]>
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 | { |
a564b8f0 | 215 | #ifndef CONFIG_NFS_SWAP |
6de1472f | 216 | dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n", |
d8d3d94b | 217 | iocb->ki_filp, (long long) pos, iter->nr_segs); |
b8a32e2b CL |
218 | |
219 | return -EINVAL; | |
a564b8f0 | 220 | #else |
a564b8f0 MG |
221 | VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE); |
222 | ||
e19a8a0a MP |
223 | if (rw == READ) |
224 | return nfs_file_direct_read(iocb, iter, pos); | |
225 | return nfs_file_direct_write(iocb, iter, pos); | |
a564b8f0 | 226 | #endif /* CONFIG_NFS_SWAP */ |
b8a32e2b CL |
227 | } |
228 | ||
749e146e | 229 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 230 | { |
749e146e | 231 | unsigned int i; |
607f31e8 TM |
232 | for (i = 0; i < npages; i++) |
233 | page_cache_release(pages[i]); | |
6b45d858 TM |
234 | } |
235 | ||
1763da12 FI |
236 | void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, |
237 | struct nfs_direct_req *dreq) | |
238 | { | |
239 | cinfo->lock = &dreq->lock; | |
240 | cinfo->mds = &dreq->mds_cinfo; | |
241 | cinfo->ds = &dreq->ds_cinfo; | |
242 | cinfo->dreq = dreq; | |
243 | cinfo->completion_ops = &nfs_direct_commit_completion_ops; | |
244 | } | |
245 | ||
93619e59 | 246 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 247 | { |
93619e59 CL |
248 | struct nfs_direct_req *dreq; |
249 | ||
292f3eee | 250 | dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
251 | if (!dreq) |
252 | return NULL; | |
253 | ||
254 | kref_init(&dreq->kref); | |
607f31e8 | 255 | kref_get(&dreq->kref); |
d72b7a6b | 256 | init_completion(&dreq->completion); |
1763da12 | 257 | INIT_LIST_HEAD(&dreq->mds_cinfo.list); |
5002c586 | 258 | dreq->verf.committed = NFS_INVALID_STABLE_HOW; /* not set yet */ |
1763da12 | 259 | INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); |
15ce4a0c | 260 | spin_lock_init(&dreq->lock); |
93619e59 CL |
261 | |
262 | return dreq; | |
1da177e4 LT |
263 | } |
264 | ||
b4946ffb | 265 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
266 | { |
267 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a | 268 | |
8c393f9a | 269 | nfs_free_pnfs_ds_cinfo(&dreq->ds_cinfo); |
f11ac8db TM |
270 | if (dreq->l_ctx != NULL) |
271 | nfs_put_lock_context(dreq->l_ctx); | |
a8881f5a TM |
272 | if (dreq->ctx != NULL) |
273 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
274 | kmem_cache_free(nfs_direct_cachep, dreq); |
275 | } | |
276 | ||
b4946ffb TM |
277 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
278 | { | |
279 | kref_put(&dreq->kref, nfs_direct_req_free); | |
280 | } | |
281 | ||
6296556f PT |
282 | ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq) |
283 | { | |
284 | return dreq->bytes_left; | |
285 | } | |
286 | EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left); | |
287 | ||
bc0fb201 CL |
288 | /* |
289 | * Collects and returns the final error value/byte-count. | |
290 | */ | |
291 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
292 | { | |
15ce4a0c | 293 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
294 | |
295 | /* Async requests don't wait here */ | |
296 | if (dreq->iocb) | |
297 | goto out; | |
298 | ||
150030b7 | 299 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 CL |
300 | |
301 | if (!result) | |
15ce4a0c | 302 | result = dreq->error; |
bc0fb201 | 303 | if (!result) |
15ce4a0c | 304 | result = dreq->count; |
bc0fb201 CL |
305 | |
306 | out: | |
bc0fb201 CL |
307 | return (ssize_t) result; |
308 | } | |
309 | ||
63ab46ab | 310 | /* |
607f31e8 TM |
311 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
312 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab | 313 | */ |
9811cd57 | 314 | static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write) |
63ab46ab | 315 | { |
9811cd57 CH |
316 | struct inode *inode = dreq->inode; |
317 | ||
2a009ec9 | 318 | if (dreq->iocb && write) { |
9811cd57 | 319 | loff_t pos = dreq->iocb->ki_pos + dreq->count; |
2a009ec9 CH |
320 | |
321 | spin_lock(&inode->i_lock); | |
322 | if (i_size_read(inode) < pos) | |
323 | i_size_write(inode, pos); | |
324 | spin_unlock(&inode->i_lock); | |
325 | } | |
326 | ||
1f90ee27 | 327 | if (write) |
2a009ec9 | 328 | nfs_zap_mapping(inode, inode->i_mapping); |
1f90ee27 CH |
329 | |
330 | inode_dio_done(inode); | |
2a009ec9 CH |
331 | |
332 | if (dreq->iocb) { | |
15ce4a0c | 333 | long res = (long) dreq->error; |
63ab46ab | 334 | if (!res) |
15ce4a0c | 335 | res = (long) dreq->count; |
63ab46ab | 336 | aio_complete(dreq->iocb, res, 0); |
d72b7a6b | 337 | } |
2a009ec9 | 338 | |
d72b7a6b | 339 | complete_all(&dreq->completion); |
63ab46ab | 340 | |
b4946ffb | 341 | nfs_direct_req_release(dreq); |
63ab46ab CL |
342 | } |
343 | ||
1385b811 | 344 | static void nfs_direct_readpage_release(struct nfs_page *req) |
1da177e4 | 345 | { |
1e8968c5 | 346 | dprintk("NFS: direct read done (%s/%llu %d@%lld)\n", |
584aa810 | 347 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1e8968c5 | 348 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
584aa810 FI |
349 | req->wb_bytes, |
350 | (long long)req_offset(req)); | |
351 | nfs_release_request(req); | |
fdd1e74c TM |
352 | } |
353 | ||
584aa810 | 354 | static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) |
fdd1e74c | 355 | { |
584aa810 FI |
356 | unsigned long bytes = 0; |
357 | struct nfs_direct_req *dreq = hdr->dreq; | |
fdd1e74c | 358 | |
584aa810 FI |
359 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) |
360 | goto out_put; | |
15ce4a0c CL |
361 | |
362 | spin_lock(&dreq->lock); | |
584aa810 FI |
363 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0)) |
364 | dreq->error = hdr->error; | |
365 | else | |
366 | dreq->count += hdr->good_bytes; | |
367 | spin_unlock(&dreq->lock); | |
368 | ||
4bd8b010 TM |
369 | while (!list_empty(&hdr->pages)) { |
370 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
371 | struct page *page = req->wb_page; | |
372 | ||
be7e9858 JL |
373 | if (!PageCompound(page) && bytes < hdr->good_bytes) |
374 | set_page_dirty(page); | |
4bd8b010 TM |
375 | bytes += req->wb_bytes; |
376 | nfs_list_remove_request(req); | |
377 | nfs_direct_readpage_release(req); | |
d4a8f367 | 378 | } |
584aa810 | 379 | out_put: |
607f31e8 | 380 | if (put_dreq(dreq)) |
9811cd57 | 381 | nfs_direct_complete(dreq, false); |
584aa810 | 382 | hdr->release(hdr); |
1da177e4 LT |
383 | } |
384 | ||
3e9e0ca3 | 385 | static void nfs_read_sync_pgio_error(struct list_head *head) |
cd841605 | 386 | { |
584aa810 | 387 | struct nfs_page *req; |
cd841605 | 388 | |
584aa810 FI |
389 | while (!list_empty(head)) { |
390 | req = nfs_list_entry(head->next); | |
391 | nfs_list_remove_request(req); | |
392 | nfs_release_request(req); | |
393 | } | |
cd841605 FI |
394 | } |
395 | ||
584aa810 FI |
396 | static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) |
397 | { | |
398 | get_dreq(hdr->dreq); | |
399 | } | |
400 | ||
401 | static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { | |
3e9e0ca3 | 402 | .error_cleanup = nfs_read_sync_pgio_error, |
584aa810 FI |
403 | .init_hdr = nfs_direct_pgio_init, |
404 | .completion = nfs_direct_read_completion, | |
405 | }; | |
406 | ||
d4cc948b | 407 | /* |
607f31e8 TM |
408 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
409 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
410 | * bail and stop sending more reads. Read length accounting is | |
411 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
412 | * no requests have been sent, just return an error. | |
1da177e4 | 413 | */ |
1da177e4 | 414 | |
91f79c43 AV |
415 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
416 | struct iov_iter *iter, | |
417 | loff_t pos) | |
418 | { | |
419 | struct nfs_pageio_descriptor desc; | |
420 | struct inode *inode = dreq->inode; | |
421 | ssize_t result = -EINVAL; | |
422 | size_t requested_bytes = 0; | |
423 | size_t rsize = max_t(size_t, NFS_SERVER(inode)->rsize, PAGE_SIZE); | |
e9f7bee1 | 424 | |
16b90578 | 425 | nfs_pageio_init_read(&desc, dreq->inode, false, |
91f79c43 AV |
426 | &nfs_direct_read_completion_ops); |
427 | get_dreq(dreq); | |
428 | desc.pg_dreq = dreq; | |
429 | atomic_inc(&inode->i_dio_count); | |
a564b8f0 | 430 | |
91f79c43 AV |
431 | while (iov_iter_count(iter)) { |
432 | struct page **pagevec; | |
433 | size_t bytes; | |
434 | size_t pgbase; | |
435 | unsigned npages, i; | |
607f31e8 | 436 | |
91f79c43 AV |
437 | result = iov_iter_get_pages_alloc(iter, &pagevec, |
438 | rsize, &pgbase); | |
439 | if (result < 0) | |
440 | break; | |
441 | ||
442 | bytes = result; | |
443 | iov_iter_advance(iter, bytes); | |
444 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; | |
584aa810 FI |
445 | for (i = 0; i < npages; i++) { |
446 | struct nfs_page *req; | |
bf5fc402 | 447 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); |
584aa810 | 448 | /* XXX do we need to do the eof zeroing found in async_filler? */ |
2bfc6e56 | 449 | req = nfs_create_request(dreq->ctx, pagevec[i], NULL, |
584aa810 FI |
450 | pgbase, req_len); |
451 | if (IS_ERR(req)) { | |
584aa810 FI |
452 | result = PTR_ERR(req); |
453 | break; | |
454 | } | |
455 | req->wb_index = pos >> PAGE_SHIFT; | |
456 | req->wb_offset = pos & ~PAGE_MASK; | |
91f79c43 AV |
457 | if (!nfs_pageio_add_request(&desc, req)) { |
458 | result = desc.pg_error; | |
584aa810 | 459 | nfs_release_request(req); |
584aa810 FI |
460 | break; |
461 | } | |
462 | pgbase = 0; | |
463 | bytes -= req_len; | |
91f79c43 | 464 | requested_bytes += req_len; |
584aa810 | 465 | pos += req_len; |
35754bc0 | 466 | dreq->bytes_left -= req_len; |
584aa810 | 467 | } |
6d74743b | 468 | nfs_direct_release_pages(pagevec, npages); |
91f79c43 | 469 | kvfree(pagevec); |
19f73787 CL |
470 | if (result < 0) |
471 | break; | |
19f73787 CL |
472 | } |
473 | ||
584aa810 FI |
474 | nfs_pageio_complete(&desc); |
475 | ||
839f7ad6 CL |
476 | /* |
477 | * If no bytes were started, return the error, and let the | |
478 | * generic layer handle the completion. | |
479 | */ | |
480 | if (requested_bytes == 0) { | |
1f90ee27 | 481 | inode_dio_done(inode); |
839f7ad6 CL |
482 | nfs_direct_req_release(dreq); |
483 | return result < 0 ? result : -EIO; | |
484 | } | |
485 | ||
19f73787 | 486 | if (put_dreq(dreq)) |
9811cd57 | 487 | nfs_direct_complete(dreq, false); |
839f7ad6 | 488 | return 0; |
19f73787 CL |
489 | } |
490 | ||
14a3ec79 CH |
491 | /** |
492 | * nfs_file_direct_read - file direct read operation for NFS files | |
493 | * @iocb: target I/O control block | |
619d30b4 | 494 | * @iter: vector of user buffers into which to read data |
14a3ec79 CH |
495 | * @pos: byte offset in file where reading starts |
496 | * | |
497 | * We use this function for direct reads instead of calling | |
498 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
499 | * the request starts before the end of the file. For that check | |
500 | * to work, we must generate a GETATTR before each direct read, and | |
501 | * even then there is a window between the GETATTR and the subsequent | |
502 | * READ where the file size could change. Our preference is simply | |
503 | * to do all reads the application wants, and the server will take | |
504 | * care of managing the end of file boundary. | |
505 | * | |
506 | * This function also eliminates unnecessarily updating the file's | |
507 | * atime locally, as the NFS server sets the file's atime, and this | |
508 | * client must read the updated atime from the server back into its | |
509 | * cache. | |
510 | */ | |
619d30b4 | 511 | ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter, |
e19a8a0a | 512 | loff_t pos) |
1da177e4 | 513 | { |
14a3ec79 CH |
514 | struct file *file = iocb->ki_filp; |
515 | struct address_space *mapping = file->f_mapping; | |
516 | struct inode *inode = mapping->host; | |
1da177e4 | 517 | struct nfs_direct_req *dreq; |
b3c54de6 | 518 | struct nfs_lock_context *l_ctx; |
14a3ec79 | 519 | ssize_t result = -EINVAL; |
a6cbcd4a | 520 | size_t count = iov_iter_count(iter); |
14a3ec79 CH |
521 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); |
522 | ||
523 | dfprintk(FILE, "NFS: direct read(%pD2, %zd@%Ld)\n", | |
524 | file, count, (long long) pos); | |
525 | ||
526 | result = 0; | |
527 | if (!count) | |
528 | goto out; | |
529 | ||
d0b9875d | 530 | mutex_lock(&inode->i_mutex); |
14a3ec79 CH |
531 | result = nfs_sync_mapping(mapping); |
532 | if (result) | |
d0b9875d | 533 | goto out_unlock; |
1da177e4 | 534 | |
14a3ec79 CH |
535 | task_io_account_read(count); |
536 | ||
537 | result = -ENOMEM; | |
607f31e8 | 538 | dreq = nfs_direct_req_alloc(); |
f11ac8db | 539 | if (dreq == NULL) |
d0b9875d | 540 | goto out_unlock; |
1da177e4 | 541 | |
91d5b470 | 542 | dreq->inode = inode; |
619d30b4 | 543 | dreq->bytes_left = count; |
cd3758e3 | 544 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
b3c54de6 TM |
545 | l_ctx = nfs_get_lock_context(dreq->ctx); |
546 | if (IS_ERR(l_ctx)) { | |
547 | result = PTR_ERR(l_ctx); | |
f11ac8db | 548 | goto out_release; |
b3c54de6 TM |
549 | } |
550 | dreq->l_ctx = l_ctx; | |
487b8372 CL |
551 | if (!is_sync_kiocb(iocb)) |
552 | dreq->iocb = iocb; | |
1da177e4 | 553 | |
619d30b4 | 554 | NFS_I(inode)->read_io += count; |
91f79c43 | 555 | result = nfs_direct_read_schedule_iovec(dreq, iter, pos); |
d0b9875d CH |
556 | |
557 | mutex_unlock(&inode->i_mutex); | |
558 | ||
14a3ec79 | 559 | if (!result) { |
607f31e8 | 560 | result = nfs_direct_wait(dreq); |
14a3ec79 CH |
561 | if (result > 0) |
562 | iocb->ki_pos = pos + result; | |
563 | } | |
d0b9875d CH |
564 | |
565 | nfs_direct_req_release(dreq); | |
566 | return result; | |
567 | ||
f11ac8db | 568 | out_release: |
b4946ffb | 569 | nfs_direct_req_release(dreq); |
d0b9875d CH |
570 | out_unlock: |
571 | mutex_unlock(&inode->i_mutex); | |
f11ac8db | 572 | out: |
1da177e4 LT |
573 | return result; |
574 | } | |
575 | ||
085d1e33 TH |
576 | static void |
577 | nfs_direct_write_scan_commit_list(struct inode *inode, | |
578 | struct list_head *list, | |
579 | struct nfs_commit_info *cinfo) | |
580 | { | |
581 | spin_lock(cinfo->lock); | |
582 | #ifdef CONFIG_NFS_V4_1 | |
583 | if (cinfo->ds != NULL && cinfo->ds->nwritten != 0) | |
584 | NFS_SERVER(inode)->pnfs_curr_ld->recover_commit_reqs(list, cinfo); | |
585 | #endif | |
586 | nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0); | |
587 | spin_unlock(cinfo->lock); | |
588 | } | |
589 | ||
fad61490 TM |
590 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) |
591 | { | |
1763da12 FI |
592 | struct nfs_pageio_descriptor desc; |
593 | struct nfs_page *req, *tmp; | |
594 | LIST_HEAD(reqs); | |
595 | struct nfs_commit_info cinfo; | |
596 | LIST_HEAD(failed); | |
597 | ||
598 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
085d1e33 | 599 | nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); |
1da177e4 | 600 | |
fad61490 | 601 | dreq->count = 0; |
607f31e8 TM |
602 | get_dreq(dreq); |
603 | ||
a20c93e3 | 604 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false, |
1763da12 FI |
605 | &nfs_direct_write_completion_ops); |
606 | desc.pg_dreq = dreq; | |
fedb595c | 607 | |
1763da12 FI |
608 | list_for_each_entry_safe(req, tmp, &reqs, wb_list) { |
609 | if (!nfs_pageio_add_request(&desc, req)) { | |
4035c248 | 610 | nfs_list_remove_request(req); |
1763da12 FI |
611 | nfs_list_add_request(req, &failed); |
612 | spin_lock(cinfo.lock); | |
613 | dreq->flags = 0; | |
614 | dreq->error = -EIO; | |
615 | spin_unlock(cinfo.lock); | |
616 | } | |
5a695da2 | 617 | nfs_release_request(req); |
1763da12 FI |
618 | } |
619 | nfs_pageio_complete(&desc); | |
fad61490 | 620 | |
4035c248 TM |
621 | while (!list_empty(&failed)) { |
622 | req = nfs_list_entry(failed.next); | |
623 | nfs_list_remove_request(req); | |
1d1afcbc | 624 | nfs_unlock_and_release_request(req); |
4035c248 | 625 | } |
fad61490 | 626 | |
1763da12 FI |
627 | if (put_dreq(dreq)) |
628 | nfs_direct_write_complete(dreq, dreq->inode); | |
c9d8f89d TM |
629 | } |
630 | ||
1763da12 | 631 | static void nfs_direct_commit_complete(struct nfs_commit_data *data) |
c9d8f89d | 632 | { |
0b7c0153 | 633 | struct nfs_direct_req *dreq = data->dreq; |
1763da12 FI |
634 | struct nfs_commit_info cinfo; |
635 | struct nfs_page *req; | |
c9d8f89d TM |
636 | int status = data->task.tk_status; |
637 | ||
1763da12 | 638 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
c9d8f89d | 639 | if (status < 0) { |
60fa3f76 | 640 | dprintk("NFS: %5u commit failed with error %d.\n", |
1763da12 | 641 | data->task.tk_pid, status); |
fad61490 | 642 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
5002c586 | 643 | } else if (nfs_direct_cmp_commit_data_verf(dreq, data)) { |
c9d8f89d | 644 | dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid); |
fad61490 | 645 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
1da177e4 LT |
646 | } |
647 | ||
c9d8f89d | 648 | dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); |
1763da12 FI |
649 | while (!list_empty(&data->pages)) { |
650 | req = nfs_list_entry(data->pages.next); | |
651 | nfs_list_remove_request(req); | |
652 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) { | |
653 | /* Note the rewrite will go through mds */ | |
654 | nfs_mark_request_commit(req, NULL, &cinfo); | |
906369e4 FI |
655 | } else |
656 | nfs_release_request(req); | |
1d1afcbc | 657 | nfs_unlock_and_release_request(req); |
1763da12 FI |
658 | } |
659 | ||
660 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
661 | nfs_direct_write_complete(dreq, data->inode); | |
1da177e4 LT |
662 | } |
663 | ||
1763da12 FI |
664 | static void nfs_direct_error_cleanup(struct nfs_inode *nfsi) |
665 | { | |
666 | /* There is no lock to clear */ | |
667 | } | |
668 | ||
669 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { | |
670 | .completion = nfs_direct_commit_complete, | |
671 | .error_cleanup = nfs_direct_error_cleanup, | |
fad61490 TM |
672 | }; |
673 | ||
674 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 675 | { |
1763da12 FI |
676 | int res; |
677 | struct nfs_commit_info cinfo; | |
678 | LIST_HEAD(mds_list); | |
679 | ||
680 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
681 | nfs_scan_commit(dreq->inode, &mds_list, &cinfo); | |
682 | res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); | |
683 | if (res < 0) /* res == -ENOMEM */ | |
684 | nfs_direct_write_reschedule(dreq); | |
fad61490 | 685 | } |
1da177e4 | 686 | |
1763da12 | 687 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
fad61490 | 688 | { |
1763da12 | 689 | struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); |
fad61490 | 690 | int flags = dreq->flags; |
1da177e4 | 691 | |
fad61490 TM |
692 | dreq->flags = 0; |
693 | switch (flags) { | |
694 | case NFS_ODIRECT_DO_COMMIT: | |
695 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 696 | break; |
fad61490 TM |
697 | case NFS_ODIRECT_RESCHED_WRITES: |
698 | nfs_direct_write_reschedule(dreq); | |
699 | break; | |
700 | default: | |
9811cd57 | 701 | nfs_direct_complete(dreq, true); |
fad61490 TM |
702 | } |
703 | } | |
1da177e4 | 704 | |
1763da12 | 705 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
fad61490 | 706 | { |
1763da12 | 707 | schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */ |
fad61490 | 708 | } |
1763da12 | 709 | |
1763da12 FI |
710 | static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) |
711 | { | |
712 | struct nfs_direct_req *dreq = hdr->dreq; | |
713 | struct nfs_commit_info cinfo; | |
c65e6254 | 714 | bool request_commit = false; |
1763da12 FI |
715 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); |
716 | ||
717 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
718 | goto out_put; | |
719 | ||
720 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
721 | ||
722 | spin_lock(&dreq->lock); | |
723 | ||
724 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
725 | dreq->flags = 0; | |
726 | dreq->error = hdr->error; | |
727 | } | |
c65e6254 | 728 | if (dreq->error == 0) { |
1763da12 | 729 | dreq->count += hdr->good_bytes; |
c65e6254 | 730 | if (nfs_write_need_commit(hdr)) { |
1763da12 | 731 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) |
c65e6254 | 732 | request_commit = true; |
1763da12 | 733 | else if (dreq->flags == 0) { |
5002c586 | 734 | nfs_direct_set_hdr_verf(dreq, hdr); |
c65e6254 | 735 | request_commit = true; |
1763da12 FI |
736 | dreq->flags = NFS_ODIRECT_DO_COMMIT; |
737 | } else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) { | |
c65e6254 WAA |
738 | request_commit = true; |
739 | if (nfs_direct_set_or_cmp_hdr_verf(dreq, hdr)) | |
5002c586 WAA |
740 | dreq->flags = |
741 | NFS_ODIRECT_RESCHED_WRITES; | |
1763da12 FI |
742 | } |
743 | } | |
744 | } | |
745 | spin_unlock(&dreq->lock); | |
746 | ||
747 | while (!list_empty(&hdr->pages)) { | |
2bfc6e56 | 748 | |
1763da12 FI |
749 | req = nfs_list_entry(hdr->pages.next); |
750 | nfs_list_remove_request(req); | |
c65e6254 | 751 | if (request_commit) { |
04277086 | 752 | kref_get(&req->wb_kref); |
1763da12 | 753 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
1763da12 | 754 | } |
1d1afcbc | 755 | nfs_unlock_and_release_request(req); |
1763da12 FI |
756 | } |
757 | ||
758 | out_put: | |
759 | if (put_dreq(dreq)) | |
760 | nfs_direct_write_complete(dreq, hdr->inode); | |
761 | hdr->release(hdr); | |
762 | } | |
763 | ||
3e9e0ca3 TM |
764 | static void nfs_write_sync_pgio_error(struct list_head *head) |
765 | { | |
766 | struct nfs_page *req; | |
767 | ||
768 | while (!list_empty(head)) { | |
769 | req = nfs_list_entry(head->next); | |
770 | nfs_list_remove_request(req); | |
1d1afcbc | 771 | nfs_unlock_and_release_request(req); |
3e9e0ca3 TM |
772 | } |
773 | } | |
774 | ||
1763da12 | 775 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { |
3e9e0ca3 | 776 | .error_cleanup = nfs_write_sync_pgio_error, |
1763da12 FI |
777 | .init_hdr = nfs_direct_pgio_init, |
778 | .completion = nfs_direct_write_completion, | |
779 | }; | |
780 | ||
91f79c43 AV |
781 | |
782 | /* | |
783 | * NB: Return the value of the first error return code. Subsequent | |
784 | * errors after the first one are ignored. | |
785 | */ | |
786 | /* | |
787 | * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE | |
788 | * operation. If nfs_writedata_alloc() or get_user_pages() fails, | |
789 | * bail and stop sending more writes. Write length accounting is | |
790 | * handled automatically by nfs_direct_write_result(). Otherwise, if | |
791 | * no requests have been sent, just return an error. | |
792 | */ | |
19f73787 | 793 | static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, |
619d30b4 | 794 | struct iov_iter *iter, |
91f79c43 | 795 | loff_t pos) |
19f73787 | 796 | { |
1763da12 | 797 | struct nfs_pageio_descriptor desc; |
1d59d61f | 798 | struct inode *inode = dreq->inode; |
19f73787 CL |
799 | ssize_t result = 0; |
800 | size_t requested_bytes = 0; | |
91f79c43 | 801 | size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE); |
19f73787 | 802 | |
a20c93e3 | 803 | nfs_pageio_init_write(&desc, inode, FLUSH_COND_STABLE, false, |
1763da12 FI |
804 | &nfs_direct_write_completion_ops); |
805 | desc.pg_dreq = dreq; | |
19f73787 | 806 | get_dreq(dreq); |
1d59d61f | 807 | atomic_inc(&inode->i_dio_count); |
19f73787 | 808 | |
91f79c43 AV |
809 | NFS_I(inode)->write_io += iov_iter_count(iter); |
810 | while (iov_iter_count(iter)) { | |
811 | struct page **pagevec; | |
812 | size_t bytes; | |
813 | size_t pgbase; | |
814 | unsigned npages, i; | |
815 | ||
816 | result = iov_iter_get_pages_alloc(iter, &pagevec, | |
817 | wsize, &pgbase); | |
19f73787 CL |
818 | if (result < 0) |
819 | break; | |
91f79c43 AV |
820 | |
821 | bytes = result; | |
822 | iov_iter_advance(iter, bytes); | |
823 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; | |
824 | for (i = 0; i < npages; i++) { | |
825 | struct nfs_page *req; | |
826 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); | |
827 | ||
16b90578 | 828 | req = nfs_create_request(dreq->ctx, pagevec[i], NULL, |
91f79c43 AV |
829 | pgbase, req_len); |
830 | if (IS_ERR(req)) { | |
831 | result = PTR_ERR(req); | |
832 | break; | |
833 | } | |
834 | nfs_lock_request(req); | |
835 | req->wb_index = pos >> PAGE_SHIFT; | |
836 | req->wb_offset = pos & ~PAGE_MASK; | |
837 | if (!nfs_pageio_add_request(&desc, req)) { | |
838 | result = desc.pg_error; | |
839 | nfs_unlock_and_release_request(req); | |
840 | break; | |
841 | } | |
842 | pgbase = 0; | |
843 | bytes -= req_len; | |
844 | requested_bytes += req_len; | |
845 | pos += req_len; | |
846 | dreq->bytes_left -= req_len; | |
847 | } | |
848 | nfs_direct_release_pages(pagevec, npages); | |
849 | kvfree(pagevec); | |
850 | if (result < 0) | |
19f73787 | 851 | break; |
19f73787 | 852 | } |
1763da12 | 853 | nfs_pageio_complete(&desc); |
19f73787 | 854 | |
839f7ad6 CL |
855 | /* |
856 | * If no bytes were started, return the error, and let the | |
857 | * generic layer handle the completion. | |
858 | */ | |
859 | if (requested_bytes == 0) { | |
1d59d61f | 860 | inode_dio_done(inode); |
839f7ad6 CL |
861 | nfs_direct_req_release(dreq); |
862 | return result < 0 ? result : -EIO; | |
863 | } | |
864 | ||
19f73787 CL |
865 | if (put_dreq(dreq)) |
866 | nfs_direct_write_complete(dreq, dreq->inode); | |
839f7ad6 | 867 | return 0; |
19f73787 CL |
868 | } |
869 | ||
1da177e4 LT |
870 | /** |
871 | * nfs_file_direct_write - file direct write operation for NFS files | |
872 | * @iocb: target I/O control block | |
619d30b4 | 873 | * @iter: vector of user buffers from which to write data |
88467055 | 874 | * @pos: byte offset in file where writing starts |
1da177e4 LT |
875 | * |
876 | * We use this function for direct writes instead of calling | |
877 | * generic_file_aio_write() in order to avoid taking the inode | |
878 | * semaphore and updating the i_size. The NFS server will set | |
879 | * the new i_size and this client must read the updated size | |
880 | * back into its cache. We let the server do generic write | |
881 | * parameter checking and report problems. | |
882 | * | |
1da177e4 LT |
883 | * We eliminate local atime updates, see direct read above. |
884 | * | |
885 | * We avoid unnecessary page cache invalidations for normal cached | |
886 | * readers of this file. | |
887 | * | |
888 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
889 | * is no atomic O_APPEND write facility in the NFS protocol. | |
890 | */ | |
619d30b4 | 891 | ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter, |
e19a8a0a | 892 | loff_t pos) |
1da177e4 | 893 | { |
22cd1bf1 | 894 | ssize_t result = -EINVAL; |
1da177e4 | 895 | struct file *file = iocb->ki_filp; |
1da177e4 | 896 | struct address_space *mapping = file->f_mapping; |
22cd1bf1 CH |
897 | struct inode *inode = mapping->host; |
898 | struct nfs_direct_req *dreq; | |
899 | struct nfs_lock_context *l_ctx; | |
a9ab5e84 | 900 | loff_t end; |
a6cbcd4a | 901 | size_t count = iov_iter_count(iter); |
a9ab5e84 CH |
902 | end = (pos + count - 1) >> PAGE_CACHE_SHIFT; |
903 | ||
c216fd70 CL |
904 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); |
905 | ||
6de1472f AV |
906 | dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n", |
907 | file, count, (long long) pos); | |
027445c3 | 908 | |
22cd1bf1 CH |
909 | result = generic_write_checks(file, &pos, &count, 0); |
910 | if (result) | |
1da177e4 | 911 | goto out; |
ce1a8e67 | 912 | |
22cd1bf1 | 913 | result = -EINVAL; |
ce1a8e67 | 914 | if ((ssize_t) count < 0) |
1da177e4 | 915 | goto out; |
22cd1bf1 | 916 | result = 0; |
1da177e4 LT |
917 | if (!count) |
918 | goto out; | |
ce1a8e67 | 919 | |
a9ab5e84 CH |
920 | mutex_lock(&inode->i_mutex); |
921 | ||
22cd1bf1 CH |
922 | result = nfs_sync_mapping(mapping); |
923 | if (result) | |
a9ab5e84 CH |
924 | goto out_unlock; |
925 | ||
926 | if (mapping->nrpages) { | |
927 | result = invalidate_inode_pages2_range(mapping, | |
928 | pos >> PAGE_CACHE_SHIFT, end); | |
929 | if (result) | |
930 | goto out_unlock; | |
931 | } | |
1da177e4 | 932 | |
7ec10f26 KK |
933 | task_io_account_write(count); |
934 | ||
22cd1bf1 CH |
935 | result = -ENOMEM; |
936 | dreq = nfs_direct_req_alloc(); | |
937 | if (!dreq) | |
a9ab5e84 | 938 | goto out_unlock; |
9eafa8cc | 939 | |
22cd1bf1 CH |
940 | dreq->inode = inode; |
941 | dreq->bytes_left = count; | |
942 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); | |
943 | l_ctx = nfs_get_lock_context(dreq->ctx); | |
944 | if (IS_ERR(l_ctx)) { | |
945 | result = PTR_ERR(l_ctx); | |
946 | goto out_release; | |
947 | } | |
948 | dreq->l_ctx = l_ctx; | |
949 | if (!is_sync_kiocb(iocb)) | |
950 | dreq->iocb = iocb; | |
951 | ||
91f79c43 | 952 | result = nfs_direct_write_schedule_iovec(dreq, iter, pos); |
a9ab5e84 CH |
953 | |
954 | if (mapping->nrpages) { | |
955 | invalidate_inode_pages2_range(mapping, | |
956 | pos >> PAGE_CACHE_SHIFT, end); | |
957 | } | |
958 | ||
959 | mutex_unlock(&inode->i_mutex); | |
960 | ||
22cd1bf1 CH |
961 | if (!result) { |
962 | result = nfs_direct_wait(dreq); | |
963 | if (result > 0) { | |
964 | struct inode *inode = mapping->host; | |
965 | ||
966 | iocb->ki_pos = pos + result; | |
967 | spin_lock(&inode->i_lock); | |
968 | if (i_size_read(inode) < iocb->ki_pos) | |
969 | i_size_write(inode, iocb->ki_pos); | |
970 | spin_unlock(&inode->i_lock); | |
971 | } | |
1763da12 | 972 | } |
a9ab5e84 CH |
973 | nfs_direct_req_release(dreq); |
974 | return result; | |
975 | ||
22cd1bf1 CH |
976 | out_release: |
977 | nfs_direct_req_release(dreq); | |
a9ab5e84 CH |
978 | out_unlock: |
979 | mutex_unlock(&inode->i_mutex); | |
1da177e4 | 980 | out: |
22cd1bf1 | 981 | return result; |
1da177e4 LT |
982 | } |
983 | ||
88467055 CL |
984 | /** |
985 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
986 | * | |
987 | */ | |
f7b422b1 | 988 | int __init nfs_init_directcache(void) |
1da177e4 LT |
989 | { |
990 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
991 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
992 | 0, (SLAB_RECLAIM_ACCOUNT| |
993 | SLAB_MEM_SPREAD), | |
20c2df83 | 994 | NULL); |
1da177e4 LT |
995 | if (nfs_direct_cachep == NULL) |
996 | return -ENOMEM; | |
997 | ||
998 | return 0; | |
999 | } | |
1000 | ||
88467055 | 1001 | /** |
f7b422b1 | 1002 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
1003 | * |
1004 | */ | |
266bee88 | 1005 | void nfs_destroy_directcache(void) |
1da177e4 | 1006 | { |
1a1d92c1 | 1007 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 1008 | } |