if (subreq
->wb_offset
== (head
->wb_offset
+ total_bytes
)) {
481 /* keep track of how many bytes this group covers */
482 total_bytes
+= subreq
->wb_bytes
;
483 } else if (WARN_ON_ONCE(subreq
->wb_offset
< head
->wb_offset
||
484 ((subreq
->wb_offset
+ subreq
->wb_bytes
) >
485 (head
->wb_offset
+ total_bytes
)))) {
486 nfs_page_group_unlock(head
);
487 spin_unlock(&inode
->i_lock
);
488 return ERR_PTR(-EIO
);
491 if (!nfs_lock_request(subreq
)) {
492 /* releases page group bit lock and
493 * inode spin lock and all references */
494 ret
= nfs_unroll_locks_and_wait(inode
, head
,
503 subreq
= subreq
->wb_this_page
;
504 } while (subreq
!= head
);
506 /* Now that all requests are locked, make sure they aren't on any list.
507 * Commit list removal accounting is done after locks are dropped */
510 nfs_clear_request_commit(subreq
);
511 subreq
= subreq
->wb_this_page
;
512 } while (subreq
!= head
);
514 /* unlink subrequests from head, destroy them later */
515 if (head
->wb_this_page
!= head
) {
516 /* destroy list will be terminated by head */
517 destroy_list
= head
->wb_this_page
;
518 head
->wb_this_page
= head
;
520 /* change head request to cover whole range that
521 * the former page group covered */
522 head
->wb_bytes
= total_bytes
;
526 * prepare head request to be added to new pgio descriptor
528 nfs_page_group_clear_bits(head
);
531 * some part of the group was still on the inode list - otherwise
532 * the group wouldn't be involved in async write.
533 * grab a reference for the head request, iff it needs one.
535 if (!test_and_set_bit(PG_INODE_REF
, &head
->wb_flags
))
536 kref_get(&head
->wb_kref
);
538 nfs_page_group_unlock(head
);
540 /* drop lock to clean uprequests on destroy list */
541 spin_unlock(&inode
->i_lock
);
543 nfs_destroy_unlinked_subrequests(destroy_list
, head
);
545 /* still holds ref on head from nfs_page_find_head_request_locked
546 * and still has lock on head from lock loop */
551 * Find an associated nfs write request, and prepare to flush it out
552 * May return an error if the user signalled nfs_wait_on_request().
554 static int nfs_page_async_flush(struct nfs_pageio_descriptor
*pgio
,
555 struct page
*page
, bool nonblock
)
557 struct nfs_page
*req
;
560 req
= nfs_lock_and_join_requests(page
, nonblock
);
567 nfs_set_page_writeback(page
);
568 WARN_ON_ONCE(test_bit(PG_CLEAN
, &req
->wb_flags
));
571 if (!nfs_pageio_add_request(pgio
, req
)) {
572 nfs_redirty_request(req
);
573 ret
= pgio
->pg_error
;
575 nfs_add_stats(page_file_mapping(page
)->host
,
576 NFSIOS_WRITEPAGES
, 1);
581 static int nfs_do_writepage(struct page
*page
, struct writeback_control
*wbc
, struct nfs_pageio_descriptor
*pgio
)
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
;
600 struct inode
*inode
= page_file_mapping(page
)->host
;
603 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGE
);
604 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
),
605 false, &nfs_async_write_completion_ops
);
606 err
= nfs_do_writepage(page
, wbc
, &pgio
);
607 nfs_pageio_complete(&pgio
);
610 if (pgio
.pg_error
< 0)
611 return pgio
.pg_error
;
615 int nfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
619 ret
= nfs_writepage_locked(page
, wbc
);
624 static int nfs_writepages_callback(struct page
*page
, struct writeback_control
*wbc
, void *data
)
628 ret
= nfs_do_writepage(page
, wbc
, data
);
633 int nfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
635 struct inode
*inode
= mapping
->host
;
636 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
637 struct nfs_pageio_descriptor pgio
;
640 /* Stop dirtying of new pages while we sync */
641 err
= wait_on_bit_lock_action(bitlock
, NFS_INO_FLUSHING
,
642 nfs_wait_bit_killable
, TASK_KILLABLE
);
646 nfs_inc_stats(inode
, NFSIOS_VFSWRITEPAGES
);
648 nfs_pageio_init_write(&pgio
, inode
, wb_priority(wbc
), false,
649 &nfs_async_write_completion_ops
);
650 err
= write_cache_pages(mapping
, wbc
, nfs_writepages_callback
, &pgio
);
651 nfs_pageio_complete(&pgio
);
653 clear_bit_unlock(NFS_INO_FLUSHING
, bitlock
);
654 smp_mb__after_atomic();
655 wake_up_bit(bitlock
, NFS_INO_FLUSHING
);
668 * Insert a write request into an inode
670 static void nfs_inode_add_request(struct inode
*inode
, struct nfs_page
*req
)
672 struct nfs_inode
*nfsi
= NFS_I(inode
);
674 WARN_ON_ONCE(req
->wb_this_page
!= req
);
676 /* Lock the request! */
677 nfs_lock_request(req
);
679 spin_lock(&inode
->i_lock
);
680 if (!nfsi
->nrequests
&&
681 NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
684 * Swap-space should not get truncated. Hence no need to plug the race
685 * with invalidate/truncate.
687 if (likely(!PageSwapCache(req
->wb_page
))) {
688 set_bit(PG_MAPPED
, &req
->wb_flags
);
689 SetPagePrivate(req
->wb_page
);
690 set_page_private(req
->wb_page
, (unsigned long)req
);
693 /* this a head request for a page group - mark it as having an
694 * extra reference so sub groups can follow suit.
695 * This flag also informs pgio layer when to bump nrequests when
696 * adding subrequests. */
697 WARN_ON(test_and_set_bit(PG_INODE_REF
, &req
->wb_flags
));
698 kref_get(&req
->wb_kref
);
699 spin_unlock(&inode
->i_lock
);
703 * Remove a write request from an inode
705 static void nfs_inode_remove_request(struct nfs_page
*req
)
707 struct inode
*inode
= d_inode(req
->wb_context
->dentry
);
708 struct nfs_inode
*nfsi
= NFS_I(inode
);
709 struct nfs_page
*head
;
711 if (nfs_page_group_sync_on_bit(req
, PG_REMOVE
)) {
714 spin_lock(&inode
->i_lock
);
715 if (likely(!PageSwapCache(head
->wb_page
))) {
716 set_page_private(head
->wb_page
, 0);
717 ClearPagePrivate(head
->wb_page
);
718 smp_mb__after_atomic();
719 wake_up_page(head
->wb_page
, PG_private
);
720 clear_bit(PG_MAPPED
, &head
->wb_flags
);
723 spin_unlock(&inode
->i_lock
);
725 spin_lock(&inode
->i_lock
);
727 spin_unlock(&inode
->i_lock
);
730 if (test_and_clear_bit(PG_INODE_REF
, &req
->wb_flags
))
731 nfs_release_request(req
);
735 nfs_mark_request_dirty(struct nfs_page
*req
)
737 __set_page_dirty_nobuffers(req
->wb_page
);
741 * nfs_page_search_commits_for_head_request_locked
743 * Search through commit lists on @inode for the head request for @page.
744 * Must be called while holding the inode (which is cinfo) lock.
746 * Returns the head request if found, or NULL if not found.
748 static struct nfs_page
*
749 nfs_page_search_commits_for_head_request_locked(struct nfs_inode
*nfsi
,
752 struct nfs_page
*freq
, *t
;
753 struct nfs_commit_info cinfo
;
754 struct inode
*inode
= &nfsi
->vfs_inode
;
756 nfs_init_cinfo_from_inode(&cinfo
, inode
);
758 /* search through pnfs commit lists */
759 freq
= pnfs_search_commit_reqs(inode
, &cinfo
, page
);
761 return freq
->wb_head
;
763 /* Linearly search the commit list for the correct request */
764 list_for_each_entry_safe(freq
, t
, &cinfo
.mds
->list
, wb_list
) {
765 if (freq
->wb_page
== page
)
766 return freq
->wb_head
;
773 * nfs_request_add_commit_list_locked - add request to a commit list
774 * @req: pointer to a struct nfs_page
775 * @dst: commit list head
776 * @cinfo: holds list lock and accounting info
778 * This sets the PG_CLEAN bit, updates the cinfo count of
779 * number of outstanding requests requiring a commit as well as
782 * The caller must hold the cinfo->lock, and the nfs_page lock.
785 nfs_request_add_commit_list_locked(struct nfs_page
*req
, struct list_head
*dst
,
786 struct nfs_commit_info
*cinfo
)
788 set_bit(PG_CLEAN
, &req
->wb_flags
);
789 nfs_list_add_request(req
, dst
);
790 cinfo
->mds
->ncommit
++;
792 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked
);
795 * nfs_request_add_commit_list - add request to a commit list
796 * @req: pointer to a struct nfs_page
797 * @dst: commit list head
798 * @cinfo: holds list lock and accounting info
800 * This sets the PG_CLEAN bit, updates the cinfo count of
801 * number of outstanding requests requiring a commit as well as
804 * The caller must _not_ hold the cinfo->lock, but must be
805 * holding the nfs_page lock.
808 nfs_request_add_commit_list(struct nfs_page
*req
, struct list_head
*dst
,
809 struct nfs_commit_info
*cinfo
)
811 spin_lock(cinfo
->lock
);
812 nfs_request_add_commit_list_locked(req
, dst
, cinfo
);
813 spin_unlock(cinfo
->lock
);
814 nfs_mark_page_unstable(req
->wb_page
, cinfo
);
816 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list
);
819 * nfs_request_remove_commit_list - Remove request from a commit list
820 * @req: pointer to a nfs_page
821 * @cinfo: holds list lock and accounting info
823 * This clears the PG_CLEAN bit, and updates the cinfo's count of
824 * number of outstanding requests requiring a commit
825 * It does not update the MM page stats.
827 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
830 nfs_request_remove_commit_list(struct nfs_page
*req
,
831 struct nfs_commit_info
*cinfo
)
833 if (!test_and_clear_bit(PG_CLEAN
, &(req
)->wb_flags
))
835 nfs_list_remove_request(req
);
836 cinfo
->mds
->ncommit
--;
838 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list
);
840 static void nfs_init_cinfo_from_inode(struct nfs_commit_info
*cinfo
,
843 cinfo
->lock
= &inode
->i_lock
;
844 cinfo
->mds
= &NFS_I(inode
)->commit_info
;
845 cinfo
->ds
= pnfs_get_ds_info(inode
);
847 cinfo
->completion_ops
= &nfs_commit_completion_ops
;
850 void nfs_init_cinfo(struct nfs_commit_info
*cinfo
,
852 struct nfs_direct_req
*dreq
)
855 nfs_init_cinfo_from_dreq(cinfo
, dreq
);
857 nfs_init_cinfo_from_inode(cinfo
, inode
);
859 EXPORT_SYMBOL_GPL(nfs_init_cinfo
);
862 * Add a request to the inode's commit list.
865 nfs_mark_request_commit(struct nfs_page
*req
, struct pnfs_layout_segment
*lseg
,
866 struct nfs_commit_info
*cinfo
, u32 ds_commit_idx
)
868 if (pnfs_mark_request_commit(req
, lseg
, cinfo
, ds_commit_idx
))
870 nfs_request_add_commit_list(req
, &cinfo
->mds
->list
, cinfo
);
874 nfs_clear_page_commit(struct page
*page
)
876 dec_zone_page_state(page
, NR_UNSTABLE_NFS
);
877 dec_wb_stat(&inode_to_bdi(page_file_mapping(page
)->host
)->wb
,
881 /* Called holding inode (/cinfo) lock */
883 nfs_clear_request_commit(struct nfs_page
*req
)
885 if (test_bit(PG_CLEAN
, &req
->wb_flags
)) {
886 struct inode
*inode
= d_inode(req
->wb_context
->dentry
);
887 struct nfs_commit_info cinfo
;
889 nfs_init_cinfo_from_inode(&cinfo
, inode
);
890 if (!pnfs_clear_request_commit(req
, &cinfo
)) {
891 nfs_request_remove_commit_list(req
, &cinfo
);
893 nfs_clear_page_commit(req
->wb_page
);
897 int nfs_write_need_commit(struct nfs_pgio_header
*hdr
)
899 if (hdr
->verf
.committed
== NFS_DATA_SYNC
)
900 return hdr
->lseg
== NULL
;
901 return hdr
->verf
.committed
!= NFS_FILE_SYNC
;
904 static void nfs_write_completion(struct nfs_pgio_header
*hdr
)
906 struct nfs_commit_info cinfo
;
907 unsigned long bytes
= 0;
909 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
911 nfs_init_cinfo_from_inode(&cinfo
, hdr
->inode
);
912 while (!list_empty(&hdr
->pages
)) {
913 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
915 bytes
+= req
->wb_bytes
;
916 nfs_list_remove_request(req
);
917 if (test_bit(NFS_IOHDR_ERROR
, &hdr
->flags
) &&
918 (hdr
->good_bytes
< bytes
)) {
919 nfs_set_pageerror(req
->wb_page
);
920 nfs_context_set_write_error(req
->wb_context
, hdr
->error
);
923 if (nfs_write_need_commit(hdr
)) {
924 memcpy(&req
->wb_verf
, &hdr
->verf
.verifier
, sizeof(req
->wb_verf
));
925 nfs_mark_request_commit(req
, hdr
->lseg
, &cinfo
,
926 hdr
->pgio_mirror_idx
);
930 nfs_inode_remove_request(req
);
932 nfs_unlock_request(req
);
933 nfs_end_page_writeback(req
);
934 nfs_release_request(req
);
941 nfs_reqs_to_commit(struct nfs_commit_info
*cinfo
)
943 return cinfo
->mds
->ncommit
;
946 /* cinfo->lock held by caller */
948 nfs_scan_commit_list(struct list_head
*src
, struct list_head
*dst
,
949 struct nfs_commit_info
*cinfo
, int max
)
951 struct nfs_page
*req
, *tmp
;
954 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
955 if (!nfs_lock_request(req
))
957 kref_get(&req
->wb_kref
);
958 if (cond_resched_lock(cinfo
->lock
))
959 list_safe_reset_next(req
, tmp
, wb_list
);
960 nfs_request_remove_commit_list(req
, cinfo
);
961 nfs_list_add_request(req
, dst
);
963 if ((ret
== max
) && !cinfo
->dreq
)
970 * nfs_scan_commit - Scan an inode for commit requests
971 * @inode: NFS inode to scan
972 * @dst: mds destination list
973 * @cinfo: mds and ds lists of reqs ready to commit
975 * Moves requests from the inode's 'commit' request list.
976 * The requests are *not* checked to ensure that they form a contiguous set.
979 nfs_scan_commit(struct inode
*inode
, struct list_head
*dst
,
980 struct nfs_commit_info
*cinfo
)
984 spin_lock(cinfo
->lock
);
985 if (cinfo
->mds
->ncommit
> 0) {
986 const int max
= INT_MAX
;
988 ret
= nfs_scan_commit_list(&cinfo
->mds
->list
, dst
,
990 ret
+= pnfs_scan_commit_lists(inode
, cinfo
, max
- ret
);
992 spin_unlock(cinfo
->lock
);
997 * Search for an existing write request, and attempt to update
998 * it to reflect a new dirty region on a given page.
1000 * If the attempt fails, then the existing request is flushed out
1003 static struct nfs_page
*nfs_try_to_update_request(struct inode
*inode
,
1005 unsigned int offset
,
1008 struct nfs_page
*req
;
1013 if (!PagePrivate(page
))
1016 end
= offset
+ bytes
;
1017 spin_lock(&inode
->i_lock
);
1020 req
= nfs_page_find_head_request_locked(NFS_I(inode
), page
);
1024 /* should be handled by nfs_flush_incompatible */
1025 WARN_ON_ONCE(req
->wb_head
!= req
);
1026 WARN_ON_ONCE(req
->wb_this_page
!= req
);
1028 rqend
= req
->wb_offset
+ req
->wb_bytes
;
1030 * Tell the caller to flush out the request if
1031 * the offsets are non-contiguous.
1032 * Note: nfs_flush_incompatible() will already
1033 * have flushed out requests having wrong owners.
1036 || end
< req
->wb_offset
)
1039 if (nfs_lock_request(req
))
1042 /* The request is locked, so wait and then retry */
1043 spin_unlock(&inode
->i_lock
);
1044 error
= nfs_wait_on_request(req
);
1045 nfs_release_request(req
);
1048 spin_lock(&inode
->i_lock
);
1051 /* Okay, the request matches. Update the region */
1052 if (offset
< req
->wb_offset
) {
1053 req
->wb_offset
= offset
;
1054 req
->wb_pgbase
= offset
;
1057 req
->wb_bytes
= end
- req
->wb_offset
;
1059 req
->wb_bytes
= rqend
- req
->wb_offset
;
1062 nfs_clear_request_commit(req
);
1063 spin_unlock(&inode
->i_lock
);
1066 spin_unlock(&inode
->i_lock
);
1067 nfs_release_request(req
);
1068 error
= nfs_wb_page(inode
, page
);
1070 return ERR_PTR(error
);
1074 * Try to update an existing write request, or create one if there is none.
1076 * Note: Should always be called with the Page Lock held to prevent races
1077 * if we have to add a new request. Also assumes that the caller has
1078 * already called nfs_flush_incompatible() if necessary.
1080 static struct nfs_page
* nfs_setup_write_request(struct nfs_open_context
* ctx
,
1081 struct page
*page
, unsigned int offset
, unsigned int bytes
)
1083 struct inode
*inode
= page_file_mapping(page
)->host
;
1084 struct nfs_page
*req
;
1086 req
= nfs_try_to_update_request(inode
, page
, offset
, bytes
);
1089 req
= nfs_create_request(ctx
, page
, NULL
, offset
, bytes
);
1092 nfs_inode_add_request(inode
, req
);
1097 static int nfs_writepage_setup(struct nfs_open_context
*ctx
, struct page
*page
,
1098 unsigned int offset
, unsigned int count
)
1100 struct nfs_page
*req
;
1102 req
= nfs_setup_write_request(ctx
, page
, offset
, count
);
1104 return PTR_ERR(req
);
1105 /* Update file length */
1106 nfs_grow_file(page
, offset
, count
);
1107 nfs_mark_uptodate(req
);
1108 nfs_mark_request_dirty(req
);
1109 nfs_unlock_and_release_request(req
);
1113 int nfs_flush_incompatible(struct file
*file
, struct page
*page
)
1115 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
1116 struct nfs_lock_context
*l_ctx
;
1117 struct file_lock_context
*flctx
= file_inode(file
)->i_flctx
;
1118 struct nfs_page
*req
;
1119 int do_flush
, status
;
1121 * Look for a request corresponding to this page. If there
1122 * is one, and it belongs to another file, we flush it out
1123 * before we try to copy anything into the page. Do this
1124 * due to the lack of an ACCESS-type call in NFSv2.
1125 * Also do the same if we find a request from an existing
1129 req
= nfs_page_find_head_request(page
);
1132 l_ctx
= req
->wb_lock_context
;
1133 do_flush
= req
->wb_page
!= page
||
1134 !nfs_match_open_context(req
->wb_context
, ctx
);
1135 /* for now, flush if more than 1 request in page_group */
1136 do_flush
|= req
->wb_this_page
!= req
;
1137 if (l_ctx
&& flctx
&&
1138 !(list_empty_careful(&flctx
->flc_posix
) &&
1139 list_empty_careful(&flctx
->flc_flock
))) {
1140 do_flush
|= l_ctx
->lockowner
.l_owner
!= current
->files
1141 || l_ctx
->lockowner
.l_pid
!= current
->tgid
;
1143 nfs_release_request(req
);
1146 status
= nfs_wb_page(page_file_mapping(page
)->host
, page
);
1147 } while (status
== 0);
1152 * Avoid buffered writes when a open context credential's key would
1155 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1157 * Return 0 and set a credential flag which triggers the inode to flush
1158 * and performs NFS_FILE_SYNC writes if the key will expired within
1159 * RPC_KEY_EXPIRE_TIMEO.
1162 nfs_key_timeout_notify(struct file
*filp
, struct inode
*inode
)
1164 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
1165 struct rpc_auth
*auth
= NFS_SERVER(inode
)->client
->cl_auth
;
1167 return rpcauth_key_timeout_notify(auth
, ctx
->cred
);
1171 * Test if the open context credential key is marked to expire soon.
1173 bool nfs_ctx_key_to_expire(struct nfs_open_context
*ctx
)
1175 return rpcauth_cred_key_to_expire(ctx
->cred
);
1179 * If the page cache is marked as unsafe or invalid, then we can't rely on
1180 * the PageUptodate() flag. In this case, we will need to turn off
1181 * write optimisations that depend on the page contents being correct.
1183 static bool nfs_write_pageuptodate(struct page
*page
, struct inode
*inode
)
1185 struct nfs_inode
*nfsi
= NFS_I(inode
);
1187 if (nfs_have_delegated_attributes(inode
))
1189 if (nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
1192 if (test_bit(NFS_INO_INVALIDATING
, &nfsi
->flags
))
1195 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1197 return PageUptodate(page
) != 0;
1201 is_whole_file_wrlock(struct file_lock
*fl
)
1203 return fl
->fl_start
== 0 && fl
->fl_end
== OFFSET_MAX
&&
1204 fl
->fl_type
== F_WRLCK
;
1207 /* If we know the page is up to date, and we're not using byte range locks (or
1208 * if we have the whole file locked for writing), it may be more efficient to
1209 * extend the write to cover the entire page in order to avoid fragmentation
1212 * If the file is opened for synchronous writes then we can just skip the rest
1215 static int nfs_can_extend_write(struct file
*file
, struct page
*page
, struct inode
*inode
)
1218 struct file_lock_context
*flctx
= inode
->i_flctx
;
1219 struct file_lock
*fl
;
1221 if (file
->f_flags
& O_DSYNC
)
1223 if (!nfs_write_pageuptodate(page
, inode
))
1225 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_WRITE
))
1227 if (!flctx
|| (list_empty_careful(&flctx
->flc_flock
) &&
1228 list_empty_careful(&flctx
->flc_posix
)))
1231 /* Check to see if there are whole file write locks */
1233 spin_lock(&flctx
->flc_lock
);
1234 if (!list_empty(&flctx
->flc_posix
)) {
1235 fl
= list_first_entry(&flctx
->flc_posix
, struct file_lock
,
1237 if (is_whole_file_wrlock(fl
))
1239 } else if (!list_empty(&flctx
->flc_flock
)) {
1240 fl
= list_first_entry(&flctx
->flc_flock
, struct file_lock
,
1242 if (fl
->fl_type
== F_WRLCK
)
1245 spin_unlock(&flctx
->flc_lock
);
1250 * Update and possibly write a cached page of an NFS file.
1252 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1253 * things with a page scheduled for an RPC call (e.g. invalidate it).
1255 int nfs_updatepage(struct file
*file
, struct page
*page
,
1256 unsigned int offset
, unsigned int count
)
1258 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
1259 struct inode
*inode
= page_file_mapping(page
)->host
;
1262 nfs_inc_stats(inode
, NFSIOS_VFSUPDATEPAGE
);
1264 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1265 file
, count
, (long long)(page_file_offset(page
) + offset
));
1267 if (nfs_can_extend_write(file
, page
, inode
)) {
1268 count
= max(count
+ offset
, nfs_page_length(page
));
1272 status
= nfs_writepage_setup(ctx
, page
, offset
, count
);
1274 nfs_set_pageerror(page
);
1276 __set_page_dirty_nobuffers(page
);
1278 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1279 status
, (long long)i_size_read(inode
));
1283 static int flush_task_priority(int how
)
1285 switch (how
& (FLUSH_HIGHPRI
|FLUSH_LOWPRI
)) {
1287 return RPC_PRIORITY_HIGH
;
1289 return RPC_PRIORITY_LOW
;
1291 return RPC_PRIORITY_NORMAL
;
1294 static void nfs_initiate_write(struct nfs_pgio_header
*hdr
,
1295 struct rpc_message
*msg
,
1296 const struct nfs_rpc_ops
*rpc_ops
,
1297 struct rpc_task_setup
*task_setup_data
, int how
)
1299 int priority
= flush_task_priority(how
);
1301 task_setup_data
->priority
= priority
;
1302 rpc_ops
->write_setup(hdr
, msg
);
1304 nfs4_state_protect_write(NFS_SERVER(hdr
->inode
)->nfs_client
,
1305 &task_setup_data
->rpc_client
, msg
, hdr
);
1308 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1309 * call this on each, which will prepare them to be retried on next
1310 * writeback using standard nfs.
1312 static void nfs_redirty_request(struct nfs_page
*req
)
1314 nfs_mark_request_dirty(req
);
1315 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1316 nfs_unlock_request(req
);
1317 nfs_end_page_writeback(req
);
1318 nfs_release_request(req
);
1321 static void nfs_async_write_error(struct list_head
*head
)
1323 struct nfs_page
*req
;
1325 while (!list_empty(head
)) {
1326 req
= nfs_list_entry(head
->next
);
1327 nfs_list_remove_request(req
);
1328 nfs_redirty_request(req
);
1332 static void nfs_async_write_reschedule_io(struct nfs_pgio_header
*hdr
)
1334 nfs_async_write_error(&hdr
->pages
);
1337 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops
= {
1338 .error_cleanup
= nfs_async_write_error
,
1339 .completion
= nfs_write_completion
,
1340 .reschedule_io
= nfs_async_write_reschedule_io
,
1343 void nfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
,
1344 struct inode
*inode
, int ioflags
, bool force_mds
,
1345 const struct nfs_pgio_completion_ops
*compl_ops
)
1347 struct nfs_server
*server
= NFS_SERVER(inode
);
1348 const struct nfs_pageio_ops
*pg_ops
= &nfs_pgio_rw_ops
;
1350 #ifdef CONFIG_NFS_V4_1
1351 if (server
->pnfs_curr_ld
&& !force_mds
)
1352 pg_ops
= server
->pnfs_curr_ld
->pg_write_ops
;
1354 nfs_pageio_init(pgio
, inode
, pg_ops
, compl_ops
, &nfs_rw_write_ops
,
1355 server
->wsize
, ioflags
);
1357 EXPORT_SYMBOL_GPL(nfs_pageio_init_write
);
1359 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor
*pgio
)
1361 struct nfs_pgio_mirror
*mirror
;
1363 if (pgio
->pg_ops
&& pgio
->pg_ops
->pg_cleanup
)
1364 pgio
->pg_ops
->pg_cleanup(pgio
);
1366 pgio
->pg_ops
= &nfs_pgio_rw_ops
;
1368 nfs_pageio_stop_mirroring(pgio
);
1370 mirror
= &pgio
->pg_mirrors
[0];
1371 mirror
->pg_bsize
= NFS_SERVER(pgio
->pg_inode
)->wsize
;
1373 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds
);
1376 void nfs_commit_prepare(struct rpc_task
*task
, void *calldata
)
1378 struct nfs_commit_data
*data
= calldata
;
1380 NFS_PROTO(data
->inode
)->commit_rpc_prepare(task
, data
);
1384 * Special version of should_remove_suid() that ignores capabilities.
1386 static int nfs_should_remove_suid(const struct inode
*inode
)
1388 umode_t mode
= inode
->i_mode
;
1391 /* suid always must be killed */
1392 if (unlikely(mode
& S_ISUID
))
1393 kill
= ATTR_KILL_SUID
;
1396 * sgid without any exec bits is just a mandatory locking mark; leave
1397 * it alone. If some exec bits are set, it's a real sgid; kill it.
1399 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1400 kill
|= ATTR_KILL_SGID
;
1402 if (unlikely(kill
&& S_ISREG(mode
)))
1408 static void nfs_writeback_check_extend(struct nfs_pgio_header
*hdr
,
1409 struct nfs_fattr
*fattr
)
1411 struct nfs_pgio_args
*argp
= &hdr
->args
;
1412 struct nfs_pgio_res
*resp
= &hdr
->res
;
1413 u64 size
= argp
->offset
+ resp
->count
;
1415 if (!(fattr
->valid
& NFS_ATTR_FATTR_SIZE
))
1417 if (nfs_size_to_loff_t(fattr
->size
) < i_size_read(hdr
->inode
)) {
1418 fattr
->valid
&= ~NFS_ATTR_FATTR_SIZE
;
1421 if (size
!= fattr
->size
)
1423 /* Set attribute barrier */
1424 nfs_fattr_set_barrier(fattr
);
1425 /* ...and update size */
1426 fattr
->valid
|= NFS_ATTR_FATTR_SIZE
;
1429 void nfs_writeback_update_inode(struct nfs_pgio_header
*hdr
)
1431 struct nfs_fattr
*fattr
= &hdr
->fattr
;
1432 struct inode
*inode
= hdr
->inode
;
1434 spin_lock(&inode
->i_lock
);
1435 nfs_writeback_check_extend(hdr
, fattr
);
1436 nfs_post_op_update_inode_force_wcc_locked(inode
, fattr
);
1437 spin_unlock(&inode
->i_lock
);
1439 EXPORT_SYMBOL_GPL(nfs_writeback_update_inode
);
1442 * This function is called when the WRITE call is complete.
1444 static int nfs_writeback_done(struct rpc_task
*task
,
1445 struct nfs_pgio_header
*hdr
,
1446 struct inode
*inode
)
1451 * ->write_done will attempt to use post-op attributes to detect
1452 * conflicting writes by other clients. A strict interpretation
1453 * of close-to-open would allow us to continue caching even if
1454 * another writer had changed the file, but some applications
1455 * depend on tighter cache coherency when writing.
1457 status
= NFS_PROTO(inode
)->write_done(task
, hdr
);
1460 nfs_add_stats(inode
, NFSIOS_SERVERWRITTENBYTES
, hdr
->res
.count
);
1462 if (hdr
->res
.verf
->committed
< hdr
->args
.stable
&&
1463 task
->tk_status
>= 0) {
1464 /* We tried a write call, but the server did not
1465 * commit data to stable storage even though we
1467 * Note: There is a known bug in Tru64 < 5.0 in which
1468 * the server reports NFS_DATA_SYNC, but performs
1469 * NFS_FILE_SYNC. We therefore implement this checking
1470 * as a dprintk() in order to avoid filling syslog.
1472 static unsigned long complain
;
1474 /* Note this will print the MDS for a DS write */
1475 if (time_before(complain
, jiffies
)) {
1476 dprintk("NFS: faulty NFS server %s:"
1477 " (committed = %d) != (stable = %d)\n",
1478 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1479 hdr
->res
.verf
->committed
, hdr
->args
.stable
);
1480 complain
= jiffies
+ 300 * HZ
;
1484 /* Deal with the suid/sgid bit corner case */
1485 if (nfs_should_remove_suid(inode
))
1486 nfs_mark_for_revalidate(inode
);
1491 * This function is called when the WRITE call is complete.
1493 static void nfs_writeback_result(struct rpc_task
*task
,
1494 struct nfs_pgio_header
*hdr
)
1496 struct nfs_pgio_args
*argp
= &hdr
->args
;
1497 struct nfs_pgio_res
*resp
= &hdr
->res
;
1499 if (resp
->count
< argp
->count
) {
1500 static unsigned long complain
;
1502 /* This a short write! */
1503 nfs_inc_stats(hdr
->inode
, NFSIOS_SHORTWRITE
);
1505 /* Has the server at least made some progress? */
1506 if (resp
->count
== 0) {
1507 if (time_before(complain
, jiffies
)) {
1509 "NFS: Server wrote zero bytes, expected %u.\n",
1511 complain
= jiffies
+ 300 * HZ
;
1513 nfs_set_pgio_error(hdr
, -EIO
, argp
->offset
);
1514 task
->tk_status
= -EIO
;
1518 /* For non rpc-based layout drivers, retry-through-MDS */
1519 if (!task
->tk_ops
) {
1520 hdr
->pnfs_error
= -EAGAIN
;
1524 /* Was this an NFSv2 write or an NFSv3 stable write? */
1525 if (resp
->verf
->committed
!= NFS_UNSTABLE
) {
1526 /* Resend from where the server left off */
1527 hdr
->mds_offset
+= resp
->count
;
1528 argp
->offset
+= resp
->count
;
1529 argp
->pgbase
+= resp
->count
;
1530 argp
->count
-= resp
->count
;
1532 /* Resend as a stable write in order to avoid
1533 * headaches in the case of a server crash.
1535 argp
->stable
= NFS_FILE_SYNC
;
1537 rpc_restart_call_prepare(task
);
1541 static int nfs_wait_atomic_killable(atomic_t
*key
)
1543 if (fatal_signal_pending(current
))
1544 return -ERESTARTSYS
;
1545 freezable_schedule_unsafe();
1549 static int wait_on_commit(struct nfs_mds_commit_info
*cinfo
)
1551 return wait_on_atomic_t(&cinfo
->rpcs_out
,
1552 nfs_wait_atomic_killable
, TASK_KILLABLE
);
1555 static void nfs_commit_begin(struct nfs_mds_commit_info
*cinfo
)
1557 atomic_inc(&cinfo
->rpcs_out
);
1560 static void nfs_commit_end(struct nfs_mds_commit_info
*cinfo
)
1562 if (atomic_dec_and_test(&cinfo
->rpcs_out
))
1563 wake_up_atomic_t(&cinfo
->rpcs_out
);
1566 void nfs_commitdata_release(struct nfs_commit_data
*data
)
1568 put_nfs_open_context(data
->context
);
1569 nfs_commit_free(data
);
1571 EXPORT_SYMBOL_GPL(nfs_commitdata_release
);
1573 int nfs_initiate_commit(struct rpc_clnt
*clnt
, struct nfs_commit_data
*data
,
1574 const struct nfs_rpc_ops
*nfs_ops
,
1575 const struct rpc_call_ops
*call_ops
,
1578 struct rpc_task
*task
;
1579 int priority
= flush_task_priority(how
);
1580 struct rpc_message msg
= {
1581 .rpc_argp
= &data
->args
,
1582 .rpc_resp
= &data
->res
,
1583 .rpc_cred
= data
->cred
,
1585 struct rpc_task_setup task_setup_data
= {
1586 .task
= &data
->task
,
1588 .rpc_message
= &msg
,
1589 .callback_ops
= call_ops
,
1590 .callback_data
= data
,
1591 .workqueue
= nfsiod_workqueue
,
1592 .flags
= RPC_TASK_ASYNC
| flags
,
1593 .priority
= priority
,
1595 /* Set up the initial task struct. */
1596 nfs_ops
->commit_setup(data
, &msg
);
1598 dprintk("NFS: initiated commit call\n");
1600 nfs4_state_protect(NFS_SERVER(data
->inode
)->nfs_client
,
1601 NFS_SP4_MACH_CRED_COMMIT
, &task_setup_data
.rpc_client
, &msg
);
1603 task
= rpc_run_task(&task_setup_data
);
1605 return PTR_ERR(task
);
1606 if (how
& FLUSH_SYNC
)
1607 rpc_wait_for_completion_task(task
);
1611 EXPORT_SYMBOL_GPL(nfs_initiate_commit
);
1613 static loff_t
nfs_get_lwb(struct list_head
*head
)
1616 struct nfs_page
*req
;
1618 list_for_each_entry(req
, head
, wb_list
)
1619 if (lwb
< (req_offset(req
) + req
->wb_bytes
))
1620 lwb
= req_offset(req
) + req
->wb_bytes
;
1626 * Set up the argument/result storage required for the RPC call.
1628 void nfs_init_commit(struct nfs_commit_data
*data
,
1629 struct list_head
*head
,
1630 struct pnfs_layout_segment
*lseg
,
1631 struct nfs_commit_info
*cinfo
)
1633 struct nfs_page
*first
= nfs_list_entry(head
->next
);
1634 struct inode
*inode
= d_inode(first
->wb_context
->dentry
);
1636 /* Set up the RPC argument and reply structs
1637 * NB: take care not to mess about with data->commit et al. */
1639 list_splice_init(head
, &data
->pages
);
1641 data
->inode
= inode
;
1642 data
->cred
= first
->wb_context
->cred
;
1643 data
->lseg
= lseg
; /* reference transferred */
1644 /* only set lwb for pnfs commit */
1646 data
->lwb
= nfs_get_lwb(&data
->pages
);
1647 data
->mds_ops
= &nfs_commit_ops
;
1648 data
->completion_ops
= cinfo
->completion_ops
;
1649 data
->dreq
= cinfo
->dreq
;
1651 data
->args
.fh
= NFS_FH(data
->inode
);
1652 /* Note: we always request a commit of the entire inode */
1653 data
->args
.offset
= 0;
1654 data
->args
.count
= 0;
1655 data
->context
= get_nfs_open_context(first
->wb_context
);
1656 data
->res
.fattr
= &data
->fattr
;
1657 data
->res
.verf
= &data
->verf
;
1658 nfs_fattr_init(&data
->fattr
);
1660 EXPORT_SYMBOL_GPL(nfs_init_commit
);
1662 void nfs_retry_commit(struct list_head
*page_list
,
1663 struct pnfs_layout_segment
*lseg
,
1664 struct nfs_commit_info
*cinfo
,
1667 struct nfs_page
*req
;
1669 while (!list_empty(page_list
)) {
1670 req
= nfs_list_entry(page_list
->next
);
1671 nfs_list_remove_request(req
);
1672 nfs_mark_request_commit(req
, lseg
, cinfo
, ds_commit_idx
);
1674 nfs_clear_page_commit(req
->wb_page
);
1675 nfs_unlock_and_release_request(req
);
1678 EXPORT_SYMBOL_GPL(nfs_retry_commit
);
1681 nfs_commit_resched_write(struct nfs_commit_info
*cinfo
,
1682 struct nfs_page
*req
)
1684 __set_page_dirty_nobuffers(req
->wb_page
);
1688 * Commit dirty pages
1691 nfs_commit_list(struct inode
*inode
, struct list_head
*head
, int how
,
1692 struct nfs_commit_info
*cinfo
)
1694 struct nfs_commit_data
*data
;
1696 data
= nfs_commitdata_alloc();
1701 /* Set up the argument struct */
1702 nfs_init_commit(data
, head
, NULL
, cinfo
);
1703 atomic_inc(&cinfo
->mds
->rpcs_out
);
1704 return nfs_initiate_commit(NFS_CLIENT(inode
), data
, NFS_PROTO(inode
),
1705 data
->mds_ops
, how
, 0);
1707 nfs_retry_commit(head
, NULL
, cinfo
, 0);
1712 * COMMIT call returned
1714 static void nfs_commit_done(struct rpc_task
*task
, void *calldata
)
1716 struct nfs_commit_data
*data
= calldata
;
1718 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1719 task
->tk_pid
, task
->tk_status
);
1721 /* Call the NFS version-specific code */
1722 NFS_PROTO(data
->inode
)->commit_done(task
, data
);
1725 static void nfs_commit_release_pages(struct nfs_commit_data
*data
)
1727 struct nfs_page
*req
;
1728 int status
= data
->task
.tk_status
;
1729 struct nfs_commit_info cinfo
;
1730 struct nfs_server
*nfss
;
1732 while (!list_empty(&data
->pages
)) {
1733 req
= nfs_list_entry(data
->pages
.next
);
1734 nfs_list_remove_request(req
);
1735 nfs_clear_page_commit(req
->wb_page
);
1737 dprintk("NFS: commit (%s/%llu %d@%lld)",
1738 req
->wb_context
->dentry
->d_sb
->s_id
,
1739 (unsigned long long)NFS_FILEID(d_inode(req
->wb_context
->dentry
)),
1741 (long long)req_offset(req
));
1743 nfs_context_set_write_error(req
->wb_context
, status
);
1744 nfs_inode_remove_request(req
);
1745 dprintk(", error = %d\n", status
);
1749 /* Okay, COMMIT succeeded, apparently. Check the verifier
1750 * returned by the server against all stored verfs. */
1751 if (!memcmp(&req
->wb_verf
, &data
->verf
.verifier
, sizeof(req
->wb_verf
))) {
1752 /* We have a match */
1753 nfs_inode_remove_request(req
);
1757 /* We have a mismatch. Write the page again */
1758 dprintk(" mismatch\n");
1759 nfs_mark_request_dirty(req
);
1760 set_bit(NFS_CONTEXT_RESEND_WRITES
, &req
->wb_context
->flags
);
1762 nfs_unlock_and_release_request(req
);
1764 nfss
= NFS_SERVER(data
->inode
);
1765 if (atomic_long_read(&nfss
->writeback
) < NFS_CONGESTION_OFF_THRESH
)
1766 clear_bdi_congested(&nfss
->backing_dev_info
, BLK_RW_ASYNC
);
1768 nfs_init_cinfo(&cinfo
, data
->inode
, data
->dreq
);
1769 nfs_commit_end(cinfo
.mds
);
1772 static void nfs_commit_release(void *calldata
)
1774 struct nfs_commit_data
*data
= calldata
;
1776 data
->completion_ops
->completion(data
);
1777 nfs_commitdata_release(calldata
);
1780 static const struct rpc_call_ops nfs_commit_ops
= {
1781 .rpc_call_prepare
= nfs_commit_prepare
,
1782 .rpc_call_done
= nfs_commit_done
,
1783 .rpc_release
= nfs_commit_release
,
1786 static const struct nfs_commit_completion_ops nfs_commit_completion_ops
= {
1787 .completion
= nfs_commit_release_pages
,
1788 .resched_write
= nfs_commit_resched_write
,
1791 int nfs_generic_commit_list(struct inode
*inode
, struct list_head
*head
,
1792 int how
, struct nfs_commit_info
*cinfo
)
1796 status
= pnfs_commit_list(inode
, head
, how
, cinfo
);
1797 if (status
== PNFS_NOT_ATTEMPTED
)
1798 status
= nfs_commit_list(inode
, head
, how
, cinfo
);
1802 int nfs_commit_inode(struct inode
*inode
, int how
)
1805 struct nfs_commit_info cinfo
;
1806 int may_wait
= how
& FLUSH_SYNC
;
1810 nfs_init_cinfo_from_inode(&cinfo
, inode
);
1811 nfs_commit_begin(cinfo
.mds
);
1812 res
= nfs_scan_commit(inode
, &head
, &cinfo
);
1814 error
= nfs_generic_commit_list(inode
, &head
, how
, &cinfo
);
1815 nfs_commit_end(cinfo
.mds
);
1819 goto out_mark_dirty
;
1820 error
= wait_on_commit(cinfo
.mds
);
1826 /* Note: If we exit without ensuring that the commit is complete,
1827 * we must mark the inode as dirty. Otherwise, future calls to
1828 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1829 * that the data is on the disk.
1832 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1835 EXPORT_SYMBOL_GPL(nfs_commit_inode
);
1837 int nfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1839 struct nfs_inode
*nfsi
= NFS_I(inode
);
1840 int flags
= FLUSH_SYNC
;
1843 /* no commits means nothing needs to be done */
1844 if (!nfsi
->commit_info
.ncommit
)
1847 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1848 /* Don't commit yet if this is a non-blocking flush and there
1849 * are a lot of outstanding writes for this mapping.
1851 if (nfsi
->commit_info
.ncommit
<= (nfsi
->nrequests
>> 1))
1852 goto out_mark_dirty
;
1854 /* don't wait for the COMMIT response */
1858 ret
= nfs_commit_inode(inode
, flags
);
1860 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1861 if (ret
< wbc
->nr_to_write
)
1862 wbc
->nr_to_write
-= ret
;
1864 wbc
->nr_to_write
= 0;
1869 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
1872 EXPORT_SYMBOL_GPL(nfs_write_inode
);
1875 * flush the inode to disk.
1877 int nfs_wb_all(struct inode
*inode
)
1881 trace_nfs_writeback_inode_enter(inode
);
1883 ret
= filemap_write_and_wait(inode
->i_mapping
);
1886 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1889 pnfs_sync_inode(inode
, true);
1893 trace_nfs_writeback_inode_exit(inode
, ret
);
1896 EXPORT_SYMBOL_GPL(nfs_wb_all
);
1898 int nfs_wb_page_cancel(struct inode
*inode
, struct page
*page
)
1900 struct nfs_page
*req
;
1903 wait_on_page_writeback(page
);
1905 /* blocking call to cancel all requests and join to a single (head)
1907 req
= nfs_lock_and_join_requests(page
, false);
1912 /* all requests from this page have been cancelled by
1913 * nfs_lock_and_join_requests, so just remove the head
1914 * request from the inode / page_private pointer and
1916 nfs_inode_remove_request(req
);
1917 nfs_unlock_and_release_request(req
);
1924 * Write back all requests on one page - we do this before reading it.
1926 int nfs_wb_page(struct inode
*inode
, struct page
*page
)
1928 loff_t range_start
= page_file_offset(page
);
1929 loff_t range_end
= range_start
+ (loff_t
)(PAGE_CACHE_SIZE
- 1);
1930 struct writeback_control wbc
= {
1931 .sync_mode
= WB_SYNC_ALL
,
1933 .range_start
= range_start
,
1934 .range_end
= range_end
,
1938 trace_nfs_writeback_page_enter(inode
);
1941 wait_on_page_writeback(page
);
1942 if (clear_page_dirty_for_io(page
)) {
1943 ret
= nfs_writepage_locked(page
, &wbc
);
1949 if (!PagePrivate(page
))
1951 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
1956 trace_nfs_writeback_page_exit(inode
, ret
);
1960 #ifdef CONFIG_MIGRATION
1961 int nfs_migrate_page(struct address_space
*mapping
, struct page
*newpage
,
1962 struct page
*page
, enum migrate_mode mode
)
1965 * If PagePrivate is set, then the page is currently associated with
1966 * an in-progress read or write request. Don't try to migrate it.
1968 * FIXME: we could do this in principle, but we'll need a way to ensure
1969 * that we can safely release the inode reference while holding
1972 if (PagePrivate(page
))
1975 if (!nfs_fscache_release_page(page
, GFP_KERNEL
))
1978 return migrate_page(mapping
, newpage
, page
, mode
);
1982 int __init
nfs_init_writepagecache(void)
1984 nfs_wdata_cachep
= kmem_cache_create("nfs_write_data",
1985 sizeof(struct nfs_pgio_header
),
1986 0, SLAB_HWCACHE_ALIGN
,
1988 if (nfs_wdata_cachep
== NULL
)
1991 nfs_wdata_mempool
= mempool_create_slab_pool(MIN_POOL_WRITE
,
1993 if (nfs_wdata_mempool
== NULL
)
1994 goto out_destroy_write_cache
;
1996 nfs_cdata_cachep
= kmem_cache_create("nfs_commit_data",
1997 sizeof(struct nfs_commit_data
),
1998 0, SLAB_HWCACHE_ALIGN
,
2000 if (nfs_cdata_cachep
== NULL
)
2001 goto out_destroy_write_mempool
;
2003 nfs_commit_mempool
= mempool_create_slab_pool(MIN_POOL_COMMIT
,
2005 if (nfs_commit_mempool
== NULL
)
2006 goto out_destroy_commit_cache
;
2009 * NFS congestion size, scale with available memory.
2021 * This allows larger machines to have larger/more transfers.
2022 * Limit the default to 256M
2024 nfs_congestion_kb
= (16*int_sqrt(totalram_pages
)) << (PAGE_SHIFT
-10);
2025 if (nfs_congestion_kb
> 256*1024)
2026 nfs_congestion_kb
= 256*1024;
2030 out_destroy_commit_cache
:
2031 kmem_cache_destroy(nfs_cdata_cachep
);
2032 out_destroy_write_mempool
:
2033 mempool_destroy(nfs_wdata_mempool
);
2034 out_destroy_write_cache
:
2035 kmem_cache_destroy(nfs_wdata_cachep
);
2039 void nfs_destroy_writepagecache(void)
2041 mempool_destroy(nfs_commit_mempool
);
2042 kmem_cache_destroy(nfs_cdata_cachep
);
2043 mempool_destroy(nfs_wdata_mempool
);
2044 kmem_cache_destroy(nfs_wdata_cachep
);
2047 static const struct nfs_rw_ops nfs_rw_write_ops
= {
2048 .rw_mode
= FMODE_WRITE
,
2049 .rw_alloc_header
= nfs_writehdr_alloc
,
2050 .rw_free_header
= nfs_writehdr_free
,
2051 .rw_done
= nfs_writeback_done
,
2052 .rw_result
= nfs_writeback_result
,
2053 .rw_initiate
= nfs_initiate_write
,