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> |
1da177e4 LT |
49 | |
50 | #include <linux/nfs_fs.h> | |
51 | #include <linux/nfs_page.h> | |
52 | #include <linux/sunrpc/clnt.h> | |
53 | ||
1da177e4 | 54 | #include <asm/uaccess.h> |
60063497 | 55 | #include <linux/atomic.h> |
1da177e4 | 56 | |
8d5658c9 | 57 | #include "internal.h" |
91d5b470 | 58 | #include "iostat.h" |
1763da12 | 59 | #include "pnfs.h" |
91d5b470 | 60 | |
1da177e4 | 61 | #define NFSDBG_FACILITY NFSDBG_VFS |
1da177e4 | 62 | |
e18b890b | 63 | static struct kmem_cache *nfs_direct_cachep; |
1da177e4 LT |
64 | |
65 | /* | |
66 | * This represents a set of asynchronous requests that we're waiting on | |
67 | */ | |
68 | struct nfs_direct_req { | |
69 | struct kref kref; /* release manager */ | |
15ce4a0c CL |
70 | |
71 | /* I/O parameters */ | |
a8881f5a | 72 | struct nfs_open_context *ctx; /* file open context info */ |
f11ac8db | 73 | struct nfs_lock_context *l_ctx; /* Lock context info */ |
99514f8f | 74 | struct kiocb * iocb; /* controlling i/o request */ |
88467055 | 75 | struct inode * inode; /* target file of i/o */ |
15ce4a0c CL |
76 | |
77 | /* completion state */ | |
607f31e8 | 78 | atomic_t io_count; /* i/os we're waiting for */ |
15ce4a0c | 79 | spinlock_t lock; /* protect completion state */ |
15ce4a0c | 80 | ssize_t count, /* bytes actually processed */ |
1da177e4 | 81 | error; /* any reported error */ |
d72b7a6b | 82 | struct completion completion; /* wait for i/o completion */ |
fad61490 TM |
83 | |
84 | /* commit state */ | |
1763da12 FI |
85 | struct nfs_mds_commit_info mds_cinfo; /* Storage for cinfo */ |
86 | struct pnfs_ds_commit_info ds_cinfo; /* Storage for cinfo */ | |
87 | struct work_struct work; | |
fad61490 TM |
88 | int flags; |
89 | #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */ | |
90 | #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */ | |
91 | struct nfs_writeverf verf; /* unstable write verifier */ | |
1da177e4 LT |
92 | }; |
93 | ||
1763da12 FI |
94 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops; |
95 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops; | |
fad61490 | 96 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode); |
1763da12 | 97 | static void nfs_direct_write_schedule_work(struct work_struct *work); |
607f31e8 TM |
98 | |
99 | static inline void get_dreq(struct nfs_direct_req *dreq) | |
100 | { | |
101 | atomic_inc(&dreq->io_count); | |
102 | } | |
103 | ||
104 | static inline int put_dreq(struct nfs_direct_req *dreq) | |
105 | { | |
106 | return atomic_dec_and_test(&dreq->io_count); | |
107 | } | |
108 | ||
1da177e4 | 109 | /** |
b8a32e2b CL |
110 | * nfs_direct_IO - NFS address space operation for direct I/O |
111 | * @rw: direction (read or write) | |
112 | * @iocb: target I/O control block | |
113 | * @iov: array of vectors that define I/O buffer | |
114 | * @pos: offset in file to begin the operation | |
115 | * @nr_segs: size of iovec array | |
116 | * | |
117 | * The presence of this routine in the address space ops vector means | |
118 | * the NFS client supports direct I/O. However, we shunt off direct | |
119 | * read and write requests before the VFS gets them, so this method | |
120 | * should never be called. | |
1da177e4 | 121 | */ |
b8a32e2b CL |
122 | ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs) |
123 | { | |
b8a32e2b | 124 | dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n", |
01cce933 | 125 | iocb->ki_filp->f_path.dentry->d_name.name, |
e99170ff | 126 | (long long) pos, nr_segs); |
b8a32e2b CL |
127 | |
128 | return -EINVAL; | |
129 | } | |
130 | ||
749e146e | 131 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 132 | { |
749e146e | 133 | unsigned int i; |
607f31e8 TM |
134 | for (i = 0; i < npages; i++) |
135 | page_cache_release(pages[i]); | |
6b45d858 TM |
136 | } |
137 | ||
1763da12 FI |
138 | void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, |
139 | struct nfs_direct_req *dreq) | |
140 | { | |
141 | cinfo->lock = &dreq->lock; | |
142 | cinfo->mds = &dreq->mds_cinfo; | |
143 | cinfo->ds = &dreq->ds_cinfo; | |
144 | cinfo->dreq = dreq; | |
145 | cinfo->completion_ops = &nfs_direct_commit_completion_ops; | |
146 | } | |
147 | ||
93619e59 | 148 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 149 | { |
93619e59 CL |
150 | struct nfs_direct_req *dreq; |
151 | ||
e94b1766 | 152 | dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
153 | if (!dreq) |
154 | return NULL; | |
155 | ||
156 | kref_init(&dreq->kref); | |
607f31e8 | 157 | kref_get(&dreq->kref); |
d72b7a6b | 158 | init_completion(&dreq->completion); |
1763da12 FI |
159 | dreq->mds_cinfo.ncommit = 0; |
160 | atomic_set(&dreq->mds_cinfo.rpcs_out, 0); | |
161 | INIT_LIST_HEAD(&dreq->mds_cinfo.list); | |
162 | INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); | |
163 | memset(&dreq->ds_cinfo, 0, sizeof(dreq->ds_cinfo)); | |
93619e59 | 164 | dreq->iocb = NULL; |
a8881f5a | 165 | dreq->ctx = NULL; |
f11ac8db | 166 | dreq->l_ctx = NULL; |
15ce4a0c | 167 | spin_lock_init(&dreq->lock); |
607f31e8 | 168 | atomic_set(&dreq->io_count, 0); |
15ce4a0c CL |
169 | dreq->count = 0; |
170 | dreq->error = 0; | |
fad61490 | 171 | dreq->flags = 0; |
93619e59 CL |
172 | |
173 | return dreq; | |
1da177e4 LT |
174 | } |
175 | ||
b4946ffb | 176 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
177 | { |
178 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a | 179 | |
f11ac8db TM |
180 | if (dreq->l_ctx != NULL) |
181 | nfs_put_lock_context(dreq->l_ctx); | |
a8881f5a TM |
182 | if (dreq->ctx != NULL) |
183 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
184 | kmem_cache_free(nfs_direct_cachep, dreq); |
185 | } | |
186 | ||
b4946ffb TM |
187 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
188 | { | |
189 | kref_put(&dreq->kref, nfs_direct_req_free); | |
190 | } | |
191 | ||
bc0fb201 CL |
192 | /* |
193 | * Collects and returns the final error value/byte-count. | |
194 | */ | |
195 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
196 | { | |
15ce4a0c | 197 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
198 | |
199 | /* Async requests don't wait here */ | |
200 | if (dreq->iocb) | |
201 | goto out; | |
202 | ||
150030b7 | 203 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 CL |
204 | |
205 | if (!result) | |
15ce4a0c | 206 | result = dreq->error; |
bc0fb201 | 207 | if (!result) |
15ce4a0c | 208 | result = dreq->count; |
bc0fb201 CL |
209 | |
210 | out: | |
bc0fb201 CL |
211 | return (ssize_t) result; |
212 | } | |
213 | ||
63ab46ab | 214 | /* |
607f31e8 TM |
215 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
216 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab CL |
217 | */ |
218 | static void nfs_direct_complete(struct nfs_direct_req *dreq) | |
219 | { | |
63ab46ab | 220 | if (dreq->iocb) { |
15ce4a0c | 221 | long res = (long) dreq->error; |
63ab46ab | 222 | if (!res) |
15ce4a0c | 223 | res = (long) dreq->count; |
63ab46ab | 224 | aio_complete(dreq->iocb, res, 0); |
d72b7a6b TM |
225 | } |
226 | complete_all(&dreq->completion); | |
63ab46ab | 227 | |
b4946ffb | 228 | nfs_direct_req_release(dreq); |
63ab46ab CL |
229 | } |
230 | ||
584aa810 | 231 | void nfs_direct_readpage_release(struct nfs_page *req) |
1da177e4 | 232 | { |
584aa810 FI |
233 | dprintk("NFS: direct read done (%s/%lld %d@%lld)\n", |
234 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
235 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
236 | req->wb_bytes, | |
237 | (long long)req_offset(req)); | |
238 | nfs_release_request(req); | |
fdd1e74c TM |
239 | } |
240 | ||
584aa810 | 241 | static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) |
fdd1e74c | 242 | { |
584aa810 FI |
243 | unsigned long bytes = 0; |
244 | struct nfs_direct_req *dreq = hdr->dreq; | |
fdd1e74c | 245 | |
584aa810 FI |
246 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) |
247 | goto out_put; | |
15ce4a0c CL |
248 | |
249 | spin_lock(&dreq->lock); | |
584aa810 FI |
250 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0)) |
251 | dreq->error = hdr->error; | |
252 | else | |
253 | dreq->count += hdr->good_bytes; | |
254 | spin_unlock(&dreq->lock); | |
255 | ||
256 | if (!test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
257 | while (!list_empty(&hdr->pages)) { | |
258 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
259 | struct page *page = req->wb_page; | |
260 | ||
261 | if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) { | |
262 | if (bytes > hdr->good_bytes) | |
263 | zero_user(page, 0, PAGE_SIZE); | |
264 | else if (hdr->good_bytes - bytes < PAGE_SIZE) | |
265 | zero_user_segment(page, | |
266 | hdr->good_bytes & ~PAGE_MASK, | |
267 | PAGE_SIZE); | |
268 | } | |
269 | bytes += req->wb_bytes; | |
270 | nfs_list_remove_request(req); | |
584aa810 FI |
271 | if (!PageCompound(page)) |
272 | set_page_dirty(page); | |
6d74743b | 273 | nfs_direct_readpage_release(req); |
584aa810 | 274 | } |
d4a8f367 | 275 | } else { |
584aa810 FI |
276 | while (!list_empty(&hdr->pages)) { |
277 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
278 | ||
279 | if (bytes < hdr->good_bytes) | |
280 | if (!PageCompound(req->wb_page)) | |
281 | set_page_dirty(req->wb_page); | |
282 | bytes += req->wb_bytes; | |
584aa810 FI |
283 | nfs_list_remove_request(req); |
284 | nfs_direct_readpage_release(req); | |
285 | } | |
d4a8f367 | 286 | } |
584aa810 | 287 | out_put: |
607f31e8 TM |
288 | if (put_dreq(dreq)) |
289 | nfs_direct_complete(dreq); | |
584aa810 | 290 | hdr->release(hdr); |
1da177e4 LT |
291 | } |
292 | ||
3e9e0ca3 | 293 | static void nfs_read_sync_pgio_error(struct list_head *head) |
cd841605 | 294 | { |
584aa810 | 295 | struct nfs_page *req; |
cd841605 | 296 | |
584aa810 FI |
297 | while (!list_empty(head)) { |
298 | req = nfs_list_entry(head->next); | |
299 | nfs_list_remove_request(req); | |
300 | nfs_release_request(req); | |
301 | } | |
cd841605 FI |
302 | } |
303 | ||
584aa810 FI |
304 | static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) |
305 | { | |
306 | get_dreq(hdr->dreq); | |
307 | } | |
308 | ||
309 | static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { | |
3e9e0ca3 | 310 | .error_cleanup = nfs_read_sync_pgio_error, |
584aa810 FI |
311 | .init_hdr = nfs_direct_pgio_init, |
312 | .completion = nfs_direct_read_completion, | |
313 | }; | |
314 | ||
d4cc948b | 315 | /* |
607f31e8 TM |
316 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
317 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
318 | * bail and stop sending more reads. Read length accounting is | |
319 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
320 | * no requests have been sent, just return an error. | |
1da177e4 | 321 | */ |
584aa810 | 322 | static ssize_t nfs_direct_read_schedule_segment(struct nfs_pageio_descriptor *desc, |
02fe4946 CL |
323 | const struct iovec *iov, |
324 | loff_t pos) | |
1da177e4 | 325 | { |
584aa810 | 326 | struct nfs_direct_req *dreq = desc->pg_dreq; |
a8881f5a | 327 | struct nfs_open_context *ctx = dreq->ctx; |
3d4ff43d | 328 | struct inode *inode = ctx->dentry->d_inode; |
02fe4946 CL |
329 | unsigned long user_addr = (unsigned long)iov->iov_base; |
330 | size_t count = iov->iov_len; | |
5dd602f2 | 331 | size_t rsize = NFS_SERVER(inode)->rsize; |
607f31e8 TM |
332 | unsigned int pgbase; |
333 | int result; | |
334 | ssize_t started = 0; | |
584aa810 FI |
335 | struct page **pagevec = NULL; |
336 | unsigned int npages; | |
607f31e8 | 337 | |
1da177e4 | 338 | do { |
5dd602f2 | 339 | size_t bytes; |
584aa810 | 340 | int i; |
1da177e4 | 341 | |
e9f7bee1 | 342 | pgbase = user_addr & ~PAGE_MASK; |
584aa810 | 343 | bytes = min(max(rsize, PAGE_SIZE), count); |
e9f7bee1 | 344 | |
607f31e8 | 345 | result = -ENOMEM; |
584aa810 FI |
346 | npages = nfs_page_array_len(pgbase, bytes); |
347 | if (!pagevec) | |
348 | pagevec = kmalloc(npages * sizeof(struct page *), | |
349 | GFP_KERNEL); | |
350 | if (!pagevec) | |
4db6e0b7 | 351 | break; |
607f31e8 TM |
352 | down_read(¤t->mm->mmap_sem); |
353 | result = get_user_pages(current, current->mm, user_addr, | |
584aa810 | 354 | npages, 1, 0, pagevec, NULL); |
607f31e8 | 355 | up_read(¤t->mm->mmap_sem); |
584aa810 | 356 | if (result < 0) |
749e146e | 357 | break; |
584aa810 | 358 | if ((unsigned)result < npages) { |
d9df8d6b TM |
359 | bytes = result * PAGE_SIZE; |
360 | if (bytes <= pgbase) { | |
584aa810 | 361 | nfs_direct_release_pages(pagevec, result); |
d9df8d6b TM |
362 | break; |
363 | } | |
364 | bytes -= pgbase; | |
584aa810 | 365 | npages = result; |
607f31e8 TM |
366 | } |
367 | ||
584aa810 FI |
368 | for (i = 0; i < npages; i++) { |
369 | struct nfs_page *req; | |
370 | unsigned int req_len = min(bytes, PAGE_SIZE - pgbase); | |
371 | /* XXX do we need to do the eof zeroing found in async_filler? */ | |
372 | req = nfs_create_request(dreq->ctx, dreq->inode, | |
373 | pagevec[i], | |
374 | pgbase, req_len); | |
375 | if (IS_ERR(req)) { | |
584aa810 FI |
376 | result = PTR_ERR(req); |
377 | break; | |
378 | } | |
379 | req->wb_index = pos >> PAGE_SHIFT; | |
380 | req->wb_offset = pos & ~PAGE_MASK; | |
381 | if (!nfs_pageio_add_request(desc, req)) { | |
382 | result = desc->pg_error; | |
383 | nfs_release_request(req); | |
584aa810 FI |
384 | break; |
385 | } | |
386 | pgbase = 0; | |
387 | bytes -= req_len; | |
388 | started += req_len; | |
389 | user_addr += req_len; | |
390 | pos += req_len; | |
391 | count -= req_len; | |
392 | } | |
6d74743b TM |
393 | /* The nfs_page now hold references to these pages */ |
394 | nfs_direct_release_pages(pagevec, npages); | |
71e8cc00 | 395 | } while (count != 0 && result >= 0); |
607f31e8 | 396 | |
584aa810 FI |
397 | kfree(pagevec); |
398 | ||
607f31e8 | 399 | if (started) |
c216fd70 | 400 | return started; |
607f31e8 | 401 | return result < 0 ? (ssize_t) result : -EFAULT; |
1da177e4 LT |
402 | } |
403 | ||
19f73787 CL |
404 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
405 | const struct iovec *iov, | |
406 | unsigned long nr_segs, | |
407 | loff_t pos) | |
408 | { | |
584aa810 | 409 | struct nfs_pageio_descriptor desc; |
19f73787 CL |
410 | ssize_t result = -EINVAL; |
411 | size_t requested_bytes = 0; | |
412 | unsigned long seg; | |
413 | ||
584aa810 FI |
414 | nfs_pageio_init_read(&desc, dreq->inode, |
415 | &nfs_direct_read_completion_ops); | |
19f73787 | 416 | get_dreq(dreq); |
584aa810 | 417 | desc.pg_dreq = dreq; |
19f73787 CL |
418 | |
419 | for (seg = 0; seg < nr_segs; seg++) { | |
420 | const struct iovec *vec = &iov[seg]; | |
584aa810 | 421 | result = nfs_direct_read_schedule_segment(&desc, vec, pos); |
19f73787 CL |
422 | if (result < 0) |
423 | break; | |
424 | requested_bytes += result; | |
425 | if ((size_t)result < vec->iov_len) | |
426 | break; | |
427 | pos += vec->iov_len; | |
428 | } | |
429 | ||
584aa810 FI |
430 | nfs_pageio_complete(&desc); |
431 | ||
839f7ad6 CL |
432 | /* |
433 | * If no bytes were started, return the error, and let the | |
434 | * generic layer handle the completion. | |
435 | */ | |
436 | if (requested_bytes == 0) { | |
437 | nfs_direct_req_release(dreq); | |
438 | return result < 0 ? result : -EIO; | |
439 | } | |
440 | ||
19f73787 CL |
441 | if (put_dreq(dreq)) |
442 | nfs_direct_complete(dreq); | |
839f7ad6 | 443 | return 0; |
19f73787 CL |
444 | } |
445 | ||
c216fd70 CL |
446 | static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov, |
447 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 448 | { |
f11ac8db | 449 | ssize_t result = -ENOMEM; |
99514f8f | 450 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1da177e4 LT |
451 | struct nfs_direct_req *dreq; |
452 | ||
607f31e8 | 453 | dreq = nfs_direct_req_alloc(); |
f11ac8db TM |
454 | if (dreq == NULL) |
455 | goto out; | |
1da177e4 | 456 | |
91d5b470 | 457 | dreq->inode = inode; |
cd3758e3 | 458 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
f11ac8db TM |
459 | dreq->l_ctx = nfs_get_lock_context(dreq->ctx); |
460 | if (dreq->l_ctx == NULL) | |
461 | goto out_release; | |
487b8372 CL |
462 | if (!is_sync_kiocb(iocb)) |
463 | dreq->iocb = iocb; | |
1da177e4 | 464 | |
c216fd70 | 465 | result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos); |
607f31e8 TM |
466 | if (!result) |
467 | result = nfs_direct_wait(dreq); | |
f11ac8db | 468 | out_release: |
b4946ffb | 469 | nfs_direct_req_release(dreq); |
f11ac8db | 470 | out: |
1da177e4 LT |
471 | return result; |
472 | } | |
473 | ||
fad61490 TM |
474 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
475 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) | |
476 | { | |
1763da12 FI |
477 | struct nfs_pageio_descriptor desc; |
478 | struct nfs_page *req, *tmp; | |
479 | LIST_HEAD(reqs); | |
480 | struct nfs_commit_info cinfo; | |
481 | LIST_HEAD(failed); | |
482 | ||
483 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
484 | pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo); | |
485 | spin_lock(cinfo.lock); | |
486 | nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0); | |
487 | spin_unlock(cinfo.lock); | |
1da177e4 | 488 | |
fad61490 | 489 | dreq->count = 0; |
607f31e8 TM |
490 | get_dreq(dreq); |
491 | ||
1763da12 FI |
492 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, |
493 | &nfs_direct_write_completion_ops); | |
494 | desc.pg_dreq = dreq; | |
fedb595c | 495 | |
1763da12 FI |
496 | list_for_each_entry_safe(req, tmp, &reqs, wb_list) { |
497 | if (!nfs_pageio_add_request(&desc, req)) { | |
498 | nfs_list_add_request(req, &failed); | |
499 | spin_lock(cinfo.lock); | |
500 | dreq->flags = 0; | |
501 | dreq->error = -EIO; | |
502 | spin_unlock(cinfo.lock); | |
503 | } | |
504 | } | |
505 | nfs_pageio_complete(&desc); | |
fad61490 | 506 | |
1763da12 | 507 | while (!list_empty(&failed)) { |
1763da12 FI |
508 | nfs_release_request(req); |
509 | nfs_unlock_request(req); | |
510 | } | |
fad61490 | 511 | |
1763da12 FI |
512 | if (put_dreq(dreq)) |
513 | nfs_direct_write_complete(dreq, dreq->inode); | |
c9d8f89d TM |
514 | } |
515 | ||
1763da12 | 516 | static void nfs_direct_commit_complete(struct nfs_commit_data *data) |
c9d8f89d | 517 | { |
0b7c0153 | 518 | struct nfs_direct_req *dreq = data->dreq; |
1763da12 FI |
519 | struct nfs_commit_info cinfo; |
520 | struct nfs_page *req; | |
c9d8f89d TM |
521 | int status = data->task.tk_status; |
522 | ||
1763da12 | 523 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
c9d8f89d | 524 | if (status < 0) { |
60fa3f76 | 525 | dprintk("NFS: %5u commit failed with error %d.\n", |
1763da12 | 526 | data->task.tk_pid, status); |
fad61490 | 527 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
60fa3f76 | 528 | } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) { |
c9d8f89d | 529 | dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid); |
fad61490 | 530 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; |
1da177e4 LT |
531 | } |
532 | ||
c9d8f89d | 533 | dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); |
1763da12 FI |
534 | while (!list_empty(&data->pages)) { |
535 | req = nfs_list_entry(data->pages.next); | |
536 | nfs_list_remove_request(req); | |
537 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) { | |
538 | /* Note the rewrite will go through mds */ | |
539 | nfs_mark_request_commit(req, NULL, &cinfo); | |
6d74743b | 540 | } else |
1763da12 | 541 | nfs_release_request(req); |
1763da12 FI |
542 | nfs_unlock_request(req); |
543 | } | |
544 | ||
545 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
546 | nfs_direct_write_complete(dreq, data->inode); | |
1da177e4 LT |
547 | } |
548 | ||
1763da12 FI |
549 | static void nfs_direct_error_cleanup(struct nfs_inode *nfsi) |
550 | { | |
551 | /* There is no lock to clear */ | |
552 | } | |
553 | ||
554 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { | |
555 | .completion = nfs_direct_commit_complete, | |
556 | .error_cleanup = nfs_direct_error_cleanup, | |
fad61490 TM |
557 | }; |
558 | ||
559 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 560 | { |
1763da12 FI |
561 | int res; |
562 | struct nfs_commit_info cinfo; | |
563 | LIST_HEAD(mds_list); | |
564 | ||
565 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
566 | nfs_scan_commit(dreq->inode, &mds_list, &cinfo); | |
567 | res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); | |
568 | if (res < 0) /* res == -ENOMEM */ | |
569 | nfs_direct_write_reschedule(dreq); | |
fad61490 | 570 | } |
1da177e4 | 571 | |
1763da12 | 572 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
fad61490 | 573 | { |
1763da12 | 574 | struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); |
fad61490 | 575 | int flags = dreq->flags; |
1da177e4 | 576 | |
fad61490 TM |
577 | dreq->flags = 0; |
578 | switch (flags) { | |
579 | case NFS_ODIRECT_DO_COMMIT: | |
580 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 581 | break; |
fad61490 TM |
582 | case NFS_ODIRECT_RESCHED_WRITES: |
583 | nfs_direct_write_reschedule(dreq); | |
584 | break; | |
585 | default: | |
1763da12 | 586 | nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping); |
fad61490 TM |
587 | nfs_direct_complete(dreq); |
588 | } | |
589 | } | |
1da177e4 | 590 | |
1763da12 | 591 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
fad61490 | 592 | { |
1763da12 | 593 | schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */ |
fad61490 | 594 | } |
1763da12 | 595 | |
fad61490 | 596 | #else |
24fc9211 BS |
597 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
598 | { | |
599 | } | |
1da177e4 | 600 | |
fad61490 TM |
601 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) |
602 | { | |
cd9ae2b6 | 603 | nfs_zap_mapping(inode, inode->i_mapping); |
fad61490 TM |
604 | nfs_direct_complete(dreq); |
605 | } | |
606 | #endif | |
1da177e4 | 607 | |
c9d8f89d TM |
608 | /* |
609 | * NB: Return the value of the first error return code. Subsequent | |
610 | * errors after the first one are ignored. | |
611 | */ | |
462d5b32 | 612 | /* |
607f31e8 TM |
613 | * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE |
614 | * operation. If nfs_writedata_alloc() or get_user_pages() fails, | |
615 | * bail and stop sending more writes. Write length accounting is | |
616 | * handled automatically by nfs_direct_write_result(). Otherwise, if | |
617 | * no requests have been sent, just return an error. | |
462d5b32 | 618 | */ |
1763da12 | 619 | static ssize_t nfs_direct_write_schedule_segment(struct nfs_pageio_descriptor *desc, |
02fe4946 | 620 | const struct iovec *iov, |
1763da12 | 621 | loff_t pos) |
462d5b32 | 622 | { |
1763da12 | 623 | struct nfs_direct_req *dreq = desc->pg_dreq; |
a8881f5a | 624 | struct nfs_open_context *ctx = dreq->ctx; |
3d4ff43d | 625 | struct inode *inode = ctx->dentry->d_inode; |
02fe4946 CL |
626 | unsigned long user_addr = (unsigned long)iov->iov_base; |
627 | size_t count = iov->iov_len; | |
462d5b32 | 628 | size_t wsize = NFS_SERVER(inode)->wsize; |
607f31e8 TM |
629 | unsigned int pgbase; |
630 | int result; | |
631 | ssize_t started = 0; | |
1763da12 FI |
632 | struct page **pagevec = NULL; |
633 | unsigned int npages; | |
82b145c5 | 634 | |
1da177e4 | 635 | do { |
462d5b32 | 636 | size_t bytes; |
1763da12 | 637 | int i; |
462d5b32 | 638 | |
e9f7bee1 | 639 | pgbase = user_addr & ~PAGE_MASK; |
1763da12 | 640 | bytes = min(max(wsize, PAGE_SIZE), count); |
e9f7bee1 | 641 | |
607f31e8 | 642 | result = -ENOMEM; |
1763da12 FI |
643 | npages = nfs_page_array_len(pgbase, bytes); |
644 | if (!pagevec) | |
645 | pagevec = kmalloc(npages * sizeof(struct page *), GFP_KERNEL); | |
646 | if (!pagevec) | |
607f31e8 TM |
647 | break; |
648 | ||
607f31e8 TM |
649 | down_read(¤t->mm->mmap_sem); |
650 | result = get_user_pages(current, current->mm, user_addr, | |
1763da12 | 651 | npages, 0, 0, pagevec, NULL); |
607f31e8 | 652 | up_read(¤t->mm->mmap_sem); |
1763da12 | 653 | if (result < 0) |
749e146e | 654 | break; |
1763da12 FI |
655 | |
656 | if ((unsigned)result < npages) { | |
d9df8d6b TM |
657 | bytes = result * PAGE_SIZE; |
658 | if (bytes <= pgbase) { | |
1763da12 | 659 | nfs_direct_release_pages(pagevec, result); |
d9df8d6b TM |
660 | break; |
661 | } | |
662 | bytes -= pgbase; | |
1763da12 | 663 | npages = result; |
607f31e8 TM |
664 | } |
665 | ||
1763da12 FI |
666 | for (i = 0; i < npages; i++) { |
667 | struct nfs_page *req; | |
668 | unsigned int req_len = min(bytes, PAGE_SIZE - pgbase); | |
1da177e4 | 669 | |
1763da12 FI |
670 | req = nfs_create_request(dreq->ctx, dreq->inode, |
671 | pagevec[i], | |
672 | pgbase, req_len); | |
673 | if (IS_ERR(req)) { | |
1763da12 FI |
674 | result = PTR_ERR(req); |
675 | break; | |
676 | } | |
677 | nfs_lock_request(req); | |
678 | req->wb_index = pos >> PAGE_SHIFT; | |
679 | req->wb_offset = pos & ~PAGE_MASK; | |
680 | if (!nfs_pageio_add_request(desc, req)) { | |
681 | result = desc->pg_error; | |
682 | nfs_unlock_request(req); | |
683 | nfs_release_request(req); | |
71e8cc00 | 684 | break; |
1763da12 FI |
685 | } |
686 | pgbase = 0; | |
687 | bytes -= req_len; | |
688 | started += req_len; | |
689 | user_addr += req_len; | |
690 | pos += req_len; | |
691 | count -= req_len; | |
692 | } | |
6d74743b TM |
693 | /* The nfs_page now hold references to these pages */ |
694 | nfs_direct_release_pages(pagevec, npages); | |
71e8cc00 | 695 | } while (count != 0 && result >= 0); |
607f31e8 | 696 | |
1763da12 FI |
697 | kfree(pagevec); |
698 | ||
607f31e8 | 699 | if (started) |
c216fd70 | 700 | return started; |
607f31e8 | 701 | return result < 0 ? (ssize_t) result : -EFAULT; |
462d5b32 | 702 | } |
1da177e4 | 703 | |
1763da12 FI |
704 | static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) |
705 | { | |
706 | struct nfs_direct_req *dreq = hdr->dreq; | |
707 | struct nfs_commit_info cinfo; | |
708 | int bit = -1; | |
709 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
710 | ||
711 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
712 | goto out_put; | |
713 | ||
714 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
715 | ||
716 | spin_lock(&dreq->lock); | |
717 | ||
718 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
719 | dreq->flags = 0; | |
720 | dreq->error = hdr->error; | |
721 | } | |
722 | if (dreq->error != 0) | |
723 | bit = NFS_IOHDR_ERROR; | |
724 | else { | |
725 | dreq->count += hdr->good_bytes; | |
726 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
727 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
728 | bit = NFS_IOHDR_NEED_RESCHED; | |
729 | } else if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
730 | if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) | |
731 | bit = NFS_IOHDR_NEED_RESCHED; | |
732 | else if (dreq->flags == 0) { | |
733 | memcpy(&dreq->verf, &req->wb_verf, | |
734 | sizeof(dreq->verf)); | |
735 | bit = NFS_IOHDR_NEED_COMMIT; | |
736 | dreq->flags = NFS_ODIRECT_DO_COMMIT; | |
737 | } else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) { | |
738 | if (memcmp(&dreq->verf, &req->wb_verf, sizeof(dreq->verf))) { | |
739 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
740 | bit = NFS_IOHDR_NEED_RESCHED; | |
741 | } else | |
742 | bit = NFS_IOHDR_NEED_COMMIT; | |
743 | } | |
744 | } | |
745 | } | |
746 | spin_unlock(&dreq->lock); | |
747 | ||
748 | while (!list_empty(&hdr->pages)) { | |
749 | req = nfs_list_entry(hdr->pages.next); | |
750 | nfs_list_remove_request(req); | |
751 | switch (bit) { | |
752 | case NFS_IOHDR_NEED_RESCHED: | |
753 | case NFS_IOHDR_NEED_COMMIT: | |
754 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); | |
755 | break; | |
756 | default: | |
1763da12 FI |
757 | nfs_release_request(req); |
758 | } | |
759 | nfs_unlock_request(req); | |
760 | } | |
761 | ||
762 | out_put: | |
763 | if (put_dreq(dreq)) | |
764 | nfs_direct_write_complete(dreq, hdr->inode); | |
765 | hdr->release(hdr); | |
766 | } | |
767 | ||
3e9e0ca3 TM |
768 | static void nfs_write_sync_pgio_error(struct list_head *head) |
769 | { | |
770 | struct nfs_page *req; | |
771 | ||
772 | while (!list_empty(head)) { | |
773 | req = nfs_list_entry(head->next); | |
774 | nfs_list_remove_request(req); | |
775 | nfs_release_request(req); | |
776 | nfs_unlock_request(req); | |
777 | } | |
778 | } | |
779 | ||
1763da12 | 780 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { |
3e9e0ca3 | 781 | .error_cleanup = nfs_write_sync_pgio_error, |
1763da12 FI |
782 | .init_hdr = nfs_direct_pgio_init, |
783 | .completion = nfs_direct_write_completion, | |
784 | }; | |
785 | ||
19f73787 CL |
786 | static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, |
787 | const struct iovec *iov, | |
788 | unsigned long nr_segs, | |
1763da12 | 789 | loff_t pos) |
19f73787 | 790 | { |
1763da12 | 791 | struct nfs_pageio_descriptor desc; |
19f73787 CL |
792 | ssize_t result = 0; |
793 | size_t requested_bytes = 0; | |
794 | unsigned long seg; | |
795 | ||
1763da12 FI |
796 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_COND_STABLE, |
797 | &nfs_direct_write_completion_ops); | |
798 | desc.pg_dreq = dreq; | |
19f73787 CL |
799 | get_dreq(dreq); |
800 | ||
801 | for (seg = 0; seg < nr_segs; seg++) { | |
802 | const struct iovec *vec = &iov[seg]; | |
1763da12 | 803 | result = nfs_direct_write_schedule_segment(&desc, vec, pos); |
19f73787 CL |
804 | if (result < 0) |
805 | break; | |
806 | requested_bytes += result; | |
807 | if ((size_t)result < vec->iov_len) | |
808 | break; | |
809 | pos += vec->iov_len; | |
810 | } | |
1763da12 | 811 | nfs_pageio_complete(&desc); |
19f73787 | 812 | |
839f7ad6 CL |
813 | /* |
814 | * If no bytes were started, return the error, and let the | |
815 | * generic layer handle the completion. | |
816 | */ | |
817 | if (requested_bytes == 0) { | |
818 | nfs_direct_req_release(dreq); | |
819 | return result < 0 ? result : -EIO; | |
820 | } | |
821 | ||
19f73787 CL |
822 | if (put_dreq(dreq)) |
823 | nfs_direct_write_complete(dreq, dreq->inode); | |
839f7ad6 | 824 | return 0; |
19f73787 CL |
825 | } |
826 | ||
c216fd70 CL |
827 | static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov, |
828 | unsigned long nr_segs, loff_t pos, | |
829 | size_t count) | |
462d5b32 | 830 | { |
f11ac8db | 831 | ssize_t result = -ENOMEM; |
c89f2ee5 | 832 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
462d5b32 | 833 | struct nfs_direct_req *dreq; |
1da177e4 | 834 | |
607f31e8 | 835 | dreq = nfs_direct_req_alloc(); |
462d5b32 | 836 | if (!dreq) |
f11ac8db | 837 | goto out; |
1da177e4 | 838 | |
c89f2ee5 | 839 | dreq->inode = inode; |
cd3758e3 | 840 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
f11ac8db | 841 | dreq->l_ctx = nfs_get_lock_context(dreq->ctx); |
568a810d | 842 | if (dreq->l_ctx == NULL) |
f11ac8db | 843 | goto out_release; |
c89f2ee5 CL |
844 | if (!is_sync_kiocb(iocb)) |
845 | dreq->iocb = iocb; | |
1da177e4 | 846 | |
1763da12 | 847 | result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos); |
607f31e8 TM |
848 | if (!result) |
849 | result = nfs_direct_wait(dreq); | |
f11ac8db | 850 | out_release: |
b4946ffb | 851 | nfs_direct_req_release(dreq); |
f11ac8db | 852 | out: |
1da177e4 LT |
853 | return result; |
854 | } | |
855 | ||
856 | /** | |
857 | * nfs_file_direct_read - file direct read operation for NFS files | |
858 | * @iocb: target I/O control block | |
027445c3 BP |
859 | * @iov: vector of user buffers into which to read data |
860 | * @nr_segs: size of iov vector | |
88467055 | 861 | * @pos: byte offset in file where reading starts |
1da177e4 LT |
862 | * |
863 | * We use this function for direct reads instead of calling | |
864 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
865 | * the request starts before the end of the file. For that check | |
866 | * to work, we must generate a GETATTR before each direct read, and | |
867 | * even then there is a window between the GETATTR and the subsequent | |
88467055 | 868 | * READ where the file size could change. Our preference is simply |
1da177e4 LT |
869 | * to do all reads the application wants, and the server will take |
870 | * care of managing the end of file boundary. | |
88467055 | 871 | * |
1da177e4 LT |
872 | * This function also eliminates unnecessarily updating the file's |
873 | * atime locally, as the NFS server sets the file's atime, and this | |
874 | * client must read the updated atime from the server back into its | |
875 | * cache. | |
876 | */ | |
027445c3 BP |
877 | ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov, |
878 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
879 | { |
880 | ssize_t retval = -EINVAL; | |
1da177e4 | 881 | struct file *file = iocb->ki_filp; |
1da177e4 | 882 | struct address_space *mapping = file->f_mapping; |
c216fd70 CL |
883 | size_t count; |
884 | ||
885 | count = iov_length(iov, nr_segs); | |
886 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); | |
1da177e4 | 887 | |
6da24bc9 | 888 | dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n", |
01cce933 JJS |
889 | file->f_path.dentry->d_parent->d_name.name, |
890 | file->f_path.dentry->d_name.name, | |
c216fd70 | 891 | count, (long long) pos); |
1da177e4 | 892 | |
1da177e4 LT |
893 | retval = 0; |
894 | if (!count) | |
895 | goto out; | |
896 | ||
29884df0 TM |
897 | retval = nfs_sync_mapping(mapping); |
898 | if (retval) | |
899 | goto out; | |
1da177e4 | 900 | |
7ec10f26 KK |
901 | task_io_account_read(count); |
902 | ||
c216fd70 | 903 | retval = nfs_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 | 904 | if (retval > 0) |
0cdd80d0 | 905 | iocb->ki_pos = pos + retval; |
1da177e4 LT |
906 | |
907 | out: | |
908 | return retval; | |
909 | } | |
910 | ||
911 | /** | |
912 | * nfs_file_direct_write - file direct write operation for NFS files | |
913 | * @iocb: target I/O control block | |
027445c3 BP |
914 | * @iov: vector of user buffers from which to write data |
915 | * @nr_segs: size of iov vector | |
88467055 | 916 | * @pos: byte offset in file where writing starts |
1da177e4 LT |
917 | * |
918 | * We use this function for direct writes instead of calling | |
919 | * generic_file_aio_write() in order to avoid taking the inode | |
920 | * semaphore and updating the i_size. The NFS server will set | |
921 | * the new i_size and this client must read the updated size | |
922 | * back into its cache. We let the server do generic write | |
923 | * parameter checking and report problems. | |
924 | * | |
1da177e4 LT |
925 | * We eliminate local atime updates, see direct read above. |
926 | * | |
927 | * We avoid unnecessary page cache invalidations for normal cached | |
928 | * readers of this file. | |
929 | * | |
930 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
931 | * is no atomic O_APPEND write facility in the NFS protocol. | |
932 | */ | |
027445c3 BP |
933 | ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov, |
934 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 935 | { |
070ea602 | 936 | ssize_t retval = -EINVAL; |
1da177e4 | 937 | struct file *file = iocb->ki_filp; |
1da177e4 | 938 | struct address_space *mapping = file->f_mapping; |
c216fd70 | 939 | size_t count; |
1da177e4 | 940 | |
c216fd70 CL |
941 | count = iov_length(iov, nr_segs); |
942 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); | |
943 | ||
6da24bc9 | 944 | dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n", |
01cce933 JJS |
945 | file->f_path.dentry->d_parent->d_name.name, |
946 | file->f_path.dentry->d_name.name, | |
c216fd70 | 947 | count, (long long) pos); |
027445c3 | 948 | |
ce1a8e67 CL |
949 | retval = generic_write_checks(file, &pos, &count, 0); |
950 | if (retval) | |
1da177e4 | 951 | goto out; |
ce1a8e67 CL |
952 | |
953 | retval = -EINVAL; | |
954 | if ((ssize_t) count < 0) | |
1da177e4 | 955 | goto out; |
1da177e4 LT |
956 | retval = 0; |
957 | if (!count) | |
958 | goto out; | |
ce1a8e67 | 959 | |
29884df0 TM |
960 | retval = nfs_sync_mapping(mapping); |
961 | if (retval) | |
962 | goto out; | |
1da177e4 | 963 | |
7ec10f26 KK |
964 | task_io_account_write(count); |
965 | ||
c216fd70 | 966 | retval = nfs_direct_write(iocb, iov, nr_segs, pos, count); |
1763da12 FI |
967 | if (retval > 0) { |
968 | struct inode *inode = mapping->host; | |
9eafa8cc | 969 | |
ce1a8e67 | 970 | iocb->ki_pos = pos + retval; |
1763da12 FI |
971 | spin_lock(&inode->i_lock); |
972 | if (i_size_read(inode) < iocb->ki_pos) | |
973 | i_size_write(inode, iocb->ki_pos); | |
974 | spin_unlock(&inode->i_lock); | |
975 | } | |
1da177e4 LT |
976 | out: |
977 | return retval; | |
978 | } | |
979 | ||
88467055 CL |
980 | /** |
981 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
982 | * | |
983 | */ | |
f7b422b1 | 984 | int __init nfs_init_directcache(void) |
1da177e4 LT |
985 | { |
986 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
987 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
988 | 0, (SLAB_RECLAIM_ACCOUNT| |
989 | SLAB_MEM_SPREAD), | |
20c2df83 | 990 | NULL); |
1da177e4 LT |
991 | if (nfs_direct_cachep == NULL) |
992 | return -ENOMEM; | |
993 | ||
994 | return 0; | |
995 | } | |
996 | ||
88467055 | 997 | /** |
f7b422b1 | 998 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
999 | * |
1000 | */ | |
266bee88 | 1001 | void nfs_destroy_directcache(void) |
1da177e4 | 1002 | { |
1a1d92c1 | 1003 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 1004 | } |