4 * Write file data over NFS.
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
9 #include <linux/types.h>
10 #include <linux/slab.h>
12 #include <linux/pagemap.h>
13 #include <linux/file.h>
14 #include <linux/writeback.h>
15 #include <linux/swap.h>
16 #include <linux/migrate.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_mount.h>
21 #include <linux/nfs_page.h>
22 #include <linux/backing-dev.h>
23 #include <linux/export.h>
25 #include <asm/uaccess.h>
27 #include "delegation.h"
34 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
36 #define MIN_POOL_WRITE (32)
37 #define MIN_POOL_COMMIT (4)
40 * Local function declarations
42 static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
43 struct inode *inode, int ioflags);
44 static void nfs_redirty_request(struct nfs_page *req);
45 static const struct rpc_call_ops nfs_write_partial_ops;
46 static const struct rpc_call_ops nfs_write_full_ops;
47 static const struct rpc_call_ops nfs_commit_ops;
49 static struct kmem_cache *nfs_wdata_cachep;
50 static mempool_t *nfs_wdata_mempool;
51 static mempool_t *nfs_commit_mempool;
53 struct nfs_write_data *nfs_commitdata_alloc(void)
55 struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
58 memset(p, 0, sizeof(*p));
59 INIT_LIST_HEAD(&p->pages);
63 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
65 void nfs_commit_free(struct nfs_write_data *p)
67 if (p && (p->pagevec != &p->page_array[0]))
69 mempool_free(p, nfs_commit_mempool);
71 EXPORT_SYMBOL_GPL(nfs_commit_free);
73 struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
75 struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
78 memset(p, 0, sizeof(*p));
79 INIT_LIST_HEAD(&p->pages);
80 p->npages = pagecount;
81 if (pagecount <= ARRAY_SIZE(p->page_array))
82 p->pagevec = p->page_array;
84 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
86 mempool_free(p, nfs_wdata_mempool);
94 void nfs_writedata_free(struct nfs_write_data *p)
96 if (p && (p->pagevec != &p->page_array[0]))
98 mempool_free(p, nfs_wdata_mempool);
101 void nfs_writedata_release(struct nfs_write_data *wdata)
103 put_lseg(wdata->lseg);
104 put_nfs_open_context(wdata->args.context);
105 nfs_writedata_free(wdata);
108 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
112 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
115 static struct nfs_page *nfs_page_find_request_locked(struct page *page)
117 struct nfs_page *req = NULL;
119 if (PagePrivate(page)) {
120 req = (struct nfs_page *)page_private(page);
122 kref_get(&req->wb_kref);
127 static struct nfs_page *nfs_page_find_request(struct page *page)
129 struct inode *inode = page->mapping->host;
130 struct nfs_page *req = NULL;
132 spin_lock(&inode->i_lock);
133 req = nfs_page_find_request_locked(page);
134 spin_unlock(&inode->i_lock);
138 /* Adjust the file length if we're writing beyond the end */
139 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
141 struct inode *inode = page->mapping->host;
145 spin_lock(&inode->i_lock);
146 i_size = i_size_read(inode);
147 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
148 if (i_size > 0 && page->index < end_index)
150 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
153 i_size_write(inode, end);
154 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
156 spin_unlock(&inode->i_lock);
159 /* A writeback failed: mark the page as bad, and invalidate the page cache */
160 static void nfs_set_pageerror(struct page *page)
163 nfs_zap_mapping(page->mapping->host, page->mapping);
166 /* We can set the PG_uptodate flag if we see that a write request
167 * covers the full page.
169 static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
171 if (PageUptodate(page))
175 if (count != nfs_page_length(page))
177 SetPageUptodate(page);
180 static int wb_priority(struct writeback_control *wbc)
182 if (wbc->for_reclaim)
183 return FLUSH_HIGHPRI | FLUSH_STABLE;
184 if (wbc->for_kupdate || wbc->for_background)
185 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
186 return FLUSH_COND_STABLE;
190 * NFS congestion control
193 int nfs_congestion_kb;
195 #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
196 #define NFS_CONGESTION_OFF_THRESH \
197 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
199 static int nfs_set_page_writeback(struct page *page)
201 int ret = test_set_page_writeback(page);
204 struct inode *inode = page->mapping->host;
205 struct nfs_server *nfss = NFS_SERVER(inode);
207 page_cache_get(page);
208 if (atomic_long_inc_return(&nfss->writeback) >
209 NFS_CONGESTION_ON_THRESH) {
210 set_bdi_congested(&nfss->backing_dev_info,
217 static void nfs_end_page_writeback(struct page *page)
219 struct inode *inode = page->mapping->host;
220 struct nfs_server *nfss = NFS_SERVER(inode);
222 end_page_writeback(page);
223 page_cache_release(page);
224 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
225 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
228 static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
230 struct inode *inode = page->mapping->host;
231 struct nfs_page *req;
234 spin_lock(&inode->i_lock);
236 req = nfs_page_find_request_locked(page);
239 if (nfs_set_page_tag_locked(req))
241 /* Note: If we hold the page lock, as is the case in nfs_writepage,
242 * then the call to nfs_set_page_tag_locked() will always
243 * succeed provided that someone hasn't already marked the
244 * request as dirty (in which case we don't care).
246 spin_unlock(&inode->i_lock);
248 ret = nfs_wait_on_request(req);
251 nfs_release_request(req);
254 spin_lock(&inode->i_lock);
256 spin_unlock(&inode->i_lock);
261 * Find an associated nfs write request, and prepare to flush it out
262 * May return an error if the user signalled nfs_wait_on_request().
264 static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
265 struct page *page, bool nonblock)
267 struct nfs_page *req;
270 req = nfs_find_and_lock_request(page, nonblock);
277 ret = nfs_set_page_writeback(page);
279 BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));
281 if (!nfs_pageio_add_request(pgio, req)) {
282 nfs_redirty_request(req);
283 ret = pgio->pg_error;
289 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
291 struct inode *inode = page->mapping->host;
294 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
295 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
297 nfs_pageio_cond_complete(pgio, page->index);
298 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
299 if (ret == -EAGAIN) {
300 redirty_page_for_writepage(wbc, page);
307 * Write an mmapped page to the server.
309 static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
311 struct nfs_pageio_descriptor pgio;
314 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
315 err = nfs_do_writepage(page, wbc, &pgio);
316 nfs_pageio_complete(&pgio);
319 if (pgio.pg_error < 0)
320 return pgio.pg_error;
324 int nfs_writepage(struct page *page, struct writeback_control *wbc)
328 ret = nfs_writepage_locked(page, wbc);
333 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
337 ret = nfs_do_writepage(page, wbc, data);
342 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
344 struct inode *inode = mapping->host;
345 unsigned long *bitlock = &NFS_I(inode)->flags;
346 struct nfs_pageio_descriptor pgio;
349 /* Stop dirtying of new pages while we sync */
350 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
351 nfs_wait_bit_killable, TASK_KILLABLE);
355 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
357 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
358 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
359 nfs_pageio_complete(&pgio);
361 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
362 smp_mb__after_clear_bit();
363 wake_up_bit(bitlock, NFS_INO_FLUSHING);
376 * Insert a write request into an inode
378 static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
380 struct nfs_inode *nfsi = NFS_I(inode);
383 error = radix_tree_preload(GFP_NOFS);
387 /* Lock the request! */
388 nfs_lock_request_dontget(req);
390 spin_lock(&inode->i_lock);
391 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
393 if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE))
395 set_bit(PG_MAPPED, &req->wb_flags);
396 SetPagePrivate(req->wb_page);
397 set_page_private(req->wb_page, (unsigned long)req);
399 kref_get(&req->wb_kref);
400 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
401 NFS_PAGE_TAG_LOCKED);
402 spin_unlock(&inode->i_lock);
403 radix_tree_preload_end();
409 * Remove a write request from an inode
411 static void nfs_inode_remove_request(struct nfs_page *req)
413 struct inode *inode = req->wb_context->dentry->d_inode;
414 struct nfs_inode *nfsi = NFS_I(inode);
416 BUG_ON (!NFS_WBACK_BUSY(req));
418 spin_lock(&inode->i_lock);
419 set_page_private(req->wb_page, 0);
420 ClearPagePrivate(req->wb_page);
421 clear_bit(PG_MAPPED, &req->wb_flags);
422 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
424 spin_unlock(&inode->i_lock);
425 nfs_release_request(req);
429 nfs_mark_request_dirty(struct nfs_page *req)
431 __set_page_dirty_nobuffers(req->wb_page);
434 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
436 * Add a request to the inode's commit list.
439 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
441 struct inode *inode = req->wb_context->dentry->d_inode;
442 struct nfs_inode *nfsi = NFS_I(inode);
444 spin_lock(&inode->i_lock);
445 set_bit(PG_CLEAN, &(req)->wb_flags);
446 radix_tree_tag_set(&nfsi->nfs_page_tree,
448 NFS_PAGE_TAG_COMMIT);
450 spin_unlock(&inode->i_lock);
451 pnfs_mark_request_commit(req, lseg);
452 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
453 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
454 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
458 nfs_clear_request_commit(struct nfs_page *req)
460 struct page *page = req->wb_page;
462 if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
463 dec_zone_page_state(page, NR_UNSTABLE_NFS);
464 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
471 int nfs_write_need_commit(struct nfs_write_data *data)
473 if (data->verf.committed == NFS_DATA_SYNC)
474 return data->lseg == NULL;
476 return data->verf.committed != NFS_FILE_SYNC;
480 int nfs_reschedule_unstable_write(struct nfs_page *req,
481 struct nfs_write_data *data)
483 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
484 nfs_mark_request_commit(req, data->lseg);
487 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
488 nfs_mark_request_dirty(req);
495 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
500 nfs_clear_request_commit(struct nfs_page *req)
506 int nfs_write_need_commit(struct nfs_write_data *data)
512 int nfs_reschedule_unstable_write(struct nfs_page *req,
513 struct nfs_write_data *data)
519 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
521 nfs_need_commit(struct nfs_inode *nfsi)
523 return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
527 * nfs_scan_commit - Scan an inode for commit requests
528 * @inode: NFS inode to scan
529 * @dst: destination list
530 * @idx_start: lower bound of page->index to scan.
531 * @npages: idx_start + npages sets the upper bound to scan.
533 * Moves requests from the inode's 'commit' request list.
534 * The requests are *not* checked to ensure that they form a contiguous set.
537 nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
539 struct nfs_inode *nfsi = NFS_I(inode);
542 if (!nfs_need_commit(nfsi))
545 spin_lock(&inode->i_lock);
546 ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
548 nfsi->ncommit -= ret;
549 spin_unlock(&inode->i_lock);
551 if (nfs_need_commit(NFS_I(inode)))
552 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
557 static inline int nfs_need_commit(struct nfs_inode *nfsi)
562 static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
569 * Search for an existing write request, and attempt to update
570 * it to reflect a new dirty region on a given page.
572 * If the attempt fails, then the existing request is flushed out
575 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
580 struct nfs_page *req;
585 if (!PagePrivate(page))
588 end = offset + bytes;
589 spin_lock(&inode->i_lock);
592 req = nfs_page_find_request_locked(page);
596 rqend = req->wb_offset + req->wb_bytes;
598 * Tell the caller to flush out the request if
599 * the offsets are non-contiguous.
600 * Note: nfs_flush_incompatible() will already
601 * have flushed out requests having wrong owners.
604 || end < req->wb_offset)
607 if (nfs_set_page_tag_locked(req))
610 /* The request is locked, so wait and then retry */
611 spin_unlock(&inode->i_lock);
612 error = nfs_wait_on_request(req);
613 nfs_release_request(req);
616 spin_lock(&inode->i_lock);
619 if (nfs_clear_request_commit(req) &&
620 radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
621 req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) {
622 NFS_I(inode)->ncommit--;
623 pnfs_clear_request_commit(req);
626 /* Okay, the request matches. Update the region */
627 if (offset < req->wb_offset) {
628 req->wb_offset = offset;
629 req->wb_pgbase = offset;
632 req->wb_bytes = end - req->wb_offset;
634 req->wb_bytes = rqend - req->wb_offset;
636 spin_unlock(&inode->i_lock);
639 spin_unlock(&inode->i_lock);
640 nfs_release_request(req);
641 error = nfs_wb_page(inode, page);
643 return ERR_PTR(error);
647 * Try to update an existing write request, or create one if there is none.
649 * Note: Should always be called with the Page Lock held to prevent races
650 * if we have to add a new request. Also assumes that the caller has
651 * already called nfs_flush_incompatible() if necessary.
653 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
654 struct page *page, unsigned int offset, unsigned int bytes)
656 struct inode *inode = page->mapping->host;
657 struct nfs_page *req;
660 req = nfs_try_to_update_request(inode, page, offset, bytes);
663 req = nfs_create_request(ctx, inode, page, offset, bytes);
666 error = nfs_inode_add_request(inode, req);
668 nfs_release_request(req);
669 req = ERR_PTR(error);
675 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
676 unsigned int offset, unsigned int count)
678 struct nfs_page *req;
680 req = nfs_setup_write_request(ctx, page, offset, count);
683 /* Update file length */
684 nfs_grow_file(page, offset, count);
685 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
686 nfs_mark_request_dirty(req);
687 nfs_clear_page_tag_locked(req);
691 int nfs_flush_incompatible(struct file *file, struct page *page)
693 struct nfs_open_context *ctx = nfs_file_open_context(file);
694 struct nfs_page *req;
695 int do_flush, status;
697 * Look for a request corresponding to this page. If there
698 * is one, and it belongs to another file, we flush it out
699 * before we try to copy anything into the page. Do this
700 * due to the lack of an ACCESS-type call in NFSv2.
701 * Also do the same if we find a request from an existing
705 req = nfs_page_find_request(page);
708 do_flush = req->wb_page != page || req->wb_context != ctx ||
709 req->wb_lock_context->lockowner != current->files ||
710 req->wb_lock_context->pid != current->tgid;
711 nfs_release_request(req);
714 status = nfs_wb_page(page->mapping->host, page);
715 } while (status == 0);
720 * If the page cache is marked as unsafe or invalid, then we can't rely on
721 * the PageUptodate() flag. In this case, we will need to turn off
722 * write optimisations that depend on the page contents being correct.
724 static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
726 return PageUptodate(page) &&
727 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
731 * Update and possibly write a cached page of an NFS file.
733 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
734 * things with a page scheduled for an RPC call (e.g. invalidate it).
736 int nfs_updatepage(struct file *file, struct page *page,
737 unsigned int offset, unsigned int count)
739 struct nfs_open_context *ctx = nfs_file_open_context(file);
740 struct inode *inode = page->mapping->host;
743 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
745 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
746 file->f_path.dentry->d_parent->d_name.name,
747 file->f_path.dentry->d_name.name, count,
748 (long long)(page_offset(page) + offset));
753 /* If we're not using byte range locks, and we know the page
754 * is up to date, it may be more efficient to extend the write
755 * to cover the entire page in order to avoid fragmentation
758 if (nfs_write_pageuptodate(page, inode) &&
759 inode->i_flock == NULL &&
760 !(file->f_flags & O_DSYNC)) {
761 count = max(count + offset, nfs_page_length(page));
765 status = nfs_writepage_setup(ctx, page, offset, count);
767 nfs_set_pageerror(page);
769 __set_page_dirty_nobuffers(page);
771 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
772 status, (long long)i_size_read(inode));
776 static void nfs_writepage_release(struct nfs_page *req,
777 struct nfs_write_data *data)
779 struct page *page = req->wb_page;
781 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
782 nfs_inode_remove_request(req);
783 nfs_clear_page_tag_locked(req);
784 nfs_end_page_writeback(page);
787 static int flush_task_priority(int how)
789 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
791 return RPC_PRIORITY_HIGH;
793 return RPC_PRIORITY_LOW;
795 return RPC_PRIORITY_NORMAL;
798 int nfs_initiate_write(struct nfs_write_data *data,
799 struct rpc_clnt *clnt,
800 const struct rpc_call_ops *call_ops,
803 struct inode *inode = data->inode;
804 int priority = flush_task_priority(how);
805 struct rpc_task *task;
806 struct rpc_message msg = {
807 .rpc_argp = &data->args,
808 .rpc_resp = &data->res,
809 .rpc_cred = data->cred,
811 struct rpc_task_setup task_setup_data = {
815 .callback_ops = call_ops,
816 .callback_data = data,
817 .workqueue = nfsiod_workqueue,
818 .flags = RPC_TASK_ASYNC,
819 .priority = priority,
823 /* Set up the initial task struct. */
824 NFS_PROTO(inode)->write_setup(data, &msg);
826 dprintk("NFS: %5u initiated write call "
827 "(req %s/%lld, %u bytes @ offset %llu)\n",
830 (long long)NFS_FILEID(inode),
832 (unsigned long long)data->args.offset);
834 task = rpc_run_task(&task_setup_data);
839 if (how & FLUSH_SYNC) {
840 ret = rpc_wait_for_completion_task(task);
842 ret = task->tk_status;
848 EXPORT_SYMBOL_GPL(nfs_initiate_write);
851 * Set up the argument/result storage required for the RPC call.
853 static void nfs_write_rpcsetup(struct nfs_page *req,
854 struct nfs_write_data *data,
855 unsigned int count, unsigned int offset,
858 struct inode *inode = req->wb_context->dentry->d_inode;
860 /* Set up the RPC argument and reply structs
861 * NB: take care not to mess about with data->commit et al. */
864 data->inode = inode = req->wb_context->dentry->d_inode;
865 data->cred = req->wb_context->cred;
867 data->args.fh = NFS_FH(inode);
868 data->args.offset = req_offset(req) + offset;
869 /* pnfs_set_layoutcommit needs this */
870 data->mds_offset = data->args.offset;
871 data->args.pgbase = req->wb_pgbase + offset;
872 data->args.pages = data->pagevec;
873 data->args.count = count;
874 data->args.context = get_nfs_open_context(req->wb_context);
875 data->args.lock_context = req->wb_lock_context;
876 data->args.stable = NFS_UNSTABLE;
877 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
880 case FLUSH_COND_STABLE:
881 if (nfs_need_commit(NFS_I(inode)))
884 data->args.stable = NFS_FILE_SYNC;
887 data->res.fattr = &data->fattr;
888 data->res.count = count;
889 data->res.verf = &data->verf;
890 nfs_fattr_init(&data->fattr);
893 static int nfs_do_write(struct nfs_write_data *data,
894 const struct rpc_call_ops *call_ops,
897 struct inode *inode = data->args.context->dentry->d_inode;
899 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
902 static int nfs_do_multiple_writes(struct list_head *head,
903 const struct rpc_call_ops *call_ops,
906 struct nfs_write_data *data;
909 while (!list_empty(head)) {
912 data = list_entry(head->next, struct nfs_write_data, list);
913 list_del_init(&data->list);
915 ret2 = nfs_do_write(data, call_ops, how);
922 /* If a nfs_flush_* function fails, it should remove reqs from @head and
923 * call this on each, which will prepare them to be retried on next
924 * writeback using standard nfs.
926 static void nfs_redirty_request(struct nfs_page *req)
928 struct page *page = req->wb_page;
930 nfs_mark_request_dirty(req);
931 nfs_clear_page_tag_locked(req);
932 nfs_end_page_writeback(page);
936 * Generate multiple small requests to write out a single
937 * contiguous dirty area on one page.
939 static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
941 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
942 struct page *page = req->wb_page;
943 struct nfs_write_data *data;
944 size_t wsize = desc->pg_bsize, nbytes;
949 nfs_list_remove_request(req);
951 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
952 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
953 desc->pg_count > wsize))
954 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
958 nbytes = desc->pg_count;
960 size_t len = min(nbytes, wsize);
962 data = nfs_writedata_alloc(1);
965 data->pagevec[0] = page;
966 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
967 list_add(&data->list, res);
971 } while (nbytes != 0);
972 atomic_set(&req->wb_complete, requests);
973 desc->pg_rpc_callops = &nfs_write_partial_ops;
977 while (!list_empty(res)) {
978 data = list_entry(res->next, struct nfs_write_data, list);
979 list_del(&data->list);
980 nfs_writedata_release(data);
982 nfs_redirty_request(req);
987 * Create an RPC task for the given write request and kick it.
988 * The page must have been locked by the caller.
990 * It may happen that the page we're passed is not marked dirty.
991 * This is the case if nfs_updatepage detects a conflicting request
992 * that has been written but not committed.
994 static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
996 struct nfs_page *req;
998 struct nfs_write_data *data;
999 struct list_head *head = &desc->pg_list;
1002 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
1005 while (!list_empty(head)) {
1006 req = nfs_list_entry(head->next);
1007 nfs_list_remove_request(req);
1008 nfs_redirty_request(req);
1013 pages = data->pagevec;
1014 while (!list_empty(head)) {
1015 req = nfs_list_entry(head->next);
1016 nfs_list_remove_request(req);
1017 nfs_list_add_request(req, &data->pages);
1018 *pages++ = req->wb_page;
1020 req = nfs_list_entry(data->pages.next);
1022 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1023 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1024 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1026 /* Set up the argument struct */
1027 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
1028 list_add(&data->list, res);
1029 desc->pg_rpc_callops = &nfs_write_full_ops;
1034 int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1036 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1037 return nfs_flush_multi(desc, head);
1038 return nfs_flush_one(desc, head);
1041 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1046 ret = nfs_generic_flush(desc, &head);
1048 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
1053 static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1054 .pg_test = nfs_generic_pg_test,
1055 .pg_doio = nfs_generic_pg_writepages,
1058 static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
1059 struct inode *inode, int ioflags)
1061 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1062 NFS_SERVER(inode)->wsize, ioflags);
1065 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1067 pgio->pg_ops = &nfs_pageio_write_ops;
1068 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1070 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1072 static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1073 struct inode *inode, int ioflags)
1075 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1076 nfs_pageio_init_write_mds(pgio, inode, ioflags);
1080 * Handle a write reply that flushed part of a page.
1082 static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
1084 struct nfs_write_data *data = calldata;
1086 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1088 data->req->wb_context->dentry->d_inode->i_sb->s_id,
1090 NFS_FILEID(data->req->wb_context->dentry->d_inode),
1091 data->req->wb_bytes, (long long)req_offset(data->req));
1093 nfs_writeback_done(task, data);
1096 static void nfs_writeback_release_partial(void *calldata)
1098 struct nfs_write_data *data = calldata;
1099 struct nfs_page *req = data->req;
1100 struct page *page = req->wb_page;
1101 int status = data->task.tk_status;
1104 nfs_set_pageerror(page);
1105 nfs_context_set_write_error(req->wb_context, status);
1106 dprintk(", error = %d\n", status);
1110 if (nfs_write_need_commit(data)) {
1111 struct inode *inode = page->mapping->host;
1113 spin_lock(&inode->i_lock);
1114 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1115 /* Do nothing we need to resend the writes */
1116 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1117 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1118 dprintk(" defer commit\n");
1119 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1120 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1121 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1122 dprintk(" server reboot detected\n");
1124 spin_unlock(&inode->i_lock);
1129 if (atomic_dec_and_test(&req->wb_complete))
1130 nfs_writepage_release(req, data);
1131 nfs_writedata_release(calldata);
1134 #if defined(CONFIG_NFS_V4_1)
1135 void nfs_write_prepare(struct rpc_task *task, void *calldata)
1137 struct nfs_write_data *data = calldata;
1139 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
1140 &data->args.seq_args,
1141 &data->res.seq_res, 1, task))
1143 rpc_call_start(task);
1145 #endif /* CONFIG_NFS_V4_1 */
1147 static const struct rpc_call_ops nfs_write_partial_ops = {
1148 #if defined(CONFIG_NFS_V4_1)
1149 .rpc_call_prepare = nfs_write_prepare,
1150 #endif /* CONFIG_NFS_V4_1 */
1151 .rpc_call_done = nfs_writeback_done_partial,
1152 .rpc_release = nfs_writeback_release_partial,
1156 * Handle a write reply that flushes a whole page.
1158 * FIXME: There is an inherent race with invalidate_inode_pages and
1159 * writebacks since the page->count is kept > 1 for as long
1160 * as the page has a write request pending.
1162 static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
1164 struct nfs_write_data *data = calldata;
1166 nfs_writeback_done(task, data);
1169 static void nfs_writeback_release_full(void *calldata)
1171 struct nfs_write_data *data = calldata;
1172 int ret, status = data->task.tk_status;
1173 struct nfs_pageio_descriptor pgio;
1175 if (data->pnfs_error) {
1176 nfs_pageio_init_write_mds(&pgio, data->inode, FLUSH_STABLE);
1177 pgio.pg_recoalesce = 1;
1180 /* Update attributes as result of writeback. */
1181 while (!list_empty(&data->pages)) {
1182 struct nfs_page *req = nfs_list_entry(data->pages.next);
1183 struct page *page = req->wb_page;
1185 nfs_list_remove_request(req);
1187 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1189 req->wb_context->dentry->d_inode->i_sb->s_id,
1190 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
1192 (long long)req_offset(req));
1194 if (data->pnfs_error) {
1195 dprintk(", pnfs error = %d\n", data->pnfs_error);
1200 nfs_set_pageerror(page);
1201 nfs_context_set_write_error(req->wb_context, status);
1202 dprintk(", error = %d\n", status);
1203 goto remove_request;
1206 if (nfs_write_need_commit(data)) {
1207 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1208 nfs_mark_request_commit(req, data->lseg);
1209 dprintk(" marked for commit\n");
1214 nfs_inode_remove_request(req);
1216 nfs_clear_page_tag_locked(req);
1217 nfs_end_page_writeback(page);
1218 if (data->pnfs_error) {
1220 nfs_pageio_cond_complete(&pgio, page->index);
1221 ret = nfs_page_async_flush(&pgio, page, 0);
1223 nfs_set_pageerror(page);
1224 dprintk("rewrite to MDS error = %d\n", ret);
1229 if (data->pnfs_error)
1230 nfs_pageio_complete(&pgio);
1231 nfs_writedata_release(calldata);
1234 static const struct rpc_call_ops nfs_write_full_ops = {
1235 #if defined(CONFIG_NFS_V4_1)
1236 .rpc_call_prepare = nfs_write_prepare,
1237 #endif /* CONFIG_NFS_V4_1 */
1238 .rpc_call_done = nfs_writeback_done_full,
1239 .rpc_release = nfs_writeback_release_full,
1244 * This function is called when the WRITE call is complete.
1246 void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
1248 struct nfs_writeargs *argp = &data->args;
1249 struct nfs_writeres *resp = &data->res;
1252 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1253 task->tk_pid, task->tk_status);
1256 * ->write_done will attempt to use post-op attributes to detect
1257 * conflicting writes by other clients. A strict interpretation
1258 * of close-to-open would allow us to continue caching even if
1259 * another writer had changed the file, but some applications
1260 * depend on tighter cache coherency when writing.
1262 status = NFS_PROTO(data->inode)->write_done(task, data);
1265 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1267 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1268 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1269 /* We tried a write call, but the server did not
1270 * commit data to stable storage even though we
1272 * Note: There is a known bug in Tru64 < 5.0 in which
1273 * the server reports NFS_DATA_SYNC, but performs
1274 * NFS_FILE_SYNC. We therefore implement this checking
1275 * as a dprintk() in order to avoid filling syslog.
1277 static unsigned long complain;
1279 /* Note this will print the MDS for a DS write */
1280 if (time_before(complain, jiffies)) {
1281 dprintk("NFS: faulty NFS server %s:"
1282 " (committed = %d) != (stable = %d)\n",
1283 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
1284 resp->verf->committed, argp->stable);
1285 complain = jiffies + 300 * HZ;
1289 /* Is this a short write? */
1290 if (task->tk_status >= 0 && resp->count < argp->count) {
1291 static unsigned long complain;
1293 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1295 /* Has the server at least made some progress? */
1296 if (resp->count != 0) {
1297 /* Was this an NFSv2 write or an NFSv3 stable write? */
1298 if (resp->verf->committed != NFS_UNSTABLE) {
1299 /* Resend from where the server left off */
1300 data->mds_offset += resp->count;
1301 argp->offset += resp->count;
1302 argp->pgbase += resp->count;
1303 argp->count -= resp->count;
1305 /* Resend as a stable write in order to avoid
1306 * headaches in the case of a server crash.
1308 argp->stable = NFS_FILE_SYNC;
1310 rpc_restart_call_prepare(task);
1313 if (time_before(complain, jiffies)) {
1315 "NFS: Server wrote zero bytes, expected %u.\n",
1317 complain = jiffies + 300 * HZ;
1319 /* Can't do anything about it except throw an error. */
1320 task->tk_status = -EIO;
1326 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1327 static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1331 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1335 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1337 nfs_wait_bit_killable,
1339 return (ret < 0) ? ret : 1;
1342 void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1344 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1345 smp_mb__after_clear_bit();
1346 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1348 EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
1350 void nfs_commitdata_release(void *data)
1352 struct nfs_write_data *wdata = data;
1354 put_lseg(wdata->lseg);
1355 put_nfs_open_context(wdata->args.context);
1356 nfs_commit_free(wdata);
1358 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1360 int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
1361 const struct rpc_call_ops *call_ops,
1364 struct rpc_task *task;
1365 int priority = flush_task_priority(how);
1366 struct rpc_message msg = {
1367 .rpc_argp = &data->args,
1368 .rpc_resp = &data->res,
1369 .rpc_cred = data->cred,
1371 struct rpc_task_setup task_setup_data = {
1372 .task = &data->task,
1374 .rpc_message = &msg,
1375 .callback_ops = call_ops,
1376 .callback_data = data,
1377 .workqueue = nfsiod_workqueue,
1378 .flags = RPC_TASK_ASYNC,
1379 .priority = priority,
1381 /* Set up the initial task struct. */
1382 NFS_PROTO(data->inode)->commit_setup(data, &msg);
1384 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1386 task = rpc_run_task(&task_setup_data);
1388 return PTR_ERR(task);
1389 if (how & FLUSH_SYNC)
1390 rpc_wait_for_completion_task(task);
1394 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1397 * Set up the argument/result storage required for the RPC call.
1399 void nfs_init_commit(struct nfs_write_data *data,
1400 struct list_head *head,
1401 struct pnfs_layout_segment *lseg)
1403 struct nfs_page *first = nfs_list_entry(head->next);
1404 struct inode *inode = first->wb_context->dentry->d_inode;
1406 /* Set up the RPC argument and reply structs
1407 * NB: take care not to mess about with data->commit et al. */
1409 list_splice_init(head, &data->pages);
1411 data->inode = inode;
1412 data->cred = first->wb_context->cred;
1413 data->lseg = lseg; /* reference transferred */
1414 data->mds_ops = &nfs_commit_ops;
1416 data->args.fh = NFS_FH(data->inode);
1417 /* Note: we always request a commit of the entire inode */
1418 data->args.offset = 0;
1419 data->args.count = 0;
1420 data->args.context = get_nfs_open_context(first->wb_context);
1421 data->res.count = 0;
1422 data->res.fattr = &data->fattr;
1423 data->res.verf = &data->verf;
1424 nfs_fattr_init(&data->fattr);
1426 EXPORT_SYMBOL_GPL(nfs_init_commit);
1428 void nfs_retry_commit(struct list_head *page_list,
1429 struct pnfs_layout_segment *lseg)
1431 struct nfs_page *req;
1433 while (!list_empty(page_list)) {
1434 req = nfs_list_entry(page_list->next);
1435 nfs_list_remove_request(req);
1436 nfs_mark_request_commit(req, lseg);
1437 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1438 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1440 nfs_clear_page_tag_locked(req);
1443 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1446 * Commit dirty pages
1449 nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1451 struct nfs_write_data *data;
1453 data = nfs_commitdata_alloc();
1458 /* Set up the argument struct */
1459 nfs_init_commit(data, head, NULL);
1460 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
1462 nfs_retry_commit(head, NULL);
1463 nfs_commit_clear_lock(NFS_I(inode));
1468 * COMMIT call returned
1470 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1472 struct nfs_write_data *data = calldata;
1474 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1475 task->tk_pid, task->tk_status);
1477 /* Call the NFS version-specific code */
1478 NFS_PROTO(data->inode)->commit_done(task, data);
1481 void nfs_commit_release_pages(struct nfs_write_data *data)
1483 struct nfs_page *req;
1484 int status = data->task.tk_status;
1486 while (!list_empty(&data->pages)) {
1487 req = nfs_list_entry(data->pages.next);
1488 nfs_list_remove_request(req);
1489 nfs_clear_request_commit(req);
1491 dprintk("NFS: commit (%s/%lld %d@%lld)",
1492 req->wb_context->dentry->d_sb->s_id,
1493 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
1495 (long long)req_offset(req));
1497 nfs_context_set_write_error(req->wb_context, status);
1498 nfs_inode_remove_request(req);
1499 dprintk(", error = %d\n", status);
1503 /* Okay, COMMIT succeeded, apparently. Check the verifier
1504 * returned by the server against all stored verfs. */
1505 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1506 /* We have a match */
1507 nfs_inode_remove_request(req);
1511 /* We have a mismatch. Write the page again */
1512 dprintk(" mismatch\n");
1513 nfs_mark_request_dirty(req);
1515 nfs_clear_page_tag_locked(req);
1518 EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
1520 static void nfs_commit_release(void *calldata)
1522 struct nfs_write_data *data = calldata;
1524 nfs_commit_release_pages(data);
1525 nfs_commit_clear_lock(NFS_I(data->inode));
1526 nfs_commitdata_release(calldata);
1529 static const struct rpc_call_ops nfs_commit_ops = {
1530 #if defined(CONFIG_NFS_V4_1)
1531 .rpc_call_prepare = nfs_write_prepare,
1532 #endif /* CONFIG_NFS_V4_1 */
1533 .rpc_call_done = nfs_commit_done,
1534 .rpc_release = nfs_commit_release,
1537 int nfs_commit_inode(struct inode *inode, int how)
1540 int may_wait = how & FLUSH_SYNC;
1543 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1545 goto out_mark_dirty;
1546 res = nfs_scan_commit(inode, &head, 0, 0);
1550 error = pnfs_commit_list(inode, &head, how);
1551 if (error == PNFS_NOT_ATTEMPTED)
1552 error = nfs_commit_list(inode, &head, how);
1556 goto out_mark_dirty;
1557 error = wait_on_bit(&NFS_I(inode)->flags,
1559 nfs_wait_bit_killable,
1564 nfs_commit_clear_lock(NFS_I(inode));
1566 /* Note: If we exit without ensuring that the commit is complete,
1567 * we must mark the inode as dirty. Otherwise, future calls to
1568 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1569 * that the data is on the disk.
1572 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1576 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1578 struct nfs_inode *nfsi = NFS_I(inode);
1579 int flags = FLUSH_SYNC;
1582 /* no commits means nothing needs to be done */
1586 if (wbc->sync_mode == WB_SYNC_NONE) {
1587 /* Don't commit yet if this is a non-blocking flush and there
1588 * are a lot of outstanding writes for this mapping.
1590 if (nfsi->ncommit <= (nfsi->npages >> 1))
1591 goto out_mark_dirty;
1593 /* don't wait for the COMMIT response */
1597 ret = nfs_commit_inode(inode, flags);
1599 if (wbc->sync_mode == WB_SYNC_NONE) {
1600 if (ret < wbc->nr_to_write)
1601 wbc->nr_to_write -= ret;
1603 wbc->nr_to_write = 0;
1608 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1612 static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1618 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1622 ret = nfs_commit_unstable_pages(inode, wbc);
1623 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
1627 if (wbc->sync_mode == WB_SYNC_NONE)
1630 status = pnfs_layoutcommit_inode(inode, sync);
1638 * flush the inode to disk.
1640 int nfs_wb_all(struct inode *inode)
1642 struct writeback_control wbc = {
1643 .sync_mode = WB_SYNC_ALL,
1644 .nr_to_write = LONG_MAX,
1646 .range_end = LLONG_MAX,
1649 return sync_inode(inode, &wbc);
1652 int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1654 struct nfs_page *req;
1657 BUG_ON(!PageLocked(page));
1659 wait_on_page_writeback(page);
1660 req = nfs_page_find_request(page);
1663 if (nfs_lock_request_dontget(req)) {
1664 nfs_inode_remove_request(req);
1666 * In case nfs_inode_remove_request has marked the
1667 * page as being dirty
1669 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1670 nfs_unlock_request(req);
1673 ret = nfs_wait_on_request(req);
1674 nfs_release_request(req);
1682 * Write back all requests on one page - we do this before reading it.
1684 int nfs_wb_page(struct inode *inode, struct page *page)
1686 loff_t range_start = page_offset(page);
1687 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1688 struct writeback_control wbc = {
1689 .sync_mode = WB_SYNC_ALL,
1691 .range_start = range_start,
1692 .range_end = range_end,
1697 wait_on_page_writeback(page);
1698 if (clear_page_dirty_for_io(page)) {
1699 ret = nfs_writepage_locked(page, &wbc);
1704 if (!PagePrivate(page))
1706 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1715 #ifdef CONFIG_MIGRATION
1716 int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1717 struct page *page, enum migrate_mode mode)
1720 * If PagePrivate is set, then the page is currently associated with
1721 * an in-progress read or write request. Don't try to migrate it.
1723 * FIXME: we could do this in principle, but we'll need a way to ensure
1724 * that we can safely release the inode reference while holding
1727 if (PagePrivate(page))
1730 nfs_fscache_release_page(page, GFP_KERNEL);
1732 return migrate_page(mapping, newpage, page, mode);
1736 int __init nfs_init_writepagecache(void)
1738 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1739 sizeof(struct nfs_write_data),
1740 0, SLAB_HWCACHE_ALIGN,
1742 if (nfs_wdata_cachep == NULL)
1745 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1747 if (nfs_wdata_mempool == NULL)
1748 goto out_destroy_write_cache;
1750 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1752 if (nfs_commit_mempool == NULL)
1753 goto out_destroy_write_mempool;
1756 * NFS congestion size, scale with available memory.
1768 * This allows larger machines to have larger/more transfers.
1769 * Limit the default to 256M
1771 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1772 if (nfs_congestion_kb > 256*1024)
1773 nfs_congestion_kb = 256*1024;
1777 out_destroy_write_mempool:
1778 mempool_destroy(nfs_wdata_mempool);
1779 out_destroy_write_cache:
1780 kmem_cache_destroy(nfs_wdata_cachep);
1784 void nfs_destroy_writepagecache(void)
1786 mempool_destroy(nfs_commit_mempool);
1787 mempool_destroy(nfs_wdata_mempool);
1788 kmem_cache_destroy(nfs_wdata_cachep);