]>
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> |
7ec10f26 | 48 | #include <linux/task_io_accounting_ops.h> |
6296556f | 49 | #include <linux/module.h> |
1da177e4 LT |
50 | |
51 | #include <linux/nfs_fs.h> | |
52 | #include <linux/nfs_page.h> | |
53 | #include <linux/sunrpc/clnt.h> | |
54 | ||
1da177e4 | 55 | #include <asm/uaccess.h> |
60063497 | 56 | #include <linux/atomic.h> |
1da177e4 | 57 | |
8d5658c9 | 58 | #include "internal.h" |
91d5b470 | 59 | #include "iostat.h" |
1763da12 | 60 | #include "pnfs.h" |
91d5b470 | 61 | |
1da177e4 | 62 | #define NFSDBG_FACILITY NFSDBG_VFS |
1da177e4 | 63 | |
e18b890b | 64 | static struct kmem_cache *nfs_direct_cachep; |
1da177e4 LT |
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 */ | |
15ce4a0c CL |
71 | |
72 | /* I/O parameters */ | |
a8881f5a | 73 | struct nfs_open_context *ctx; /* file open context info */ |
f11ac8db | 74 | struct nfs_lock_context *l_ctx; /* Lock context info */ |
99514f8f | 75 | struct kiocb * iocb; /* controlling i/o request */ |
88467055 | 76 | struct inode * inode; /* target file of i/o */ |
15ce4a0c CL |
77 | |
78 | /* completion state */ | |
607f31e8 | 79 | atomic_t io_count; /* i/os we're waiting for */ |
15ce4a0c | 80 | spinlock_t lock; /* protect completion state */ |
15ce4a0c | 81 | ssize_t count, /* bytes actually processed */ |
35754bc0 | 82 | bytes_left, /* bytes left to be sent */ |
1da177e4 | 83 | error; /* any reported error */ |
d72b7a6b | 84 | struct completion completion; /* wait for i/o completion */ |
fad61490 TM |
85 | |
86 | /* commit state */ | |
1763da12 FI |
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; | |
fad61490 TM |
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 */ | |
1da177e4 LT |
94 | }; |
95 | ||
1763da12 FI |
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; | |
fad61490 | 98 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode); |
1763da12 | 99 | static void nfs_direct_write_schedule_work(struct work_struct *work); |
607f31e8 TM |
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 | ||
5002c586 WAA |
111 | /* |
112 | * nfs_direct_select_verf - select the right verifier | |
113 | * @dreq - direct request possibly spanning multiple servers | |
114 | * @ds_clp - nfs_client of data server or NULL if MDS / non-pnfs | |
115 | * @ds_idx - index of data server in data server list, only valid if ds_clp set | |
116 | * | |
117 | * returns the correct verifier to use given the role of the server | |
118 | */ | |
119 | static struct nfs_writeverf * | |
120 | nfs_direct_select_verf(struct nfs_direct_req *dreq, | |
121 | struct nfs_client *ds_clp, | |
122 | int ds_idx) | |
123 | { | |
124 | struct nfs_writeverf *verfp = &dreq->verf; | |
125 | ||
126 | #ifdef CONFIG_NFS_V4_1 | |
127 | if (ds_clp) { | |
128 | /* pNFS is in use, use the DS verf */ | |
129 | if (ds_idx >= 0 && ds_idx < dreq->ds_cinfo.nbuckets) | |
130 | verfp = &dreq->ds_cinfo.buckets[ds_idx].direct_verf; | |
131 | else | |
132 | WARN_ON_ONCE(1); | |
133 | } | |
134 | #endif | |
135 | return verfp; | |
136 | } | |
137 | ||
138 | ||
139 | /* | |
140 | * nfs_direct_set_hdr_verf - set the write/commit verifier | |
141 | * @dreq - direct request possibly spanning multiple servers | |
142 | * @hdr - pageio header to validate against previously seen verfs | |
143 | * | |
144 | * Set the server's (MDS or DS) "seen" verifier | |
145 | */ | |
146 | static void nfs_direct_set_hdr_verf(struct nfs_direct_req *dreq, | |
147 | struct nfs_pgio_header *hdr) | |
148 | { | |
149 | struct nfs_writeverf *verfp; | |
150 | ||
d45f60c6 WAA |
151 | verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, |
152 | hdr->ds_idx); | |
5002c586 WAA |
153 | WARN_ON_ONCE(verfp->committed >= 0); |
154 | memcpy(verfp, &hdr->verf, sizeof(struct nfs_writeverf)); | |
155 | WARN_ON_ONCE(verfp->committed < 0); | |
156 | } | |
157 | ||
158 | /* | |
159 | * nfs_direct_cmp_hdr_verf - compare verifier for pgio header | |
160 | * @dreq - direct request possibly spanning multiple servers | |
161 | * @hdr - pageio header to validate against previously seen verf | |
162 | * | |
163 | * set the server's "seen" verf if not initialized. | |
164 | * returns result of comparison between @hdr->verf and the "seen" | |
165 | * verf of the server used by @hdr (DS or MDS) | |
166 | */ | |
167 | static int nfs_direct_set_or_cmp_hdr_verf(struct nfs_direct_req *dreq, | |
168 | struct nfs_pgio_header *hdr) | |
169 | { | |
170 | struct nfs_writeverf *verfp; | |
171 | ||
d45f60c6 WAA |
172 | verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, |
173 | hdr->ds_idx); | |
5002c586 WAA |
174 | if (verfp->committed < 0) { |
175 | nfs_direct_set_hdr_verf(dreq, hdr); | |
176 | return 0; | |
177 | } | |
178 | return memcmp(verfp, &hdr->verf, sizeof(struct nfs_writeverf)); | |
179 | } | |
180 | ||
5002c586 WAA |
181 | /* |
182 | * nfs_direct_cmp_commit_data_verf - compare verifier for commit data | |
183 | * @dreq - direct request possibly spanning multiple servers | |
184 | * @data - commit data to validate against previously seen verf | |
185 | * | |
186 | * returns result of comparison between @data->verf and the verf of | |
187 | * the server used by @data (DS or MDS) | |
188 | */ | |
189 | static int nfs_direct_cmp_commit_data_verf(struct nfs_direct_req *dreq, | |
190 | struct nfs_commit_data *data) | |
191 | { | |
192 | struct nfs_writeverf *verfp; | |
193 | ||
194 | verfp = nfs_direct_select_verf(dreq, data->ds_clp, | |
195 | data->ds_commit_index); | |
196 | WARN_ON_ONCE(verfp->committed < 0); | |
197 | return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf)); | |
198 | } | |
5002c586 | 199 | |
1da177e4 | 200 | /** |
b8a32e2b CL |
201 | * nfs_direct_IO - NFS address space operation for direct I/O |
202 | * @rw: direction (read or write) | |
203 | * @iocb: target I/O control block | |
204 | * @iov: array of vectors that define I/O buffer | |
205 | * @pos: offset in file to begin the operation | |
206 | * @nr_segs: size of iovec array | |
207 | * | |
208 | * The presence of this routine in the address space ops vector means | |
a564b8f0 MG |
209 | * the NFS client supports direct I/O. However, for most direct IO, we |
210 | * shunt off direct read and write requests before the VFS gets them, | |
211 | * so this method is only ever called for swap. | |
1da177e4 | 212 | */ |
d8d3d94b | 213 | ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t pos) |
b8a32e2b | 214 | { |
ee8a1a8b PT |
215 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
216 | ||
217 | /* we only support swap file calling nfs_direct_IO */ | |
218 | if (!IS_SWAPFILE(inode)) | |
219 | return 0; | |
220 | ||
a564b8f0 | 221 | #ifndef CONFIG_NFS_SWAP |
6de1472f | 222 | dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n", |
d8d3d94b | 223 | iocb->ki_filp, (long long) pos, iter->nr_segs); |
b8a32e2b CL |
224 | |
225 | return -EINVAL; | |
a564b8f0 | 226 | #else |
a564b8f0 MG |
227 | VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE); |
228 | ||
e19a8a0a MP |
229 | if (rw == READ) |
230 | return nfs_file_direct_read(iocb, iter, pos); | |
231 | return nfs_file_direct_write(iocb, iter, pos); | |
a564b8f0 | 232 | #endif /* CONFIG_NFS_SWAP */ |
b8a32e2b CL |
233 | } |
234 | ||
749e146e | 235 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 236 | { |
749e146e | 237 | unsigned int i; |
607f31e8 TM |
238 | for (i = 0; i < npages; i++) |
239 | page_cache_release(pages[i]); | |
6b45d858 TM |
240 | } |
241 | ||
1763da12 FI |
242 | void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, |
243 | struct nfs_direct_req *dreq) | |
244 | { | |
245 | cinfo->lock = &dreq->lock; | |
246 | cinfo->mds = &dreq->mds_cinfo; | |
247 | cinfo->ds = &dreq->ds_cinfo; | |
248 | cinfo->dreq = dreq; | |
249 | cinfo->completion_ops = &nfs_direct_commit_completion_ops; | |
250 | } | |
251 | ||
93619e59 | 252 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 253 | { |
93619e59 CL |
254 | struct nfs_direct_req *dreq; |
255 | ||
292f3eee | 256 | dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
257 | if (!dreq) |
258 | return NULL; | |
259 | ||
260 | kref_init(&dreq->kref); | |
607f31e8 | 261 | kref_get(&dreq->kref); |
d72b7a6b | 262 | init_completion(&dreq->completion); |
1763da12 | 263 | INIT_LIST_HEAD(&dreq->mds_cinfo.list); |
5002c586 | 264 | dreq->verf.committed = NFS_INVALID_STABLE_HOW; /* not set yet */ |
1763da12 | 265 | INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); |
15ce4a0c | 266 | spin_lock_init(&dreq->lock); |
93619e59 CL |
267 | |
268 | return dreq; | |
1da177e4 LT |
269 | } |
270 | ||
b4946ffb | 271 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
272 | { |
273 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a | 274 | |
8c393f9a | 275 | nfs_free_pnfs_ds_cinfo(&dreq->ds_cinfo); |
f11ac8db TM |
276 | if (dreq->l_ctx != NULL) |
277 | nfs_put_lock_context(dreq->l_ctx); | |
a8881f5a TM |
278 | if (dreq->ctx != NULL) |
279 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
280 | kmem_cache_free(nfs_direct_cachep, dreq); |
281 | } | |
282 | ||
b4946ffb TM |
283 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
284 | { | |
285 | kref_put(&dreq->kref, nfs_direct_req_free); | |
286 | } | |
287 | ||
6296556f PT |
288 | ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq) |
289 | { | |
290 | return dreq->bytes_left; | |
291 | } | |
292 | EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left); | |
293 | ||
bc0fb201 CL |
294 | /* |
295 | * Collects and returns the final error value/byte-count. | |
296 | */ | |
297 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
298 | { | |
15ce4a0c | 299 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
300 | |
301 | /* Async requests don't wait here */ | |
302 | if (dreq->iocb) | |
303 | goto out; | |
304 | ||
150030b7 | 305 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 CL |
306 | |
307 | if (!result) | |
15ce4a0c | 308 | result = dreq->error; |
bc0fb201 | 309 | if (!result) |
15ce4a0c | 310 | result = dreq->count; |
bc0fb201 CL |
311 | |
312 | out: | |
bc0fb201 CL |
313 | return (ssize_t) result; |
314 | } | |
315 | ||
63ab46ab | 316 | /* |
607f31e8 TM |
317 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
318 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab | 319 | */ |
9811cd57 | 320 | static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write) |
63ab46ab | 321 | { |
9811cd57 CH |
322 | struct inode *inode = dreq->inode; |
323 | ||
2a009ec9 | 324 | if (dreq->iocb && write) { |
9811cd57 | 325 | loff_t pos = dreq->iocb->ki_pos + dreq->count; |
2a009ec9 CH |
326 | |
327 | spin_lock(&inode->i_lock); | |
328 | if (i_size_read(inode) < pos) | |
329 | i_size_write(inode, pos); | |
330 | spin_unlock(&inode->i_lock); | |
331 | } | |
332 | ||
1f90ee27 | 333 | if (write) |
2a009ec9 | 334 | nfs_zap_mapping(inode, inode->i_mapping); |
1f90ee27 CH |
335 | |
336 | inode_dio_done(inode); | |
2a009ec9 CH |
337 | |
338 | if (dreq->iocb) { | |
15ce4a0c | 339 | long res = (long) dreq->error; |
63ab46ab | 340 | if (!res) |
15ce4a0c | 341 | res = (long) dreq->count; |
63ab46ab | 342 | aio_complete(dreq->iocb, res, 0); |
d72b7a6b | 343 | } |
2a009ec9 | 344 | |
d72b7a6b | 345 | complete_all(&dreq->completion); |
63ab46ab | 346 | |
b4946ffb | 347 | nfs_direct_req_release(dreq); |
63ab46ab CL |
348 | } |
349 | ||
1385b811 | 350 | static void nfs_direct_readpage_release(struct nfs_page *req) |
1da177e4 | 351 | { |
1e8968c5 | 352 | dprintk("NFS: direct read done (%s/%llu %d@%lld)\n", |
584aa810 | 353 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1e8968c5 | 354 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
584aa810 FI |
355 | req->wb_bytes, |
356 | (long long)req_offset(req)); | |
357 | nfs_release_request(req); | |
fdd1e74c TM |
358 | } |
359 | ||
584aa810 | 360 | static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) |
fdd1e74c | 361 | { |
584aa810 FI |
362 | unsigned long bytes = 0; |
363 | struct nfs_direct_req *dreq = hdr->dreq; | |
fdd1e74c | 364 | |
584aa810 FI |
365 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) |
366 | goto out_put; | |
15ce4a0c CL |
367 | |
368 | spin_lock(&dreq->lock); | |
584aa810 FI |
369 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0)) |
370 | dreq->error = hdr->error; | |
371 | else | |
372 | dreq->count += hdr->good_bytes; | |
373 | spin_unlock(&dreq->lock); | |
374 | ||
4bd8b010 TM |
375 | while (!list_empty(&hdr->pages)) { |
376 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
377 | struct page *page = req->wb_page; | |
378 | ||
be7e9858 JL |
379 | if (!PageCompound(page) && bytes < hdr->good_bytes) |
380 | set_page_dirty(page); | |
4bd8b010 TM |
381 | bytes += req->wb_bytes; |
382 | nfs_list_remove_request(req); | |
383 | nfs_direct_readpage_release(req); | |
d4a8f367 | 384 | } |
584aa810 | 385 | out_put: |
607f31e8 | 386 | if (put_dreq(dreq)) |
9811cd57 | 387 | nfs_direct_complete(dreq, false); |
584aa810 | 388 | hdr->release(hdr); |
1da177e4 LT |
389 | } |
390 | ||
3e9e0ca3 | 391 | static void nfs_read_sync_pgio_error(struct list_head *head) |
cd841605 | 392 | { |
584aa810 | 393 | struct nfs_page *req; |
cd841605 | 394 | |
584aa810 FI |
395 | while (!list_empty(head)) { |
396 | req = nfs_list_entry(head->next); | |
397 | nfs_list_remove_request(req); | |
398 | nfs_release_request(req); | |
399 | } | |
cd841605 FI |
400 | } |
401 | ||
584aa810 FI |
402 | static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) |
403 | { | |
404 | get_dreq(hdr->dreq); | |
405 | } | |
406 | ||
407 | static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { | |
3e9e0ca3 | 408 | .error_cleanup = nfs_read_sync_pgio_error, |
584aa810 FI |
409 | .init_hdr = nfs_direct_pgio_init, |
410 | .completion = nfs_direct_read_completion, | |
411 | }; | |
412 | ||
d4cc948b | 413 | /* |
607f31e8 TM |
414 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
415 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
416 | * bail and stop sending more reads. Read length accounting is | |
417 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
418 | * no requests have been sent, just return an error. | |
1da177e4 | 419 | */ |
1da177e4 | 420 | |
91f79c43 AV |
421 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
422 | struct iov_iter *iter, | |
423 | loff_t pos) | |
424 | { | |
425 | struct nfs_pageio_descriptor desc; | |
426 | struct inode *inode = dreq->inode; | |
427 | ssize_t result = -EINVAL; | |
428 | size_t requested_bytes = 0; | |
429 | size_t rsize = max_t(size_t, NFS_SERVER(inode)->rsize, PAGE_SIZE); | |
e9f7bee1 | 430 | |
16b90578 | 431 | nfs_pageio_init_read(&desc, dreq->inode, false, |
91f79c43 AV |
432 | &nfs_direct_read_completion_ops); |
433 | get_dreq(dreq); | |
434 | desc.pg_dreq = dreq; | |
435 | atomic_inc(&inode->i_dio_count); | |
a564b8f0 | 436 | |
91f79c43 AV |
437 | while (iov_iter_count(iter)) { |
438 | struct page **pagevec; | |
439 | size_t bytes; | |
440 | size_t pgbase; | |
441 | unsigned npages, i; | |
607f31e8 | 442 | |
91f79c43 AV |
443 | result = iov_iter_get_pages_alloc(iter, &pagevec, |
444 | rsize, &pgbase); | |
445 | if (result < 0) | |
446 | break; | |
447 | ||
448 | bytes = result; | |
449 | iov_iter_advance(iter, bytes); | |
450 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; | |
584aa810 FI |
451 | for (i = 0; i < npages; i++) { |
452 | struct nfs_page *req; | |
bf5fc402 | 453 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); |
584aa810 | 454 | /* XXX do we need to do the eof zeroing found in async_filler? */ |
2bfc6e56 | 455 | req = nfs_create_request(dreq->ctx, pagevec[i], NULL, |
584aa810 FI |
456 | pgbase, req_len); |
457 | if (IS_ERR(req)) { | |
584aa810 FI |
458 | result = PTR_ERR(req); |
459 | break; | |
460 | } | |
461 | req->wb_index = pos >> PAGE_SHIFT; | |
462 | req->wb_offset = pos & ~PAGE_MASK; | |
91f79c43 AV |
463 | if (!nfs_pageio_add_request(&desc, req)) { |
464 | result = desc.pg_error; | |
584aa810 | 465 | nfs_release_request(req); |
584aa810 FI |
466 | break; |
467 | } | |
468 | pgbase = 0; | |
469 | bytes -= req_len; | |
91f79c43 | 470 | requested_bytes += req_len; |
584aa810 | 471 | pos += req_len; |
35754bc0 | 472 | dreq->bytes_left -= req_len; |
584aa810 | 473 | } |
6d74743b | 474 | nfs_direct_release_pages(pagevec, npages); |
91f79c43 | 475 | kvfree(pagevec); |
19f73787 CL |
476 | if (result < 0) |
477 | break; | |
19f73787 CL |
478 | } |
479 | ||
584aa810 FI |
480 | nfs_pageio_complete(&desc); |
481 | ||
839f7ad6 CL |
482 | /* |
483 | * If no bytes were started, return the error, and let the | |
484 | * generic layer handle the completion. | |
485 | */ | |
486 | if (requested_bytes == 0) { | |
1f90ee27 | 487 | inode_dio_done(inode); |
839f7ad6 CL |
488 | nfs_direct_req_release(dreq); |
489 | return result < 0 ? result : -EIO; | |
490 | } | |
491 | ||
19f73787 | 492 | if (put_dreq(dreq)) |
9811cd57 | 493 | nfs_direct_complete(dreq, false); |
839f7ad6 | 494 | return 0; |
19f73787 CL |
495 | } |
496 | ||
14a3ec79 CH |
497 | /** |
498 | * nfs_file_direct_read - file direct read operation for NFS files | |
499 | * @iocb: target I/O control block | |
619d30b4 | 500 | * @iter: vector of user buffers into which to read data |
14a3ec79 CH |
501 | * @pos: byte offset in file where reading starts |
502 | * | |
503 | * We use this function for direct reads instead of calling | |
504 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
505 | * the request starts before the end of the file. For that check | |
506 | * to work, we must generate a GETATTR before each direct read, and | |
507 | * even then there is a window between the GETATTR and the subsequent | |
508 | * READ where the file size could change. Our preference is simply | |
509 | * to do all reads the application wants, and the server will take | |
510 | * care of managing the end of file boundary. | |
511 | * | |
512 | * This function also eliminates unnecessarily updating the file's | |
513 | * atime locally, as the NFS server sets the file's atime, and this | |
514 | * client must read the updated atime from the server back into its | |
515 | * cache. | |
516 | */ | |
619d30b4 | 517 | ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter, |
e19a8a0a | 518 | loff_t pos) |
1da177e4 | 519 | { |
14a3ec79 CH |
520 | struct file *file = iocb->ki_filp; |
521 | struct address_space *mapping = file->f_mapping; | |
522 | struct inode *inode = mapping->host; | |
1da177e4 | 523 | struct nfs_direct_req *dreq; |
b3c54de6 | 524 | struct nfs_lock_context *l_ctx; |
14a3ec79 | 525 | ssize_t result = -EINVAL; |
a6cbcd4a | 526 | size_t count = iov_iter_count(iter); |
14a3ec79 CH |
527 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); |
528 | ||
529 | dfprintk(FILE, "NFS: direct read(%pD2, %zd@%Ld)\n", | |
530 | file, count, (long long) pos); | |
531 | ||
532 | result = 0; | |
533 | if (!count) | |
534 | goto out; | |
535 | ||
d0b9875d | 536 | mutex_lock(&inode->i_mutex); |
14a3ec79 CH |
537 | result = nfs_sync_mapping(mapping); |
538 | if (result) | |
d0b9875d | 539 | goto out_unlock; |
1da177e4 | 540 | |
14a3ec79 CH |
541 | task_io_account_read(count); |
542 | ||
543 | result = -ENOMEM; | |
607f31e8 | 544 | dreq = nfs_direct_req_alloc(); |
f11ac8db | 545 | if (dreq == NULL) |
d0b9875d | 546 | goto out_unlock; |
1da177e4 | 547 | |
91d5b470 | 548 | dreq->inode = inode; |
619d30b4 | 549 | dreq->bytes_left = count; |
cd3758e3 | 550 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
b3c54de6 TM |
551 | l_ctx = nfs_get_lock_context(dreq->ctx); |
552 | if (IS_ERR(l_ctx)) { | |
553 | result = PTR_ERR(l_ctx); | |
f11ac8db | 554 | goto out_release; |
b3c54de6 TM |
555 | } |
556 | dreq->l_ctx = l_ctx; | |
487b8372 CL |
557 | if (!is_sync_kiocb(iocb)) |
558 | dreq->iocb = iocb; | |
1da177e4 | 559 | |
619d30b4 | 560 | NFS_I(inode)->read_io += count; |
91f79c43 | 561 | result = nfs_direct_read_schedule_iovec(dreq, iter, pos); |
d0b9875d CH |
562 | |
563 | mutex_unlock(&inode->i_mutex); | |
564 | ||
14a3ec79 | 565 | if (!result) { |
607f31e8 | 566 | result = nfs_direct_wait(dreq); |
14a3ec79 CH |
567 | if (result > 0) |
568 | iocb->ki_pos = pos + result; | |
569 | } | |
d0b9875d CH |
570 | |
571 | nfs_direct_req_release(dreq); | |
572 | return result; | |
573 | ||
f11ac8db | 574 | out_release: |
b4946ffb | 575 | nfs_direct_req_release(dreq); |
d0b9875d CH |
576 | out_unlock: |
577 | mutex_unlock(&inode->i_mutex); | |
f11ac8db | 578 | out: |
1da177e4 LT |
579 | return result; |
580 | } | |
581 | ||
fad61490 TM |
582 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) |
583 | { | |
1763da12 FI |
584 | struct nfs_pageio_descriptor desc; |
585 | struct nfs_page *req, *tmp; | |
586 | LIST_HEAD(reqs); | |
587 | struct nfs_commit_info cinfo; | |
588 | LIST_HEAD(failed); | |
589 | ||
590 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
591 | pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo); | |
592 | spin_lock(cinfo.lock); | |
593 | nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0); | |
594 | spin_unlock(cinfo.lock); | |
1da177e4 | 595 | |
fad61490 | 596 | dreq->count = 0; |
607f31e8 TM |
597 | get_dreq(dreq); |
598 | ||
a20c93e3 | 599 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false, |
1763da12 FI |
600 | &nfs_direct_write_completion_ops); |
601 | desc.pg_dreq = dreq; | |
fedb595c | 602 | |
1763da12 FI |
603 | list_for_each_entry_safe(req, tmp, &reqs, wb_list) { |
604 | if (!nfs_pageio_add_request(&desc, req)) { | |
4035c248 | 605 | nfs_list_remove_request(req); |
1763da12 FI |
606 | nfs_list_add_request(req, &failed); |
607 | spin_lock(cinfo.lock); | |
608 | dreq->flags = 0; | |
609 | dreq->error = -EIO; | |
610 | spin_unlock(cinfo.lock); | |
611 | } | |
5a695da2 | 612 | nfs_release_request(req); |
1763da12 FI |
613 | } |
614 | nfs_pageio_complete(&desc); | |
fad61490 | 615 | |
4035c248 TM |
616 | while (!list_empty(&failed)) { |
617 | req = nfs_list_entry(failed.next); | |
618 | nfs_list_remove_request(req); | |
1d1afcbc | 619 | nfs_unlock_and_release_request(req); |
4035c248 | 620 | } |
fad61490 | 621 | |
1763da12 FI |
622 | if (put_dreq(dreq)) |
623 | nfs_direct_write_complete(dreq, dreq->inode); | |
c9d8f89d TM |
624 | } |
625 | ||
1763da12 | 626 | static void nfs_direct_commit_complete(struct nfs_commit_data *data) |
c9d8f89d | 627 | { |
0b7c0153 | 628 | struct nfs_direct_req *dreq = data->dreq; |
1763da12 FI |
629 | struct nfs_commit_info cinfo; |
630 | struct nfs_page *req; | |
c9d8f89d TM |
631 | int status = data->task.tk_status; |
632 | ||
1763da12 | 633 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
c9d8f89d | 634 | if (status < 0) { |
60fa3f76 | 635 | dprintk("NFS: %5u commit failed with error %d.\n", |
1763da12 | 636 | data->task.tk_pid, status); |
fad61490 | 637 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
5002c586 | 638 | } else if (nfs_direct_cmp_commit_data_verf(dreq, data)) { |
c9d8f89d | 639 | dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid); |
fad61490 | 640 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
1da177e4 LT |
641 | } |
642 | ||
c9d8f89d | 643 | dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); |
1763da12 FI |
644 | while (!list_empty(&data->pages)) { |
645 | req = nfs_list_entry(data->pages.next); | |
646 | nfs_list_remove_request(req); | |
647 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) { | |
648 | /* Note the rewrite will go through mds */ | |
649 | nfs_mark_request_commit(req, NULL, &cinfo); | |
906369e4 FI |
650 | } else |
651 | nfs_release_request(req); | |
1d1afcbc | 652 | nfs_unlock_and_release_request(req); |
1763da12 FI |
653 | } |
654 | ||
655 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
656 | nfs_direct_write_complete(dreq, data->inode); | |
1da177e4 LT |
657 | } |
658 | ||
1763da12 FI |
659 | static void nfs_direct_error_cleanup(struct nfs_inode *nfsi) |
660 | { | |
661 | /* There is no lock to clear */ | |
662 | } | |
663 | ||
664 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { | |
665 | .completion = nfs_direct_commit_complete, | |
666 | .error_cleanup = nfs_direct_error_cleanup, | |
fad61490 TM |
667 | }; |
668 | ||
669 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 670 | { |
1763da12 FI |
671 | int res; |
672 | struct nfs_commit_info cinfo; | |
673 | LIST_HEAD(mds_list); | |
674 | ||
675 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
676 | nfs_scan_commit(dreq->inode, &mds_list, &cinfo); | |
677 | res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); | |
678 | if (res < 0) /* res == -ENOMEM */ | |
679 | nfs_direct_write_reschedule(dreq); | |
fad61490 | 680 | } |
1da177e4 | 681 | |
1763da12 | 682 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
fad61490 | 683 | { |
1763da12 | 684 | struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); |
fad61490 | 685 | int flags = dreq->flags; |
1da177e4 | 686 | |
fad61490 TM |
687 | dreq->flags = 0; |
688 | switch (flags) { | |
689 | case NFS_ODIRECT_DO_COMMIT: | |
690 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 691 | break; |
fad61490 TM |
692 | case NFS_ODIRECT_RESCHED_WRITES: |
693 | nfs_direct_write_reschedule(dreq); | |
694 | break; | |
695 | default: | |
9811cd57 | 696 | nfs_direct_complete(dreq, true); |
fad61490 TM |
697 | } |
698 | } | |
1da177e4 | 699 | |
1763da12 | 700 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
fad61490 | 701 | { |
1763da12 | 702 | schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */ |
fad61490 | 703 | } |
1763da12 | 704 | |
1763da12 FI |
705 | static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) |
706 | { | |
707 | struct nfs_direct_req *dreq = hdr->dreq; | |
708 | struct nfs_commit_info cinfo; | |
c65e6254 | 709 | bool request_commit = false; |
1763da12 FI |
710 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); |
711 | ||
712 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
713 | goto out_put; | |
714 | ||
715 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
716 | ||
717 | spin_lock(&dreq->lock); | |
718 | ||
719 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
720 | dreq->flags = 0; | |
721 | dreq->error = hdr->error; | |
722 | } | |
c65e6254 | 723 | if (dreq->error == 0) { |
1763da12 | 724 | dreq->count += hdr->good_bytes; |
c65e6254 | 725 | if (nfs_write_need_commit(hdr)) { |
1763da12 | 726 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) |
c65e6254 | 727 | request_commit = true; |
1763da12 | 728 | else if (dreq->flags == 0) { |
5002c586 | 729 | nfs_direct_set_hdr_verf(dreq, hdr); |
c65e6254 | 730 | request_commit = true; |
1763da12 FI |
731 | dreq->flags = NFS_ODIRECT_DO_COMMIT; |
732 | } else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) { | |
c65e6254 WAA |
733 | request_commit = true; |
734 | if (nfs_direct_set_or_cmp_hdr_verf(dreq, hdr)) | |
5002c586 WAA |
735 | dreq->flags = |
736 | NFS_ODIRECT_RESCHED_WRITES; | |
1763da12 FI |
737 | } |
738 | } | |
739 | } | |
740 | spin_unlock(&dreq->lock); | |
741 | ||
742 | while (!list_empty(&hdr->pages)) { | |
2bfc6e56 | 743 | |
1763da12 FI |
744 | req = nfs_list_entry(hdr->pages.next); |
745 | nfs_list_remove_request(req); | |
c65e6254 | 746 | if (request_commit) { |
04277086 | 747 | kref_get(&req->wb_kref); |
1763da12 | 748 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
1763da12 | 749 | } |
1d1afcbc | 750 | nfs_unlock_and_release_request(req); |
1763da12 FI |
751 | } |
752 | ||
753 | out_put: | |
754 | if (put_dreq(dreq)) | |
755 | nfs_direct_write_complete(dreq, hdr->inode); | |
756 | hdr->release(hdr); | |
757 | } | |
758 | ||
3e9e0ca3 TM |
759 | static void nfs_write_sync_pgio_error(struct list_head *head) |
760 | { | |
761 | struct nfs_page *req; | |
762 | ||
763 | while (!list_empty(head)) { | |
764 | req = nfs_list_entry(head->next); | |
765 | nfs_list_remove_request(req); | |
1d1afcbc | 766 | nfs_unlock_and_release_request(req); |
3e9e0ca3 TM |
767 | } |
768 | } | |
769 | ||
1763da12 | 770 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { |
3e9e0ca3 | 771 | .error_cleanup = nfs_write_sync_pgio_error, |
1763da12 FI |
772 | .init_hdr = nfs_direct_pgio_init, |
773 | .completion = nfs_direct_write_completion, | |
774 | }; | |
775 | ||
91f79c43 AV |
776 | |
777 | /* | |
778 | * NB: Return the value of the first error return code. Subsequent | |
779 | * errors after the first one are ignored. | |
780 | */ | |
781 | /* | |
782 | * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE | |
783 | * operation. If nfs_writedata_alloc() or get_user_pages() fails, | |
784 | * bail and stop sending more writes. Write length accounting is | |
785 | * handled automatically by nfs_direct_write_result(). Otherwise, if | |
786 | * no requests have been sent, just return an error. | |
787 | */ | |
19f73787 | 788 | static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, |
619d30b4 | 789 | struct iov_iter *iter, |
91f79c43 | 790 | loff_t pos) |
19f73787 | 791 | { |
1763da12 | 792 | struct nfs_pageio_descriptor desc; |
1d59d61f | 793 | struct inode *inode = dreq->inode; |
19f73787 CL |
794 | ssize_t result = 0; |
795 | size_t requested_bytes = 0; | |
91f79c43 | 796 | size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE); |
19f73787 | 797 | |
a20c93e3 | 798 | nfs_pageio_init_write(&desc, inode, FLUSH_COND_STABLE, false, |
1763da12 FI |
799 | &nfs_direct_write_completion_ops); |
800 | desc.pg_dreq = dreq; | |
19f73787 | 801 | get_dreq(dreq); |
1d59d61f | 802 | atomic_inc(&inode->i_dio_count); |
19f73787 | 803 | |
91f79c43 AV |
804 | NFS_I(inode)->write_io += iov_iter_count(iter); |
805 | while (iov_iter_count(iter)) { | |
806 | struct page **pagevec; | |
807 | size_t bytes; | |
808 | size_t pgbase; | |
809 | unsigned npages, i; | |
810 | ||
811 | result = iov_iter_get_pages_alloc(iter, &pagevec, | |
812 | wsize, &pgbase); | |
19f73787 CL |
813 | if (result < 0) |
814 | break; | |
91f79c43 AV |
815 | |
816 | bytes = result; | |
817 | iov_iter_advance(iter, bytes); | |
818 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; | |
819 | for (i = 0; i < npages; i++) { | |
820 | struct nfs_page *req; | |
821 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); | |
822 | ||
16b90578 | 823 | req = nfs_create_request(dreq->ctx, pagevec[i], NULL, |
91f79c43 AV |
824 | pgbase, req_len); |
825 | if (IS_ERR(req)) { | |
826 | result = PTR_ERR(req); | |
827 | break; | |
828 | } | |
829 | nfs_lock_request(req); | |
830 | req->wb_index = pos >> PAGE_SHIFT; | |
831 | req->wb_offset = pos & ~PAGE_MASK; | |
832 | if (!nfs_pageio_add_request(&desc, req)) { | |
833 | result = desc.pg_error; | |
834 | nfs_unlock_and_release_request(req); | |
835 | break; | |
836 | } | |
837 | pgbase = 0; | |
838 | bytes -= req_len; | |
839 | requested_bytes += req_len; | |
840 | pos += req_len; | |
841 | dreq->bytes_left -= req_len; | |
842 | } | |
843 | nfs_direct_release_pages(pagevec, npages); | |
844 | kvfree(pagevec); | |
845 | if (result < 0) | |
19f73787 | 846 | break; |
19f73787 | 847 | } |
1763da12 | 848 | nfs_pageio_complete(&desc); |
19f73787 | 849 | |
839f7ad6 CL |
850 | /* |
851 | * If no bytes were started, return the error, and let the | |
852 | * generic layer handle the completion. | |
853 | */ | |
854 | if (requested_bytes == 0) { | |
1d59d61f | 855 | inode_dio_done(inode); |
839f7ad6 CL |
856 | nfs_direct_req_release(dreq); |
857 | return result < 0 ? result : -EIO; | |
858 | } | |
859 | ||
19f73787 CL |
860 | if (put_dreq(dreq)) |
861 | nfs_direct_write_complete(dreq, dreq->inode); | |
839f7ad6 | 862 | return 0; |
19f73787 CL |
863 | } |
864 | ||
1da177e4 LT |
865 | /** |
866 | * nfs_file_direct_write - file direct write operation for NFS files | |
867 | * @iocb: target I/O control block | |
619d30b4 | 868 | * @iter: vector of user buffers from which to write data |
88467055 | 869 | * @pos: byte offset in file where writing starts |
1da177e4 LT |
870 | * |
871 | * We use this function for direct writes instead of calling | |
872 | * generic_file_aio_write() in order to avoid taking the inode | |
873 | * semaphore and updating the i_size. The NFS server will set | |
874 | * the new i_size and this client must read the updated size | |
875 | * back into its cache. We let the server do generic write | |
876 | * parameter checking and report problems. | |
877 | * | |
1da177e4 LT |
878 | * We eliminate local atime updates, see direct read above. |
879 | * | |
880 | * We avoid unnecessary page cache invalidations for normal cached | |
881 | * readers of this file. | |
882 | * | |
883 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
884 | * is no atomic O_APPEND write facility in the NFS protocol. | |
885 | */ | |
619d30b4 | 886 | ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter, |
e19a8a0a | 887 | loff_t pos) |
1da177e4 | 888 | { |
22cd1bf1 | 889 | ssize_t result = -EINVAL; |
1da177e4 | 890 | struct file *file = iocb->ki_filp; |
1da177e4 | 891 | struct address_space *mapping = file->f_mapping; |
22cd1bf1 CH |
892 | struct inode *inode = mapping->host; |
893 | struct nfs_direct_req *dreq; | |
894 | struct nfs_lock_context *l_ctx; | |
a9ab5e84 | 895 | loff_t end; |
a6cbcd4a | 896 | size_t count = iov_iter_count(iter); |
a9ab5e84 CH |
897 | end = (pos + count - 1) >> PAGE_CACHE_SHIFT; |
898 | ||
c216fd70 CL |
899 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); |
900 | ||
6de1472f AV |
901 | dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n", |
902 | file, count, (long long) pos); | |
027445c3 | 903 | |
22cd1bf1 CH |
904 | result = generic_write_checks(file, &pos, &count, 0); |
905 | if (result) | |
1da177e4 | 906 | goto out; |
ce1a8e67 | 907 | |
22cd1bf1 | 908 | result = -EINVAL; |
ce1a8e67 | 909 | if ((ssize_t) count < 0) |
1da177e4 | 910 | goto out; |
22cd1bf1 | 911 | result = 0; |
1da177e4 LT |
912 | if (!count) |
913 | goto out; | |
ce1a8e67 | 914 | |
a9ab5e84 CH |
915 | mutex_lock(&inode->i_mutex); |
916 | ||
22cd1bf1 CH |
917 | result = nfs_sync_mapping(mapping); |
918 | if (result) | |
a9ab5e84 CH |
919 | goto out_unlock; |
920 | ||
921 | if (mapping->nrpages) { | |
922 | result = invalidate_inode_pages2_range(mapping, | |
923 | pos >> PAGE_CACHE_SHIFT, end); | |
924 | if (result) | |
925 | goto out_unlock; | |
926 | } | |
1da177e4 | 927 | |
7ec10f26 KK |
928 | task_io_account_write(count); |
929 | ||
22cd1bf1 CH |
930 | result = -ENOMEM; |
931 | dreq = nfs_direct_req_alloc(); | |
932 | if (!dreq) | |
a9ab5e84 | 933 | goto out_unlock; |
9eafa8cc | 934 | |
22cd1bf1 CH |
935 | dreq->inode = inode; |
936 | dreq->bytes_left = count; | |
937 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); | |
938 | l_ctx = nfs_get_lock_context(dreq->ctx); | |
939 | if (IS_ERR(l_ctx)) { | |
940 | result = PTR_ERR(l_ctx); | |
941 | goto out_release; | |
942 | } | |
943 | dreq->l_ctx = l_ctx; | |
944 | if (!is_sync_kiocb(iocb)) | |
945 | dreq->iocb = iocb; | |
946 | ||
91f79c43 | 947 | result = nfs_direct_write_schedule_iovec(dreq, iter, pos); |
a9ab5e84 CH |
948 | |
949 | if (mapping->nrpages) { | |
950 | invalidate_inode_pages2_range(mapping, | |
951 | pos >> PAGE_CACHE_SHIFT, end); | |
952 | } | |
953 | ||
954 | mutex_unlock(&inode->i_mutex); | |
955 | ||
22cd1bf1 CH |
956 | if (!result) { |
957 | result = nfs_direct_wait(dreq); | |
958 | if (result > 0) { | |
959 | struct inode *inode = mapping->host; | |
960 | ||
961 | iocb->ki_pos = pos + result; | |
962 | spin_lock(&inode->i_lock); | |
963 | if (i_size_read(inode) < iocb->ki_pos) | |
964 | i_size_write(inode, iocb->ki_pos); | |
965 | spin_unlock(&inode->i_lock); | |
966 | } | |
1763da12 | 967 | } |
a9ab5e84 CH |
968 | nfs_direct_req_release(dreq); |
969 | return result; | |
970 | ||
22cd1bf1 CH |
971 | out_release: |
972 | nfs_direct_req_release(dreq); | |
a9ab5e84 CH |
973 | out_unlock: |
974 | mutex_unlock(&inode->i_mutex); | |
1da177e4 | 975 | out: |
22cd1bf1 | 976 | return result; |
1da177e4 LT |
977 | } |
978 | ||
88467055 CL |
979 | /** |
980 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
981 | * | |
982 | */ | |
f7b422b1 | 983 | int __init nfs_init_directcache(void) |
1da177e4 LT |
984 | { |
985 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
986 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
987 | 0, (SLAB_RECLAIM_ACCOUNT| |
988 | SLAB_MEM_SPREAD), | |
20c2df83 | 989 | NULL); |
1da177e4 LT |
990 | if (nfs_direct_cachep == NULL) |
991 | return -ENOMEM; | |
992 | ||
993 | return 0; | |
994 | } | |
995 | ||
88467055 | 996 | /** |
f7b422b1 | 997 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
998 | * |
999 | */ | |
266bee88 | 1000 | void nfs_destroy_directcache(void) |
1da177e4 | 1001 | { |
1a1d92c1 | 1002 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 1003 | } |