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
48 #include <linux/nfs_fs.h>
49 #include <linux/nfs_page.h>
50 #include <linux/sunrpc/clnt.h>
51
52 #include <asm/system.h>
53 #include <asm/uaccess.h>
54 #include <asm/atomic.h>
55
56 #include "internal.h"
57 #include "iostat.h"
58
59 #define NFSDBG_FACILITY NFSDBG_VFS
60
61 static struct kmem_cache *nfs_direct_cachep;
62
63 /*
64 * This represents a set of asynchronous requests that we're waiting on
65 */
66 struct nfs_direct_req {
67 struct kref kref; /* release manager */
68
69 /* I/O parameters */
70 struct nfs_open_context *ctx; /* file open context info */
71 struct kiocb * iocb; /* controlling i/o request */
72 struct inode * inode; /* target file of i/o */
73
74 /* completion state */
75 atomic_t io_count; /* i/os we're waiting for */
76 spinlock_t lock; /* protect completion state */
77 ssize_t count, /* bytes actually processed */
78 error; /* any reported error */
79 struct completion completion; /* wait for i/o completion */
80
81 /* commit state */
82 struct list_head rewrite_list; /* saved nfs_write_data structs */
83 struct nfs_write_data * commit_data; /* special write_data for commits */
84 int flags;
85 #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */
86 #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */
87 struct nfs_writeverf verf; /* unstable write verifier */
88 };
89
90 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
91 static const struct rpc_call_ops nfs_write_direct_ops;
92
93 static inline void get_dreq(struct nfs_direct_req *dreq)
94 {
95 atomic_inc(&dreq->io_count);
96 }
97
98 static inline int put_dreq(struct nfs_direct_req *dreq)
99 {
100 return atomic_dec_and_test(&dreq->io_count);
101 }
102
103 /**
104 * nfs_direct_IO - NFS address space operation for direct I/O
105 * @rw: direction (read or write)
106 * @iocb: target I/O control block
107 * @iov: array of vectors that define I/O buffer
108 * @pos: offset in file to begin the operation
109 * @nr_segs: size of iovec array
110 *
111 * The presence of this routine in the address space ops vector means
112 * the NFS client supports direct I/O. However, we shunt off direct
113 * read and write requests before the VFS gets them, so this method
114 * should never be called.
115 */
116 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
117 {
118 dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
119 iocb->ki_filp->f_path.dentry->d_name.name,
120 (long long) pos, nr_segs);
121
122 return -EINVAL;
123 }
124
125 static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
126 {
127 unsigned int npages;
128 unsigned int i;
129
130 if (count == 0)
131 return;
132 pages += (pgbase >> PAGE_SHIFT);
133 npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
134 for (i = 0; i < npages; i++) {
135 struct page *page = pages[i];
136 if (!PageCompound(page))
137 set_page_dirty(page);
138 }
139 }
140
141 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
142 {
143 unsigned int i;
144 for (i = 0; i < npages; i++)
145 page_cache_release(pages[i]);
146 }
147
148 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
149 {
150 struct nfs_direct_req *dreq;
151
152 dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
153 if (!dreq)
154 return NULL;
155
156 kref_init(&dreq->kref);
157 kref_get(&dreq->kref);
158 init_completion(&dreq->completion);
159 INIT_LIST_HEAD(&dreq->rewrite_list);
160 dreq->iocb = NULL;
161 dreq->ctx = NULL;
162 spin_lock_init(&dreq->lock);
163 atomic_set(&dreq->io_count, 0);
164 dreq->count = 0;
165 dreq->error = 0;
166 dreq->flags = 0;
167
168 return dreq;
169 }
170
171 static void nfs_direct_req_free(struct kref *kref)
172 {
173 struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
174
175 if (dreq->ctx != NULL)
176 put_nfs_open_context(dreq->ctx);
177 kmem_cache_free(nfs_direct_cachep, dreq);
178 }
179
180 static void nfs_direct_req_release(struct nfs_direct_req *dreq)
181 {
182 kref_put(&dreq->kref, nfs_direct_req_free);
183 }
184
185 /*
186 * Collects and returns the final error value/byte-count.
187 */
188 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
189 {
190 ssize_t result = -EIOCBQUEUED;
191
192 /* Async requests don't wait here */
193 if (dreq->iocb)
194 goto out;
195
196 result = wait_for_completion_killable(&dreq->completion);
197
198 if (!result)
199 result = dreq->error;
200 if (!result)
201 result = dreq->count;
202
203 out:
204 return (ssize_t) result;
205 }
206
207 /*
208 * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust
209 * the iocb is still valid here if this is a synchronous request.
210 */
211 static void nfs_direct_complete(struct nfs_direct_req *dreq)
212 {
213 if (dreq->iocb) {
214 long res = (long) dreq->error;
215 if (!res)
216 res = (long) dreq->count;
217 aio_complete(dreq->iocb, res, 0);
218 }
219 complete_all(&dreq->completion);
220
221 nfs_direct_req_release(dreq);
222 }
223
224 /*
225 * We must hold a reference to all the pages in this direct read request
226 * until the RPCs complete. This could be long *after* we are woken up in
227 * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
228 */
229 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
230 {
231 struct nfs_read_data *data = calldata;
232
233 nfs_readpage_result(task, data);
234 }
235
236 static void nfs_direct_read_release(void *calldata)
237 {
238
239 struct nfs_read_data *data = calldata;
240 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
241 int status = data->task.tk_status;
242
243 spin_lock(&dreq->lock);
244 if (unlikely(status < 0)) {
245 dreq->error = status;
246 spin_unlock(&dreq->lock);
247 } else {
248 dreq->count += data->res.count;
249 spin_unlock(&dreq->lock);
250 nfs_direct_dirty_pages(data->pagevec,
251 data->args.pgbase,
252 data->res.count);
253 }
254 nfs_direct_release_pages(data->pagevec, data->npages);
255
256 if (put_dreq(dreq))
257 nfs_direct_complete(dreq);
258 nfs_readdata_free(data);
259 }
260
261 static const struct rpc_call_ops nfs_read_direct_ops = {
262 #if defined(CONFIG_NFS_V4_1)
263 .rpc_call_prepare = nfs_read_prepare,
264 #endif /* CONFIG_NFS_V4_1 */
265 .rpc_call_done = nfs_direct_read_result,
266 .rpc_release = nfs_direct_read_release,
267 };
268
269 /*
270 * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
271 * operation. If nfs_readdata_alloc() or get_user_pages() fails,
272 * bail and stop sending more reads. Read length accounting is
273 * handled automatically by nfs_direct_read_result(). Otherwise, if
274 * no requests have been sent, just return an error.
275 */
276 static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
277 const struct iovec *iov,
278 loff_t pos)
279 {
280 struct nfs_open_context *ctx = dreq->ctx;
281 struct inode *inode = ctx->path.dentry->d_inode;
282 unsigned long user_addr = (unsigned long)iov->iov_base;
283 size_t count = iov->iov_len;
284 size_t rsize = NFS_SERVER(inode)->rsize;
285 struct rpc_task *task;
286 struct rpc_message msg = {
287 .rpc_cred = ctx->cred,
288 };
289 struct rpc_task_setup task_setup_data = {
290 .rpc_client = NFS_CLIENT(inode),
291 .rpc_message = &msg,
292 .callback_ops = &nfs_read_direct_ops,
293 .workqueue = nfsiod_workqueue,
294 .flags = RPC_TASK_ASYNC,
295 };
296 unsigned int pgbase;
297 int result;
298 ssize_t started = 0;
299
300 do {
301 struct nfs_read_data *data;
302 size_t bytes;
303
304 pgbase = user_addr & ~PAGE_MASK;
305 bytes = min(rsize,count);
306
307 result = -ENOMEM;
308 data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
309 if (unlikely(!data))
310 break;
311
312 down_read(¤t->mm->mmap_sem);
313 result = get_user_pages(current, current->mm, user_addr,
314 data->npages, 1, 0, data->pagevec, NULL);
315 up_read(¤t->mm->mmap_sem);
316 if (result < 0) {
317 nfs_readdata_free(data);
318 break;
319 }
320 if ((unsigned)result < data->npages) {
321 bytes = result * PAGE_SIZE;
322 if (bytes <= pgbase) {
323 nfs_direct_release_pages(data->pagevec, result);
324 nfs_readdata_free(data);
325 break;
326 }
327 bytes -= pgbase;
328 data->npages = result;
329 }
330
331 get_dreq(dreq);
332
333 data->req = (struct nfs_page *) dreq;
334 data->inode = inode;
335 data->cred = msg.rpc_cred;
336 data->args.fh = NFS_FH(inode);
337 data->args.context = ctx;
338 data->args.offset = pos;
339 data->args.pgbase = pgbase;
340 data->args.pages = data->pagevec;
341 data->args.count = bytes;
342 data->res.fattr = &data->fattr;
343 data->res.eof = 0;
344 data->res.count = bytes;
345 msg.rpc_argp = &data->args;
346 msg.rpc_resp = &data->res;
347
348 task_setup_data.task = &data->task;
349 task_setup_data.callback_data = data;
350 NFS_PROTO(inode)->read_setup(data, &msg);
351
352 task = rpc_run_task(&task_setup_data);
353 if (IS_ERR(task))
354 break;
355 rpc_put_task(task);
356
357 dprintk("NFS: %5u initiated direct read call "
358 "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
359 data->task.tk_pid,
360 inode->i_sb->s_id,
361 (long long)NFS_FILEID(inode),
362 bytes,
363 (unsigned long long)data->args.offset);
364
365 started += bytes;
366 user_addr += bytes;
367 pos += bytes;
368 /* FIXME: Remove this unnecessary math from final patch */
369 pgbase += bytes;
370 pgbase &= ~PAGE_MASK;
371 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
372
373 count -= bytes;
374 } while (count != 0);
375
376 if (started)
377 return started;
378 return result < 0 ? (ssize_t) result : -EFAULT;
379 }
380
381 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
382 const struct iovec *iov,
383 unsigned long nr_segs,
384 loff_t pos)
385 {
386 ssize_t result = -EINVAL;
387 size_t requested_bytes = 0;
388 unsigned long seg;
389
390 get_dreq(dreq);
391
392 for (seg = 0; seg < nr_segs; seg++) {
393 const struct iovec *vec = &iov[seg];
394 result = nfs_direct_read_schedule_segment(dreq, vec, pos);
395 if (result < 0)
396 break;
397 requested_bytes += result;
398 if ((size_t)result < vec->iov_len)
399 break;
400 pos += vec->iov_len;
401 }
402
403 if (put_dreq(dreq))
404 nfs_direct_complete(dreq);
405
406 if (requested_bytes != 0)
407 return 0;
408
409 if (result < 0)
410 return result;
411 return -EIO;
412 }
413
414 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
415 unsigned long nr_segs, loff_t pos)
416 {
417 ssize_t result = 0;
418 struct inode *inode = iocb->ki_filp->f_mapping->host;
419 struct nfs_direct_req *dreq;
420
421 dreq = nfs_direct_req_alloc();
422 if (!dreq)
423 return -ENOMEM;
424
425 dreq->inode = inode;
426 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
427 if (!is_sync_kiocb(iocb))
428 dreq->iocb = iocb;
429
430 result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
431 if (!result)
432 result = nfs_direct_wait(dreq);
433 nfs_direct_req_release(dreq);
434
435 return result;
436 }
437
438 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
439 {
440 while (!list_empty(&dreq->rewrite_list)) {
441 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
442 list_del(&data->pages);
443 nfs_direct_release_pages(data->pagevec, data->npages);
444 nfs_writedata_free(data);
445 }
446 }
447
448 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
449 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
450 {
451 struct inode *inode = dreq->inode;
452 struct list_head *p;
453 struct nfs_write_data *data;
454 struct rpc_task *task;
455 struct rpc_message msg = {
456 .rpc_cred = dreq->ctx->cred,
457 };
458 struct rpc_task_setup task_setup_data = {
459 .rpc_client = NFS_CLIENT(inode),
460 .rpc_message = &msg,
461 .callback_ops = &nfs_write_direct_ops,
462 .workqueue = nfsiod_workqueue,
463 .flags = RPC_TASK_ASYNC,
464 };
465
466 dreq->count = 0;
467 get_dreq(dreq);
468
469 list_for_each(p, &dreq->rewrite_list) {
470 data = list_entry(p, struct nfs_write_data, pages);
471
472 get_dreq(dreq);
473
474 /* Use stable writes */
475 data->args.stable = NFS_FILE_SYNC;
476
477 /*
478 * Reset data->res.
479 */
480 nfs_fattr_init(&data->fattr);
481 data->res.count = data->args.count;
482 memset(&data->verf, 0, sizeof(data->verf));
483
484 /*
485 * Reuse data->task; data->args should not have changed
486 * since the original request was sent.
487 */
488 task_setup_data.task = &data->task;
489 task_setup_data.callback_data = data;
490 msg.rpc_argp = &data->args;
491 msg.rpc_resp = &data->res;
492 NFS_PROTO(inode)->write_setup(data, &msg);
493
494 /*
495 * We're called via an RPC callback, so BKL is already held.
496 */
497 task = rpc_run_task(&task_setup_data);
498 if (!IS_ERR(task))
499 rpc_put_task(task);
500
501 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
502 data->task.tk_pid,
503 inode->i_sb->s_id,
504 (long long)NFS_FILEID(inode),
505 data->args.count,
506 (unsigned long long)data->args.offset);
507 }
508
509 if (put_dreq(dreq))
510 nfs_direct_write_complete(dreq, inode);
511 }
512
513 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
514 {
515 struct nfs_write_data *data = calldata;
516
517 /* Call the NFS version-specific code */
518 NFS_PROTO(data->inode)->commit_done(task, data);
519 }
520
521 static void nfs_direct_commit_release(void *calldata)
522 {
523 struct nfs_write_data *data = calldata;
524 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
525 int status = data->task.tk_status;
526
527 if (status < 0) {
528 dprintk("NFS: %5u commit failed with error %d.\n",
529 data->task.tk_pid, status);
530 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
531 } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
532 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
533 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
534 }
535
536 dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
537 nfs_direct_write_complete(dreq, data->inode);
538 nfs_commit_free(data);
539 }
540
541 static const struct rpc_call_ops nfs_commit_direct_ops = {
542 #if defined(CONFIG_NFS_V4_1)
543 .rpc_call_prepare = nfs_write_prepare,
544 #endif /* CONFIG_NFS_V4_1 */
545 .rpc_call_done = nfs_direct_commit_result,
546 .rpc_release = nfs_direct_commit_release,
547 };
548
549 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
550 {
551 struct nfs_write_data *data = dreq->commit_data;
552 struct rpc_task *task;
553 struct rpc_message msg = {
554 .rpc_argp = &data->args,
555 .rpc_resp = &data->res,
556 .rpc_cred = dreq->ctx->cred,
557 };
558 struct rpc_task_setup task_setup_data = {
559 .task = &data->task,
560 .rpc_client = NFS_CLIENT(dreq->inode),
561 .rpc_message = &msg,
562 .callback_ops = &nfs_commit_direct_ops,
563 .callback_data = data,
564 .workqueue = nfsiod_workqueue,
565 .flags = RPC_TASK_ASYNC,
566 };
567
568 data->inode = dreq->inode;
569 data->cred = msg.rpc_cred;
570
571 data->args.fh = NFS_FH(data->inode);
572 data->args.offset = 0;
573 data->args.count = 0;
574 data->args.context = dreq->ctx;
575 data->res.count = 0;
576 data->res.fattr = &data->fattr;
577 data->res.verf = &data->verf;
578
579 NFS_PROTO(data->inode)->commit_setup(data, &msg);
580
581 /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
582 dreq->commit_data = NULL;
583
584 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
585
586 task = rpc_run_task(&task_setup_data);
587 if (!IS_ERR(task))
588 rpc_put_task(task);
589 }
590
591 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
592 {
593 int flags = dreq->flags;
594
595 dreq->flags = 0;
596 switch (flags) {
597 case NFS_ODIRECT_DO_COMMIT:
598 nfs_direct_commit_schedule(dreq);
599 break;
600 case NFS_ODIRECT_RESCHED_WRITES:
601 nfs_direct_write_reschedule(dreq);
602 break;
603 default:
604 if (dreq->commit_data != NULL)
605 nfs_commit_free(dreq->commit_data);
606 nfs_direct_free_writedata(dreq);
607 nfs_zap_mapping(inode, inode->i_mapping);
608 nfs_direct_complete(dreq);
609 }
610 }
611
612 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
613 {
614 dreq->commit_data = nfs_commitdata_alloc();
615 if (dreq->commit_data != NULL)
616 dreq->commit_data->req = (struct nfs_page *) dreq;
617 }
618 #else
619 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
620 {
621 dreq->commit_data = NULL;
622 }
623
624 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
625 {
626 nfs_direct_free_writedata(dreq);
627 nfs_zap_mapping(inode, inode->i_mapping);
628 nfs_direct_complete(dreq);
629 }
630 #endif
631
632 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
633 {
634 struct nfs_write_data *data = calldata;
635
636 if (nfs_writeback_done(task, data) != 0)
637 return;
638 }
639
640 /*
641 * NB: Return the value of the first error return code. Subsequent
642 * errors after the first one are ignored.
643 */
644 static void nfs_direct_write_release(void *calldata)
645 {
646 struct nfs_write_data *data = calldata;
647 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
648 int status = data->task.tk_status;
649
650 spin_lock(&dreq->lock);
651
652 if (unlikely(status < 0)) {
653 /* An error has occurred, so we should not commit */
654 dreq->flags = 0;
655 dreq->error = status;
656 }
657 if (unlikely(dreq->error != 0))
658 goto out_unlock;
659
660 dreq->count += data->res.count;
661
662 if (data->res.verf->committed != NFS_FILE_SYNC) {
663 switch (dreq->flags) {
664 case 0:
665 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
666 dreq->flags = NFS_ODIRECT_DO_COMMIT;
667 break;
668 case NFS_ODIRECT_DO_COMMIT:
669 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
670 dprintk("NFS: %5u write verify failed\n", data->task.tk_pid);
671 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
672 }
673 }
674 }
675 out_unlock:
676 spin_unlock(&dreq->lock);
677
678 if (put_dreq(dreq))
679 nfs_direct_write_complete(dreq, data->inode);
680 }
681
682 static const struct rpc_call_ops nfs_write_direct_ops = {
683 #if defined(CONFIG_NFS_V4_1)
684 .rpc_call_prepare = nfs_write_prepare,
685 #endif /* CONFIG_NFS_V4_1 */
686 .rpc_call_done = nfs_direct_write_result,
687 .rpc_release = nfs_direct_write_release,
688 };
689
690 /*
691 * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
692 * operation. If nfs_writedata_alloc() or get_user_pages() fails,
693 * bail and stop sending more writes. Write length accounting is
694 * handled automatically by nfs_direct_write_result(). Otherwise, if
695 * no requests have been sent, just return an error.
696 */
697 static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
698 const struct iovec *iov,
699 loff_t pos, int sync)
700 {
701 struct nfs_open_context *ctx = dreq->ctx;
702 struct inode *inode = ctx->path.dentry->d_inode;
703 unsigned long user_addr = (unsigned long)iov->iov_base;
704 size_t count = iov->iov_len;
705 struct rpc_task *task;
706 struct rpc_message msg = {
707 .rpc_cred = ctx->cred,
708 };
709 struct rpc_task_setup task_setup_data = {
710 .rpc_client = NFS_CLIENT(inode),
711 .rpc_message = &msg,
712 .callback_ops = &nfs_write_direct_ops,
713 .workqueue = nfsiod_workqueue,
714 .flags = RPC_TASK_ASYNC,
715 };
716 size_t wsize = NFS_SERVER(inode)->wsize;
717 unsigned int pgbase;
718 int result;
719 ssize_t started = 0;
720
721 do {
722 struct nfs_write_data *data;
723 size_t bytes;
724
725 pgbase = user_addr & ~PAGE_MASK;
726 bytes = min(wsize,count);
727
728 result = -ENOMEM;
729 data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
730 if (unlikely(!data))
731 break;
732
733 down_read(¤t->mm->mmap_sem);
734 result = get_user_pages(current, current->mm, user_addr,
735 data->npages, 0, 0, data->pagevec, NULL);
736 up_read(¤t->mm->mmap_sem);
737 if (result < 0) {
738 nfs_writedata_free(data);
739 break;
740 }
741 if ((unsigned)result < data->npages) {
742 bytes = result * PAGE_SIZE;
743 if (bytes <= pgbase) {
744 nfs_direct_release_pages(data->pagevec, result);
745 nfs_writedata_free(data);
746 break;
747 }
748 bytes -= pgbase;
749 data->npages = result;
750 }
751
752 get_dreq(dreq);
753
754 list_move_tail(&data->pages, &dreq->rewrite_list);
755
756 data->req = (struct nfs_page *) dreq;
757 data->inode = inode;
758 data->cred = msg.rpc_cred;
759 data->args.fh = NFS_FH(inode);
760 data->args.context = ctx;
761 data->args.offset = pos;
762 data->args.pgbase = pgbase;
763 data->args.pages = data->pagevec;
764 data->args.count = bytes;
765 data->args.stable = sync;
766 data->res.fattr = &data->fattr;
767 data->res.count = bytes;
768 data->res.verf = &data->verf;
769
770 task_setup_data.task = &data->task;
771 task_setup_data.callback_data = data;
772 msg.rpc_argp = &data->args;
773 msg.rpc_resp = &data->res;
774 NFS_PROTO(inode)->write_setup(data, &msg);
775
776 task = rpc_run_task(&task_setup_data);
777 if (IS_ERR(task))
778 break;
779 rpc_put_task(task);
780
781 dprintk("NFS: %5u initiated direct write call "
782 "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
783 data->task.tk_pid,
784 inode->i_sb->s_id,
785 (long long)NFS_FILEID(inode),
786 bytes,
787 (unsigned long long)data->args.offset);
788
789 started += bytes;
790 user_addr += bytes;
791 pos += bytes;
792
793 /* FIXME: Remove this useless math from the final patch */
794 pgbase += bytes;
795 pgbase &= ~PAGE_MASK;
796 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
797
798 count -= bytes;
799 } while (count != 0);
800
801 if (started)
802 return started;
803 return result < 0 ? (ssize_t) result : -EFAULT;
804 }
805
806 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
807 const struct iovec *iov,
808 unsigned long nr_segs,
809 loff_t pos, int sync)
810 {
811 ssize_t result = 0;
812 size_t requested_bytes = 0;
813 unsigned long seg;
814
815 get_dreq(dreq);
816
817 for (seg = 0; seg < nr_segs; seg++) {
818 const struct iovec *vec = &iov[seg];
819 result = nfs_direct_write_schedule_segment(dreq, vec,
820 pos, sync);
821 if (result < 0)
822 break;
823 requested_bytes += result;
824 if ((size_t)result < vec->iov_len)
825 break;
826 pos += vec->iov_len;
827 }
828
829 if (put_dreq(dreq))
830 nfs_direct_write_complete(dreq, dreq->inode);
831
832 if (requested_bytes != 0)
833 return 0;
834
835 if (result < 0)
836 return result;
837 return -EIO;
838 }
839
840 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
841 unsigned long nr_segs, loff_t pos,
842 size_t count)
843 {
844 ssize_t result = 0;
845 struct inode *inode = iocb->ki_filp->f_mapping->host;
846 struct nfs_direct_req *dreq;
847 size_t wsize = NFS_SERVER(inode)->wsize;
848 int sync = NFS_UNSTABLE;
849
850 dreq = nfs_direct_req_alloc();
851 if (!dreq)
852 return -ENOMEM;
853 nfs_alloc_commit_data(dreq);
854
855 if (dreq->commit_data == NULL || count < wsize)
856 sync = NFS_FILE_SYNC;
857
858 dreq->inode = inode;
859 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
860 if (!is_sync_kiocb(iocb))
861 dreq->iocb = iocb;
862
863 result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
864 if (!result)
865 result = nfs_direct_wait(dreq);
866 nfs_direct_req_release(dreq);
867
868 return result;
869 }
870
871 /**
872 * nfs_file_direct_read - file direct read operation for NFS files
873 * @iocb: target I/O control block
874 * @iov: vector of user buffers into which to read data
875 * @nr_segs: size of iov vector
876 * @pos: byte offset in file where reading starts
877 *
878 * We use this function for direct reads instead of calling
879 * generic_file_aio_read() in order to avoid gfar's check to see if
880 * the request starts before the end of the file. For that check
881 * to work, we must generate a GETATTR before each direct read, and
882 * even then there is a window between the GETATTR and the subsequent
883 * READ where the file size could change. Our preference is simply
884 * to do all reads the application wants, and the server will take
885 * care of managing the end of file boundary.
886 *
887 * This function also eliminates unnecessarily updating the file's
888 * atime locally, as the NFS server sets the file's atime, and this
889 * client must read the updated atime from the server back into its
890 * cache.
891 */
892 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
893 unsigned long nr_segs, loff_t pos)
894 {
895 ssize_t retval = -EINVAL;
896 struct file *file = iocb->ki_filp;
897 struct address_space *mapping = file->f_mapping;
898 size_t count;
899
900 count = iov_length(iov, nr_segs);
901 nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
902
903 dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
904 file->f_path.dentry->d_parent->d_name.name,
905 file->f_path.dentry->d_name.name,
906 count, (long long) pos);
907
908 retval = 0;
909 if (!count)
910 goto out;
911
912 retval = nfs_sync_mapping(mapping);
913 if (retval)
914 goto out;
915
916 retval = nfs_direct_read(iocb, iov, nr_segs, pos);
917 if (retval > 0)
918 iocb->ki_pos = pos + retval;
919
920 out:
921 return retval;
922 }
923
924 /**
925 * nfs_file_direct_write - file direct write operation for NFS files
926 * @iocb: target I/O control block
927 * @iov: vector of user buffers from which to write data
928 * @nr_segs: size of iov vector
929 * @pos: byte offset in file where writing starts
930 *
931 * We use this function for direct writes instead of calling
932 * generic_file_aio_write() in order to avoid taking the inode
933 * semaphore and updating the i_size. The NFS server will set
934 * the new i_size and this client must read the updated size
935 * back into its cache. We let the server do generic write
936 * parameter checking and report problems.
937 *
938 * We also avoid an unnecessary invocation of generic_osync_inode(),
939 * as it is fairly meaningless to sync the metadata of an NFS file.
940 *
941 * We eliminate local atime updates, see direct read above.
942 *
943 * We avoid unnecessary page cache invalidations for normal cached
944 * readers of this file.
945 *
946 * Note that O_APPEND is not supported for NFS direct writes, as there
947 * is no atomic O_APPEND write facility in the NFS protocol.
948 */
949 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
950 unsigned long nr_segs, loff_t pos)
951 {
952 ssize_t retval = -EINVAL;
953 struct file *file = iocb->ki_filp;
954 struct address_space *mapping = file->f_mapping;
955 size_t count;
956
957 count = iov_length(iov, nr_segs);
958 nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
959
960 dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
961 file->f_path.dentry->d_parent->d_name.name,
962 file->f_path.dentry->d_name.name,
963 count, (long long) pos);
964
965 retval = generic_write_checks(file, &pos, &count, 0);
966 if (retval)
967 goto out;
968
969 retval = -EINVAL;
970 if ((ssize_t) count < 0)
971 goto out;
972 retval = 0;
973 if (!count)
974 goto out;
975
976 retval = nfs_sync_mapping(mapping);
977 if (retval)
978 goto out;
979
980 retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
981
982 if (retval > 0)
983 iocb->ki_pos = pos + retval;
984
985 out:
986 return retval;
987 }
988
989 /**
990 * nfs_init_directcache - create a slab cache for nfs_direct_req structures
991 *
992 */
993 int __init nfs_init_directcache(void)
994 {
995 nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
996 sizeof(struct nfs_direct_req),
997 0, (SLAB_RECLAIM_ACCOUNT|
998 SLAB_MEM_SPREAD),
999 NULL);
1000 if (nfs_direct_cachep == NULL)
1001 return -ENOMEM;
1002
1003 return 0;
1004 }
1005
1006 /**
1007 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1008 *
1009 */
1010 void nfs_destroy_directcache(void)
1011 {
1012 kmem_cache_destroy(nfs_direct_cachep);
1013 }
1014
|
This page was automatically generated by the
LXR engine.
|