4 * Write file data over NFS.
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
9 #include <linux/types.h>
10 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/file.h>
14 #include <linux/writeback.h>
15 #include <linux/swap.h>
16 #include <linux/migrate.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_mount.h>
21 #include <linux/nfs_page.h>
22 #include <linux/backing-dev.h>
23 #include <linux/export.h>
25 #include <asm/uaccess.h>
27 #include "delegation.h"
36 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
38 #define MIN_POOL_WRITE (32)
39 #define MIN_POOL_COMMIT (4)
42 * Local function declarations
44 static void nfs_redirty_request(struct nfs_page *req);
45 static const struct rpc_call_ops nfs_commit_ops;
46 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
47 static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
48 static const struct nfs_rw_ops nfs_rw_write_ops;
49 static void nfs_clear_request_commit(struct nfs_page *req);
50 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
52 static struct nfs_page *
53 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
56 static struct kmem_cache *nfs_wdata_cachep;
57 static mempool_t *nfs_wdata_mempool;
58 static struct kmem_cache *nfs_cdata_cachep;
59 static mempool_t *nfs_commit_mempool;
61 struct nfs_commit_data *nfs_commitdata_alloc(void)
63 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
66 memset(p, 0, sizeof(*p));
67 INIT_LIST_HEAD(&p->pages);
71 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
73 void nfs_commit_free(struct nfs_commit_data *p)
75 mempool_free(p, nfs_commit_mempool);
77 EXPORT_SYMBOL_GPL(nfs_commit_free);
79 static struct nfs_pgio_header *nfs_writehdr_alloc(void)
81 struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
84 memset(p, 0, sizeof(*p));
88 static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
90 mempool_free(hdr, nfs_wdata_mempool);
93 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
97 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
101 * nfs_page_find_head_request_locked - find head request associated with @page
103 * must be called while holding the inode lock.
105 * returns matching head request with reference held, or NULL if not found.
107 static struct nfs_page *
108 nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
110 struct nfs_page *req = NULL;
112 if (PagePrivate(page))
113 req = (struct nfs_page *)page_private(page);
114 else if (unlikely(PageSwapCache(page)))
115 req = nfs_page_search_commits_for_head_request_locked(nfsi,
119 WARN_ON_ONCE(req->wb_head != req);
120 kref_get(&req->wb_kref);
127 * nfs_page_find_head_request - find head request associated with @page
129 * returns matching head request with reference held, or NULL if not found.
131 static struct nfs_page *nfs_page_find_head_request(struct page *page)
133 struct inode *inode = page_file_mapping(page)->host;
134 struct nfs_page *req = NULL;
136 spin_lock(&inode->i_lock);
137 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
138 spin_unlock(&inode->i_lock);
142 /* Adjust the file length if we're writing beyond the end */
143 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
145 struct inode *inode = page_file_mapping(page)->host;
149 spin_lock(&inode->i_lock);
150 i_size = i_size_read(inode);
151 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
152 if (i_size > 0 && page_file_index(page) < end_index)
154 end = page_file_offset(page) + ((loff_t)offset+count);
157 i_size_write(inode, end);
158 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
160 spin_unlock(&inode->i_lock);
163 /* A writeback failed: mark the page as bad, and invalidate the page cache */
164 static void nfs_set_pageerror(struct page *page)
166 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
170 * nfs_page_group_search_locked
171 * @head - head request of page group
172 * @page_offset - offset into page
174 * Search page group with head @head to find a request that contains the
175 * page offset @page_offset.
177 * Returns a pointer to the first matching nfs request, or NULL if no
180 * Must be called with the page group lock held
182 static struct nfs_page *
183 nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
185 struct nfs_page *req;
187 WARN_ON_ONCE(head != head->wb_head);
188 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
192 if (page_offset >= req->wb_pgbase &&
193 page_offset < (req->wb_pgbase + req->wb_bytes))
196 req = req->wb_this_page;
197 } while (req != head);
203 * nfs_page_group_covers_page
204 * @head - head request of page group
206 * Return true if the page group with head @head covers the whole page,
207 * returns false otherwise
209 static bool nfs_page_group_covers_page(struct nfs_page *req)
211 struct nfs_page *tmp;
212 unsigned int pos = 0;
213 unsigned int len = nfs_page_length(req->wb_page);
215 nfs_page_group_lock(req, false);
218 tmp = nfs_page_group_search_locked(req->wb_head, pos);
220 /* no way this should happen */
221 WARN_ON_ONCE(tmp->wb_pgbase != pos);
222 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
224 } while (tmp && pos < len);
226 nfs_page_group_unlock(req);
227 WARN_ON_ONCE(pos > len);
231 /* We can set the PG_uptodate flag if we see that a write request
232 * covers the full page.
234 static void nfs_mark_uptodate(struct nfs_page *req)
236 if (PageUptodate(req->wb_page))
238 if (!nfs_page_group_covers_page(req))
240 SetPageUptodate(req->wb_page);
243 static int wb_priority(struct writeback_control *wbc)
245 if (wbc->for_reclaim)
246 return FLUSH_HIGHPRI | FLUSH_STABLE;
247 if (wbc->for_kupdate || wbc->for_background)
248 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
249 return FLUSH_COND_STABLE;
253 * NFS congestion control
256 int nfs_congestion_kb;
258 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
259 #define NFS_CONGESTION_OFF_THRESH \
260 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
262 static void nfs_set_page_writeback(struct page *page)
264 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
265 int ret = test_set_page_writeback(page);
267 WARN_ON_ONCE(ret != 0);
269 if (atomic_long_inc_return(&nfss->writeback) >
270 NFS_CONGESTION_ON_THRESH) {
271 set_bdi_congested(&nfss->backing_dev_info,
276 static void nfs_end_page_writeback(struct nfs_page *req)
278 struct inode *inode = page_file_mapping(req->wb_page)->host;
279 struct nfs_server *nfss = NFS_SERVER(inode);
281 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
284 end_page_writeback(req->wb_page);
285 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
286 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
290 /* nfs_page_group_clear_bits
291 * @req - an nfs request
292 * clears all page group related bits from @req
295 nfs_page_group_clear_bits(struct nfs_page *req)
297 clear_bit(PG_TEARDOWN, &req->wb_flags);
298 clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
299 clear_bit(PG_UPTODATE, &req->wb_flags);
300 clear_bit(PG_WB_END, &req->wb_flags);
301 clear_bit(PG_REMOVE, &req->wb_flags);
306 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
308 * this is a helper function for nfs_lock_and_join_requests
310 * @inode - inode associated with request page group, must be holding inode lock
311 * @head - head request of page group, must be holding head lock
312 * @req - request that couldn't lock and needs to wait on the req bit lock
313 * @nonblock - if true, don't actually wait
315 * NOTE: this must be called holding page_group bit lock and inode spin lock
316 * and BOTH will be released before returning.
318 * returns 0 on success, < 0 on error.
321 nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
322 struct nfs_page *req, bool nonblock)
323 __releases(&inode->i_lock)
325 struct nfs_page *tmp;
328 /* relinquish all the locks successfully grabbed this run */
329 for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
330 nfs_unlock_request(tmp);
332 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
334 /* grab a ref on the request that will be waited on */
335 kref_get(&req->wb_kref);
337 nfs_page_group_unlock(head);
338 spin_unlock(&inode->i_lock);
340 /* release ref from nfs_page_find_head_request_locked */
341 nfs_release_request(head);
344 ret = nfs_wait_on_request(req);
347 nfs_release_request(req);
353 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
355 * @destroy_list - request list (using wb_this_page) terminated by @old_head
356 * @old_head - the old head of the list
358 * All subrequests must be locked and removed from all lists, so at this point
359 * they are only "active" in this function, and possibly in nfs_wait_on_request
360 * with a reference held by some other context.
363 nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
364 struct nfs_page *old_head)
366 while (destroy_list) {
367 struct nfs_page *subreq = destroy_list;
369 destroy_list = (subreq->wb_this_page == old_head) ?
370 NULL : subreq->wb_this_page;
372 WARN_ON_ONCE(old_head != subreq->wb_head);
374 /* make sure old group is not used */
375 subreq->wb_head = subreq;
376 subreq->wb_this_page = subreq;
378 /* subreq is now totally disconnected from page group or any
379 * write / commit lists. last chance to wake any waiters */
380 nfs_unlock_request(subreq);
382 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
383 /* release ref on old head request */
384 nfs_release_request(old_head);
386 nfs_page_group_clear_bits(subreq);
388 /* release the PG_INODE_REF reference */
389 if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
390 nfs_release_request(subreq);
394 WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
395 /* zombie requests have already released the last
396 * reference and were waiting on the rest of the
397 * group to complete. Since it's no longer part of a
398 * group, simply free the request */
399 nfs_page_group_clear_bits(subreq);
400 nfs_free_request(subreq);
406 * nfs_lock_and_join_requests - join all subreqs to the head req and return
407 * a locked reference, cancelling any pending
408 * operations for this page.
410 * @page - the page used to lookup the "page group" of nfs_page structures
411 * @nonblock - if true, don't block waiting for request locks
413 * This function joins all sub requests to the head request by first
414 * locking all requests in the group, cancelling any pending operations
415 * and finally updating the head request to cover the whole range covered by
416 * the (former) group. All subrequests are removed from any write or commit
417 * lists, unlinked from the group and destroyed.
419 * Returns a locked, referenced pointer to the head request - which after
420 * this call is guaranteed to be the only request associated with the page.
421 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
422 * error was encountered.
424 static struct nfs_page *
425 nfs_lock_and_join_requests(struct page *page, bool nonblock)
427 struct inode *inode = page_file_mapping(page)->host;
428 struct nfs_page *head, *subreq;
429 struct nfs_page *destroy_list = NULL;
430 unsigned int total_bytes;
436 WARN_ON_ONCE(destroy_list);
438 spin_lock(&inode->i_lock);
441 * A reference is taken only on the head request which acts as a
442 * reference to the whole page group - the group will not be destroyed
443 * until the head reference is released.
445 head = nfs_page_find_head_request_locked(NFS_I(inode), page);
448 spin_unlock(&inode->i_lock);
452 /* holding inode lock, so always make a non-blocking call to try the
454 ret = nfs_page_group_lock(head, true);
456 spin_unlock(&inode->i_lock);
458 if (!nonblock && ret == -EAGAIN) {
459 nfs_page_group_lock_wait(head);
460 nfs_release_request(head);
464 nfs_release_request(head);
468 /* lock each request in the page group */
472 * Subrequests are always contiguous, non overlapping
473 * and in order. If not, it's a programming error.
475 WARN_ON_ONCE(subreq->wb_offset !=
476 (head->wb_offset + total_bytes));
478 /* keep track of how many bytes this group covers */
479 total_bytes += subreq->wb_bytes;
481 if (!nfs_lock_request(subreq)) {
482 /* releases page group bit lock and
483 * inode spin lock and all references */
484 ret = nfs_unroll_locks_and_wait(inode, head,
493 subreq = subreq->wb_this_page;
494 } while (subreq != head);
496 /* Now that all requests are locked, make sure they aren't on any list.
497 * Commit list removal accounting is done after locks are dropped */
500 nfs_clear_request_commit(subreq);
501 subreq = subreq->wb_this_page;
502 } while (subreq != head);
504 /* unlink subrequests from head, destroy them later */
505 if (head->wb_this_page != head) {
506 /* destroy list will be terminated by head */
507 destroy_list = head->wb_this_page;
508 head->wb_this_page = head;
510 /* change head request to cover whole range that
511 * the former page group covered */
512 head->wb_bytes = total_bytes;
516 * prepare head request to be added to new pgio descriptor
518 nfs_page_group_clear_bits(head);
521 * some part of the group was still on the inode list - otherwise
522 * the group wouldn't be involved in async write.
523 * grab a reference for the head request, iff it needs one.
525 if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
526 kref_get(&head->wb_kref);
528 nfs_page_group_unlock(head);
530 /* drop lock to clean uprequests on destroy list */
531 spin_unlock(&inode->i_lock);
533 nfs_destroy_unlinked_subrequests(destroy_list, head);
535 /* still holds ref on head from nfs_page_find_head_request_locked
536 * and still has lock on head from lock loop */
541 * Find an associated nfs write request, and prepare to flush it out
542 * May return an error if the user signalled nfs_wait_on_request().
544 static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
545 struct page *page, bool nonblock)
547 struct nfs_page *req;
550 req = nfs_lock_and_join_requests(page, nonblock);
557 nfs_set_page_writeback(page);
558 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
561 if (!nfs_pageio_add_request(pgio, req)) {
562 nfs_redirty_request(req);
563 ret = pgio->pg_error;
569 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
571 struct inode *inode = page_file_mapping(page)->host;
574 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
575 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
577 nfs_pageio_cond_complete(pgio, page_file_index(page));
578 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
579 if (ret == -EAGAIN) {
580 redirty_page_for_writepage(wbc, page);
587 * Write an mmapped page to the server.
589 static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
591 struct nfs_pageio_descriptor pgio;
594 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
595 false, &nfs_async_write_completion_ops);
596 err = nfs_do_writepage(page, wbc, &pgio);
597 nfs_pageio_complete(&pgio);
600 if (pgio.pg_error < 0)
601 return pgio.pg_error;
605 int nfs_writepage(struct page *page, struct writeback_control *wbc)
609 ret = nfs_writepage_locked(page, wbc);
614 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
618 ret = nfs_do_writepage(page, wbc, data);
623 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
625 struct inode *inode = mapping->host;
626 unsigned long *bitlock = &NFS_I(inode)->flags;
627 struct nfs_pageio_descriptor pgio;
630 /* Stop dirtying of new pages while we sync */
631 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
632 nfs_wait_bit_killable, TASK_KILLABLE);
636 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
638 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
639 &nfs_async_write_completion_ops);
640 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
641 nfs_pageio_complete(&pgio);
643 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
644 smp_mb__after_atomic();
645 wake_up_bit(bitlock, NFS_INO_FLUSHING);
658 * Insert a write request into an inode
660 static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
662 struct nfs_inode *nfsi = NFS_I(inode);
664 WARN_ON_ONCE(req->wb_this_page != req);
666 /* Lock the request! */
667 nfs_lock_request(req);
669 spin_lock(&inode->i_lock);
670 if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
673 * Swap-space should not get truncated. Hence no need to plug the race
674 * with invalidate/truncate.
676 if (likely(!PageSwapCache(req->wb_page))) {
677 set_bit(PG_MAPPED, &req->wb_flags);
678 SetPagePrivate(req->wb_page);
679 set_page_private(req->wb_page, (unsigned long)req);
682 /* this a head request for a page group - mark it as having an
683 * extra reference so sub groups can follow suit */
684 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
685 kref_get(&req->wb_kref);
686 spin_unlock(&inode->i_lock);
690 * Remove a write request from an inode
692 static void nfs_inode_remove_request(struct nfs_page *req)
694 struct inode *inode = req->wb_context->dentry->d_inode;
695 struct nfs_inode *nfsi = NFS_I(inode);
696 struct nfs_page *head;
698 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
701 spin_lock(&inode->i_lock);
702 if (likely(!PageSwapCache(head->wb_page))) {
703 set_page_private(head->wb_page, 0);
704 ClearPagePrivate(head->wb_page);
705 clear_bit(PG_MAPPED, &head->wb_flags);
708 spin_unlock(&inode->i_lock);
711 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
712 nfs_release_request(req);
718 nfs_mark_request_dirty(struct nfs_page *req)
720 __set_page_dirty_nobuffers(req->wb_page);
723 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
725 * nfs_page_search_commits_for_head_request_locked
727 * Search through commit lists on @inode for the head request for @page.
728 * Must be called while holding the inode (which is cinfo) lock.
730 * Returns the head request if found, or NULL if not found.
732 static struct nfs_page *
733 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
736 struct nfs_page *freq, *t;
737 struct nfs_commit_info cinfo;
738 struct inode *inode = &nfsi->vfs_inode;
740 nfs_init_cinfo_from_inode(&cinfo, inode);
742 /* search through pnfs commit lists */
743 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
745 return freq->wb_head;
747 /* Linearly search the commit list for the correct request */
748 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
749 if (freq->wb_page == page)
750 return freq->wb_head;
757 * nfs_request_add_commit_list - add request to a commit list
758 * @req: pointer to a struct nfs_page
759 * @dst: commit list head
760 * @cinfo: holds list lock and accounting info
762 * This sets the PG_CLEAN bit, updates the cinfo count of
763 * number of outstanding requests requiring a commit as well as
766 * The caller must _not_ hold the cinfo->lock, but must be
767 * holding the nfs_page lock.
770 nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
771 struct nfs_commit_info *cinfo)
773 set_bit(PG_CLEAN, &(req)->wb_flags);
774 spin_lock(cinfo->lock);
775 nfs_list_add_request(req, dst);
776 cinfo->mds->ncommit++;
777 spin_unlock(cinfo->lock);
779 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
780 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
782 __mark_inode_dirty(req->wb_context->dentry->d_inode,
786 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
789 * nfs_request_remove_commit_list - Remove request from a commit list
790 * @req: pointer to a nfs_page
791 * @cinfo: holds list lock and accounting info
793 * This clears the PG_CLEAN bit, and updates the cinfo's count of
794 * number of outstanding requests requiring a commit
795 * It does not update the MM page stats.
797 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
800 nfs_request_remove_commit_list(struct nfs_page *req,
801 struct nfs_commit_info *cinfo)
803 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
805 nfs_list_remove_request(req);
806 cinfo->mds->ncommit--;
808 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
810 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
813 cinfo->lock = &inode->i_lock;
814 cinfo->mds = &NFS_I(inode)->commit_info;
815 cinfo->ds = pnfs_get_ds_info(inode);
817 cinfo->completion_ops = &nfs_commit_completion_ops;
820 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
822 struct nfs_direct_req *dreq)
825 nfs_init_cinfo_from_dreq(cinfo, dreq);
827 nfs_init_cinfo_from_inode(cinfo, inode);
829 EXPORT_SYMBOL_GPL(nfs_init_cinfo);
832 * Add a request to the inode's commit list.
835 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
836 struct nfs_commit_info *cinfo)
838 if (pnfs_mark_request_commit(req, lseg, cinfo))
840 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
844 nfs_clear_page_commit(struct page *page)
846 dec_zone_page_state(page, NR_UNSTABLE_NFS);
847 dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
850 /* Called holding inode (/cinfo) lock */
852 nfs_clear_request_commit(struct nfs_page *req)
854 if (test_bit(PG_CLEAN, &req->wb_flags)) {
855 struct inode *inode = req->wb_context->dentry->d_inode;
856 struct nfs_commit_info cinfo;
858 nfs_init_cinfo_from_inode(&cinfo, inode);
859 if (!pnfs_clear_request_commit(req, &cinfo)) {
860 nfs_request_remove_commit_list(req, &cinfo);
862 nfs_clear_page_commit(req->wb_page);
866 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
868 if (hdr->verf.committed == NFS_DATA_SYNC)
869 return hdr->lseg == NULL;
870 return hdr->verf.committed != NFS_FILE_SYNC;
874 static struct nfs_page *
875 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
881 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
886 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
888 struct nfs_direct_req *dreq)
893 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
894 struct nfs_commit_info *cinfo)
899 nfs_clear_request_commit(struct nfs_page *req)
903 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
910 static void nfs_write_completion(struct nfs_pgio_header *hdr)
912 struct nfs_commit_info cinfo;
913 unsigned long bytes = 0;
915 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
917 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
918 while (!list_empty(&hdr->pages)) {
919 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
921 bytes += req->wb_bytes;
922 nfs_list_remove_request(req);
923 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
924 (hdr->good_bytes < bytes)) {
925 nfs_set_pageerror(req->wb_page);
926 nfs_context_set_write_error(req->wb_context, hdr->error);
929 if (nfs_write_need_commit(hdr)) {
930 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
931 nfs_mark_request_commit(req, hdr->lseg, &cinfo);
935 nfs_inode_remove_request(req);
937 nfs_unlock_request(req);
938 nfs_end_page_writeback(req);
939 nfs_release_request(req);
945 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
947 nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
949 return cinfo->mds->ncommit;
952 /* cinfo->lock held by caller */
954 nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
955 struct nfs_commit_info *cinfo, int max)
957 struct nfs_page *req, *tmp;
960 list_for_each_entry_safe(req, tmp, src, wb_list) {
961 if (!nfs_lock_request(req))
963 kref_get(&req->wb_kref);
964 if (cond_resched_lock(cinfo->lock))
965 list_safe_reset_next(req, tmp, wb_list);
966 nfs_request_remove_commit_list(req, cinfo);
967 nfs_list_add_request(req, dst);
969 if ((ret == max) && !cinfo->dreq)
976 * nfs_scan_commit - Scan an inode for commit requests
977 * @inode: NFS inode to scan
978 * @dst: mds destination list
979 * @cinfo: mds and ds lists of reqs ready to commit
981 * Moves requests from the inode's 'commit' request list.
982 * The requests are *not* checked to ensure that they form a contiguous set.
985 nfs_scan_commit(struct inode *inode, struct list_head *dst,
986 struct nfs_commit_info *cinfo)
990 spin_lock(cinfo->lock);
991 if (cinfo->mds->ncommit > 0) {
992 const int max = INT_MAX;
994 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
996 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
998 spin_unlock(cinfo->lock);
1003 unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
1008 int nfs_scan_commit(struct inode *inode, struct list_head *dst,
1009 struct nfs_commit_info *cinfo)
1016 * Search for an existing write request, and attempt to update
1017 * it to reflect a new dirty region on a given page.
1019 * If the attempt fails, then the existing request is flushed out
1022 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1024 unsigned int offset,
1027 struct nfs_page *req;
1032 if (!PagePrivate(page))
1035 end = offset + bytes;
1036 spin_lock(&inode->i_lock);
1039 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
1043 /* should be handled by nfs_flush_incompatible */
1044 WARN_ON_ONCE(req->wb_head != req);
1045 WARN_ON_ONCE(req->wb_this_page != req);
1047 rqend = req->wb_offset + req->wb_bytes;
1049 * Tell the caller to flush out the request if
1050 * the offsets are non-contiguous.
1051 * Note: nfs_flush_incompatible() will already
1052 * have flushed out requests having wrong owners.
1055 || end < req->wb_offset)
1058 if (nfs_lock_request(req))
1061 /* The request is locked, so wait and then retry */
1062 spin_unlock(&inode->i_lock);
1063 error = nfs_wait_on_request(req);
1064 nfs_release_request(req);
1067 spin_lock(&inode->i_lock);
1070 /* Okay, the request matches. Update the region */
1071 if (offset < req->wb_offset) {
1072 req->wb_offset = offset;
1073 req->wb_pgbase = offset;
1076 req->wb_bytes = end - req->wb_offset;
1078 req->wb_bytes = rqend - req->wb_offset;
1081 nfs_clear_request_commit(req);
1082 spin_unlock(&inode->i_lock);
1085 spin_unlock(&inode->i_lock);
1086 nfs_release_request(req);
1087 error = nfs_wb_page(inode, page);
1089 return ERR_PTR(error);
1093 * Try to update an existing write request, or create one if there is none.
1095 * Note: Should always be called with the Page Lock held to prevent races
1096 * if we have to add a new request. Also assumes that the caller has
1097 * already called nfs_flush_incompatible() if necessary.
1099 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1100 struct page *page, unsigned int offset, unsigned int bytes)
1102 struct inode *inode = page_file_mapping(page)->host;
1103 struct nfs_page *req;
1105 req = nfs_try_to_update_request(inode, page, offset, bytes);
1108 req = nfs_create_request(ctx, page, NULL, offset, bytes);
1111 nfs_inode_add_request(inode, req);
1116 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1117 unsigned int offset, unsigned int count)
1119 struct nfs_page *req;
1121 req = nfs_setup_write_request(ctx, page, offset, count);
1123 return PTR_ERR(req);
1124 /* Update file length */
1125 nfs_grow_file(page, offset, count);
1126 nfs_mark_uptodate(req);
1127 nfs_mark_request_dirty(req);
1128 nfs_unlock_and_release_request(req);
1132 int nfs_flush_incompatible(struct file *file, struct page *page)
1134 struct nfs_open_context *ctx = nfs_file_open_context(file);
1135 struct nfs_lock_context *l_ctx;
1136 struct nfs_page *req;
1137 int do_flush, status;
1139 * Look for a request corresponding to this page. If there
1140 * is one, and it belongs to another file, we flush it out
1141 * before we try to copy anything into the page. Do this
1142 * due to the lack of an ACCESS-type call in NFSv2.
1143 * Also do the same if we find a request from an existing
1147 req = nfs_page_find_head_request(page);
1150 l_ctx = req->wb_lock_context;
1151 do_flush = req->wb_page != page || req->wb_context != ctx;
1152 /* for now, flush if more than 1 request in page_group */
1153 do_flush |= req->wb_this_page != req;
1154 if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
1155 do_flush |= l_ctx->lockowner.l_owner != current->files
1156 || l_ctx->lockowner.l_pid != current->tgid;
1158 nfs_release_request(req);
1161 status = nfs_wb_page(page_file_mapping(page)->host, page);
1162 } while (status == 0);
1167 * Avoid buffered writes when a open context credential's key would
1170 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1172 * Return 0 and set a credential flag which triggers the inode to flush
1173 * and performs NFS_FILE_SYNC writes if the key will expired within
1174 * RPC_KEY_EXPIRE_TIMEO.
1177 nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1179 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1180 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1182 return rpcauth_key_timeout_notify(auth, ctx->cred);
1186 * Test if the open context credential key is marked to expire soon.
1188 bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1190 return rpcauth_cred_key_to_expire(ctx->cred);
1194 * If the page cache is marked as unsafe or invalid, then we can't rely on
1195 * the PageUptodate() flag. In this case, we will need to turn off
1196 * write optimisations that depend on the page contents being correct.
1198 static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
1200 struct nfs_inode *nfsi = NFS_I(inode);
1202 if (nfs_have_delegated_attributes(inode))
1204 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1207 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
1210 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1212 return PageUptodate(page) != 0;
1215 /* If we know the page is up to date, and we're not using byte range locks (or
1216 * if we have the whole file locked for writing), it may be more efficient to
1217 * extend the write to cover the entire page in order to avoid fragmentation
1220 * If the file is opened for synchronous writes then we can just skip the rest
1223 static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1225 if (file->f_flags & O_DSYNC)
1227 if (!nfs_write_pageuptodate(page, inode))
1229 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1231 if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
1232 inode->i_flock->fl_end == OFFSET_MAX &&
1233 inode->i_flock->fl_type != F_RDLCK))
1239 * Update and possibly write a cached page of an NFS file.
1241 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1242 * things with a page scheduled for an RPC call (e.g. invalidate it).
1244 int nfs_updatepage(struct file *file, struct page *page,
1245 unsigned int offset, unsigned int count)
1247 struct nfs_open_context *ctx = nfs_file_open_context(file);
1248 struct inode *inode = page_file_mapping(page)->host;
1251 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1253 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1254 file, count, (long long)(page_file_offset(page) + offset));
1256 if (nfs_can_extend_write(file, page, inode)) {
1257 count = max(count + offset, nfs_page_length(page));
1261 status = nfs_writepage_setup(ctx, page, offset, count);
1263 nfs_set_pageerror(page);
1265 __set_page_dirty_nobuffers(page);
1267 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1268 status, (long long)i_size_read(inode));
1272 static int flush_task_priority(int how)
1274 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1276 return RPC_PRIORITY_HIGH;
1278 return RPC_PRIORITY_LOW;
1280 return RPC_PRIORITY_NORMAL;
1283 static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1284 struct rpc_message *msg,
1285 struct rpc_task_setup *task_setup_data, int how)
1287 struct inode *inode = hdr->inode;
1288 int priority = flush_task_priority(how);
1290 task_setup_data->priority = priority;
1291 NFS_PROTO(inode)->write_setup(hdr, msg);
1293 nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
1294 &task_setup_data->rpc_client, msg, hdr);
1297 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1298 * call this on each, which will prepare them to be retried on next
1299 * writeback using standard nfs.
1301 static void nfs_redirty_request(struct nfs_page *req)
1303 nfs_mark_request_dirty(req);
1304 nfs_unlock_request(req);
1305 nfs_end_page_writeback(req);
1306 nfs_release_request(req);
1309 static void nfs_async_write_error(struct list_head *head)
1311 struct nfs_page *req;
1313 while (!list_empty(head)) {
1314 req = nfs_list_entry(head->next);
1315 nfs_list_remove_request(req);
1316 nfs_redirty_request(req);
1320 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1321 .error_cleanup = nfs_async_write_error,
1322 .completion = nfs_write_completion,
1325 void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1326 struct inode *inode, int ioflags, bool force_mds,
1327 const struct nfs_pgio_completion_ops *compl_ops)
1329 struct nfs_server *server = NFS_SERVER(inode);
1330 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1332 #ifdef CONFIG_NFS_V4_1
1333 if (server->pnfs_curr_ld && !force_mds)
1334 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1336 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1337 server->wsize, ioflags);
1339 EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1341 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1343 pgio->pg_ops = &nfs_pgio_rw_ops;
1344 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1346 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1349 void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1351 struct nfs_commit_data *data = calldata;
1353 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1356 static void nfs_writeback_release_common(struct nfs_pgio_header *hdr)
1362 * Special version of should_remove_suid() that ignores capabilities.
1364 static int nfs_should_remove_suid(const struct inode *inode)
1366 umode_t mode = inode->i_mode;
1369 /* suid always must be killed */
1370 if (unlikely(mode & S_ISUID))
1371 kill = ATTR_KILL_SUID;
1374 * sgid without any exec bits is just a mandatory locking mark; leave
1375 * it alone. If some exec bits are set, it's a real sgid; kill it.
1377 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1378 kill |= ATTR_KILL_SGID;
1380 if (unlikely(kill && S_ISREG(mode)))
1387 * This function is called when the WRITE call is complete.
1389 static int nfs_writeback_done(struct rpc_task *task,
1390 struct nfs_pgio_header *hdr,
1391 struct inode *inode)
1396 * ->write_done will attempt to use post-op attributes to detect
1397 * conflicting writes by other clients. A strict interpretation
1398 * of close-to-open would allow us to continue caching even if
1399 * another writer had changed the file, but some applications
1400 * depend on tighter cache coherency when writing.
1402 status = NFS_PROTO(inode)->write_done(task, hdr);
1405 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1407 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1408 if (hdr->res.verf->committed < hdr->args.stable &&
1409 task->tk_status >= 0) {
1410 /* We tried a write call, but the server did not
1411 * commit data to stable storage even though we
1413 * Note: There is a known bug in Tru64 < 5.0 in which
1414 * the server reports NFS_DATA_SYNC, but performs
1415 * NFS_FILE_SYNC. We therefore implement this checking
1416 * as a dprintk() in order to avoid filling syslog.
1418 static unsigned long complain;
1420 /* Note this will print the MDS for a DS write */
1421 if (time_before(complain, jiffies)) {
1422 dprintk("NFS: faulty NFS server %s:"
1423 " (committed = %d) != (stable = %d)\n",
1424 NFS_SERVER(inode)->nfs_client->cl_hostname,
1425 hdr->res.verf->committed, hdr->args.stable);
1426 complain = jiffies + 300 * HZ;
1431 /* Deal with the suid/sgid bit corner case */
1432 if (nfs_should_remove_suid(inode))
1433 nfs_mark_for_revalidate(inode);
1438 * This function is called when the WRITE call is complete.
1440 static void nfs_writeback_result(struct rpc_task *task,
1441 struct nfs_pgio_header *hdr)
1443 struct nfs_pgio_args *argp = &hdr->args;
1444 struct nfs_pgio_res *resp = &hdr->res;
1446 if (resp->count < argp->count) {
1447 static unsigned long complain;
1449 /* This a short write! */
1450 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1452 /* Has the server at least made some progress? */
1453 if (resp->count == 0) {
1454 if (time_before(complain, jiffies)) {
1456 "NFS: Server wrote zero bytes, expected %u.\n",
1458 complain = jiffies + 300 * HZ;
1460 nfs_set_pgio_error(hdr, -EIO, argp->offset);
1461 task->tk_status = -EIO;
1464 /* Was this an NFSv2 write or an NFSv3 stable write? */
1465 if (resp->verf->committed != NFS_UNSTABLE) {
1466 /* Resend from where the server left off */
1467 hdr->mds_offset += resp->count;
1468 argp->offset += resp->count;
1469 argp->pgbase += resp->count;
1470 argp->count -= resp->count;
1472 /* Resend as a stable write in order to avoid
1473 * headaches in the case of a server crash.
1475 argp->stable = NFS_FILE_SYNC;
1477 rpc_restart_call_prepare(task);
1482 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
1483 static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1487 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1491 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1493 nfs_wait_bit_killable,
1495 return (ret < 0) ? ret : 1;
1498 static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1500 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1501 smp_mb__after_atomic();
1502 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1505 void nfs_commitdata_release(struct nfs_commit_data *data)
1507 put_nfs_open_context(data->context);
1508 nfs_commit_free(data);
1510 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1512 int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1513 const struct rpc_call_ops *call_ops,
1516 struct rpc_task *task;
1517 int priority = flush_task_priority(how);
1518 struct rpc_message msg = {
1519 .rpc_argp = &data->args,
1520 .rpc_resp = &data->res,
1521 .rpc_cred = data->cred,
1523 struct rpc_task_setup task_setup_data = {
1524 .task = &data->task,
1526 .rpc_message = &msg,
1527 .callback_ops = call_ops,
1528 .callback_data = data,
1529 .workqueue = nfsiod_workqueue,
1530 .flags = RPC_TASK_ASYNC | flags,
1531 .priority = priority,
1533 /* Set up the initial task struct. */
1534 NFS_PROTO(data->inode)->commit_setup(data, &msg);
1536 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1538 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1539 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1541 task = rpc_run_task(&task_setup_data);
1543 return PTR_ERR(task);
1544 if (how & FLUSH_SYNC)
1545 rpc_wait_for_completion_task(task);
1549 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1551 static loff_t nfs_get_lwb(struct list_head *head)
1554 struct nfs_page *req;
1556 list_for_each_entry(req, head, wb_list)
1557 if (lwb < (req_offset(req) + req->wb_bytes))
1558 lwb = req_offset(req) + req->wb_bytes;
1564 * Set up the argument/result storage required for the RPC call.
1566 void nfs_init_commit(struct nfs_commit_data *data,
1567 struct list_head *head,
1568 struct pnfs_layout_segment *lseg,
1569 struct nfs_commit_info *cinfo)
1571 struct nfs_page *first = nfs_list_entry(head->next);
1572 struct inode *inode = first->wb_context->dentry->d_inode;
1574 /* Set up the RPC argument and reply structs
1575 * NB: take care not to mess about with data->commit et al. */
1577 list_splice_init(head, &data->pages);
1579 data->inode = inode;
1580 data->cred = first->wb_context->cred;
1581 data->lseg = lseg; /* reference transferred */
1582 /* only set lwb for pnfs commit */
1584 data->lwb = nfs_get_lwb(&data->pages);
1585 data->mds_ops = &nfs_commit_ops;
1586 data->completion_ops = cinfo->completion_ops;
1587 data->dreq = cinfo->dreq;
1589 data->args.fh = NFS_FH(data->inode);
1590 /* Note: we always request a commit of the entire inode */
1591 data->args.offset = 0;
1592 data->args.count = 0;
1593 data->context = get_nfs_open_context(first->wb_context);
1594 data->res.fattr = &data->fattr;
1595 data->res.verf = &data->verf;
1596 nfs_fattr_init(&data->fattr);
1598 EXPORT_SYMBOL_GPL(nfs_init_commit);
1600 void nfs_retry_commit(struct list_head *page_list,
1601 struct pnfs_layout_segment *lseg,
1602 struct nfs_commit_info *cinfo)
1604 struct nfs_page *req;
1606 while (!list_empty(page_list)) {
1607 req = nfs_list_entry(page_list->next);
1608 nfs_list_remove_request(req);
1609 nfs_mark_request_commit(req, lseg, cinfo);
1611 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1612 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1615 nfs_unlock_and_release_request(req);
1618 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1621 * Commit dirty pages
1624 nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1625 struct nfs_commit_info *cinfo)
1627 struct nfs_commit_data *data;
1629 data = nfs_commitdata_alloc();
1634 /* Set up the argument struct */
1635 nfs_init_commit(data, head, NULL, cinfo);
1636 atomic_inc(&cinfo->mds->rpcs_out);
1637 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
1640 nfs_retry_commit(head, NULL, cinfo);
1641 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1646 * COMMIT call returned
1648 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1650 struct nfs_commit_data *data = calldata;
1652 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1653 task->tk_pid, task->tk_status);
1655 /* Call the NFS version-specific code */
1656 NFS_PROTO(data->inode)->commit_done(task, data);
1659 static void nfs_commit_release_pages(struct nfs_commit_data *data)
1661 struct nfs_page *req;
1662 int status = data->task.tk_status;
1663 struct nfs_commit_info cinfo;
1665 while (!list_empty(&data->pages)) {
1666 req = nfs_list_entry(data->pages.next);
1667 nfs_list_remove_request(req);
1668 nfs_clear_page_commit(req->wb_page);
1670 dprintk("NFS: commit (%s/%llu %d@%lld)",
1671 req->wb_context->dentry->d_sb->s_id,
1672 (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
1674 (long long)req_offset(req));
1676 nfs_context_set_write_error(req->wb_context, status);
1677 nfs_inode_remove_request(req);
1678 dprintk(", error = %d\n", status);
1682 /* Okay, COMMIT succeeded, apparently. Check the verifier
1683 * returned by the server against all stored verfs. */
1684 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
1685 /* We have a match */
1686 nfs_inode_remove_request(req);
1690 /* We have a mismatch. Write the page again */
1691 dprintk(" mismatch\n");
1692 nfs_mark_request_dirty(req);
1693 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1695 nfs_unlock_and_release_request(req);
1697 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1698 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1699 nfs_commit_clear_lock(NFS_I(data->inode));
1702 static void nfs_commit_release(void *calldata)
1704 struct nfs_commit_data *data = calldata;
1706 data->completion_ops->completion(data);
1707 nfs_commitdata_release(calldata);
1710 static const struct rpc_call_ops nfs_commit_ops = {
1711 .rpc_call_prepare = nfs_commit_prepare,
1712 .rpc_call_done = nfs_commit_done,
1713 .rpc_release = nfs_commit_release,
1716 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1717 .completion = nfs_commit_release_pages,
1718 .error_cleanup = nfs_commit_clear_lock,
1721 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1722 int how, struct nfs_commit_info *cinfo)
1726 status = pnfs_commit_list(inode, head, how, cinfo);
1727 if (status == PNFS_NOT_ATTEMPTED)
1728 status = nfs_commit_list(inode, head, how, cinfo);
1732 int nfs_commit_inode(struct inode *inode, int how)
1735 struct nfs_commit_info cinfo;
1736 int may_wait = how & FLUSH_SYNC;
1739 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1741 goto out_mark_dirty;
1742 nfs_init_cinfo_from_inode(&cinfo, inode);
1743 res = nfs_scan_commit(inode, &head, &cinfo);
1747 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1751 goto out_mark_dirty;
1752 error = wait_on_bit_action(&NFS_I(inode)->flags,
1754 nfs_wait_bit_killable,
1759 nfs_commit_clear_lock(NFS_I(inode));
1761 /* Note: If we exit without ensuring that the commit is complete,
1762 * we must mark the inode as dirty. Otherwise, future calls to
1763 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1764 * that the data is on the disk.
1767 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1771 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1773 struct nfs_inode *nfsi = NFS_I(inode);
1774 int flags = FLUSH_SYNC;
1777 /* no commits means nothing needs to be done */
1778 if (!nfsi->commit_info.ncommit)
1781 if (wbc->sync_mode == WB_SYNC_NONE) {
1782 /* Don't commit yet if this is a non-blocking flush and there
1783 * are a lot of outstanding writes for this mapping.
1785 if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
1786 goto out_mark_dirty;
1788 /* don't wait for the COMMIT response */
1792 ret = nfs_commit_inode(inode, flags);
1794 if (wbc->sync_mode == WB_SYNC_NONE) {
1795 if (ret < wbc->nr_to_write)
1796 wbc->nr_to_write -= ret;
1798 wbc->nr_to_write = 0;
1803 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1807 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1813 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1815 return nfs_commit_unstable_pages(inode, wbc);
1817 EXPORT_SYMBOL_GPL(nfs_write_inode);
1820 * flush the inode to disk.
1822 int nfs_wb_all(struct inode *inode)
1824 struct writeback_control wbc = {
1825 .sync_mode = WB_SYNC_ALL,
1826 .nr_to_write = LONG_MAX,
1828 .range_end = LLONG_MAX,
1832 trace_nfs_writeback_inode_enter(inode);
1834 ret = sync_inode(inode, &wbc);
1836 trace_nfs_writeback_inode_exit(inode, ret);
1839 EXPORT_SYMBOL_GPL(nfs_wb_all);
1841 int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1843 struct nfs_page *req;
1846 wait_on_page_writeback(page);
1848 /* blocking call to cancel all requests and join to a single (head)
1850 req = nfs_lock_and_join_requests(page, false);
1855 /* all requests from this page have been cancelled by
1856 * nfs_lock_and_join_requests, so just remove the head
1857 * request from the inode / page_private pointer and
1859 nfs_inode_remove_request(req);
1861 * In case nfs_inode_remove_request has marked the
1862 * page as being dirty
1864 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1865 nfs_unlock_and_release_request(req);
1872 * Write back all requests on one page - we do this before reading it.
1874 int nfs_wb_page(struct inode *inode, struct page *page)
1876 loff_t range_start = page_file_offset(page);
1877 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1878 struct writeback_control wbc = {
1879 .sync_mode = WB_SYNC_ALL,
1881 .range_start = range_start,
1882 .range_end = range_end,
1886 trace_nfs_writeback_page_enter(inode);
1889 wait_on_page_writeback(page);
1890 if (clear_page_dirty_for_io(page)) {
1891 ret = nfs_writepage_locked(page, &wbc);
1897 if (!PagePrivate(page))
1899 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1904 trace_nfs_writeback_page_exit(inode, ret);
1908 #ifdef CONFIG_MIGRATION
1909 int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1910 struct page *page, enum migrate_mode mode)
1913 * If PagePrivate is set, then the page is currently associated with
1914 * an in-progress read or write request. Don't try to migrate it.
1916 * FIXME: we could do this in principle, but we'll need a way to ensure
1917 * that we can safely release the inode reference while holding
1920 if (PagePrivate(page))
1923 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1926 return migrate_page(mapping, newpage, page, mode);
1930 int __init nfs_init_writepagecache(void)
1932 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1933 sizeof(struct nfs_pgio_header),
1934 0, SLAB_HWCACHE_ALIGN,
1936 if (nfs_wdata_cachep == NULL)
1939 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1941 if (nfs_wdata_mempool == NULL)
1942 goto out_destroy_write_cache;
1944 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1945 sizeof(struct nfs_commit_data),
1946 0, SLAB_HWCACHE_ALIGN,
1948 if (nfs_cdata_cachep == NULL)
1949 goto out_destroy_write_mempool;
1951 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1953 if (nfs_commit_mempool == NULL)
1954 goto out_destroy_commit_cache;
1957 * NFS congestion size, scale with available memory.
1969 * This allows larger machines to have larger/more transfers.
1970 * Limit the default to 256M
1972 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1973 if (nfs_congestion_kb > 256*1024)
1974 nfs_congestion_kb = 256*1024;
1978 out_destroy_commit_cache:
1979 kmem_cache_destroy(nfs_cdata_cachep);
1980 out_destroy_write_mempool:
1981 mempool_destroy(nfs_wdata_mempool);
1982 out_destroy_write_cache:
1983 kmem_cache_destroy(nfs_wdata_cachep);
1987 void nfs_destroy_writepagecache(void)
1989 mempool_destroy(nfs_commit_mempool);
1990 kmem_cache_destroy(nfs_cdata_cachep);
1991 mempool_destroy(nfs_wdata_mempool);
1992 kmem_cache_destroy(nfs_wdata_cachep);
1995 static const struct nfs_rw_ops nfs_rw_write_ops = {
1996 .rw_mode = FMODE_WRITE,
1997 .rw_alloc_header = nfs_writehdr_alloc,
1998 .rw_free_header = nfs_writehdr_free,
1999 .rw_release = nfs_writeback_release_common,
2000 .rw_done = nfs_writeback_done,
2001 .rw_result = nfs_writeback_result,
2002 .rw_initiate = nfs_initiate_write,