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,
53 static struct kmem_cache *nfs_wdata_cachep;
54 static mempool_t *nfs_wdata_mempool;
55 static struct kmem_cache *nfs_cdata_cachep;
56 static mempool_t *nfs_commit_mempool;
58 struct nfs_commit_data *nfs_commitdata_alloc(void)
60 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
63 memset(p, 0, sizeof(*p));
64 INIT_LIST_HEAD(&p->pages);
68 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
70 void nfs_commit_free(struct nfs_commit_data *p)
72 mempool_free(p, nfs_commit_mempool);
74 EXPORT_SYMBOL_GPL(nfs_commit_free);
76 static struct nfs_pgio_header *nfs_writehdr_alloc(void)
78 struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
81 memset(p, 0, sizeof(*p));
85 static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
87 mempool_free(hdr, nfs_wdata_mempool);
90 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
94 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
98 * nfs_page_search_commits_for_head_request_locked
100 * Search through commit lists on @inode for the head request for @page.
101 * Must be called while holding the inode (which is cinfo) lock.
103 * Returns the head request if found, or NULL if not found.
105 static struct nfs_page *
106 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
109 struct nfs_page *freq, *t;
110 struct nfs_commit_info cinfo;
111 struct inode *inode = &nfsi->vfs_inode;
113 nfs_init_cinfo_from_inode(&cinfo, inode);
115 /* search through pnfs commit lists */
116 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
118 return freq->wb_head;
120 /* Linearly search the commit list for the correct request */
121 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
122 if (freq->wb_page == page)
123 return freq->wb_head;
130 * nfs_page_find_head_request_locked - find head request associated with @page
132 * must be called while holding the inode lock.
134 * returns matching head request with reference held, or NULL if not found.
136 static struct nfs_page *
137 nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
139 struct nfs_page *req = NULL;
141 if (PagePrivate(page))
142 req = (struct nfs_page *)page_private(page);
143 else if (unlikely(PageSwapCache(page)))
144 req = nfs_page_search_commits_for_head_request_locked(nfsi,
148 WARN_ON_ONCE(req->wb_head != req);
149 kref_get(&req->wb_kref);
156 * nfs_page_find_head_request - find head request associated with @page
158 * returns matching head request with reference held, or NULL if not found.
160 static struct nfs_page *nfs_page_find_head_request(struct page *page)
162 struct inode *inode = page_file_mapping(page)->host;
163 struct nfs_page *req = NULL;
165 spin_lock(&inode->i_lock);
166 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
167 spin_unlock(&inode->i_lock);
171 /* Adjust the file length if we're writing beyond the end */
172 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
174 struct inode *inode = page_file_mapping(page)->host;
178 spin_lock(&inode->i_lock);
179 i_size = i_size_read(inode);
180 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
181 if (i_size > 0 && page_file_index(page) < end_index)
183 end = page_file_offset(page) + ((loff_t)offset+count);
186 i_size_write(inode, end);
187 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
189 spin_unlock(&inode->i_lock);
192 /* A writeback failed: mark the page as bad, and invalidate the page cache */
193 static void nfs_set_pageerror(struct page *page)
195 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
199 * nfs_page_group_search_locked
200 * @head - head request of page group
201 * @page_offset - offset into page
203 * Search page group with head @head to find a request that contains the
204 * page offset @page_offset.
206 * Returns a pointer to the first matching nfs request, or NULL if no
209 * Must be called with the page group lock held
211 static struct nfs_page *
212 nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
214 struct nfs_page *req;
216 WARN_ON_ONCE(head != head->wb_head);
217 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
221 if (page_offset >= req->wb_pgbase &&
222 page_offset < (req->wb_pgbase + req->wb_bytes))
225 req = req->wb_this_page;
226 } while (req != head);
232 * nfs_page_group_covers_page
233 * @head - head request of page group
235 * Return true if the page group with head @head covers the whole page,
236 * returns false otherwise
238 static bool nfs_page_group_covers_page(struct nfs_page *req)
240 struct nfs_page *tmp;
241 unsigned int pos = 0;
242 unsigned int len = nfs_page_length(req->wb_page);
244 nfs_page_group_lock(req, false);
247 tmp = nfs_page_group_search_locked(req->wb_head, pos);
249 /* no way this should happen */
250 WARN_ON_ONCE(tmp->wb_pgbase != pos);
251 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
253 } while (tmp && pos < len);
255 nfs_page_group_unlock(req);
256 WARN_ON_ONCE(pos > len);
260 /* We can set the PG_uptodate flag if we see that a write request
261 * covers the full page.
263 static void nfs_mark_uptodate(struct nfs_page *req)
265 if (PageUptodate(req->wb_page))
267 if (!nfs_page_group_covers_page(req))
269 SetPageUptodate(req->wb_page);
272 static int wb_priority(struct writeback_control *wbc)
274 if (wbc->for_reclaim)
275 return FLUSH_HIGHPRI | FLUSH_STABLE;
276 if (wbc->for_kupdate || wbc->for_background)
277 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
278 return FLUSH_COND_STABLE;
282 * NFS congestion control
285 int nfs_congestion_kb;
287 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
288 #define NFS_CONGESTION_OFF_THRESH \
289 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
291 static void nfs_set_page_writeback(struct page *page)
293 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
294 int ret = test_set_page_writeback(page);
296 WARN_ON_ONCE(ret != 0);
298 if (atomic_long_inc_return(&nfss->writeback) >
299 NFS_CONGESTION_ON_THRESH) {
300 set_bdi_congested(&nfss->backing_dev_info,
305 static void nfs_end_page_writeback(struct nfs_page *req)
307 struct inode *inode = page_file_mapping(req->wb_page)->host;
308 struct nfs_server *nfss = NFS_SERVER(inode);
310 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
313 end_page_writeback(req->wb_page);
314 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
315 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
319 /* nfs_page_group_clear_bits
320 * @req - an nfs request
321 * clears all page group related bits from @req
324 nfs_page_group_clear_bits(struct nfs_page *req)
326 clear_bit(PG_TEARDOWN, &req->wb_flags);
327 clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
328 clear_bit(PG_UPTODATE, &req->wb_flags);
329 clear_bit(PG_WB_END, &req->wb_flags);
330 clear_bit(PG_REMOVE, &req->wb_flags);
335 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
337 * this is a helper function for nfs_lock_and_join_requests
339 * @inode - inode associated with request page group, must be holding inode lock
340 * @head - head request of page group, must be holding head lock
341 * @req - request that couldn't lock and needs to wait on the req bit lock
342 * @nonblock - if true, don't actually wait
344 * NOTE: this must be called holding page_group bit lock and inode spin lock
345 * and BOTH will be released before returning.
347 * returns 0 on success, < 0 on error.
350 nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
351 struct nfs_page *req, bool nonblock)
352 __releases(&inode->i_lock)
354 struct nfs_page *tmp;
357 /* relinquish all the locks successfully grabbed this run */
358 for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
359 nfs_unlock_request(tmp);
361 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
363 /* grab a ref on the request that will be waited on */
364 kref_get(&req->wb_kref);
366 nfs_page_group_unlock(head);
367 spin_unlock(&inode->i_lock);
369 /* release ref from nfs_page_find_head_request_locked */
370 nfs_release_request(head);
373 ret = nfs_wait_on_request(req);
376 nfs_release_request(req);
382 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
384 * @destroy_list - request list (using wb_this_page) terminated by @old_head
385 * @old_head - the old head of the list
387 * All subrequests must be locked and removed from all lists, so at this point
388 * they are only "active" in this function, and possibly in nfs_wait_on_request
389 * with a reference held by some other context.
392 nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
393 struct nfs_page *old_head)
395 while (destroy_list) {
396 struct nfs_page *subreq = destroy_list;
398 destroy_list = (subreq->wb_this_page == old_head) ?
399 NULL : subreq->wb_this_page;
401 WARN_ON_ONCE(old_head != subreq->wb_head);
403 /* make sure old group is not used */
404 subreq->wb_head = subreq;
405 subreq->wb_this_page = subreq;
407 /* subreq is now totally disconnected from page group or any
408 * write / commit lists. last chance to wake any waiters */
409 nfs_unlock_request(subreq);
411 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
412 /* release ref on old head request */
413 nfs_release_request(old_head);
415 nfs_page_group_clear_bits(subreq);
417 /* release the PG_INODE_REF reference */
418 if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
419 nfs_release_request(subreq);
423 WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
424 /* zombie requests have already released the last
425 * reference and were waiting on the rest of the
426 * group to complete. Since it's no longer part of a
427 * group, simply free the request */
428 nfs_page_group_clear_bits(subreq);
429 nfs_free_request(subreq);
435 * nfs_lock_and_join_requests - join all subreqs to the head req and return
436 * a locked reference, cancelling any pending
437 * operations for this page.
439 * @page - the page used to lookup the "page group" of nfs_page structures
440 * @nonblock - if true, don't block waiting for request locks
442 * This function joins all sub requests to the head request by first
443 * locking all requests in the group, cancelling any pending operations
444 * and finally updating the head request to cover the whole range covered by
445 * the (former) group. All subrequests are removed from any write or commit
446 * lists, unlinked from the group and destroyed.
448 * Returns a locked, referenced pointer to the head request - which after
449 * this call is guaranteed to be the only request associated with the page.
450 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
451 * error was encountered.
453 static struct nfs_page *
454 nfs_lock_and_join_requests(struct page *page, bool nonblock)
456 struct inode *inode = page_file_mapping(page)->host;
457 struct nfs_page *head, *subreq;
458 struct nfs_page *destroy_list = NULL;
459 unsigned int total_bytes;
465 WARN_ON_ONCE(destroy_list);
467 spin_lock(&inode->i_lock);
470 * A reference is taken only on the head request which acts as a
471 * reference to the whole page group - the group will not be destroyed
472 * until the head reference is released.
474 head = nfs_page_find_head_request_locked(NFS_I(inode), page);
477 spin_unlock(&inode->i_lock);
481 /* lock each request in the page group */
482 ret = nfs_page_group_lock(head, true);
484 spin_unlock(&inode->i_lock);
485 nfs_release_request(head);
491 * Subrequests are always contiguous, non overlapping
492 * and in order. If not, it's a programming error.
494 WARN_ON_ONCE(subreq->wb_offset !=
495 (head->wb_offset + total_bytes));
497 /* keep track of how many bytes this group covers */
498 total_bytes += subreq->wb_bytes;
500 if (!nfs_lock_request(subreq)) {
501 /* releases page group bit lock and
502 * inode spin lock and all references */
503 ret = nfs_unroll_locks_and_wait(inode, head,
512 subreq = subreq->wb_this_page;
513 } while (subreq != head);
515 /* Now that all requests are locked, make sure they aren't on any list.
516 * Commit list removal accounting is done after locks are dropped */
519 nfs_clear_request_commit(subreq);
520 subreq = subreq->wb_this_page;
521 } while (subreq != head);
523 /* unlink subrequests from head, destroy them later */
524 if (head->wb_this_page != head) {
525 /* destroy list will be terminated by head */
526 destroy_list = head->wb_this_page;
527 head->wb_this_page = head;
529 /* change head request to cover whole range that
530 * the former page group covered */
531 head->wb_bytes = total_bytes;
535 * prepare head request to be added to new pgio descriptor
537 nfs_page_group_clear_bits(head);
540 * some part of the group was still on the inode list - otherwise
541 * the group wouldn't be involved in async write.
542 * grab a reference for the head request, iff it needs one.
544 if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
545 kref_get(&head->wb_kref);
547 nfs_page_group_unlock(head);
549 /* drop lock to clean uprequests on destroy list */
550 spin_unlock(&inode->i_lock);
552 nfs_destroy_unlinked_subrequests(destroy_list, head);
554 /* still holds ref on head from nfs_page_find_head_request_locked
555 * and still has lock on head from lock loop */
560 * Find an associated nfs write request, and prepare to flush it out
561 * May return an error if the user signalled nfs_wait_on_request().
563 static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
564 struct page *page, bool nonblock)
566 struct nfs_page *req;
569 req = nfs_lock_and_join_requests(page, nonblock);
576 nfs_set_page_writeback(page);
577 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
580 if (!nfs_pageio_add_request(pgio, req)) {
581 nfs_redirty_request(req);
582 ret = pgio->pg_error;
588 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
590 struct inode *inode = page_file_mapping(page)->host;
593 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
594 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
596 nfs_pageio_cond_complete(pgio, page_file_index(page));
597 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
598 if (ret == -EAGAIN) {
599 redirty_page_for_writepage(wbc, page);
606 * Write an mmapped page to the server.
608 static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
610 struct nfs_pageio_descriptor pgio;
613 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
614 false, &nfs_async_write_completion_ops);
615 err = nfs_do_writepage(page, wbc, &pgio);
616 nfs_pageio_complete(&pgio);
619 if (pgio.pg_error < 0)
620 return pgio.pg_error;
624 int nfs_writepage(struct page *page, struct writeback_control *wbc)
628 ret = nfs_writepage_locked(page, wbc);
633 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
637 ret = nfs_do_writepage(page, wbc, data);
642 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
644 struct inode *inode = mapping->host;
645 unsigned long *bitlock = &NFS_I(inode)->flags;
646 struct nfs_pageio_descriptor pgio;
649 /* Stop dirtying of new pages while we sync */
650 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
651 nfs_wait_bit_killable, TASK_KILLABLE);
655 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
657 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
658 &nfs_async_write_completion_ops);
659 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
660 nfs_pageio_complete(&pgio);
662 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
663 smp_mb__after_atomic();
664 wake_up_bit(bitlock, NFS_INO_FLUSHING);
677 * Insert a write request into an inode
679 static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
681 struct nfs_inode *nfsi = NFS_I(inode);
683 WARN_ON_ONCE(req->wb_this_page != req);
685 /* Lock the request! */
686 nfs_lock_request(req);
688 spin_lock(&inode->i_lock);
689 if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
692 * Swap-space should not get truncated. Hence no need to plug the race
693 * with invalidate/truncate.
695 if (likely(!PageSwapCache(req->wb_page))) {
696 set_bit(PG_MAPPED, &req->wb_flags);
697 SetPagePrivate(req->wb_page);
698 set_page_private(req->wb_page, (unsigned long)req);
701 /* this a head request for a page group - mark it as having an
702 * extra reference so sub groups can follow suit */
703 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
704 kref_get(&req->wb_kref);
705 spin_unlock(&inode->i_lock);
709 * Remove a write request from an inode
711 static void nfs_inode_remove_request(struct nfs_page *req)
713 struct inode *inode = req->wb_context->dentry->d_inode;
714 struct nfs_inode *nfsi = NFS_I(inode);
715 struct nfs_page *head;
717 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
720 spin_lock(&inode->i_lock);
721 if (likely(!PageSwapCache(head->wb_page))) {
722 set_page_private(head->wb_page, 0);
723 ClearPagePrivate(head->wb_page);
724 clear_bit(PG_MAPPED, &head->wb_flags);
727 spin_unlock(&inode->i_lock);
730 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
731 nfs_release_request(req);
737 nfs_mark_request_dirty(struct nfs_page *req)
739 __set_page_dirty_nobuffers(req->wb_page);
742 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
744 * nfs_request_add_commit_list - add request to a commit list
745 * @req: pointer to a struct nfs_page
746 * @dst: commit list head
747 * @cinfo: holds list lock and accounting info
749 * This sets the PG_CLEAN bit, updates the cinfo count of
750 * number of outstanding requests requiring a commit as well as
753 * The caller must _not_ hold the cinfo->lock, but must be
754 * holding the nfs_page lock.
757 nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
758 struct nfs_commit_info *cinfo)
760 set_bit(PG_CLEAN, &(req)->wb_flags);
761 spin_lock(cinfo->lock);
762 nfs_list_add_request(req, dst);
763 cinfo->mds->ncommit++;
764 spin_unlock(cinfo->lock);
766 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
767 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
769 __mark_inode_dirty(req->wb_context->dentry->d_inode,
773 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
776 * nfs_request_remove_commit_list - Remove request from a commit list
777 * @req: pointer to a nfs_page
778 * @cinfo: holds list lock and accounting info
780 * This clears the PG_CLEAN bit, and updates the cinfo's count of
781 * number of outstanding requests requiring a commit
782 * It does not update the MM page stats.
784 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
787 nfs_request_remove_commit_list(struct nfs_page *req,
788 struct nfs_commit_info *cinfo)
790 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
792 nfs_list_remove_request(req);
793 cinfo->mds->ncommit--;
795 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
797 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
800 cinfo->lock = &inode->i_lock;
801 cinfo->mds = &NFS_I(inode)->commit_info;
802 cinfo->ds = pnfs_get_ds_info(inode);
804 cinfo->completion_ops = &nfs_commit_completion_ops;
807 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
809 struct nfs_direct_req *dreq)
812 nfs_init_cinfo_from_dreq(cinfo, dreq);
814 nfs_init_cinfo_from_inode(cinfo, inode);
816 EXPORT_SYMBOL_GPL(nfs_init_cinfo);
819 * Add a request to the inode's commit list.
822 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
823 struct nfs_commit_info *cinfo)
825 if (pnfs_mark_request_commit(req, lseg, cinfo))
827 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
831 nfs_clear_page_commit(struct page *page)
833 dec_zone_page_state(page, NR_UNSTABLE_NFS);
834 dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
837 /* Called holding inode (/cinfo) lock */
839 nfs_clear_request_commit(struct nfs_page *req)
841 if (test_bit(PG_CLEAN, &req->wb_flags)) {
842 struct inode *inode = req->wb_context->dentry->d_inode;
843 struct nfs_commit_info cinfo;
845 nfs_init_cinfo_from_inode(&cinfo, inode);
846 if (!pnfs_clear_request_commit(req, &cinfo)) {
847 nfs_request_remove_commit_list(req, &cinfo);
849 nfs_clear_page_commit(req->wb_page);
853 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
855 if (hdr->verf.committed == NFS_DATA_SYNC)
856 return hdr->lseg == NULL;
857 return hdr->verf.committed != NFS_FILE_SYNC;
861 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
866 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
868 struct nfs_direct_req *dreq)
873 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
874 struct nfs_commit_info *cinfo)
879 nfs_clear_request_commit(struct nfs_page *req)
883 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
890 static void nfs_write_completion(struct nfs_pgio_header *hdr)
892 struct nfs_commit_info cinfo;
893 unsigned long bytes = 0;
895 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
897 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
898 while (!list_empty(&hdr->pages)) {
899 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
901 bytes += req->wb_bytes;
902 nfs_list_remove_request(req);
903 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
904 (hdr->good_bytes < bytes)) {
905 nfs_set_pageerror(req->wb_page);
906 nfs_context_set_write_error(req->wb_context, hdr->error);
909 if (nfs_write_need_commit(hdr)) {
910 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
911 nfs_mark_request_commit(req, hdr->lseg, &cinfo);
915 nfs_inode_remove_request(req);
917 nfs_unlock_request(req);
918 nfs_end_page_writeback(req);
919 nfs_release_request(req);
925 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
927 nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
929 return cinfo->mds->ncommit;
932 /* cinfo->lock held by caller */
934 nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
935 struct nfs_commit_info *cinfo, int max)
937 struct nfs_page *req, *tmp;
940 list_for_each_entry_safe(req, tmp, src, wb_list) {
941 if (!nfs_lock_request(req))
943 kref_get(&req->wb_kref);
944 if (cond_resched_lock(cinfo->lock))
945 list_safe_reset_next(req, tmp, wb_list);
946 nfs_request_remove_commit_list(req, cinfo);
947 nfs_list_add_request(req, dst);
949 if ((ret == max) && !cinfo->dreq)
956 * nfs_scan_commit - Scan an inode for commit requests
957 * @inode: NFS inode to scan
958 * @dst: mds destination list
959 * @cinfo: mds and ds lists of reqs ready to commit
961 * Moves requests from the inode's 'commit' request list.
962 * The requests are *not* checked to ensure that they form a contiguous set.
965 nfs_scan_commit(struct inode *inode, struct list_head *dst,
966 struct nfs_commit_info *cinfo)
970 spin_lock(cinfo->lock);
971 if (cinfo->mds->ncommit > 0) {
972 const int max = INT_MAX;
974 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
976 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
978 spin_unlock(cinfo->lock);
983 unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
988 int nfs_scan_commit(struct inode *inode, struct list_head *dst,
989 struct nfs_commit_info *cinfo)
996 * Search for an existing write request, and attempt to update
997 * it to reflect a new dirty region on a given page.
999 * If the attempt fails, then the existing request is flushed out
1002 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1004 unsigned int offset,
1007 struct nfs_page *req;
1012 if (!PagePrivate(page))
1015 end = offset + bytes;
1016 spin_lock(&inode->i_lock);
1019 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
1023 /* should be handled by nfs_flush_incompatible */
1024 WARN_ON_ONCE(req->wb_head != req);
1025 WARN_ON_ONCE(req->wb_this_page != req);
1027 rqend = req->wb_offset + req->wb_bytes;
1029 * Tell the caller to flush out the request if
1030 * the offsets are non-contiguous.
1031 * Note: nfs_flush_incompatible() will already
1032 * have flushed out requests having wrong owners.
1035 || end < req->wb_offset)
1038 if (nfs_lock_request(req))
1041 /* The request is locked, so wait and then retry */
1042 spin_unlock(&inode->i_lock);
1043 error = nfs_wait_on_request(req);
1044 nfs_release_request(req);
1047 spin_lock(&inode->i_lock);
1050 /* Okay, the request matches. Update the region */
1051 if (offset < req->wb_offset) {
1052 req->wb_offset = offset;
1053 req->wb_pgbase = offset;
1056 req->wb_bytes = end - req->wb_offset;
1058 req->wb_bytes = rqend - req->wb_offset;
1061 nfs_clear_request_commit(req);
1062 spin_unlock(&inode->i_lock);
1065 spin_unlock(&inode->i_lock);
1066 nfs_release_request(req);
1067 error = nfs_wb_page(inode, page);
1069 return ERR_PTR(error);
1073 * Try to update an existing write request, or create one if there is none.
1075 * Note: Should always be called with the Page Lock held to prevent races
1076 * if we have to add a new request. Also assumes that the caller has
1077 * already called nfs_flush_incompatible() if necessary.
1079 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1080 struct page *page, unsigned int offset, unsigned int bytes)
1082 struct inode *inode = page_file_mapping(page)->host;
1083 struct nfs_page *req;
1085 req = nfs_try_to_update_request(inode, page, offset, bytes);
1088 req = nfs_create_request(ctx, page, NULL, offset, bytes);
1091 nfs_inode_add_request(inode, req);
1096 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1097 unsigned int offset, unsigned int count)
1099 struct nfs_page *req;
1101 req = nfs_setup_write_request(ctx, page, offset, count);
1103 return PTR_ERR(req);
1104 /* Update file length */
1105 nfs_grow_file(page, offset, count);
1106 nfs_mark_uptodate(req);
1107 nfs_mark_request_dirty(req);
1108 nfs_unlock_and_release_request(req);
1112 int nfs_flush_incompatible(struct file *file, struct page *page)
1114 struct nfs_open_context *ctx = nfs_file_open_context(file);
1115 struct nfs_lock_context *l_ctx;
1116 struct nfs_page *req;
1117 int do_flush, status;
1119 * Look for a request corresponding to this page. If there
1120 * is one, and it belongs to another file, we flush it out
1121 * before we try to copy anything into the page. Do this
1122 * due to the lack of an ACCESS-type call in NFSv2.
1123 * Also do the same if we find a request from an existing
1127 req = nfs_page_find_head_request(page);
1130 l_ctx = req->wb_lock_context;
1131 do_flush = req->wb_page != page || req->wb_context != ctx;
1132 /* for now, flush if more than 1 request in page_group */
1133 do_flush |= req->wb_this_page != req;
1134 if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
1135 do_flush |= l_ctx->lockowner.l_owner != current->files
1136 || l_ctx->lockowner.l_pid != current->tgid;
1138 nfs_release_request(req);
1141 status = nfs_wb_page(page_file_mapping(page)->host, page);
1142 } while (status == 0);
1147 * Avoid buffered writes when a open context credential's key would
1150 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1152 * Return 0 and set a credential flag which triggers the inode to flush
1153 * and performs NFS_FILE_SYNC writes if the key will expired within
1154 * RPC_KEY_EXPIRE_TIMEO.
1157 nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1159 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1160 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1162 return rpcauth_key_timeout_notify(auth, ctx->cred);
1166 * Test if the open context credential key is marked to expire soon.
1168 bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1170 return rpcauth_cred_key_to_expire(ctx->cred);
1174 * If the page cache is marked as unsafe or invalid, then we can't rely on
1175 * the PageUptodate() flag. In this case, we will need to turn off
1176 * write optimisations that depend on the page contents being correct.
1178 static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
1180 struct nfs_inode *nfsi = NFS_I(inode);
1182 if (nfs_have_delegated_attributes(inode))
1184 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1187 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
1190 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1192 return PageUptodate(page) != 0;
1195 /* If we know the page is up to date, and we're not using byte range locks (or
1196 * if we have the whole file locked for writing), it may be more efficient to
1197 * extend the write to cover the entire page in order to avoid fragmentation
1200 * If the file is opened for synchronous writes then we can just skip the rest
1203 static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1205 if (file->f_flags & O_DSYNC)
1207 if (!nfs_write_pageuptodate(page, inode))
1209 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1211 if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
1212 inode->i_flock->fl_end == OFFSET_MAX &&
1213 inode->i_flock->fl_type != F_RDLCK))
1219 * Update and possibly write a cached page of an NFS file.
1221 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1222 * things with a page scheduled for an RPC call (e.g. invalidate it).
1224 int nfs_updatepage(struct file *file, struct page *page,
1225 unsigned int offset, unsigned int count)
1227 struct nfs_open_context *ctx = nfs_file_open_context(file);
1228 struct inode *inode = page_file_mapping(page)->host;
1231 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1233 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1234 file, count, (long long)(page_file_offset(page) + offset));
1236 if (nfs_can_extend_write(file, page, inode)) {
1237 count = max(count + offset, nfs_page_length(page));
1241 status = nfs_writepage_setup(ctx, page, offset, count);
1243 nfs_set_pageerror(page);
1245 __set_page_dirty_nobuffers(page);
1247 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1248 status, (long long)i_size_read(inode));
1252 static int flush_task_priority(int how)
1254 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1256 return RPC_PRIORITY_HIGH;
1258 return RPC_PRIORITY_LOW;
1260 return RPC_PRIORITY_NORMAL;
1263 static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1264 struct rpc_message *msg,
1265 struct rpc_task_setup *task_setup_data, int how)
1267 struct inode *inode = hdr->inode;
1268 int priority = flush_task_priority(how);
1270 task_setup_data->priority = priority;
1271 NFS_PROTO(inode)->write_setup(hdr, msg);
1273 nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
1274 &task_setup_data->rpc_client, msg, hdr);
1277 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1278 * call this on each, which will prepare them to be retried on next
1279 * writeback using standard nfs.
1281 static void nfs_redirty_request(struct nfs_page *req)
1283 nfs_mark_request_dirty(req);
1284 nfs_unlock_request(req);
1285 nfs_end_page_writeback(req);
1286 nfs_release_request(req);
1289 static void nfs_async_write_error(struct list_head *head)
1291 struct nfs_page *req;
1293 while (!list_empty(head)) {
1294 req = nfs_list_entry(head->next);
1295 nfs_list_remove_request(req);
1296 nfs_redirty_request(req);
1300 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1301 .error_cleanup = nfs_async_write_error,
1302 .completion = nfs_write_completion,
1305 void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1306 struct inode *inode, int ioflags, bool force_mds,
1307 const struct nfs_pgio_completion_ops *compl_ops)
1309 struct nfs_server *server = NFS_SERVER(inode);
1310 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1312 #ifdef CONFIG_NFS_V4_1
1313 if (server->pnfs_curr_ld && !force_mds)
1314 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1316 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1317 server->wsize, ioflags);
1319 EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1321 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1323 pgio->pg_ops = &nfs_pgio_rw_ops;
1324 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1326 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1329 void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1331 struct nfs_commit_data *data = calldata;
1333 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1336 static void nfs_writeback_release_common(struct nfs_pgio_header *hdr)
1342 * Special version of should_remove_suid() that ignores capabilities.
1344 static int nfs_should_remove_suid(const struct inode *inode)
1346 umode_t mode = inode->i_mode;
1349 /* suid always must be killed */
1350 if (unlikely(mode & S_ISUID))
1351 kill = ATTR_KILL_SUID;
1354 * sgid without any exec bits is just a mandatory locking mark; leave
1355 * it alone. If some exec bits are set, it's a real sgid; kill it.
1357 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1358 kill |= ATTR_KILL_SGID;
1360 if (unlikely(kill && S_ISREG(mode)))
1367 * This function is called when the WRITE call is complete.
1369 static int nfs_writeback_done(struct rpc_task *task,
1370 struct nfs_pgio_header *hdr,
1371 struct inode *inode)
1376 * ->write_done will attempt to use post-op attributes to detect
1377 * conflicting writes by other clients. A strict interpretation
1378 * of close-to-open would allow us to continue caching even if
1379 * another writer had changed the file, but some applications
1380 * depend on tighter cache coherency when writing.
1382 status = NFS_PROTO(inode)->write_done(task, hdr);
1385 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1387 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1388 if (hdr->res.verf->committed < hdr->args.stable &&
1389 task->tk_status >= 0) {
1390 /* We tried a write call, but the server did not
1391 * commit data to stable storage even though we
1393 * Note: There is a known bug in Tru64 < 5.0 in which
1394 * the server reports NFS_DATA_SYNC, but performs
1395 * NFS_FILE_SYNC. We therefore implement this checking
1396 * as a dprintk() in order to avoid filling syslog.
1398 static unsigned long complain;
1400 /* Note this will print the MDS for a DS write */
1401 if (time_before(complain, jiffies)) {
1402 dprintk("NFS: faulty NFS server %s:"
1403 " (committed = %d) != (stable = %d)\n",
1404 NFS_SERVER(inode)->nfs_client->cl_hostname,
1405 hdr->res.verf->committed, hdr->args.stable);
1406 complain = jiffies + 300 * HZ;
1411 /* Deal with the suid/sgid bit corner case */
1412 if (nfs_should_remove_suid(inode))
1413 nfs_mark_for_revalidate(inode);
1418 * This function is called when the WRITE call is complete.
1420 static void nfs_writeback_result(struct rpc_task *task,
1421 struct nfs_pgio_header *hdr)
1423 struct nfs_pgio_args *argp = &hdr->args;
1424 struct nfs_pgio_res *resp = &hdr->res;
1426 if (resp->count < argp->count) {
1427 static unsigned long complain;
1429 /* This a short write! */
1430 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1432 /* Has the server at least made some progress? */
1433 if (resp->count == 0) {
1434 if (time_before(complain, jiffies)) {
1436 "NFS: Server wrote zero bytes, expected %u.\n",
1438 complain = jiffies + 300 * HZ;
1440 nfs_set_pgio_error(hdr, -EIO, argp->offset);
1441 task->tk_status = -EIO;
1444 /* Was this an NFSv2 write or an NFSv3 stable write? */
1445 if (resp->verf->committed != NFS_UNSTABLE) {
1446 /* Resend from where the server left off */
1447 hdr->mds_offset += resp->count;
1448 argp->offset += resp->count;
1449 argp->pgbase += resp->count;
1450 argp->count -= resp->count;
1452 /* Resend as a stable write in order to avoid
1453 * headaches in the case of a server crash.
1455 argp->stable = NFS_FILE_SYNC;
1457 rpc_restart_call_prepare(task);
1462 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1463 static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1467 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1471 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1473 nfs_wait_bit_killable,
1475 return (ret < 0) ? ret : 1;
1478 static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1480 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1481 smp_mb__after_atomic();
1482 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1485 void nfs_commitdata_release(struct nfs_commit_data *data)
1487 put_nfs_open_context(data->context);
1488 nfs_commit_free(data);
1490 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1492 int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1493 const struct rpc_call_ops *call_ops,
1496 struct rpc_task *task;
1497 int priority = flush_task_priority(how);
1498 struct rpc_message msg = {
1499 .rpc_argp = &data->args,
1500 .rpc_resp = &data->res,
1501 .rpc_cred = data->cred,
1503 struct rpc_task_setup task_setup_data = {
1504 .task = &data->task,
1506 .rpc_message = &msg,
1507 .callback_ops = call_ops,
1508 .callback_data = data,
1509 .workqueue = nfsiod_workqueue,
1510 .flags = RPC_TASK_ASYNC | flags,
1511 .priority = priority,
1513 /* Set up the initial task struct. */
1514 NFS_PROTO(data->inode)->commit_setup(data, &msg);
1516 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1518 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1519 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1521 task = rpc_run_task(&task_setup_data);
1523 return PTR_ERR(task);
1524 if (how & FLUSH_SYNC)
1525 rpc_wait_for_completion_task(task);
1529 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1532 * Set up the argument/result storage required for the RPC call.
1534 void nfs_init_commit(struct nfs_commit_data *data,
1535 struct list_head *head,
1536 struct pnfs_layout_segment *lseg,
1537 struct nfs_commit_info *cinfo)
1539 struct nfs_page *first = nfs_list_entry(head->next);
1540 struct inode *inode = first->wb_context->dentry->d_inode;
1542 /* Set up the RPC argument and reply structs
1543 * NB: take care not to mess about with data->commit et al. */
1545 list_splice_init(head, &data->pages);
1547 data->inode = inode;
1548 data->cred = first->wb_context->cred;
1549 data->lseg = lseg; /* reference transferred */
1550 data->mds_ops = &nfs_commit_ops;
1551 data->completion_ops = cinfo->completion_ops;
1552 data->dreq = cinfo->dreq;
1554 data->args.fh = NFS_FH(data->inode);
1555 /* Note: we always request a commit of the entire inode */
1556 data->args.offset = 0;
1557 data->args.count = 0;
1558 data->context = get_nfs_open_context(first->wb_context);
1559 data->res.fattr = &data->fattr;
1560 data->res.verf = &data->verf;
1561 nfs_fattr_init(&data->fattr);
1563 EXPORT_SYMBOL_GPL(nfs_init_commit);
1565 void nfs_retry_commit(struct list_head *page_list,
1566 struct pnfs_layout_segment *lseg,
1567 struct nfs_commit_info *cinfo)
1569 struct nfs_page *req;
1571 while (!list_empty(page_list)) {
1572 req = nfs_list_entry(page_list->next);
1573 nfs_list_remove_request(req);
1574 nfs_mark_request_commit(req, lseg, cinfo);
1576 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1577 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1580 nfs_unlock_and_release_request(req);
1583 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1586 * Commit dirty pages
1589 nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1590 struct nfs_commit_info *cinfo)
1592 struct nfs_commit_data *data;
1594 data = nfs_commitdata_alloc();
1599 /* Set up the argument struct */
1600 nfs_init_commit(data, head, NULL, cinfo);
1601 atomic_inc(&cinfo->mds->rpcs_out);
1602 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
1605 nfs_retry_commit(head, NULL, cinfo);
1606 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1611 * COMMIT call returned
1613 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1615 struct nfs_commit_data *data = calldata;
1617 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1618 task->tk_pid, task->tk_status);
1620 /* Call the NFS version-specific code */
1621 NFS_PROTO(data->inode)->commit_done(task, data);
1624 static void nfs_commit_release_pages(struct nfs_commit_data *data)
1626 struct nfs_page *req;
1627 int status = data->task.tk_status;
1628 struct nfs_commit_info cinfo;
1630 while (!list_empty(&data->pages)) {
1631 req = nfs_list_entry(data->pages.next);
1632 nfs_list_remove_request(req);
1633 nfs_clear_page_commit(req->wb_page);
1635 dprintk("NFS: commit (%s/%llu %d@%lld)",
1636 req->wb_context->dentry->d_sb->s_id,
1637 (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
1639 (long long)req_offset(req));
1641 nfs_context_set_write_error(req->wb_context, status);
1642 nfs_inode_remove_request(req);
1643 dprintk(", error = %d\n", status);
1647 /* Okay, COMMIT succeeded, apparently. Check the verifier
1648 * returned by the server against all stored verfs. */
1649 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
1650 /* We have a match */
1651 nfs_inode_remove_request(req);
1655 /* We have a mismatch. Write the page again */
1656 dprintk(" mismatch\n");
1657 nfs_mark_request_dirty(req);
1658 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1660 nfs_unlock_and_release_request(req);
1662 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1663 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1664 nfs_commit_clear_lock(NFS_I(data->inode));
1667 static void nfs_commit_release(void *calldata)
1669 struct nfs_commit_data *data = calldata;
1671 data->completion_ops->completion(data);
1672 nfs_commitdata_release(calldata);
1675 static const struct rpc_call_ops nfs_commit_ops = {
1676 .rpc_call_prepare = nfs_commit_prepare,
1677 .rpc_call_done = nfs_commit_done,
1678 .rpc_release = nfs_commit_release,
1681 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1682 .completion = nfs_commit_release_pages,
1683 .error_cleanup = nfs_commit_clear_lock,
1686 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1687 int how, struct nfs_commit_info *cinfo)
1691 status = pnfs_commit_list(inode, head, how, cinfo);
1692 if (status == PNFS_NOT_ATTEMPTED)
1693 status = nfs_commit_list(inode, head, how, cinfo);
1697 int nfs_commit_inode(struct inode *inode, int how)
1700 struct nfs_commit_info cinfo;
1701 int may_wait = how & FLUSH_SYNC;
1704 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1706 goto out_mark_dirty;
1707 nfs_init_cinfo_from_inode(&cinfo, inode);
1708 res = nfs_scan_commit(inode, &head, &cinfo);
1712 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1716 goto out_mark_dirty;
1717 error = wait_on_bit_action(&NFS_I(inode)->flags,
1719 nfs_wait_bit_killable,
1724 nfs_commit_clear_lock(NFS_I(inode));
1726 /* Note: If we exit without ensuring that the commit is complete,
1727 * we must mark the inode as dirty. Otherwise, future calls to
1728 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1729 * that the data is on the disk.
1732 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1736 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1738 struct nfs_inode *nfsi = NFS_I(inode);
1739 int flags = FLUSH_SYNC;
1742 /* no commits means nothing needs to be done */
1743 if (!nfsi->commit_info.ncommit)
1746 if (wbc->sync_mode == WB_SYNC_NONE) {
1747 /* Don't commit yet if this is a non-blocking flush and there
1748 * are a lot of outstanding writes for this mapping.
1750 if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
1751 goto out_mark_dirty;
1753 /* don't wait for the COMMIT response */
1757 ret = nfs_commit_inode(inode, flags);
1759 if (wbc->sync_mode == WB_SYNC_NONE) {
1760 if (ret < wbc->nr_to_write)
1761 wbc->nr_to_write -= ret;
1763 wbc->nr_to_write = 0;
1768 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1772 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1778 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1780 return nfs_commit_unstable_pages(inode, wbc);
1782 EXPORT_SYMBOL_GPL(nfs_write_inode);
1785 * flush the inode to disk.
1787 int nfs_wb_all(struct inode *inode)
1789 struct writeback_control wbc = {
1790 .sync_mode = WB_SYNC_ALL,
1791 .nr_to_write = LONG_MAX,
1793 .range_end = LLONG_MAX,
1797 trace_nfs_writeback_inode_enter(inode);
1799 ret = sync_inode(inode, &wbc);
1801 trace_nfs_writeback_inode_exit(inode, ret);
1804 EXPORT_SYMBOL_GPL(nfs_wb_all);
1806 int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1808 struct nfs_page *req;
1811 wait_on_page_writeback(page);
1813 /* blocking call to cancel all requests and join to a single (head)
1815 req = nfs_lock_and_join_requests(page, false);
1820 /* all requests from this page have been cancelled by
1821 * nfs_lock_and_join_requests, so just remove the head
1822 * request from the inode / page_private pointer and
1824 nfs_inode_remove_request(req);
1826 * In case nfs_inode_remove_request has marked the
1827 * page as being dirty
1829 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1830 nfs_unlock_and_release_request(req);
1837 * Write back all requests on one page - we do this before reading it.
1839 int nfs_wb_page(struct inode *inode, struct page *page)
1841 loff_t range_start = page_file_offset(page);
1842 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1843 struct writeback_control wbc = {
1844 .sync_mode = WB_SYNC_ALL,
1846 .range_start = range_start,
1847 .range_end = range_end,
1851 trace_nfs_writeback_page_enter(inode);
1854 wait_on_page_writeback(page);
1855 if (clear_page_dirty_for_io(page)) {
1856 ret = nfs_writepage_locked(page, &wbc);
1862 if (!PagePrivate(page))
1864 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1869 trace_nfs_writeback_page_exit(inode, ret);
1873 #ifdef CONFIG_MIGRATION
1874 int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1875 struct page *page, enum migrate_mode mode)
1878 * If PagePrivate is set, then the page is currently associated with
1879 * an in-progress read or write request. Don't try to migrate it.
1881 * FIXME: we could do this in principle, but we'll need a way to ensure
1882 * that we can safely release the inode reference while holding
1885 if (PagePrivate(page))
1888 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1891 return migrate_page(mapping, newpage, page, mode);
1895 int __init nfs_init_writepagecache(void)
1897 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1898 sizeof(struct nfs_pgio_header),
1899 0, SLAB_HWCACHE_ALIGN,
1901 if (nfs_wdata_cachep == NULL)
1904 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1906 if (nfs_wdata_mempool == NULL)
1907 goto out_destroy_write_cache;
1909 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1910 sizeof(struct nfs_commit_data),
1911 0, SLAB_HWCACHE_ALIGN,
1913 if (nfs_cdata_cachep == NULL)
1914 goto out_destroy_write_mempool;
1916 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1918 if (nfs_commit_mempool == NULL)
1919 goto out_destroy_commit_cache;
1922 * NFS congestion size, scale with available memory.
1934 * This allows larger machines to have larger/more transfers.
1935 * Limit the default to 256M
1937 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1938 if (nfs_congestion_kb > 256*1024)
1939 nfs_congestion_kb = 256*1024;
1943 out_destroy_commit_cache:
1944 kmem_cache_destroy(nfs_cdata_cachep);
1945 out_destroy_write_mempool:
1946 mempool_destroy(nfs_wdata_mempool);
1947 out_destroy_write_cache:
1948 kmem_cache_destroy(nfs_wdata_cachep);
1952 void nfs_destroy_writepagecache(void)
1954 mempool_destroy(nfs_commit_mempool);
1955 kmem_cache_destroy(nfs_cdata_cachep);
1956 mempool_destroy(nfs_wdata_mempool);
1957 kmem_cache_destroy(nfs_wdata_cachep);
1960 static const struct nfs_rw_ops nfs_rw_write_ops = {
1961 .rw_mode = FMODE_WRITE,
1962 .rw_alloc_header = nfs_writehdr_alloc,
1963 .rw_free_header = nfs_writehdr_free,
1964 .rw_release = nfs_writeback_release_common,
1965 .rw_done = nfs_writeback_done,
1966 .rw_result = nfs_writeback_result,
1967 .rw_initiate = nfs_initiate_write,