]> nv-tegra.nvidia Code Review - linux-3.10.git/blob - fs/nfs/direct.c
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-3.10.git] / fs / nfs / direct.c
1 /*
2  * linux/fs/nfs/direct.c
3  *
4  * Copyright (C) 2003 by Chuck Lever <cel@netapp.com>
5  *
6  * High-performance uncached I/O for the Linux NFS client
7  *
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.
16  *
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
22  * an application.
23  *
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.
28  *
29  * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with
30  * help from Andrew Morton.
31  *
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
38  *
39  */
40
41 #include <linux/errno.h>
42 #include <linux/sched.h>
43 #include <linux/kernel.h>
44 #include <linux/file.h>
45 #include <linux/pagemap.h>
46 #include <linux/kref.h>
47 #include <linux/slab.h>
48 #include <linux/task_io_accounting_ops.h>
49 #include <linux/module.h>
50
51 #include <linux/nfs_fs.h>
52 #include <linux/nfs_page.h>
53 #include <linux/sunrpc/clnt.h>
54
55 #include <asm/uaccess.h>
56 #include <linux/atomic.h>
57
58 #include "internal.h"
59 #include "iostat.h"
60 #include "pnfs.h"
61
62 #define NFSDBG_FACILITY         NFSDBG_VFS
63
64 static struct kmem_cache *nfs_direct_cachep;
65
66 /*
67  * This represents a set of asynchronous requests that we're waiting on
68  */
69 struct nfs_direct_req {
70         struct kref             kref;           /* release manager */
71
72         /* I/O parameters */
73         struct nfs_open_context *ctx;           /* file open context info */
74         struct nfs_lock_context *l_ctx;         /* Lock context info */
75         struct kiocb *          iocb;           /* controlling i/o request */
76         struct inode *          inode;          /* target file of i/o */
77
78         /* completion state */
79         atomic_t                io_count;       /* i/os we're waiting for */
80         spinlock_t              lock;           /* protect completion state */
81         ssize_t                 count,          /* bytes actually processed */
82                                 bytes_left,     /* bytes left to be sent */
83                                 error;          /* any reported error */
84         struct completion       completion;     /* wait for i/o completion */
85
86         /* commit state */
87         struct nfs_mds_commit_info mds_cinfo;   /* Storage for cinfo */
88         struct pnfs_ds_commit_info ds_cinfo;    /* Storage for cinfo */
89         struct work_struct      work;
90         int                     flags;
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 */
94 };
95
96 static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops;
97 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops;
98 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
99 static void nfs_direct_write_schedule_work(struct work_struct *work);
100
101 static inline void get_dreq(struct nfs_direct_req *dreq)
102 {
103         atomic_inc(&dreq->io_count);
104 }
105
106 static inline int put_dreq(struct nfs_direct_req *dreq)
107 {
108         return atomic_dec_and_test(&dreq->io_count);
109 }
110
111 /**
112  * nfs_direct_IO - NFS address space operation for direct I/O
113  * @rw: direction (read or write)
114  * @iocb: target I/O control block
115  * @iov: array of vectors that define I/O buffer
116  * @pos: offset in file to begin the operation
117  * @nr_segs: size of iovec array
118  *
119  * The presence of this routine in the address space ops vector means
120  * the NFS client supports direct I/O. However, for most direct IO, we
121  * shunt off direct read and write requests before the VFS gets them,
122  * so this method is only ever called for swap.
123  */
124 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
125 {
126 #ifndef CONFIG_NFS_SWAP
127         dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
128                         iocb->ki_filp->f_path.dentry->d_name.name,
129                         (long long) pos, nr_segs);
130
131         return -EINVAL;
132 #else
133         VM_BUG_ON(iocb->ki_left != PAGE_SIZE);
134         VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE);
135
136         if (rw == READ || rw == KERNEL_READ)
137                 return nfs_file_direct_read(iocb, iov, nr_segs, pos,
138                                 rw == READ ? true : false);
139         return nfs_file_direct_write(iocb, iov, nr_segs, pos,
140                                 rw == WRITE ? true : false);
141 #endif /* CONFIG_NFS_SWAP */
142 }
143
144 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
145 {
146         unsigned int i;
147         for (i = 0; i < npages; i++)
148                 page_cache_release(pages[i]);
149 }
150
151 void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo,
152                               struct nfs_direct_req *dreq)
153 {
154         cinfo->lock = &dreq->lock;
155         cinfo->mds = &dreq->mds_cinfo;
156         cinfo->ds = &dreq->ds_cinfo;
157         cinfo->dreq = dreq;
158         cinfo->completion_ops = &nfs_direct_commit_completion_ops;
159 }
160
161 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
162 {
163         struct nfs_direct_req *dreq;
164
165         dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL);
166         if (!dreq)
167                 return NULL;
168
169         kref_init(&dreq->kref);
170         kref_get(&dreq->kref);
171         init_completion(&dreq->completion);
172         INIT_LIST_HEAD(&dreq->mds_cinfo.list);
173         INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
174         spin_lock_init(&dreq->lock);
175
176         return dreq;
177 }
178
179 static void nfs_direct_req_free(struct kref *kref)
180 {
181         struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
182
183         if (dreq->l_ctx != NULL)
184                 nfs_put_lock_context(dreq->l_ctx);
185         if (dreq->ctx != NULL)
186                 put_nfs_open_context(dreq->ctx);
187         kmem_cache_free(nfs_direct_cachep, dreq);
188 }
189
190 static void nfs_direct_req_release(struct nfs_direct_req *dreq)
191 {
192         kref_put(&dreq->kref, nfs_direct_req_free);
193 }
194
195 ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq)
196 {
197         return dreq->bytes_left;
198 }
199 EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left);
200
201 /*
202  * Collects and returns the final error value/byte-count.
203  */
204 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
205 {
206         ssize_t result = -EIOCBQUEUED;
207
208         /* Async requests don't wait here */
209         if (dreq->iocb)
210                 goto out;
211
212         result = wait_for_completion_killable(&dreq->completion);
213
214         if (!result)
215                 result = dreq->error;
216         if (!result)
217                 result = dreq->count;
218
219 out:
220         return (ssize_t) result;
221 }
222
223 /*
224  * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
225  * the iocb is still valid here if this is a synchronous request.
226  */
227 static void nfs_direct_complete(struct nfs_direct_req *dreq)
228 {
229         if (dreq->iocb) {
230                 long res = (long) dreq->error;
231                 if (!res)
232                         res = (long) dreq->count;
233                 aio_complete(dreq->iocb, res, 0);
234         }
235         complete_all(&dreq->completion);
236
237         nfs_direct_req_release(dreq);
238 }
239
240 static void nfs_direct_readpage_release(struct nfs_page *req)
241 {
242         dprintk("NFS: direct read done (%s/%lld %d@%lld)\n",
243                 req->wb_context->dentry->d_inode->i_sb->s_id,
244                 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
245                 req->wb_bytes,
246                 (long long)req_offset(req));
247         nfs_release_request(req);
248 }
249
250 static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
251 {
252         unsigned long bytes = 0;
253         struct nfs_direct_req *dreq = hdr->dreq;
254
255         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
256                 goto out_put;
257
258         spin_lock(&dreq->lock);
259         if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0))
260                 dreq->error = hdr->error;
261         else
262                 dreq->count += hdr->good_bytes;
263         spin_unlock(&dreq->lock);
264
265         while (!list_empty(&hdr->pages)) {
266                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
267                 struct page *page = req->wb_page;
268
269                 if (!PageCompound(page) && bytes < hdr->good_bytes)
270                         set_page_dirty(page);
271                 bytes += req->wb_bytes;
272                 nfs_list_remove_request(req);
273                 nfs_direct_readpage_release(req);
274         }
275 out_put:
276         if (put_dreq(dreq))
277                 nfs_direct_complete(dreq);
278         hdr->release(hdr);
279 }
280
281 static void nfs_read_sync_pgio_error(struct list_head *head)
282 {
283         struct nfs_page *req;
284
285         while (!list_empty(head)) {
286                 req = nfs_list_entry(head->next);
287                 nfs_list_remove_request(req);
288                 nfs_release_request(req);
289         }
290 }
291
292 static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr)
293 {
294         get_dreq(hdr->dreq);
295 }
296
297 static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = {
298         .error_cleanup = nfs_read_sync_pgio_error,
299         .init_hdr = nfs_direct_pgio_init,
300         .completion = nfs_direct_read_completion,
301 };
302
303 /*
304  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
305  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
306  * bail and stop sending more reads.  Read length accounting is
307  * handled automatically by nfs_direct_read_result().  Otherwise, if
308  * no requests have been sent, just return an error.
309  */
310 static ssize_t nfs_direct_read_schedule_segment(struct nfs_pageio_descriptor *desc,
311                                                 const struct iovec *iov,
312                                                 loff_t pos, bool uio)
313 {
314         struct nfs_direct_req *dreq = desc->pg_dreq;
315         struct nfs_open_context *ctx = dreq->ctx;
316         struct inode *inode = ctx->dentry->d_inode;
317         unsigned long user_addr = (unsigned long)iov->iov_base;
318         size_t count = iov->iov_len;
319         size_t rsize = NFS_SERVER(inode)->rsize;
320         unsigned int pgbase;
321         int result;
322         ssize_t started = 0;
323         struct page **pagevec = NULL;
324         unsigned int npages;
325
326         do {
327                 size_t bytes;
328                 int i;
329
330                 pgbase = user_addr & ~PAGE_MASK;
331                 bytes = min(max_t(size_t, rsize, PAGE_SIZE), count);
332
333                 result = -ENOMEM;
334                 npages = nfs_page_array_len(pgbase, bytes);
335                 if (!pagevec)
336                         pagevec = kmalloc(npages * sizeof(struct page *),
337                                           GFP_KERNEL);
338                 if (!pagevec)
339                         break;
340                 if (uio) {
341                         down_read(&current->mm->mmap_sem);
342                         result = get_user_pages(current, current->mm, user_addr,
343                                         npages, 1, 0, pagevec, NULL);
344                         up_read(&current->mm->mmap_sem);
345                         if (result < 0)
346                                 break;
347                 } else {
348                         WARN_ON(npages != 1);
349                         result = get_kernel_page(user_addr, 1, pagevec);
350                         if (WARN_ON(result != 1))
351                                 break;
352                 }
353
354                 if ((unsigned)result < npages) {
355                         bytes = result * PAGE_SIZE;
356                         if (bytes <= pgbase) {
357                                 nfs_direct_release_pages(pagevec, result);
358                                 break;
359                         }
360                         bytes -= pgbase;
361                         npages = result;
362                 }
363
364                 for (i = 0; i < npages; i++) {
365                         struct nfs_page *req;
366                         unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
367                         /* XXX do we need to do the eof zeroing found in async_filler? */
368                         req = nfs_create_request(dreq->ctx, dreq->inode,
369                                                  pagevec[i],
370                                                  pgbase, req_len);
371                         if (IS_ERR(req)) {
372                                 result = PTR_ERR(req);
373                                 break;
374                         }
375                         req->wb_index = pos >> PAGE_SHIFT;
376                         req->wb_offset = pos & ~PAGE_MASK;
377                         if (!nfs_pageio_add_request(desc, req)) {
378                                 result = desc->pg_error;
379                                 nfs_release_request(req);
380                                 break;
381                         }
382                         pgbase = 0;
383                         bytes -= req_len;
384                         started += req_len;
385                         user_addr += req_len;
386                         pos += req_len;
387                         count -= req_len;
388                         dreq->bytes_left -= req_len;
389                 }
390                 /* The nfs_page now hold references to these pages */
391                 nfs_direct_release_pages(pagevec, npages);
392         } while (count != 0 && result >= 0);
393
394         kfree(pagevec);
395
396         if (started)
397                 return started;
398         return result < 0 ? (ssize_t) result : -EFAULT;
399 }
400
401 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
402                                               const struct iovec *iov,
403                                               unsigned long nr_segs,
404                                               loff_t pos, bool uio)
405 {
406         struct nfs_pageio_descriptor desc;
407         ssize_t result = -EINVAL;
408         size_t requested_bytes = 0;
409         unsigned long seg;
410
411         NFS_PROTO(dreq->inode)->read_pageio_init(&desc, dreq->inode,
412                              &nfs_direct_read_completion_ops);
413         get_dreq(dreq);
414         desc.pg_dreq = dreq;
415
416         for (seg = 0; seg < nr_segs; seg++) {
417                 const struct iovec *vec = &iov[seg];
418                 result = nfs_direct_read_schedule_segment(&desc, vec, pos, uio);
419                 if (result < 0)
420                         break;
421                 requested_bytes += result;
422                 if ((size_t)result < vec->iov_len)
423                         break;
424                 pos += vec->iov_len;
425         }
426
427         nfs_pageio_complete(&desc);
428
429         /*
430          * If no bytes were started, return the error, and let the
431          * generic layer handle the completion.
432          */
433         if (requested_bytes == 0) {
434                 nfs_direct_req_release(dreq);
435                 return result < 0 ? result : -EIO;
436         }
437
438         if (put_dreq(dreq))
439                 nfs_direct_complete(dreq);
440         return 0;
441 }
442
443 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
444                                unsigned long nr_segs, loff_t pos, bool uio)
445 {
446         ssize_t result = -ENOMEM;
447         struct inode *inode = iocb->ki_filp->f_mapping->host;
448         struct nfs_direct_req *dreq;
449         struct nfs_lock_context *l_ctx;
450
451         dreq = nfs_direct_req_alloc();
452         if (dreq == NULL)
453                 goto out;
454
455         dreq->inode = inode;
456         dreq->bytes_left = iov_length(iov, nr_segs);
457         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
458         l_ctx = nfs_get_lock_context(dreq->ctx);
459         if (IS_ERR(l_ctx)) {
460                 result = PTR_ERR(l_ctx);
461                 goto out_release;
462         }
463         dreq->l_ctx = l_ctx;
464         if (!is_sync_kiocb(iocb))
465                 dreq->iocb = iocb;
466
467         NFS_I(inode)->read_io += iov_length(iov, nr_segs);
468         result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos, uio);
469         if (!result)
470                 result = nfs_direct_wait(dreq);
471 out_release:
472         nfs_direct_req_release(dreq);
473 out:
474         return result;
475 }
476
477 static void nfs_inode_dio_write_done(struct inode *inode)
478 {
479         nfs_zap_mapping(inode, inode->i_mapping);
480         inode_dio_done(inode);
481 }
482
483 #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
484 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
485 {
486         struct nfs_pageio_descriptor desc;
487         struct nfs_page *req, *tmp;
488         LIST_HEAD(reqs);
489         struct nfs_commit_info cinfo;
490         LIST_HEAD(failed);
491
492         nfs_init_cinfo_from_dreq(&cinfo, dreq);
493         pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo);
494         spin_lock(cinfo.lock);
495         nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0);
496         spin_unlock(cinfo.lock);
497
498         dreq->count = 0;
499         get_dreq(dreq);
500
501         NFS_PROTO(dreq->inode)->write_pageio_init(&desc, dreq->inode, FLUSH_STABLE,
502                               &nfs_direct_write_completion_ops);
503         desc.pg_dreq = dreq;
504
505         list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
506                 if (!nfs_pageio_add_request(&desc, req)) {
507                         nfs_list_remove_request(req);
508                         nfs_list_add_request(req, &failed);
509                         spin_lock(cinfo.lock);
510                         dreq->flags = 0;
511                         dreq->error = -EIO;
512                         spin_unlock(cinfo.lock);
513                 }
514                 nfs_release_request(req);
515         }
516         nfs_pageio_complete(&desc);
517
518         while (!list_empty(&failed)) {
519                 req = nfs_list_entry(failed.next);
520                 nfs_list_remove_request(req);
521                 nfs_unlock_and_release_request(req);
522         }
523
524         if (put_dreq(dreq))
525                 nfs_direct_write_complete(dreq, dreq->inode);
526 }
527
528 static void nfs_direct_commit_complete(struct nfs_commit_data *data)
529 {
530         struct nfs_direct_req *dreq = data->dreq;
531         struct nfs_commit_info cinfo;
532         struct nfs_page *req;
533         int status = data->task.tk_status;
534
535         nfs_init_cinfo_from_dreq(&cinfo, dreq);
536         if (status < 0) {
537                 dprintk("NFS: %5u commit failed with error %d.\n",
538                         data->task.tk_pid, status);
539                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
540         } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
541                 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
542                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
543         }
544
545         dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
546         while (!list_empty(&data->pages)) {
547                 req = nfs_list_entry(data->pages.next);
548                 nfs_list_remove_request(req);
549                 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) {
550                         /* Note the rewrite will go through mds */
551                         nfs_mark_request_commit(req, NULL, &cinfo);
552                 } else
553                         nfs_release_request(req);
554                 nfs_unlock_and_release_request(req);
555         }
556
557         if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
558                 nfs_direct_write_complete(dreq, data->inode);
559 }
560
561 static void nfs_direct_error_cleanup(struct nfs_inode *nfsi)
562 {
563         /* There is no lock to clear */
564 }
565
566 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = {
567         .completion = nfs_direct_commit_complete,
568         .error_cleanup = nfs_direct_error_cleanup,
569 };
570
571 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
572 {
573         int res;
574         struct nfs_commit_info cinfo;
575         LIST_HEAD(mds_list);
576
577         nfs_init_cinfo_from_dreq(&cinfo, dreq);
578         nfs_scan_commit(dreq->inode, &mds_list, &cinfo);
579         res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo);
580         if (res < 0) /* res == -ENOMEM */
581                 nfs_direct_write_reschedule(dreq);
582 }
583
584 static void nfs_direct_write_schedule_work(struct work_struct *work)
585 {
586         struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work);
587         int flags = dreq->flags;
588
589         dreq->flags = 0;
590         switch (flags) {
591                 case NFS_ODIRECT_DO_COMMIT:
592                         nfs_direct_commit_schedule(dreq);
593                         break;
594                 case NFS_ODIRECT_RESCHED_WRITES:
595                         nfs_direct_write_reschedule(dreq);
596                         break;
597                 default:
598                         nfs_inode_dio_write_done(dreq->inode);
599                         nfs_direct_complete(dreq);
600         }
601 }
602
603 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
604 {
605         schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */
606 }
607
608 #else
609 static void nfs_direct_write_schedule_work(struct work_struct *work)
610 {
611 }
612
613 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
614 {
615         nfs_inode_dio_write_done(inode);
616         nfs_direct_complete(dreq);
617 }
618 #endif
619
620 /*
621  * NB: Return the value of the first error return code.  Subsequent
622  *     errors after the first one are ignored.
623  */
624 /*
625  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
626  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
627  * bail and stop sending more writes.  Write length accounting is
628  * handled automatically by nfs_direct_write_result().  Otherwise, if
629  * no requests have been sent, just return an error.
630  */
631 static ssize_t nfs_direct_write_schedule_segment(struct nfs_pageio_descriptor *desc,
632                                                  const struct iovec *iov,
633                                                  loff_t pos, bool uio)
634 {
635         struct nfs_direct_req *dreq = desc->pg_dreq;
636         struct nfs_open_context *ctx = dreq->ctx;
637         struct inode *inode = ctx->dentry->d_inode;
638         unsigned long user_addr = (unsigned long)iov->iov_base;
639         size_t count = iov->iov_len;
640         size_t wsize = NFS_SERVER(inode)->wsize;
641         unsigned int pgbase;
642         int result;
643         ssize_t started = 0;
644         struct page **pagevec = NULL;
645         unsigned int npages;
646
647         do {
648                 size_t bytes;
649                 int i;
650
651                 pgbase = user_addr & ~PAGE_MASK;
652                 bytes = min(max_t(size_t, wsize, PAGE_SIZE), count);
653
654                 result = -ENOMEM;
655                 npages = nfs_page_array_len(pgbase, bytes);
656                 if (!pagevec)
657                         pagevec = kmalloc(npages * sizeof(struct page *), GFP_KERNEL);
658                 if (!pagevec)
659                         break;
660
661                 if (uio) {
662                         down_read(&current->mm->mmap_sem);
663                         result = get_user_pages(current, current->mm, user_addr,
664                                                 npages, 0, 0, pagevec, NULL);
665                         up_read(&current->mm->mmap_sem);
666                         if (result < 0)
667                                 break;
668                 } else {
669                         WARN_ON(npages != 1);
670                         result = get_kernel_page(user_addr, 0, pagevec);
671                         if (WARN_ON(result != 1))
672                                 break;
673                 }
674
675                 if ((unsigned)result < npages) {
676                         bytes = result * PAGE_SIZE;
677                         if (bytes <= pgbase) {
678                                 nfs_direct_release_pages(pagevec, result);
679                                 break;
680                         }
681                         bytes -= pgbase;
682                         npages = result;
683                 }
684
685                 for (i = 0; i < npages; i++) {
686                         struct nfs_page *req;
687                         unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
688
689                         req = nfs_create_request(dreq->ctx, dreq->inode,
690                                                  pagevec[i],
691                                                  pgbase, req_len);
692                         if (IS_ERR(req)) {
693                                 result = PTR_ERR(req);
694                                 break;
695                         }
696                         nfs_lock_request(req);
697                         req->wb_index = pos >> PAGE_SHIFT;
698                         req->wb_offset = pos & ~PAGE_MASK;
699                         if (!nfs_pageio_add_request(desc, req)) {
700                                 result = desc->pg_error;
701                                 nfs_unlock_and_release_request(req);
702                                 break;
703                         }
704                         pgbase = 0;
705                         bytes -= req_len;
706                         started += req_len;
707                         user_addr += req_len;
708                         pos += req_len;
709                         count -= req_len;
710                         dreq->bytes_left -= req_len;
711                 }
712                 /* The nfs_page now hold references to these pages */
713                 nfs_direct_release_pages(pagevec, npages);
714         } while (count != 0 && result >= 0);
715
716         kfree(pagevec);
717
718         if (started)
719                 return started;
720         return result < 0 ? (ssize_t) result : -EFAULT;
721 }
722
723 static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
724 {
725         struct nfs_direct_req *dreq = hdr->dreq;
726         struct nfs_commit_info cinfo;
727         int bit = -1;
728         struct nfs_page *req = nfs_list_entry(hdr->pages.next);
729
730         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
731                 goto out_put;
732
733         nfs_init_cinfo_from_dreq(&cinfo, dreq);
734
735         spin_lock(&dreq->lock);
736
737         if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
738                 dreq->flags = 0;
739                 dreq->error = hdr->error;
740         }
741         if (dreq->error != 0)
742                 bit = NFS_IOHDR_ERROR;
743         else {
744                 dreq->count += hdr->good_bytes;
745                 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
746                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
747                         bit = NFS_IOHDR_NEED_RESCHED;
748                 } else if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
749                         if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES)
750                                 bit = NFS_IOHDR_NEED_RESCHED;
751                         else if (dreq->flags == 0) {
752                                 memcpy(&dreq->verf, hdr->verf,
753                                        sizeof(dreq->verf));
754                                 bit = NFS_IOHDR_NEED_COMMIT;
755                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
756                         } else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) {
757                                 if (memcmp(&dreq->verf, hdr->verf, sizeof(dreq->verf))) {
758                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
759                                         bit = NFS_IOHDR_NEED_RESCHED;
760                                 } else
761                                         bit = NFS_IOHDR_NEED_COMMIT;
762                         }
763                 }
764         }
765         spin_unlock(&dreq->lock);
766
767         while (!list_empty(&hdr->pages)) {
768                 req = nfs_list_entry(hdr->pages.next);
769                 nfs_list_remove_request(req);
770                 switch (bit) {
771                 case NFS_IOHDR_NEED_RESCHED:
772                 case NFS_IOHDR_NEED_COMMIT:
773                         kref_get(&req->wb_kref);
774                         nfs_mark_request_commit(req, hdr->lseg, &cinfo);
775                 }
776                 nfs_unlock_and_release_request(req);
777         }
778
779 out_put:
780         if (put_dreq(dreq))
781                 nfs_direct_write_complete(dreq, hdr->inode);
782         hdr->release(hdr);
783 }
784
785 static void nfs_write_sync_pgio_error(struct list_head *head)
786 {
787         struct nfs_page *req;
788
789         while (!list_empty(head)) {
790                 req = nfs_list_entry(head->next);
791                 nfs_list_remove_request(req);
792                 nfs_unlock_and_release_request(req);
793         }
794 }
795
796 static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
797         .error_cleanup = nfs_write_sync_pgio_error,
798         .init_hdr = nfs_direct_pgio_init,
799         .completion = nfs_direct_write_completion,
800 };
801
802 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
803                                                const struct iovec *iov,
804                                                unsigned long nr_segs,
805                                                loff_t pos, bool uio)
806 {
807         struct nfs_pageio_descriptor desc;
808         struct inode *inode = dreq->inode;
809         ssize_t result = 0;
810         size_t requested_bytes = 0;
811         unsigned long seg;
812
813         NFS_PROTO(inode)->write_pageio_init(&desc, inode, FLUSH_COND_STABLE,
814                               &nfs_direct_write_completion_ops);
815         desc.pg_dreq = dreq;
816         get_dreq(dreq);
817         atomic_inc(&inode->i_dio_count);
818
819         NFS_I(dreq->inode)->write_io += iov_length(iov, nr_segs);
820         for (seg = 0; seg < nr_segs; seg++) {
821                 const struct iovec *vec = &iov[seg];
822                 result = nfs_direct_write_schedule_segment(&desc, vec, pos, uio);
823                 if (result < 0)
824                         break;
825                 requested_bytes += result;
826                 if ((size_t)result < vec->iov_len)
827                         break;
828                 pos += vec->iov_len;
829         }
830         nfs_pageio_complete(&desc);
831
832         /*
833          * If no bytes were started, return the error, and let the
834          * generic layer handle the completion.
835          */
836         if (requested_bytes == 0) {
837                 inode_dio_done(inode);
838                 nfs_direct_req_release(dreq);
839                 return result < 0 ? result : -EIO;
840         }
841
842         if (put_dreq(dreq))
843                 nfs_direct_write_complete(dreq, dreq->inode);
844         return 0;
845 }
846
847 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
848                                 unsigned long nr_segs, loff_t pos,
849                                 size_t count, bool uio)
850 {
851         ssize_t result = -ENOMEM;
852         struct inode *inode = iocb->ki_filp->f_mapping->host;
853         struct nfs_direct_req *dreq;
854         struct nfs_lock_context *l_ctx;
855
856         dreq = nfs_direct_req_alloc();
857         if (!dreq)
858                 goto out;
859
860         dreq->inode = inode;
861         dreq->bytes_left = count;
862         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
863         l_ctx = nfs_get_lock_context(dreq->ctx);
864         if (IS_ERR(l_ctx)) {
865                 result = PTR_ERR(l_ctx);
866                 goto out_release;
867         }
868         dreq->l_ctx = l_ctx;
869         if (!is_sync_kiocb(iocb))
870                 dreq->iocb = iocb;
871
872         result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, uio);
873         if (!result)
874                 result = nfs_direct_wait(dreq);
875 out_release:
876         nfs_direct_req_release(dreq);
877 out:
878         return result;
879 }
880
881 /**
882  * nfs_file_direct_read - file direct read operation for NFS files
883  * @iocb: target I/O control block
884  * @iov: vector of user buffers into which to read data
885  * @nr_segs: size of iov vector
886  * @pos: byte offset in file where reading starts
887  *
888  * We use this function for direct reads instead of calling
889  * generic_file_aio_read() in order to avoid gfar's check to see if
890  * the request starts before the end of the file.  For that check
891  * to work, we must generate a GETATTR before each direct read, and
892  * even then there is a window between the GETATTR and the subsequent
893  * READ where the file size could change.  Our preference is simply
894  * to do all reads the application wants, and the server will take
895  * care of managing the end of file boundary.
896  *
897  * This function also eliminates unnecessarily updating the file's
898  * atime locally, as the NFS server sets the file's atime, and this
899  * client must read the updated atime from the server back into its
900  * cache.
901  */
902 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
903                                 unsigned long nr_segs, loff_t pos, bool uio)
904 {
905         ssize_t retval = -EINVAL;
906         struct file *file = iocb->ki_filp;
907         struct address_space *mapping = file->f_mapping;
908         size_t count;
909
910         count = iov_length(iov, nr_segs);
911         nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
912
913         dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
914                 file->f_path.dentry->d_parent->d_name.name,
915                 file->f_path.dentry->d_name.name,
916                 count, (long long) pos);
917
918         retval = 0;
919         if (!count)
920                 goto out;
921
922         retval = nfs_sync_mapping(mapping);
923         if (retval)
924                 goto out;
925
926         task_io_account_read(count);
927
928         retval = nfs_direct_read(iocb, iov, nr_segs, pos, uio);
929         if (retval > 0)
930                 iocb->ki_pos = pos + retval;
931
932 out:
933         return retval;
934 }
935
936 /**
937  * nfs_file_direct_write - file direct write operation for NFS files
938  * @iocb: target I/O control block
939  * @iov: vector of user buffers from which to write data
940  * @nr_segs: size of iov vector
941  * @pos: byte offset in file where writing starts
942  *
943  * We use this function for direct writes instead of calling
944  * generic_file_aio_write() in order to avoid taking the inode
945  * semaphore and updating the i_size.  The NFS server will set
946  * the new i_size and this client must read the updated size
947  * back into its cache.  We let the server do generic write
948  * parameter checking and report problems.
949  *
950  * We eliminate local atime updates, see direct read above.
951  *
952  * We avoid unnecessary page cache invalidations for normal cached
953  * readers of this file.
954  *
955  * Note that O_APPEND is not supported for NFS direct writes, as there
956  * is no atomic O_APPEND write facility in the NFS protocol.
957  */
958 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
959                                 unsigned long nr_segs, loff_t pos, bool uio)
960 {
961         ssize_t retval = -EINVAL;
962         struct file *file = iocb->ki_filp;
963         struct address_space *mapping = file->f_mapping;
964         size_t count;
965
966         count = iov_length(iov, nr_segs);
967         nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
968
969         dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
970                 file->f_path.dentry->d_parent->d_name.name,
971                 file->f_path.dentry->d_name.name,
972                 count, (long long) pos);
973
974         retval = generic_write_checks(file, &pos, &count, 0);
975         if (retval)
976                 goto out;
977
978         retval = -EINVAL;
979         if ((ssize_t) count < 0)
980                 goto out;
981         retval = 0;
982         if (!count)
983                 goto out;
984
985         retval = nfs_sync_mapping(mapping);
986         if (retval)
987                 goto out;
988
989         task_io_account_write(count);
990
991         retval = nfs_direct_write(iocb, iov, nr_segs, pos, count, uio);
992         if (retval > 0) {
993                 struct inode *inode = mapping->host;
994
995                 iocb->ki_pos = pos + retval;
996                 spin_lock(&inode->i_lock);
997                 if (i_size_read(inode) < iocb->ki_pos)
998                         i_size_write(inode, iocb->ki_pos);
999                 spin_unlock(&inode->i_lock);
1000         }
1001 out:
1002         return retval;
1003 }
1004
1005 /**
1006  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
1007  *
1008  */
1009 int __init nfs_init_directcache(void)
1010 {
1011         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
1012                                                 sizeof(struct nfs_direct_req),
1013                                                 0, (SLAB_RECLAIM_ACCOUNT|
1014                                                         SLAB_MEM_SPREAD),
1015                                                 NULL);
1016         if (nfs_direct_cachep == NULL)
1017                 return -ENOMEM;
1018
1019         return 0;
1020 }
1021
1022 /**
1023  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1024  *
1025  */
1026 void nfs_destroy_directcache(void)
1027 {
1028         kmem_cache_destroy(nfs_direct_cachep);
1029 }