Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
88467055 | 10 | * (multiple copies of the same instance running on separate hosts) |
1da177e4 | 11 | * implement their own cache coherency protocol that subsumes file |
88467055 CL |
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 | |
1da177e4 LT |
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 | |
88467055 | 37 | * 04 May 2005 support O_DIRECT with aio --cel |
1da177e4 LT |
38 | * |
39 | */ | |
40 | ||
1da177e4 LT |
41 | #include <linux/errno.h> |
42 | #include <linux/sched.h> | |
43 | #include <linux/kernel.h> | |
1da177e4 LT |
44 | #include <linux/file.h> |
45 | #include <linux/pagemap.h> | |
46 | #include <linux/kref.h> | |
5a0e3ad6 | 47 | #include <linux/slab.h> |
1da177e4 LT |
48 | |
49 | #include <linux/nfs_fs.h> | |
50 | #include <linux/nfs_page.h> | |
51 | #include <linux/sunrpc/clnt.h> | |
52 | ||
53 | #include <asm/system.h> | |
54 | #include <asm/uaccess.h> | |
55 | #include <asm/atomic.h> | |
56 | ||
8d5658c9 | 57 | #include "internal.h" |
91d5b470 CL |
58 | #include "iostat.h" |
59 | ||
1da177e4 | 60 | #define NFSDBG_FACILITY NFSDBG_VFS |
1da177e4 | 61 | |
e18b890b | 62 | static struct kmem_cache *nfs_direct_cachep; |
1da177e4 LT |
63 | |
64 | /* | |
65 | * This represents a set of asynchronous requests that we're waiting on | |
66 | */ | |
67 | struct nfs_direct_req { | |
68 | struct kref kref; /* release manager */ | |
15ce4a0c CL |
69 | |
70 | /* I/O parameters */ | |
a8881f5a | 71 | struct nfs_open_context *ctx; /* file open context info */ |
99514f8f | 72 | struct kiocb * iocb; /* controlling i/o request */ |
88467055 | 73 | struct inode * inode; /* target file of i/o */ |
15ce4a0c CL |
74 | |
75 | /* completion state */ | |
607f31e8 | 76 | atomic_t io_count; /* i/os we're waiting for */ |
15ce4a0c | 77 | spinlock_t lock; /* protect completion state */ |
15ce4a0c | 78 | ssize_t count, /* bytes actually processed */ |
1da177e4 | 79 | error; /* any reported error */ |
d72b7a6b | 80 | struct completion completion; /* wait for i/o completion */ |
fad61490 TM |
81 | |
82 | /* commit state */ | |
607f31e8 | 83 | struct list_head rewrite_list; /* saved nfs_write_data structs */ |
fad61490 TM |
84 | struct nfs_write_data * commit_data; /* special write_data for commits */ |
85 | int flags; | |
86 | #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */ | |
87 | #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */ | |
88 | struct nfs_writeverf verf; /* unstable write verifier */ | |
1da177e4 LT |
89 | }; |
90 | ||
fad61490 | 91 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode); |
607f31e8 TM |
92 | static const struct rpc_call_ops nfs_write_direct_ops; |
93 | ||
94 | static inline void get_dreq(struct nfs_direct_req *dreq) | |
95 | { | |
96 | atomic_inc(&dreq->io_count); | |
97 | } | |
98 | ||
99 | static inline int put_dreq(struct nfs_direct_req *dreq) | |
100 | { | |
101 | return atomic_dec_and_test(&dreq->io_count); | |
102 | } | |
103 | ||
1da177e4 | 104 | /** |
b8a32e2b CL |
105 | * nfs_direct_IO - NFS address space operation for direct I/O |
106 | * @rw: direction (read or write) | |
107 | * @iocb: target I/O control block | |
108 | * @iov: array of vectors that define I/O buffer | |
109 | * @pos: offset in file to begin the operation | |
110 | * @nr_segs: size of iovec array | |
111 | * | |
112 | * The presence of this routine in the address space ops vector means | |
113 | * the NFS client supports direct I/O. However, we shunt off direct | |
114 | * read and write requests before the VFS gets them, so this method | |
115 | * should never be called. | |
1da177e4 | 116 | */ |
b8a32e2b CL |
117 | ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs) |
118 | { | |
b8a32e2b | 119 | dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n", |
01cce933 | 120 | iocb->ki_filp->f_path.dentry->d_name.name, |
e99170ff | 121 | (long long) pos, nr_segs); |
b8a32e2b CL |
122 | |
123 | return -EINVAL; | |
124 | } | |
125 | ||
d4a8f367 | 126 | static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count) |
6b45d858 | 127 | { |
d4a8f367 | 128 | unsigned int npages; |
749e146e | 129 | unsigned int i; |
d4a8f367 TM |
130 | |
131 | if (count == 0) | |
132 | return; | |
133 | pages += (pgbase >> PAGE_SHIFT); | |
134 | npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
6b45d858 TM |
135 | for (i = 0; i < npages; i++) { |
136 | struct page *page = pages[i]; | |
607f31e8 | 137 | if (!PageCompound(page)) |
d4a8f367 | 138 | set_page_dirty(page); |
6b45d858 | 139 | } |
9c93ab7d CL |
140 | } |
141 | ||
749e146e | 142 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 143 | { |
749e146e | 144 | unsigned int i; |
607f31e8 TM |
145 | for (i = 0; i < npages; i++) |
146 | page_cache_release(pages[i]); | |
6b45d858 TM |
147 | } |
148 | ||
93619e59 | 149 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 150 | { |
93619e59 CL |
151 | struct nfs_direct_req *dreq; |
152 | ||
e94b1766 | 153 | dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
154 | if (!dreq) |
155 | return NULL; | |
156 | ||
157 | kref_init(&dreq->kref); | |
607f31e8 | 158 | kref_get(&dreq->kref); |
d72b7a6b | 159 | init_completion(&dreq->completion); |
fad61490 | 160 | INIT_LIST_HEAD(&dreq->rewrite_list); |
93619e59 | 161 | dreq->iocb = NULL; |
a8881f5a | 162 | dreq->ctx = NULL; |
15ce4a0c | 163 | spin_lock_init(&dreq->lock); |
607f31e8 | 164 | atomic_set(&dreq->io_count, 0); |
15ce4a0c CL |
165 | dreq->count = 0; |
166 | dreq->error = 0; | |
fad61490 | 167 | dreq->flags = 0; |
93619e59 CL |
168 | |
169 | return dreq; | |
1da177e4 LT |
170 | } |
171 | ||
b4946ffb | 172 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
173 | { |
174 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a TM |
175 | |
176 | if (dreq->ctx != NULL) | |
177 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
178 | kmem_cache_free(nfs_direct_cachep, dreq); |
179 | } | |
180 | ||
b4946ffb TM |
181 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
182 | { | |
183 | kref_put(&dreq->kref, nfs_direct_req_free); | |
184 | } | |
185 | ||
bc0fb201 CL |
186 | /* |
187 | * Collects and returns the final error value/byte-count. | |
188 | */ | |
189 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
190 | { | |
15ce4a0c | 191 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
192 | |
193 | /* Async requests don't wait here */ | |
194 | if (dreq->iocb) | |
195 | goto out; | |
196 | ||
150030b7 | 197 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 CL |
198 | |
199 | if (!result) | |
15ce4a0c | 200 | result = dreq->error; |
bc0fb201 | 201 | if (!result) |
15ce4a0c | 202 | result = dreq->count; |
bc0fb201 CL |
203 | |
204 | out: | |
bc0fb201 CL |
205 | return (ssize_t) result; |
206 | } | |
207 | ||
63ab46ab | 208 | /* |
607f31e8 TM |
209 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
210 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab CL |
211 | */ |
212 | static void nfs_direct_complete(struct nfs_direct_req *dreq) | |
213 | { | |
63ab46ab | 214 | if (dreq->iocb) { |
15ce4a0c | 215 | long res = (long) dreq->error; |
63ab46ab | 216 | if (!res) |
15ce4a0c | 217 | res = (long) dreq->count; |
63ab46ab | 218 | aio_complete(dreq->iocb, res, 0); |
d72b7a6b TM |
219 | } |
220 | complete_all(&dreq->completion); | |
63ab46ab | 221 | |
b4946ffb | 222 | nfs_direct_req_release(dreq); |
63ab46ab CL |
223 | } |
224 | ||
06cf6f2e | 225 | /* |
607f31e8 TM |
226 | * We must hold a reference to all the pages in this direct read request |
227 | * until the RPCs complete. This could be long *after* we are woken up in | |
228 | * nfs_direct_wait (for instance, if someone hits ^C on a slow server). | |
06cf6f2e | 229 | */ |
ec06c096 | 230 | static void nfs_direct_read_result(struct rpc_task *task, void *calldata) |
1da177e4 | 231 | { |
ec06c096 | 232 | struct nfs_read_data *data = calldata; |
1da177e4 | 233 | |
fdd1e74c TM |
234 | nfs_readpage_result(task, data); |
235 | } | |
236 | ||
237 | static void nfs_direct_read_release(void *calldata) | |
238 | { | |
239 | ||
240 | struct nfs_read_data *data = calldata; | |
241 | struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req; | |
242 | int status = data->task.tk_status; | |
15ce4a0c CL |
243 | |
244 | spin_lock(&dreq->lock); | |
fdd1e74c TM |
245 | if (unlikely(status < 0)) { |
246 | dreq->error = status; | |
d4a8f367 TM |
247 | spin_unlock(&dreq->lock); |
248 | } else { | |
249 | dreq->count += data->res.count; | |
250 | spin_unlock(&dreq->lock); | |
251 | nfs_direct_dirty_pages(data->pagevec, | |
252 | data->args.pgbase, | |
253 | data->res.count); | |
254 | } | |
255 | nfs_direct_release_pages(data->pagevec, data->npages); | |
607f31e8 TM |
256 | |
257 | if (put_dreq(dreq)) | |
258 | nfs_direct_complete(dreq); | |
1ae88b2e | 259 | nfs_readdata_free(data); |
1da177e4 LT |
260 | } |
261 | ||
ec06c096 | 262 | static const struct rpc_call_ops nfs_read_direct_ops = { |
f11c88af AA |
263 | #if defined(CONFIG_NFS_V4_1) |
264 | .rpc_call_prepare = nfs_read_prepare, | |
265 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 266 | .rpc_call_done = nfs_direct_read_result, |
fdd1e74c | 267 | .rpc_release = nfs_direct_read_release, |
ec06c096 TM |
268 | }; |
269 | ||
d4cc948b | 270 | /* |
607f31e8 TM |
271 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
272 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
273 | * bail and stop sending more reads. Read length accounting is | |
274 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
275 | * no requests have been sent, just return an error. | |
1da177e4 | 276 | */ |
02fe4946 CL |
277 | static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq, |
278 | const struct iovec *iov, | |
279 | loff_t pos) | |
1da177e4 | 280 | { |
a8881f5a | 281 | struct nfs_open_context *ctx = dreq->ctx; |
88be9f99 | 282 | struct inode *inode = ctx->path.dentry->d_inode; |
02fe4946 CL |
283 | unsigned long user_addr = (unsigned long)iov->iov_base; |
284 | size_t count = iov->iov_len; | |
5dd602f2 | 285 | size_t rsize = NFS_SERVER(inode)->rsize; |
07737691 | 286 | struct rpc_task *task; |
bdc7f021 TM |
287 | struct rpc_message msg = { |
288 | .rpc_cred = ctx->cred, | |
289 | }; | |
84115e1c TM |
290 | struct rpc_task_setup task_setup_data = { |
291 | .rpc_client = NFS_CLIENT(inode), | |
bdc7f021 | 292 | .rpc_message = &msg, |
84115e1c | 293 | .callback_ops = &nfs_read_direct_ops, |
101070ca | 294 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
295 | .flags = RPC_TASK_ASYNC, |
296 | }; | |
607f31e8 TM |
297 | unsigned int pgbase; |
298 | int result; | |
299 | ssize_t started = 0; | |
300 | ||
1da177e4 | 301 | do { |
82b145c5 | 302 | struct nfs_read_data *data; |
5dd602f2 | 303 | size_t bytes; |
1da177e4 | 304 | |
e9f7bee1 TM |
305 | pgbase = user_addr & ~PAGE_MASK; |
306 | bytes = min(rsize,count); | |
307 | ||
607f31e8 | 308 | result = -ENOMEM; |
8d5658c9 | 309 | data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes)); |
607f31e8 TM |
310 | if (unlikely(!data)) |
311 | break; | |
312 | ||
607f31e8 TM |
313 | down_read(¤t->mm->mmap_sem); |
314 | result = get_user_pages(current, current->mm, user_addr, | |
315 | data->npages, 1, 0, data->pagevec, NULL); | |
316 | up_read(¤t->mm->mmap_sem); | |
749e146e | 317 | if (result < 0) { |
1ae88b2e | 318 | nfs_readdata_free(data); |
749e146e CL |
319 | break; |
320 | } | |
321 | if ((unsigned)result < data->npages) { | |
d9df8d6b TM |
322 | bytes = result * PAGE_SIZE; |
323 | if (bytes <= pgbase) { | |
324 | nfs_direct_release_pages(data->pagevec, result); | |
1ae88b2e | 325 | nfs_readdata_free(data); |
d9df8d6b TM |
326 | break; |
327 | } | |
328 | bytes -= pgbase; | |
329 | data->npages = result; | |
607f31e8 TM |
330 | } |
331 | ||
332 | get_dreq(dreq); | |
82b145c5 | 333 | |
607f31e8 | 334 | data->req = (struct nfs_page *) dreq; |
1da177e4 | 335 | data->inode = inode; |
bdc7f021 | 336 | data->cred = msg.rpc_cred; |
1da177e4 | 337 | data->args.fh = NFS_FH(inode); |
1ae88b2e | 338 | data->args.context = ctx; |
88467055 | 339 | data->args.offset = pos; |
1da177e4 | 340 | data->args.pgbase = pgbase; |
607f31e8 | 341 | data->args.pages = data->pagevec; |
1da177e4 LT |
342 | data->args.count = bytes; |
343 | data->res.fattr = &data->fattr; | |
344 | data->res.eof = 0; | |
345 | data->res.count = bytes; | |
65d26953 | 346 | nfs_fattr_init(&data->fattr); |
bdc7f021 TM |
347 | msg.rpc_argp = &data->args; |
348 | msg.rpc_resp = &data->res; | |
1da177e4 | 349 | |
07737691 | 350 | task_setup_data.task = &data->task; |
84115e1c | 351 | task_setup_data.callback_data = data; |
bdc7f021 | 352 | NFS_PROTO(inode)->read_setup(data, &msg); |
1da177e4 | 353 | |
07737691 | 354 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
355 | if (IS_ERR(task)) |
356 | break; | |
357 | rpc_put_task(task); | |
1da177e4 | 358 | |
a3f565b1 CL |
359 | dprintk("NFS: %5u initiated direct read call " |
360 | "(req %s/%Ld, %zu bytes @ offset %Lu)\n", | |
1da177e4 LT |
361 | data->task.tk_pid, |
362 | inode->i_sb->s_id, | |
363 | (long long)NFS_FILEID(inode), | |
364 | bytes, | |
365 | (unsigned long long)data->args.offset); | |
366 | ||
607f31e8 TM |
367 | started += bytes; |
368 | user_addr += bytes; | |
88467055 | 369 | pos += bytes; |
e9f7bee1 | 370 | /* FIXME: Remove this unnecessary math from final patch */ |
1da177e4 | 371 | pgbase += bytes; |
1da177e4 | 372 | pgbase &= ~PAGE_MASK; |
e9f7bee1 | 373 | BUG_ON(pgbase != (user_addr & ~PAGE_MASK)); |
1da177e4 LT |
374 | |
375 | count -= bytes; | |
376 | } while (count != 0); | |
607f31e8 | 377 | |
607f31e8 | 378 | if (started) |
c216fd70 | 379 | return started; |
607f31e8 | 380 | return result < 0 ? (ssize_t) result : -EFAULT; |
1da177e4 LT |
381 | } |
382 | ||
19f73787 CL |
383 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
384 | const struct iovec *iov, | |
385 | unsigned long nr_segs, | |
386 | loff_t pos) | |
387 | { | |
388 | ssize_t result = -EINVAL; | |
389 | size_t requested_bytes = 0; | |
390 | unsigned long seg; | |
391 | ||
392 | get_dreq(dreq); | |
393 | ||
394 | for (seg = 0; seg < nr_segs; seg++) { | |
395 | const struct iovec *vec = &iov[seg]; | |
02fe4946 | 396 | result = nfs_direct_read_schedule_segment(dreq, vec, pos); |
19f73787 CL |
397 | if (result < 0) |
398 | break; | |
399 | requested_bytes += result; | |
400 | if ((size_t)result < vec->iov_len) | |
401 | break; | |
402 | pos += vec->iov_len; | |
403 | } | |
404 | ||
405 | if (put_dreq(dreq)) | |
406 | nfs_direct_complete(dreq); | |
407 | ||
408 | if (requested_bytes != 0) | |
409 | return 0; | |
410 | ||
411 | if (result < 0) | |
412 | return result; | |
413 | return -EIO; | |
414 | } | |
415 | ||
c216fd70 CL |
416 | static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov, |
417 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 418 | { |
607f31e8 | 419 | ssize_t result = 0; |
99514f8f | 420 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1da177e4 LT |
421 | struct nfs_direct_req *dreq; |
422 | ||
607f31e8 | 423 | dreq = nfs_direct_req_alloc(); |
1da177e4 LT |
424 | if (!dreq) |
425 | return -ENOMEM; | |
426 | ||
91d5b470 | 427 | dreq->inode = inode; |
cd3758e3 | 428 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
487b8372 CL |
429 | if (!is_sync_kiocb(iocb)) |
430 | dreq->iocb = iocb; | |
1da177e4 | 431 | |
c216fd70 | 432 | result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos); |
607f31e8 TM |
433 | if (!result) |
434 | result = nfs_direct_wait(dreq); | |
b4946ffb | 435 | nfs_direct_req_release(dreq); |
1da177e4 LT |
436 | |
437 | return result; | |
438 | } | |
439 | ||
fad61490 | 440 | static void nfs_direct_free_writedata(struct nfs_direct_req *dreq) |
1da177e4 | 441 | { |
607f31e8 TM |
442 | while (!list_empty(&dreq->rewrite_list)) { |
443 | struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages); | |
fad61490 | 444 | list_del(&data->pages); |
607f31e8 | 445 | nfs_direct_release_pages(data->pagevec, data->npages); |
1ae88b2e | 446 | nfs_writedata_free(data); |
fad61490 TM |
447 | } |
448 | } | |
1da177e4 | 449 | |
fad61490 TM |
450 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
451 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) | |
452 | { | |
607f31e8 TM |
453 | struct inode *inode = dreq->inode; |
454 | struct list_head *p; | |
455 | struct nfs_write_data *data; | |
07737691 | 456 | struct rpc_task *task; |
bdc7f021 TM |
457 | struct rpc_message msg = { |
458 | .rpc_cred = dreq->ctx->cred, | |
459 | }; | |
84115e1c TM |
460 | struct rpc_task_setup task_setup_data = { |
461 | .rpc_client = NFS_CLIENT(inode), | |
a8b40bc7 | 462 | .rpc_message = &msg, |
84115e1c | 463 | .callback_ops = &nfs_write_direct_ops, |
101070ca | 464 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
465 | .flags = RPC_TASK_ASYNC, |
466 | }; | |
1da177e4 | 467 | |
fad61490 | 468 | dreq->count = 0; |
607f31e8 TM |
469 | get_dreq(dreq); |
470 | ||
471 | list_for_each(p, &dreq->rewrite_list) { | |
472 | data = list_entry(p, struct nfs_write_data, pages); | |
473 | ||
474 | get_dreq(dreq); | |
475 | ||
bdc7f021 TM |
476 | /* Use stable writes */ |
477 | data->args.stable = NFS_FILE_SYNC; | |
478 | ||
607f31e8 TM |
479 | /* |
480 | * Reset data->res. | |
481 | */ | |
482 | nfs_fattr_init(&data->fattr); | |
483 | data->res.count = data->args.count; | |
484 | memset(&data->verf, 0, sizeof(data->verf)); | |
485 | ||
486 | /* | |
487 | * Reuse data->task; data->args should not have changed | |
488 | * since the original request was sent. | |
489 | */ | |
07737691 | 490 | task_setup_data.task = &data->task; |
84115e1c | 491 | task_setup_data.callback_data = data; |
bdc7f021 TM |
492 | msg.rpc_argp = &data->args; |
493 | msg.rpc_resp = &data->res; | |
494 | NFS_PROTO(inode)->write_setup(data, &msg); | |
607f31e8 | 495 | |
607f31e8 TM |
496 | /* |
497 | * We're called via an RPC callback, so BKL is already held. | |
498 | */ | |
07737691 TM |
499 | task = rpc_run_task(&task_setup_data); |
500 | if (!IS_ERR(task)) | |
501 | rpc_put_task(task); | |
607f31e8 TM |
502 | |
503 | dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n", | |
504 | data->task.tk_pid, | |
505 | inode->i_sb->s_id, | |
506 | (long long)NFS_FILEID(inode), | |
507 | data->args.count, | |
508 | (unsigned long long)data->args.offset); | |
509 | } | |
fedb595c | 510 | |
607f31e8 TM |
511 | if (put_dreq(dreq)) |
512 | nfs_direct_write_complete(dreq, inode); | |
fad61490 TM |
513 | } |
514 | ||
515 | static void nfs_direct_commit_result(struct rpc_task *task, void *calldata) | |
516 | { | |
517 | struct nfs_write_data *data = calldata; | |
fad61490 TM |
518 | |
519 | /* Call the NFS version-specific code */ | |
c9d8f89d TM |
520 | NFS_PROTO(data->inode)->commit_done(task, data); |
521 | } | |
522 | ||
523 | static void nfs_direct_commit_release(void *calldata) | |
524 | { | |
525 | struct nfs_write_data *data = calldata; | |
526 | struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req; | |
527 | int status = data->task.tk_status; | |
528 | ||
529 | if (status < 0) { | |
60fa3f76 | 530 | dprintk("NFS: %5u commit failed with error %d.\n", |
c9d8f89d | 531 | data->task.tk_pid, status); |
fad61490 | 532 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
60fa3f76 | 533 | } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) { |
c9d8f89d | 534 | dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid); |
fad61490 | 535 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
1da177e4 LT |
536 | } |
537 | ||
c9d8f89d | 538 | dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); |
fad61490 | 539 | nfs_direct_write_complete(dreq, data->inode); |
1ae88b2e | 540 | nfs_commit_free(data); |
1da177e4 LT |
541 | } |
542 | ||
fad61490 | 543 | static const struct rpc_call_ops nfs_commit_direct_ops = { |
21d9a851 AA |
544 | #if defined(CONFIG_NFS_V4_1) |
545 | .rpc_call_prepare = nfs_write_prepare, | |
546 | #endif /* CONFIG_NFS_V4_1 */ | |
fad61490 | 547 | .rpc_call_done = nfs_direct_commit_result, |
c9d8f89d | 548 | .rpc_release = nfs_direct_commit_release, |
fad61490 TM |
549 | }; |
550 | ||
551 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 552 | { |
fad61490 | 553 | struct nfs_write_data *data = dreq->commit_data; |
07737691 | 554 | struct rpc_task *task; |
bdc7f021 TM |
555 | struct rpc_message msg = { |
556 | .rpc_argp = &data->args, | |
557 | .rpc_resp = &data->res, | |
558 | .rpc_cred = dreq->ctx->cred, | |
559 | }; | |
84115e1c | 560 | struct rpc_task_setup task_setup_data = { |
07737691 | 561 | .task = &data->task, |
84115e1c | 562 | .rpc_client = NFS_CLIENT(dreq->inode), |
bdc7f021 | 563 | .rpc_message = &msg, |
84115e1c TM |
564 | .callback_ops = &nfs_commit_direct_ops, |
565 | .callback_data = data, | |
101070ca | 566 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
567 | .flags = RPC_TASK_ASYNC, |
568 | }; | |
1da177e4 | 569 | |
fad61490 | 570 | data->inode = dreq->inode; |
bdc7f021 | 571 | data->cred = msg.rpc_cred; |
1da177e4 | 572 | |
fad61490 | 573 | data->args.fh = NFS_FH(data->inode); |
607f31e8 TM |
574 | data->args.offset = 0; |
575 | data->args.count = 0; | |
1ae88b2e | 576 | data->args.context = dreq->ctx; |
fad61490 TM |
577 | data->res.count = 0; |
578 | data->res.fattr = &data->fattr; | |
579 | data->res.verf = &data->verf; | |
65d26953 | 580 | nfs_fattr_init(&data->fattr); |
1da177e4 | 581 | |
bdc7f021 | 582 | NFS_PROTO(data->inode)->commit_setup(data, &msg); |
1da177e4 | 583 | |
fad61490 TM |
584 | /* Note: task.tk_ops->rpc_release will free dreq->commit_data */ |
585 | dreq->commit_data = NULL; | |
1da177e4 | 586 | |
e99170ff | 587 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); |
1da177e4 | 588 | |
07737691 TM |
589 | task = rpc_run_task(&task_setup_data); |
590 | if (!IS_ERR(task)) | |
591 | rpc_put_task(task); | |
fad61490 | 592 | } |
1da177e4 | 593 | |
fad61490 TM |
594 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
595 | { | |
596 | int flags = dreq->flags; | |
1da177e4 | 597 | |
fad61490 TM |
598 | dreq->flags = 0; |
599 | switch (flags) { | |
600 | case NFS_ODIRECT_DO_COMMIT: | |
601 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 602 | break; |
fad61490 TM |
603 | case NFS_ODIRECT_RESCHED_WRITES: |
604 | nfs_direct_write_reschedule(dreq); | |
605 | break; | |
606 | default: | |
fad61490 TM |
607 | if (dreq->commit_data != NULL) |
608 | nfs_commit_free(dreq->commit_data); | |
609 | nfs_direct_free_writedata(dreq); | |
cd9ae2b6 | 610 | nfs_zap_mapping(inode, inode->i_mapping); |
fad61490 TM |
611 | nfs_direct_complete(dreq); |
612 | } | |
613 | } | |
1da177e4 | 614 | |
fad61490 TM |
615 | static void nfs_alloc_commit_data(struct nfs_direct_req *dreq) |
616 | { | |
c9d8f89d | 617 | dreq->commit_data = nfs_commitdata_alloc(); |
fad61490 TM |
618 | if (dreq->commit_data != NULL) |
619 | dreq->commit_data->req = (struct nfs_page *) dreq; | |
620 | } | |
621 | #else | |
622 | static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq) | |
623 | { | |
624 | dreq->commit_data = NULL; | |
625 | } | |
1da177e4 | 626 | |
fad61490 TM |
627 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
628 | { | |
fad61490 | 629 | nfs_direct_free_writedata(dreq); |
cd9ae2b6 | 630 | nfs_zap_mapping(inode, inode->i_mapping); |
fad61490 TM |
631 | nfs_direct_complete(dreq); |
632 | } | |
633 | #endif | |
1da177e4 | 634 | |
462d5b32 | 635 | static void nfs_direct_write_result(struct rpc_task *task, void *calldata) |
1da177e4 | 636 | { |
462d5b32 | 637 | struct nfs_write_data *data = calldata; |
462d5b32 CL |
638 | |
639 | if (nfs_writeback_done(task, data) != 0) | |
640 | return; | |
c9d8f89d TM |
641 | } |
642 | ||
643 | /* | |
644 | * NB: Return the value of the first error return code. Subsequent | |
645 | * errors after the first one are ignored. | |
646 | */ | |
647 | static void nfs_direct_write_release(void *calldata) | |
648 | { | |
649 | struct nfs_write_data *data = calldata; | |
650 | struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req; | |
651 | int status = data->task.tk_status; | |
462d5b32 | 652 | |
15ce4a0c | 653 | spin_lock(&dreq->lock); |
1da177e4 | 654 | |
eda3cef8 | 655 | if (unlikely(status < 0)) { |
432409ee | 656 | /* An error has occurred, so we should not commit */ |
60fa3f76 | 657 | dreq->flags = 0; |
eda3cef8 | 658 | dreq->error = status; |
eda3cef8 | 659 | } |
432409ee NB |
660 | if (unlikely(dreq->error != 0)) |
661 | goto out_unlock; | |
eda3cef8 TM |
662 | |
663 | dreq->count += data->res.count; | |
1da177e4 | 664 | |
fad61490 TM |
665 | if (data->res.verf->committed != NFS_FILE_SYNC) { |
666 | switch (dreq->flags) { | |
667 | case 0: | |
668 | memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf)); | |
669 | dreq->flags = NFS_ODIRECT_DO_COMMIT; | |
1da177e4 | 670 | break; |
fad61490 TM |
671 | case NFS_ODIRECT_DO_COMMIT: |
672 | if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) { | |
c9d8f89d | 673 | dprintk("NFS: %5u write verify failed\n", data->task.tk_pid); |
fad61490 TM |
674 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
675 | } | |
1da177e4 | 676 | } |
1da177e4 | 677 | } |
eda3cef8 | 678 | out_unlock: |
fad61490 | 679 | spin_unlock(&dreq->lock); |
1da177e4 | 680 | |
607f31e8 TM |
681 | if (put_dreq(dreq)) |
682 | nfs_direct_write_complete(dreq, data->inode); | |
462d5b32 CL |
683 | } |
684 | ||
685 | static const struct rpc_call_ops nfs_write_direct_ops = { | |
def6ed7e AA |
686 | #if defined(CONFIG_NFS_V4_1) |
687 | .rpc_call_prepare = nfs_write_prepare, | |
688 | #endif /* CONFIG_NFS_V4_1 */ | |
462d5b32 | 689 | .rpc_call_done = nfs_direct_write_result, |
fad61490 | 690 | .rpc_release = nfs_direct_write_release, |
462d5b32 CL |
691 | }; |
692 | ||
693 | /* | |
607f31e8 TM |
694 | * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE |
695 | * operation. If nfs_writedata_alloc() or get_user_pages() fails, | |
696 | * bail and stop sending more writes. Write length accounting is | |
697 | * handled automatically by nfs_direct_write_result(). Otherwise, if | |
698 | * no requests have been sent, just return an error. | |
462d5b32 | 699 | */ |
02fe4946 CL |
700 | static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq, |
701 | const struct iovec *iov, | |
702 | loff_t pos, int sync) | |
462d5b32 | 703 | { |
a8881f5a | 704 | struct nfs_open_context *ctx = dreq->ctx; |
88be9f99 | 705 | struct inode *inode = ctx->path.dentry->d_inode; |
02fe4946 CL |
706 | unsigned long user_addr = (unsigned long)iov->iov_base; |
707 | size_t count = iov->iov_len; | |
07737691 | 708 | struct rpc_task *task; |
bdc7f021 TM |
709 | struct rpc_message msg = { |
710 | .rpc_cred = ctx->cred, | |
711 | }; | |
84115e1c TM |
712 | struct rpc_task_setup task_setup_data = { |
713 | .rpc_client = NFS_CLIENT(inode), | |
bdc7f021 | 714 | .rpc_message = &msg, |
84115e1c | 715 | .callback_ops = &nfs_write_direct_ops, |
101070ca | 716 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
717 | .flags = RPC_TASK_ASYNC, |
718 | }; | |
462d5b32 | 719 | size_t wsize = NFS_SERVER(inode)->wsize; |
607f31e8 TM |
720 | unsigned int pgbase; |
721 | int result; | |
722 | ssize_t started = 0; | |
82b145c5 | 723 | |
1da177e4 | 724 | do { |
82b145c5 | 725 | struct nfs_write_data *data; |
462d5b32 CL |
726 | size_t bytes; |
727 | ||
e9f7bee1 TM |
728 | pgbase = user_addr & ~PAGE_MASK; |
729 | bytes = min(wsize,count); | |
730 | ||
607f31e8 | 731 | result = -ENOMEM; |
8d5658c9 | 732 | data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes)); |
607f31e8 TM |
733 | if (unlikely(!data)) |
734 | break; | |
735 | ||
607f31e8 TM |
736 | down_read(¤t->mm->mmap_sem); |
737 | result = get_user_pages(current, current->mm, user_addr, | |
738 | data->npages, 0, 0, data->pagevec, NULL); | |
739 | up_read(¤t->mm->mmap_sem); | |
749e146e | 740 | if (result < 0) { |
1ae88b2e | 741 | nfs_writedata_free(data); |
749e146e CL |
742 | break; |
743 | } | |
744 | if ((unsigned)result < data->npages) { | |
d9df8d6b TM |
745 | bytes = result * PAGE_SIZE; |
746 | if (bytes <= pgbase) { | |
747 | nfs_direct_release_pages(data->pagevec, result); | |
1ae88b2e | 748 | nfs_writedata_free(data); |
d9df8d6b TM |
749 | break; |
750 | } | |
751 | bytes -= pgbase; | |
752 | data->npages = result; | |
607f31e8 TM |
753 | } |
754 | ||
755 | get_dreq(dreq); | |
756 | ||
fad61490 | 757 | list_move_tail(&data->pages, &dreq->rewrite_list); |
462d5b32 | 758 | |
607f31e8 | 759 | data->req = (struct nfs_page *) dreq; |
462d5b32 | 760 | data->inode = inode; |
bdc7f021 | 761 | data->cred = msg.rpc_cred; |
462d5b32 | 762 | data->args.fh = NFS_FH(inode); |
1ae88b2e | 763 | data->args.context = ctx; |
88467055 | 764 | data->args.offset = pos; |
462d5b32 | 765 | data->args.pgbase = pgbase; |
607f31e8 | 766 | data->args.pages = data->pagevec; |
462d5b32 | 767 | data->args.count = bytes; |
bdc7f021 | 768 | data->args.stable = sync; |
462d5b32 CL |
769 | data->res.fattr = &data->fattr; |
770 | data->res.count = bytes; | |
47989d74 | 771 | data->res.verf = &data->verf; |
65d26953 | 772 | nfs_fattr_init(&data->fattr); |
462d5b32 | 773 | |
07737691 | 774 | task_setup_data.task = &data->task; |
84115e1c | 775 | task_setup_data.callback_data = data; |
bdc7f021 TM |
776 | msg.rpc_argp = &data->args; |
777 | msg.rpc_resp = &data->res; | |
778 | NFS_PROTO(inode)->write_setup(data, &msg); | |
1da177e4 | 779 | |
07737691 | 780 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
781 | if (IS_ERR(task)) |
782 | break; | |
783 | rpc_put_task(task); | |
1da177e4 | 784 | |
a3f565b1 CL |
785 | dprintk("NFS: %5u initiated direct write call " |
786 | "(req %s/%Ld, %zu bytes @ offset %Lu)\n", | |
462d5b32 CL |
787 | data->task.tk_pid, |
788 | inode->i_sb->s_id, | |
789 | (long long)NFS_FILEID(inode), | |
790 | bytes, | |
791 | (unsigned long long)data->args.offset); | |
1da177e4 | 792 | |
607f31e8 TM |
793 | started += bytes; |
794 | user_addr += bytes; | |
88467055 | 795 | pos += bytes; |
e9f7bee1 TM |
796 | |
797 | /* FIXME: Remove this useless math from the final patch */ | |
462d5b32 | 798 | pgbase += bytes; |
462d5b32 | 799 | pgbase &= ~PAGE_MASK; |
e9f7bee1 | 800 | BUG_ON(pgbase != (user_addr & ~PAGE_MASK)); |
1da177e4 | 801 | |
462d5b32 CL |
802 | count -= bytes; |
803 | } while (count != 0); | |
607f31e8 | 804 | |
607f31e8 | 805 | if (started) |
c216fd70 | 806 | return started; |
607f31e8 | 807 | return result < 0 ? (ssize_t) result : -EFAULT; |
462d5b32 | 808 | } |
1da177e4 | 809 | |
19f73787 CL |
810 | static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, |
811 | const struct iovec *iov, | |
812 | unsigned long nr_segs, | |
813 | loff_t pos, int sync) | |
814 | { | |
815 | ssize_t result = 0; | |
816 | size_t requested_bytes = 0; | |
817 | unsigned long seg; | |
818 | ||
819 | get_dreq(dreq); | |
820 | ||
821 | for (seg = 0; seg < nr_segs; seg++) { | |
822 | const struct iovec *vec = &iov[seg]; | |
02fe4946 CL |
823 | result = nfs_direct_write_schedule_segment(dreq, vec, |
824 | pos, sync); | |
19f73787 CL |
825 | if (result < 0) |
826 | break; | |
827 | requested_bytes += result; | |
828 | if ((size_t)result < vec->iov_len) | |
829 | break; | |
830 | pos += vec->iov_len; | |
831 | } | |
832 | ||
833 | if (put_dreq(dreq)) | |
834 | nfs_direct_write_complete(dreq, dreq->inode); | |
835 | ||
836 | if (requested_bytes != 0) | |
837 | return 0; | |
838 | ||
839 | if (result < 0) | |
840 | return result; | |
841 | return -EIO; | |
842 | } | |
843 | ||
c216fd70 CL |
844 | static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov, |
845 | unsigned long nr_segs, loff_t pos, | |
846 | size_t count) | |
462d5b32 | 847 | { |
607f31e8 | 848 | ssize_t result = 0; |
c89f2ee5 | 849 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
462d5b32 | 850 | struct nfs_direct_req *dreq; |
fad61490 | 851 | size_t wsize = NFS_SERVER(inode)->wsize; |
bdc7f021 | 852 | int sync = NFS_UNSTABLE; |
1da177e4 | 853 | |
607f31e8 | 854 | dreq = nfs_direct_req_alloc(); |
462d5b32 CL |
855 | if (!dreq) |
856 | return -ENOMEM; | |
607f31e8 TM |
857 | nfs_alloc_commit_data(dreq); |
858 | ||
fad61490 | 859 | if (dreq->commit_data == NULL || count < wsize) |
bdc7f021 | 860 | sync = NFS_FILE_SYNC; |
1da177e4 | 861 | |
c89f2ee5 | 862 | dreq->inode = inode; |
cd3758e3 | 863 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
c89f2ee5 CL |
864 | if (!is_sync_kiocb(iocb)) |
865 | dreq->iocb = iocb; | |
1da177e4 | 866 | |
c216fd70 | 867 | result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync); |
607f31e8 TM |
868 | if (!result) |
869 | result = nfs_direct_wait(dreq); | |
b4946ffb | 870 | nfs_direct_req_release(dreq); |
1da177e4 | 871 | |
1da177e4 LT |
872 | return result; |
873 | } | |
874 | ||
875 | /** | |
876 | * nfs_file_direct_read - file direct read operation for NFS files | |
877 | * @iocb: target I/O control block | |
027445c3 BP |
878 | * @iov: vector of user buffers into which to read data |
879 | * @nr_segs: size of iov vector | |
88467055 | 880 | * @pos: byte offset in file where reading starts |
1da177e4 LT |
881 | * |
882 | * We use this function for direct reads instead of calling | |
883 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
884 | * the request starts before the end of the file. For that check | |
885 | * to work, we must generate a GETATTR before each direct read, and | |
886 | * even then there is a window between the GETATTR and the subsequent | |
88467055 | 887 | * READ where the file size could change. Our preference is simply |
1da177e4 LT |
888 | * to do all reads the application wants, and the server will take |
889 | * care of managing the end of file boundary. | |
88467055 | 890 | * |
1da177e4 LT |
891 | * This function also eliminates unnecessarily updating the file's |
892 | * atime locally, as the NFS server sets the file's atime, and this | |
893 | * client must read the updated atime from the server back into its | |
894 | * cache. | |
895 | */ | |
027445c3 BP |
896 | ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov, |
897 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
898 | { |
899 | ssize_t retval = -EINVAL; | |
1da177e4 | 900 | struct file *file = iocb->ki_filp; |
1da177e4 | 901 | struct address_space *mapping = file->f_mapping; |
c216fd70 CL |
902 | size_t count; |
903 | ||
904 | count = iov_length(iov, nr_segs); | |
905 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); | |
1da177e4 | 906 | |
6da24bc9 | 907 | dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n", |
01cce933 JJS |
908 | file->f_path.dentry->d_parent->d_name.name, |
909 | file->f_path.dentry->d_name.name, | |
c216fd70 | 910 | count, (long long) pos); |
1da177e4 | 911 | |
1da177e4 LT |
912 | retval = 0; |
913 | if (!count) | |
914 | goto out; | |
915 | ||
29884df0 TM |
916 | retval = nfs_sync_mapping(mapping); |
917 | if (retval) | |
918 | goto out; | |
1da177e4 | 919 | |
c216fd70 | 920 | retval = nfs_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 | 921 | if (retval > 0) |
0cdd80d0 | 922 | iocb->ki_pos = pos + retval; |
1da177e4 LT |
923 | |
924 | out: | |
925 | return retval; | |
926 | } | |
927 | ||
928 | /** | |
929 | * nfs_file_direct_write - file direct write operation for NFS files | |
930 | * @iocb: target I/O control block | |
027445c3 BP |
931 | * @iov: vector of user buffers from which to write data |
932 | * @nr_segs: size of iov vector | |
88467055 | 933 | * @pos: byte offset in file where writing starts |
1da177e4 LT |
934 | * |
935 | * We use this function for direct writes instead of calling | |
936 | * generic_file_aio_write() in order to avoid taking the inode | |
937 | * semaphore and updating the i_size. The NFS server will set | |
938 | * the new i_size and this client must read the updated size | |
939 | * back into its cache. We let the server do generic write | |
940 | * parameter checking and report problems. | |
941 | * | |
1da177e4 LT |
942 | * We eliminate local atime updates, see direct read above. |
943 | * | |
944 | * We avoid unnecessary page cache invalidations for normal cached | |
945 | * readers of this file. | |
946 | * | |
947 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
948 | * is no atomic O_APPEND write facility in the NFS protocol. | |
949 | */ | |
027445c3 BP |
950 | ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov, |
951 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 952 | { |
070ea602 | 953 | ssize_t retval = -EINVAL; |
1da177e4 | 954 | struct file *file = iocb->ki_filp; |
1da177e4 | 955 | struct address_space *mapping = file->f_mapping; |
c216fd70 | 956 | size_t count; |
1da177e4 | 957 | |
c216fd70 CL |
958 | count = iov_length(iov, nr_segs); |
959 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); | |
960 | ||
6da24bc9 | 961 | dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n", |
01cce933 JJS |
962 | file->f_path.dentry->d_parent->d_name.name, |
963 | file->f_path.dentry->d_name.name, | |
c216fd70 | 964 | count, (long long) pos); |
027445c3 | 965 | |
ce1a8e67 CL |
966 | retval = generic_write_checks(file, &pos, &count, 0); |
967 | if (retval) | |
1da177e4 | 968 | goto out; |
ce1a8e67 CL |
969 | |
970 | retval = -EINVAL; | |
971 | if ((ssize_t) count < 0) | |
1da177e4 | 972 | goto out; |
1da177e4 LT |
973 | retval = 0; |
974 | if (!count) | |
975 | goto out; | |
ce1a8e67 | 976 | |
29884df0 TM |
977 | retval = nfs_sync_mapping(mapping); |
978 | if (retval) | |
979 | goto out; | |
1da177e4 | 980 | |
c216fd70 | 981 | retval = nfs_direct_write(iocb, iov, nr_segs, pos, count); |
9eafa8cc | 982 | |
1da177e4 | 983 | if (retval > 0) |
ce1a8e67 | 984 | iocb->ki_pos = pos + retval; |
1da177e4 LT |
985 | |
986 | out: | |
987 | return retval; | |
988 | } | |
989 | ||
88467055 CL |
990 | /** |
991 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
992 | * | |
993 | */ | |
f7b422b1 | 994 | int __init nfs_init_directcache(void) |
1da177e4 LT |
995 | { |
996 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
997 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
998 | 0, (SLAB_RECLAIM_ACCOUNT| |
999 | SLAB_MEM_SPREAD), | |
20c2df83 | 1000 | NULL); |
1da177e4 LT |
1001 | if (nfs_direct_cachep == NULL) |
1002 | return -ENOMEM; | |
1003 | ||
1004 | return 0; | |
1005 | } | |
1006 | ||
88467055 | 1007 | /** |
f7b422b1 | 1008 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
1009 | * |
1010 | */ | |
266bee88 | 1011 | void nfs_destroy_directcache(void) |
1da177e4 | 1012 | { |
1a1d92c1 | 1013 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 1014 | } |