2 * linux/fs/nfs/direct.c
4 * Copyright (C) 2003 by Chuck Lever <cel@netapp.com>
6 * High-performance uncached I/O for the Linux NFS client
8 * There are important applications whose performance or correctness
9 * depends on uncached access to file data. Database clusters
10 * (multiple copies of the same instance running on separate hosts)
11 * implement their own cache coherency protocol that subsumes file
12 * system cache protocols. Applications that process datasets
13 * considerably larger than the client's memory do not always benefit
14 * from a local cache. A streaming video server, for instance, has no
15 * need to cache the contents of a file.
17 * When an application requests uncached I/O, all read and write requests
18 * are made directly to the server; data stored or fetched via these
19 * requests is not cached in the Linux page cache. The client does not
20 * correct unaligned requests from applications. All requested bytes are
21 * held on permanent storage before a direct write system call returns to
24 * Solaris implements an uncached I/O facility called directio() that
25 * is used for backups and sequential I/O to very large files. Solaris
26 * also supports uncaching whole NFS partitions with "-o forcedirectio,"
27 * an undocumented mount option.
29 * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with
30 * help from Andrew Morton.
32 * 18 Dec 2001 Initial implementation for 2.4 --cel
33 * 08 Jul 2002 Version for 2.4.19, with bug fixes --trondmy
34 * 08 Jun 2003 Port to 2.5 APIs --cel
35 * 31 Mar 2004 Handle direct I/O without VFS support --cel
36 * 15 Sep 2004 Parallel async reads --cel
37 * 04 May 2005 support O_DIRECT with aio --cel
41 #include <linux/errno.h>
42 #include <linux/sched.h>
43 #include <linux/kernel.h>
44 #include <linux/smp_lock.h>
45 #include <linux/file.h>
46 #include <linux/pagemap.h>
47 #include <linux/kref.h>
49 #include <linux/nfs_fs.h>
50 #include <linux/nfs_page.h>
51 #include <linux/sunrpc/clnt.h>
53 #include <asm/system.h>
54 #include <asm/uaccess.h>
55 #include <asm/atomic.h>
59 #define NFSDBG_FACILITY NFSDBG_VFS
61 static kmem_cache_t *nfs_direct_cachep;
64 * This represents a set of asynchronous requests that we're waiting on
66 struct nfs_direct_req {
67 struct kref kref; /* release manager */
70 struct list_head list, /* nfs_read/write_data structs */
71 rewrite_list; /* saved nfs_write_data structs */
72 struct nfs_open_context *ctx; /* file open context info */
73 struct kiocb * iocb; /* controlling i/o request */
74 struct inode * inode; /* target file of i/o */
75 unsigned long user_addr; /* location of user's buffer */
76 size_t user_count; /* total bytes to move */
77 loff_t pos; /* starting offset in file */
78 struct page ** pages; /* pages in our buffer */
79 unsigned int npages; /* count of pages */
81 /* completion state */
82 spinlock_t lock; /* protect completion state */
83 int outstanding; /* i/os we're waiting for */
84 ssize_t count, /* bytes actually processed */
85 error; /* any reported error */
86 struct completion completion; /* wait for i/o completion */
89 struct nfs_write_data * commit_data; /* special write_data for commits */
91 #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */
92 #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */
93 struct nfs_writeverf verf; /* unstable write verifier */
96 static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync);
97 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
100 * nfs_direct_IO - NFS address space operation for direct I/O
101 * @rw: direction (read or write)
102 * @iocb: target I/O control block
103 * @iov: array of vectors that define I/O buffer
104 * @pos: offset in file to begin the operation
105 * @nr_segs: size of iovec array
107 * The presence of this routine in the address space ops vector means
108 * the NFS client supports direct I/O. However, we shunt off direct
109 * read and write requests before the VFS gets them, so this method
110 * should never be called.
112 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
114 dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
115 iocb->ki_filp->f_dentry->d_name.name,
116 (long long) pos, nr_segs);
121 static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty)
124 for (i = 0; i < npages; i++) {
125 struct page *page = pages[i];
126 if (do_dirty && !PageCompound(page))
127 set_page_dirty_lock(page);
128 page_cache_release(page);
133 static inline int nfs_get_user_pages(int rw, unsigned long user_addr, size_t size, struct page ***pages)
135 int result = -ENOMEM;
136 unsigned long page_count;
139 page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
140 page_count -= user_addr >> PAGE_SHIFT;
142 array_size = (page_count * sizeof(struct page *));
143 *pages = kmalloc(array_size, GFP_KERNEL);
145 down_read(¤t->mm->mmap_sem);
146 result = get_user_pages(current, current->mm, user_addr,
147 page_count, (rw == READ), 0,
149 up_read(¤t->mm->mmap_sem);
150 if (result != page_count) {
152 * If we got fewer pages than expected from
153 * get_user_pages(), the user buffer runs off the
154 * end of a mapping; return EFAULT.
157 nfs_free_user_pages(*pages, result, 0);
167 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
169 struct nfs_direct_req *dreq;
171 dreq = kmem_cache_alloc(nfs_direct_cachep, SLAB_KERNEL);
175 kref_init(&dreq->kref);
176 init_completion(&dreq->completion);
177 INIT_LIST_HEAD(&dreq->list);
178 INIT_LIST_HEAD(&dreq->rewrite_list);
181 spin_lock_init(&dreq->lock);
182 dreq->outstanding = 0;
190 static void nfs_direct_req_release(struct kref *kref)
192 struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
194 if (dreq->ctx != NULL)
195 put_nfs_open_context(dreq->ctx);
196 kmem_cache_free(nfs_direct_cachep, dreq);
200 * Collects and returns the final error value/byte-count.
202 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
204 ssize_t result = -EIOCBQUEUED;
206 /* Async requests don't wait here */
210 result = wait_for_completion_interruptible(&dreq->completion);
213 result = dreq->error;
215 result = dreq->count;
218 kref_put(&dreq->kref, nfs_direct_req_release);
219 return (ssize_t) result;
223 * We must hold a reference to all the pages in this direct read request
224 * until the RPCs complete. This could be long *after* we are woken up in
225 * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
227 * In addition, synchronous I/O uses a stack-allocated iocb. Thus we
228 * can't trust the iocb is still valid here if this is a synchronous
229 * request. If the waiter is woken prematurely, the iocb is long gone.
231 static void nfs_direct_complete(struct nfs_direct_req *dreq)
233 nfs_free_user_pages(dreq->pages, dreq->npages, 1);
236 long res = (long) dreq->error;
238 res = (long) dreq->count;
239 aio_complete(dreq->iocb, res, 0);
241 complete_all(&dreq->completion);
243 kref_put(&dreq->kref, nfs_direct_req_release);
247 * Note we also set the number of requests we have in the dreq when we are
248 * done. This prevents races with I/O completion so we will always wait
249 * until all requests have been dispatched and completed.
251 static struct nfs_direct_req *nfs_direct_read_alloc(size_t nbytes, size_t rsize)
253 struct list_head *list;
254 struct nfs_direct_req *dreq;
255 unsigned int rpages = (rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
257 dreq = nfs_direct_req_alloc();
263 struct nfs_read_data *data = nfs_readdata_alloc(rpages);
265 if (unlikely(!data)) {
266 while (!list_empty(list)) {
267 data = list_entry(list->next,
268 struct nfs_read_data, pages);
269 list_del(&data->pages);
270 nfs_readdata_free(data);
272 kref_put(&dreq->kref, nfs_direct_req_release);
276 INIT_LIST_HEAD(&data->pages);
277 list_add(&data->pages, list);
279 data->req = (struct nfs_page *) dreq;
285 kref_get(&dreq->kref);
289 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
291 struct nfs_read_data *data = calldata;
292 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
294 if (nfs_readpage_result(task, data) != 0)
297 spin_lock(&dreq->lock);
299 if (likely(task->tk_status >= 0))
300 dreq->count += data->res.count;
302 dreq->error = task->tk_status;
304 if (--dreq->outstanding) {
305 spin_unlock(&dreq->lock);
309 spin_unlock(&dreq->lock);
310 nfs_direct_complete(dreq);
313 static const struct rpc_call_ops nfs_read_direct_ops = {
314 .rpc_call_done = nfs_direct_read_result,
315 .rpc_release = nfs_readdata_release,
319 * For each nfs_read_data struct that was allocated on the list, dispatch
320 * an NFS READ operation
322 static void nfs_direct_read_schedule(struct nfs_direct_req *dreq)
324 struct nfs_open_context *ctx = dreq->ctx;
325 struct inode *inode = ctx->dentry->d_inode;
326 struct list_head *list = &dreq->list;
327 struct page **pages = dreq->pages;
328 size_t count = dreq->user_count;
329 loff_t pos = dreq->pos;
330 size_t rsize = NFS_SERVER(inode)->rsize;
331 unsigned int curpage, pgbase;
334 pgbase = dreq->user_addr & ~PAGE_MASK;
336 struct nfs_read_data *data;
343 BUG_ON(list_empty(list));
344 data = list_entry(list->next, struct nfs_read_data, pages);
345 list_del_init(&data->pages);
348 data->cred = ctx->cred;
349 data->args.fh = NFS_FH(inode);
350 data->args.context = ctx;
351 data->args.offset = pos;
352 data->args.pgbase = pgbase;
353 data->args.pages = &pages[curpage];
354 data->args.count = bytes;
355 data->res.fattr = &data->fattr;
357 data->res.count = bytes;
359 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
360 &nfs_read_direct_ops, data);
361 NFS_PROTO(inode)->read_setup(data);
363 data->task.tk_cookie = (unsigned long) inode;
366 rpc_execute(&data->task);
369 dfprintk(VFS, "NFS: %5u initiated direct read call (req %s/%Ld, %zu bytes @ offset %Lu)\n",
372 (long long)NFS_FILEID(inode),
374 (unsigned long long)data->args.offset);
378 curpage += pgbase >> PAGE_SHIFT;
379 pgbase &= ~PAGE_MASK;
382 } while (count != 0);
383 BUG_ON(!list_empty(list));
386 static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, unsigned int nr_pages)
390 struct inode *inode = iocb->ki_filp->f_mapping->host;
391 struct rpc_clnt *clnt = NFS_CLIENT(inode);
392 struct nfs_direct_req *dreq;
394 dreq = nfs_direct_read_alloc(count, NFS_SERVER(inode)->rsize);
398 dreq->user_addr = user_addr;
399 dreq->user_count = count;
402 dreq->npages = nr_pages;
404 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
405 if (!is_sync_kiocb(iocb))
408 nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
409 rpc_clnt_sigmask(clnt, &oldset);
410 nfs_direct_read_schedule(dreq);
411 result = nfs_direct_wait(dreq);
412 rpc_clnt_sigunmask(clnt, &oldset);
417 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
419 list_splice_init(&dreq->rewrite_list, &dreq->list);
420 while (!list_empty(&dreq->list)) {
421 struct nfs_write_data *data = list_entry(dreq->list.next, struct nfs_write_data, pages);
422 list_del(&data->pages);
423 nfs_writedata_release(data);
427 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
428 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
430 struct list_head *pos;
432 list_splice_init(&dreq->rewrite_list, &dreq->list);
433 list_for_each(pos, &dreq->list)
437 nfs_direct_write_schedule(dreq, FLUSH_STABLE);
440 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
442 struct nfs_write_data *data = calldata;
443 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
445 /* Call the NFS version-specific code */
446 if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
448 if (unlikely(task->tk_status < 0)) {
449 dreq->error = task->tk_status;
450 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
452 if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
453 dprintk("NFS: %5u commit verify failed\n", task->tk_pid);
454 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
457 dprintk("NFS: %5u commit returned %d\n", task->tk_pid, task->tk_status);
458 nfs_direct_write_complete(dreq, data->inode);
461 static const struct rpc_call_ops nfs_commit_direct_ops = {
462 .rpc_call_done = nfs_direct_commit_result,
463 .rpc_release = nfs_commit_release,
466 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
468 struct nfs_write_data *data = dreq->commit_data;
470 data->inode = dreq->inode;
471 data->cred = dreq->ctx->cred;
473 data->args.fh = NFS_FH(data->inode);
474 data->args.offset = dreq->pos;
475 data->args.count = dreq->user_count;
477 data->res.fattr = &data->fattr;
478 data->res.verf = &data->verf;
480 rpc_init_task(&data->task, NFS_CLIENT(dreq->inode), RPC_TASK_ASYNC,
481 &nfs_commit_direct_ops, data);
482 NFS_PROTO(data->inode)->commit_setup(data, 0);
484 data->task.tk_priority = RPC_PRIORITY_NORMAL;
485 data->task.tk_cookie = (unsigned long)data->inode;
486 /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
487 dreq->commit_data = NULL;
489 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
492 rpc_execute(&data->task);
496 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
498 int flags = dreq->flags;
502 case NFS_ODIRECT_DO_COMMIT:
503 nfs_direct_commit_schedule(dreq);
505 case NFS_ODIRECT_RESCHED_WRITES:
506 nfs_direct_write_reschedule(dreq);
509 nfs_end_data_update(inode);
510 if (dreq->commit_data != NULL)
511 nfs_commit_free(dreq->commit_data);
512 nfs_direct_free_writedata(dreq);
513 nfs_direct_complete(dreq);
517 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
519 dreq->commit_data = nfs_commit_alloc(0);
520 if (dreq->commit_data != NULL)
521 dreq->commit_data->req = (struct nfs_page *) dreq;
524 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
526 dreq->commit_data = NULL;
529 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
531 nfs_end_data_update(inode);
532 nfs_direct_free_writedata(dreq);
533 nfs_direct_complete(dreq);
537 static struct nfs_direct_req *nfs_direct_write_alloc(size_t nbytes, size_t wsize)
539 struct list_head *list;
540 struct nfs_direct_req *dreq;
541 unsigned int wpages = (wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
543 dreq = nfs_direct_req_alloc();
549 struct nfs_write_data *data = nfs_writedata_alloc(wpages);
551 if (unlikely(!data)) {
552 while (!list_empty(list)) {
553 data = list_entry(list->next,
554 struct nfs_write_data, pages);
555 list_del(&data->pages);
556 nfs_writedata_free(data);
558 kref_put(&dreq->kref, nfs_direct_req_release);
562 INIT_LIST_HEAD(&data->pages);
563 list_add(&data->pages, list);
565 data->req = (struct nfs_page *) dreq;
572 nfs_alloc_commit_data(dreq);
574 kref_get(&dreq->kref);
578 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
580 struct nfs_write_data *data = calldata;
581 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
582 int status = task->tk_status;
584 if (nfs_writeback_done(task, data) != 0)
587 spin_lock(&dreq->lock);
589 if (likely(status >= 0))
590 dreq->count += data->res.count;
592 dreq->error = task->tk_status;
594 if (data->res.verf->committed != NFS_FILE_SYNC) {
595 switch (dreq->flags) {
597 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
598 dreq->flags = NFS_ODIRECT_DO_COMMIT;
600 case NFS_ODIRECT_DO_COMMIT:
601 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
602 dprintk("NFS: %5u write verify failed\n", task->tk_pid);
603 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
607 /* In case we have to resend */
608 data->args.stable = NFS_FILE_SYNC;
610 spin_unlock(&dreq->lock);
614 * NB: Return the value of the first error return code. Subsequent
615 * errors after the first one are ignored.
617 static void nfs_direct_write_release(void *calldata)
619 struct nfs_write_data *data = calldata;
620 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
622 spin_lock(&dreq->lock);
623 if (--dreq->outstanding) {
624 spin_unlock(&dreq->lock);
627 spin_unlock(&dreq->lock);
629 nfs_direct_write_complete(dreq, data->inode);
632 static const struct rpc_call_ops nfs_write_direct_ops = {
633 .rpc_call_done = nfs_direct_write_result,
634 .rpc_release = nfs_direct_write_release,
638 * For each nfs_write_data struct that was allocated on the list, dispatch
639 * an NFS WRITE operation
641 static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync)
643 struct nfs_open_context *ctx = dreq->ctx;
644 struct inode *inode = ctx->dentry->d_inode;
645 struct list_head *list = &dreq->list;
646 struct page **pages = dreq->pages;
647 size_t count = dreq->user_count;
648 loff_t pos = dreq->pos;
649 size_t wsize = NFS_SERVER(inode)->wsize;
650 unsigned int curpage, pgbase;
653 pgbase = dreq->user_addr & ~PAGE_MASK;
655 struct nfs_write_data *data;
662 BUG_ON(list_empty(list));
663 data = list_entry(list->next, struct nfs_write_data, pages);
664 list_move_tail(&data->pages, &dreq->rewrite_list);
667 data->cred = ctx->cred;
668 data->args.fh = NFS_FH(inode);
669 data->args.context = ctx;
670 data->args.offset = pos;
671 data->args.pgbase = pgbase;
672 data->args.pages = &pages[curpage];
673 data->args.count = bytes;
674 data->res.fattr = &data->fattr;
675 data->res.count = bytes;
676 data->res.verf = &data->verf;
678 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
679 &nfs_write_direct_ops, data);
680 NFS_PROTO(inode)->write_setup(data, sync);
682 data->task.tk_priority = RPC_PRIORITY_NORMAL;
683 data->task.tk_cookie = (unsigned long) inode;
686 rpc_execute(&data->task);
689 dfprintk(VFS, "NFS: %5u initiated direct write call (req %s/%Ld, %zu bytes @ offset %Lu)\n",
692 (long long)NFS_FILEID(inode),
694 (unsigned long long)data->args.offset);
698 curpage += pgbase >> PAGE_SHIFT;
699 pgbase &= ~PAGE_MASK;
702 } while (count != 0);
703 BUG_ON(!list_empty(list));
706 static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, int nr_pages)
710 struct inode *inode = iocb->ki_filp->f_mapping->host;
711 struct rpc_clnt *clnt = NFS_CLIENT(inode);
712 struct nfs_direct_req *dreq;
713 size_t wsize = NFS_SERVER(inode)->wsize;
716 dreq = nfs_direct_write_alloc(count, wsize);
719 if (dreq->commit_data == NULL || count < wsize)
722 dreq->user_addr = user_addr;
723 dreq->user_count = count;
726 dreq->npages = nr_pages;
728 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
729 if (!is_sync_kiocb(iocb))
732 nfs_add_stats(inode, NFSIOS_DIRECTWRITTENBYTES, count);
734 nfs_begin_data_update(inode);
736 rpc_clnt_sigmask(clnt, &oldset);
737 nfs_direct_write_schedule(dreq, sync);
738 result = nfs_direct_wait(dreq);
739 rpc_clnt_sigunmask(clnt, &oldset);
745 * nfs_file_direct_read - file direct read operation for NFS files
746 * @iocb: target I/O control block
747 * @buf: user's buffer into which to read data
748 * @count: number of bytes to read
749 * @pos: byte offset in file where reading starts
751 * We use this function for direct reads instead of calling
752 * generic_file_aio_read() in order to avoid gfar's check to see if
753 * the request starts before the end of the file. For that check
754 * to work, we must generate a GETATTR before each direct read, and
755 * even then there is a window between the GETATTR and the subsequent
756 * READ where the file size could change. Our preference is simply
757 * to do all reads the application wants, and the server will take
758 * care of managing the end of file boundary.
760 * This function also eliminates unnecessarily updating the file's
761 * atime locally, as the NFS server sets the file's atime, and this
762 * client must read the updated atime from the server back into its
765 ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
767 ssize_t retval = -EINVAL;
770 struct file *file = iocb->ki_filp;
771 struct address_space *mapping = file->f_mapping;
773 dprintk("nfs: direct read(%s/%s, %lu@%Ld)\n",
774 file->f_dentry->d_parent->d_name.name,
775 file->f_dentry->d_name.name,
776 (unsigned long) count, (long long) pos);
781 if (!access_ok(VERIFY_WRITE, buf, count))
787 retval = nfs_sync_mapping(mapping);
791 retval = nfs_get_user_pages(READ, (unsigned long) buf,
797 retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos,
800 iocb->ki_pos = pos + retval;
807 * nfs_file_direct_write - file direct write operation for NFS files
808 * @iocb: target I/O control block
809 * @buf: user's buffer from which to write data
810 * @count: number of bytes to write
811 * @pos: byte offset in file where writing starts
813 * We use this function for direct writes instead of calling
814 * generic_file_aio_write() in order to avoid taking the inode
815 * semaphore and updating the i_size. The NFS server will set
816 * the new i_size and this client must read the updated size
817 * back into its cache. We let the server do generic write
818 * parameter checking and report problems.
820 * We also avoid an unnecessary invocation of generic_osync_inode(),
821 * as it is fairly meaningless to sync the metadata of an NFS file.
823 * We eliminate local atime updates, see direct read above.
825 * We avoid unnecessary page cache invalidations for normal cached
826 * readers of this file.
828 * Note that O_APPEND is not supported for NFS direct writes, as there
829 * is no atomic O_APPEND write facility in the NFS protocol.
831 ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
836 struct file *file = iocb->ki_filp;
837 struct address_space *mapping = file->f_mapping;
839 dfprintk(VFS, "nfs: direct write(%s/%s, %lu@%Ld)\n",
840 file->f_dentry->d_parent->d_name.name,
841 file->f_dentry->d_name.name,
842 (unsigned long) count, (long long) pos);
844 retval = generic_write_checks(file, &pos, &count, 0);
849 if ((ssize_t) count < 0)
856 if (!access_ok(VERIFY_READ, buf, count))
859 retval = nfs_sync_mapping(mapping);
863 retval = nfs_get_user_pages(WRITE, (unsigned long) buf,
869 retval = nfs_direct_write(iocb, (unsigned long) buf, count,
870 pos, pages, page_count);
873 * XXX: nfs_end_data_update() already ensures this file's
874 * cached data is subsequently invalidated. Do we really
875 * need to call invalidate_inode_pages2() again here?
877 * For aio writes, this invalidation will almost certainly
878 * occur before the writes complete. Kind of racey.
880 if (mapping->nrpages)
881 invalidate_inode_pages2(mapping);
884 iocb->ki_pos = pos + retval;
891 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
894 int __init nfs_init_directcache(void)
896 nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
897 sizeof(struct nfs_direct_req),
898 0, (SLAB_RECLAIM_ACCOUNT|
901 if (nfs_direct_cachep == NULL)
908 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
911 void nfs_destroy_directcache(void)
913 if (kmem_cache_destroy(nfs_direct_cachep))
914 printk(KERN_INFO "nfs_direct_cache: not all structures were freed\n");