]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/nfs/direct.c | |
4 | * | |
5 | * Copyright (C) 2003 by Chuck Lever <cel@netapp.com> | |
6 | * | |
7 | * High-performance uncached I/O for the Linux NFS client | |
8 | * | |
9 | * There are important applications whose performance or correctness | |
10 | * depends on uncached access to file data. Database clusters | |
88467055 | 11 | * (multiple copies of the same instance running on separate hosts) |
1da177e4 | 12 | * implement their own cache coherency protocol that subsumes file |
88467055 CL |
13 | * system cache protocols. Applications that process datasets |
14 | * considerably larger than the client's memory do not always benefit | |
15 | * from a local cache. A streaming video server, for instance, has no | |
1da177e4 LT |
16 | * need to cache the contents of a file. |
17 | * | |
18 | * When an application requests uncached I/O, all read and write requests | |
19 | * are made directly to the server; data stored or fetched via these | |
20 | * requests is not cached in the Linux page cache. The client does not | |
21 | * correct unaligned requests from applications. All requested bytes are | |
22 | * held on permanent storage before a direct write system call returns to | |
23 | * an application. | |
24 | * | |
25 | * Solaris implements an uncached I/O facility called directio() that | |
26 | * is used for backups and sequential I/O to very large files. Solaris | |
27 | * also supports uncaching whole NFS partitions with "-o forcedirectio," | |
28 | * an undocumented mount option. | |
29 | * | |
30 | * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with | |
31 | * help from Andrew Morton. | |
32 | * | |
33 | * 18 Dec 2001 Initial implementation for 2.4 --cel | |
34 | * 08 Jul 2002 Version for 2.4.19, with bug fixes --trondmy | |
35 | * 08 Jun 2003 Port to 2.5 APIs --cel | |
36 | * 31 Mar 2004 Handle direct I/O without VFS support --cel | |
37 | * 15 Sep 2004 Parallel async reads --cel | |
88467055 | 38 | * 04 May 2005 support O_DIRECT with aio --cel |
1da177e4 LT |
39 | * |
40 | */ | |
41 | ||
1da177e4 LT |
42 | #include <linux/errno.h> |
43 | #include <linux/sched.h> | |
44 | #include <linux/kernel.h> | |
1da177e4 LT |
45 | #include <linux/file.h> |
46 | #include <linux/pagemap.h> | |
47 | #include <linux/kref.h> | |
5a0e3ad6 | 48 | #include <linux/slab.h> |
7ec10f26 | 49 | #include <linux/task_io_accounting_ops.h> |
6296556f | 50 | #include <linux/module.h> |
1da177e4 LT |
51 | |
52 | #include <linux/nfs_fs.h> | |
53 | #include <linux/nfs_page.h> | |
54 | #include <linux/sunrpc/clnt.h> | |
55 | ||
7c0f6ba6 | 56 | #include <linux/uaccess.h> |
60063497 | 57 | #include <linux/atomic.h> |
1da177e4 | 58 | |
8d5658c9 | 59 | #include "internal.h" |
91d5b470 | 60 | #include "iostat.h" |
1763da12 | 61 | #include "pnfs.h" |
a6b5a28e | 62 | #include "fscache.h" |
8efc4bbe | 63 | #include "nfstrace.h" |
91d5b470 | 64 | |
1da177e4 | 65 | #define NFSDBG_FACILITY NFSDBG_VFS |
1da177e4 | 66 | |
e18b890b | 67 | static struct kmem_cache *nfs_direct_cachep; |
1da177e4 | 68 | |
1763da12 FI |
69 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops; |
70 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops; | |
4d3b55d3 | 71 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq); |
1763da12 | 72 | static void nfs_direct_write_schedule_work(struct work_struct *work); |
607f31e8 TM |
73 | |
74 | static inline void get_dreq(struct nfs_direct_req *dreq) | |
75 | { | |
76 | atomic_inc(&dreq->io_count); | |
77 | } | |
78 | ||
79 | static inline int put_dreq(struct nfs_direct_req *dreq) | |
80 | { | |
81 | return atomic_dec_and_test(&dreq->io_count); | |
82 | } | |
83 | ||
0a00b77b | 84 | static void |
031d73ed TM |
85 | nfs_direct_handle_truncated(struct nfs_direct_req *dreq, |
86 | const struct nfs_pgio_header *hdr, | |
87 | ssize_t dreq_len) | |
0a00b77b | 88 | { |
031d73ed TM |
89 | if (!(test_bit(NFS_IOHDR_ERROR, &hdr->flags) || |
90 | test_bit(NFS_IOHDR_EOF, &hdr->flags))) | |
91 | return; | |
92 | if (dreq->max_count >= dreq_len) { | |
93 | dreq->max_count = dreq_len; | |
94 | if (dreq->count > dreq_len) | |
95 | dreq->count = dreq_len; | |
96 | ||
97 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) | |
98 | dreq->error = hdr->error; | |
99 | else /* Clear outstanding error if this is EOF */ | |
100 | dreq->error = 0; | |
101 | } | |
031d73ed | 102 | } |
0a00b77b | 103 | |
031d73ed TM |
104 | static void |
105 | nfs_direct_count_bytes(struct nfs_direct_req *dreq, | |
106 | const struct nfs_pgio_header *hdr) | |
107 | { | |
031d73ed TM |
108 | loff_t hdr_end = hdr->io_start + hdr->good_bytes; |
109 | ssize_t dreq_len = 0; | |
ed3743a6 | 110 | |
031d73ed TM |
111 | if (hdr_end > dreq->io_start) |
112 | dreq_len = hdr_end - dreq->io_start; | |
0a00b77b | 113 | |
031d73ed | 114 | nfs_direct_handle_truncated(dreq, hdr, dreq_len); |
0a00b77b | 115 | |
031d73ed TM |
116 | if (dreq_len > dreq->max_count) |
117 | dreq_len = dreq->max_count; | |
118 | ||
031d73ed TM |
119 | if (dreq->count < dreq_len) |
120 | dreq->count = dreq_len; | |
0a00b77b WAA |
121 | } |
122 | ||
1da177e4 | 123 | /** |
eb79f3af | 124 | * nfs_swap_rw - NFS address space operation for swap I/O |
b8a32e2b | 125 | * @iocb: target I/O control block |
90090ae6 | 126 | * @iter: I/O buffer |
b8a32e2b | 127 | * |
eb79f3af | 128 | * Perform IO to the swap-file. This is much like direct IO. |
1da177e4 | 129 | */ |
eb79f3af | 130 | int nfs_swap_rw(struct kiocb *iocb, struct iov_iter *iter) |
b8a32e2b | 131 | { |
eb79f3af | 132 | ssize_t ret; |
ee8a1a8b | 133 | |
66ee59af | 134 | VM_BUG_ON(iov_iter_count(iter) != PAGE_SIZE); |
a564b8f0 | 135 | |
6f673763 | 136 | if (iov_iter_rw(iter) == READ) |
eb79f3af N |
137 | ret = nfs_file_direct_read(iocb, iter, true); |
138 | else | |
139 | ret = nfs_file_direct_write(iocb, iter, true); | |
140 | if (ret < 0) | |
141 | return ret; | |
142 | return 0; | |
b8a32e2b CL |
143 | } |
144 | ||
749e146e | 145 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 146 | { |
749e146e | 147 | unsigned int i; |
607f31e8 | 148 | for (i = 0; i < npages; i++) |
09cbfeaf | 149 | put_page(pages[i]); |
6b45d858 TM |
150 | } |
151 | ||
1763da12 FI |
152 | void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, |
153 | struct nfs_direct_req *dreq) | |
154 | { | |
fe238e60 | 155 | cinfo->inode = dreq->inode; |
1763da12 FI |
156 | cinfo->mds = &dreq->mds_cinfo; |
157 | cinfo->ds = &dreq->ds_cinfo; | |
158 | cinfo->dreq = dreq; | |
159 | cinfo->completion_ops = &nfs_direct_commit_completion_ops; | |
160 | } | |
161 | ||
93619e59 | 162 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 163 | { |
93619e59 CL |
164 | struct nfs_direct_req *dreq; |
165 | ||
292f3eee | 166 | dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
167 | if (!dreq) |
168 | return NULL; | |
169 | ||
170 | kref_init(&dreq->kref); | |
607f31e8 | 171 | kref_get(&dreq->kref); |
d72b7a6b | 172 | init_completion(&dreq->completion); |
1763da12 | 173 | INIT_LIST_HEAD(&dreq->mds_cinfo.list); |
c21e7168 | 174 | pnfs_init_ds_commit_info(&dreq->ds_cinfo); |
1763da12 | 175 | INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); |
15ce4a0c | 176 | spin_lock_init(&dreq->lock); |
93619e59 CL |
177 | |
178 | return dreq; | |
1da177e4 LT |
179 | } |
180 | ||
b4946ffb | 181 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
182 | { |
183 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a | 184 | |
18f41296 | 185 | pnfs_release_ds_info(&dreq->ds_cinfo, dreq->inode); |
f11ac8db TM |
186 | if (dreq->l_ctx != NULL) |
187 | nfs_put_lock_context(dreq->l_ctx); | |
a8881f5a TM |
188 | if (dreq->ctx != NULL) |
189 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
190 | kmem_cache_free(nfs_direct_cachep, dreq); |
191 | } | |
192 | ||
b4946ffb TM |
193 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
194 | { | |
195 | kref_put(&dreq->kref, nfs_direct_req_free); | |
196 | } | |
197 | ||
6296556f PT |
198 | ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq) |
199 | { | |
200 | return dreq->bytes_left; | |
201 | } | |
202 | EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left); | |
203 | ||
bc0fb201 CL |
204 | /* |
205 | * Collects and returns the final error value/byte-count. | |
206 | */ | |
207 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
208 | { | |
15ce4a0c | 209 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
210 | |
211 | /* Async requests don't wait here */ | |
212 | if (dreq->iocb) | |
213 | goto out; | |
214 | ||
150030b7 | 215 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 | 216 | |
d2a7de0b TM |
217 | if (!result) { |
218 | result = dreq->count; | |
219 | WARN_ON_ONCE(dreq->count < 0); | |
220 | } | |
bc0fb201 | 221 | if (!result) |
15ce4a0c | 222 | result = dreq->error; |
bc0fb201 CL |
223 | |
224 | out: | |
bc0fb201 CL |
225 | return (ssize_t) result; |
226 | } | |
227 | ||
63ab46ab | 228 | /* |
607f31e8 TM |
229 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
230 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab | 231 | */ |
f7b5c340 | 232 | static void nfs_direct_complete(struct nfs_direct_req *dreq) |
63ab46ab | 233 | { |
9811cd57 CH |
234 | struct inode *inode = dreq->inode; |
235 | ||
65caafd0 OK |
236 | inode_dio_end(inode); |
237 | ||
2a009ec9 | 238 | if (dreq->iocb) { |
15ce4a0c | 239 | long res = (long) dreq->error; |
d2a7de0b | 240 | if (dreq->count != 0) { |
15ce4a0c | 241 | res = (long) dreq->count; |
d2a7de0b TM |
242 | WARN_ON_ONCE(dreq->count < 0); |
243 | } | |
6b19b766 | 244 | dreq->iocb->ki_complete(dreq->iocb, res); |
d72b7a6b | 245 | } |
2a009ec9 | 246 | |
024de8f1 | 247 | complete(&dreq->completion); |
63ab46ab | 248 | |
b4946ffb | 249 | nfs_direct_req_release(dreq); |
63ab46ab CL |
250 | } |
251 | ||
584aa810 | 252 | static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) |
fdd1e74c | 253 | { |
584aa810 FI |
254 | unsigned long bytes = 0; |
255 | struct nfs_direct_req *dreq = hdr->dreq; | |
fdd1e74c | 256 | |
15ce4a0c | 257 | spin_lock(&dreq->lock); |
eb2c50da TM |
258 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
259 | spin_unlock(&dreq->lock); | |
260 | goto out_put; | |
261 | } | |
262 | ||
031d73ed | 263 | nfs_direct_count_bytes(dreq, hdr); |
584aa810 FI |
264 | spin_unlock(&dreq->lock); |
265 | ||
4bd8b010 TM |
266 | while (!list_empty(&hdr->pages)) { |
267 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
268 | struct page *page = req->wb_page; | |
269 | ||
ad3cba22 DK |
270 | if (!PageCompound(page) && bytes < hdr->good_bytes && |
271 | (dreq->flags == NFS_ODIRECT_SHOULD_DIRTY)) | |
be7e9858 | 272 | set_page_dirty(page); |
4bd8b010 TM |
273 | bytes += req->wb_bytes; |
274 | nfs_list_remove_request(req); | |
beeb5338 | 275 | nfs_release_request(req); |
d4a8f367 | 276 | } |
584aa810 | 277 | out_put: |
607f31e8 | 278 | if (put_dreq(dreq)) |
f7b5c340 | 279 | nfs_direct_complete(dreq); |
584aa810 | 280 | hdr->release(hdr); |
1da177e4 LT |
281 | } |
282 | ||
df3accb8 | 283 | static void nfs_read_sync_pgio_error(struct list_head *head, int error) |
cd841605 | 284 | { |
584aa810 | 285 | struct nfs_page *req; |
cd841605 | 286 | |
584aa810 FI |
287 | while (!list_empty(head)) { |
288 | req = nfs_list_entry(head->next); | |
289 | nfs_list_remove_request(req); | |
290 | nfs_release_request(req); | |
291 | } | |
cd841605 FI |
292 | } |
293 | ||
584aa810 FI |
294 | static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) |
295 | { | |
296 | get_dreq(hdr->dreq); | |
297 | } | |
298 | ||
299 | static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { | |
3e9e0ca3 | 300 | .error_cleanup = nfs_read_sync_pgio_error, |
584aa810 FI |
301 | .init_hdr = nfs_direct_pgio_init, |
302 | .completion = nfs_direct_read_completion, | |
303 | }; | |
304 | ||
d4cc948b | 305 | /* |
607f31e8 TM |
306 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
307 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
308 | * bail and stop sending more reads. Read length accounting is | |
309 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
310 | * no requests have been sent, just return an error. | |
1da177e4 | 311 | */ |
1da177e4 | 312 | |
91f79c43 AV |
313 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
314 | struct iov_iter *iter, | |
315 | loff_t pos) | |
316 | { | |
317 | struct nfs_pageio_descriptor desc; | |
318 | struct inode *inode = dreq->inode; | |
319 | ssize_t result = -EINVAL; | |
320 | size_t requested_bytes = 0; | |
321 | size_t rsize = max_t(size_t, NFS_SERVER(inode)->rsize, PAGE_SIZE); | |
e9f7bee1 | 322 | |
16b90578 | 323 | nfs_pageio_init_read(&desc, dreq->inode, false, |
91f79c43 AV |
324 | &nfs_direct_read_completion_ops); |
325 | get_dreq(dreq); | |
326 | desc.pg_dreq = dreq; | |
fe0f07d0 | 327 | inode_dio_begin(inode); |
a564b8f0 | 328 | |
91f79c43 AV |
329 | while (iov_iter_count(iter)) { |
330 | struct page **pagevec; | |
331 | size_t bytes; | |
332 | size_t pgbase; | |
333 | unsigned npages, i; | |
607f31e8 | 334 | |
1ef255e2 | 335 | result = iov_iter_get_pages_alloc2(iter, &pagevec, |
91f79c43 AV |
336 | rsize, &pgbase); |
337 | if (result < 0) | |
338 | break; | |
339 | ||
340 | bytes = result; | |
91f79c43 | 341 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; |
584aa810 FI |
342 | for (i = 0; i < npages; i++) { |
343 | struct nfs_page *req; | |
bf5fc402 | 344 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); |
584aa810 | 345 | /* XXX do we need to do the eof zeroing found in async_filler? */ |
28b1d3f5 | 346 | req = nfs_create_request(dreq->ctx, pagevec[i], |
584aa810 FI |
347 | pgbase, req_len); |
348 | if (IS_ERR(req)) { | |
584aa810 FI |
349 | result = PTR_ERR(req); |
350 | break; | |
351 | } | |
352 | req->wb_index = pos >> PAGE_SHIFT; | |
353 | req->wb_offset = pos & ~PAGE_MASK; | |
91f79c43 AV |
354 | if (!nfs_pageio_add_request(&desc, req)) { |
355 | result = desc.pg_error; | |
584aa810 | 356 | nfs_release_request(req); |
584aa810 FI |
357 | break; |
358 | } | |
359 | pgbase = 0; | |
360 | bytes -= req_len; | |
91f79c43 | 361 | requested_bytes += req_len; |
584aa810 | 362 | pos += req_len; |
35754bc0 | 363 | dreq->bytes_left -= req_len; |
584aa810 | 364 | } |
6d74743b | 365 | nfs_direct_release_pages(pagevec, npages); |
91f79c43 | 366 | kvfree(pagevec); |
19f73787 CL |
367 | if (result < 0) |
368 | break; | |
19f73787 CL |
369 | } |
370 | ||
584aa810 FI |
371 | nfs_pageio_complete(&desc); |
372 | ||
839f7ad6 CL |
373 | /* |
374 | * If no bytes were started, return the error, and let the | |
375 | * generic layer handle the completion. | |
376 | */ | |
377 | if (requested_bytes == 0) { | |
d03727b2 | 378 | inode_dio_end(inode); |
65caafd0 | 379 | nfs_direct_req_release(dreq); |
839f7ad6 CL |
380 | return result < 0 ? result : -EIO; |
381 | } | |
382 | ||
19f73787 | 383 | if (put_dreq(dreq)) |
f7b5c340 | 384 | nfs_direct_complete(dreq); |
85128b2b | 385 | return requested_bytes; |
19f73787 CL |
386 | } |
387 | ||
14a3ec79 CH |
388 | /** |
389 | * nfs_file_direct_read - file direct read operation for NFS files | |
390 | * @iocb: target I/O control block | |
619d30b4 | 391 | * @iter: vector of user buffers into which to read data |
64158668 | 392 | * @swap: flag indicating this is swap IO, not O_DIRECT IO |
14a3ec79 CH |
393 | * |
394 | * We use this function for direct reads instead of calling | |
395 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
396 | * the request starts before the end of the file. For that check | |
397 | * to work, we must generate a GETATTR before each direct read, and | |
398 | * even then there is a window between the GETATTR and the subsequent | |
399 | * READ where the file size could change. Our preference is simply | |
400 | * to do all reads the application wants, and the server will take | |
401 | * care of managing the end of file boundary. | |
402 | * | |
403 | * This function also eliminates unnecessarily updating the file's | |
404 | * atime locally, as the NFS server sets the file's atime, and this | |
405 | * client must read the updated atime from the server back into its | |
406 | * cache. | |
407 | */ | |
64158668 N |
408 | ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter, |
409 | bool swap) | |
1da177e4 | 410 | { |
14a3ec79 CH |
411 | struct file *file = iocb->ki_filp; |
412 | struct address_space *mapping = file->f_mapping; | |
413 | struct inode *inode = mapping->host; | |
1da177e4 | 414 | struct nfs_direct_req *dreq; |
b3c54de6 | 415 | struct nfs_lock_context *l_ctx; |
86b93667 | 416 | ssize_t result, requested; |
a6cbcd4a | 417 | size_t count = iov_iter_count(iter); |
14a3ec79 CH |
418 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); |
419 | ||
420 | dfprintk(FILE, "NFS: direct read(%pD2, %zd@%Ld)\n", | |
c8b8e32d | 421 | file, count, (long long) iocb->ki_pos); |
14a3ec79 CH |
422 | |
423 | result = 0; | |
424 | if (!count) | |
425 | goto out; | |
426 | ||
14a3ec79 CH |
427 | task_io_account_read(count); |
428 | ||
429 | result = -ENOMEM; | |
607f31e8 | 430 | dreq = nfs_direct_req_alloc(); |
f11ac8db | 431 | if (dreq == NULL) |
a5864c99 | 432 | goto out; |
1da177e4 | 433 | |
91d5b470 | 434 | dreq->inode = inode; |
ed3743a6 | 435 | dreq->bytes_left = dreq->max_count = count; |
c8b8e32d | 436 | dreq->io_start = iocb->ki_pos; |
cd3758e3 | 437 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
b3c54de6 TM |
438 | l_ctx = nfs_get_lock_context(dreq->ctx); |
439 | if (IS_ERR(l_ctx)) { | |
440 | result = PTR_ERR(l_ctx); | |
8605cf0e | 441 | nfs_direct_req_release(dreq); |
f11ac8db | 442 | goto out_release; |
b3c54de6 TM |
443 | } |
444 | dreq->l_ctx = l_ctx; | |
487b8372 CL |
445 | if (!is_sync_kiocb(iocb)) |
446 | dreq->iocb = iocb; | |
1da177e4 | 447 | |
fcb14cb1 | 448 | if (user_backed_iter(iter)) |
ad3cba22 DK |
449 | dreq->flags = NFS_ODIRECT_SHOULD_DIRTY; |
450 | ||
64158668 N |
451 | if (!swap) |
452 | nfs_start_io_direct(inode); | |
a5864c99 | 453 | |
619d30b4 | 454 | NFS_I(inode)->read_io += count; |
85128b2b | 455 | requested = nfs_direct_read_schedule_iovec(dreq, iter, iocb->ki_pos); |
d0b9875d | 456 | |
64158668 N |
457 | if (!swap) |
458 | nfs_end_io_direct(inode); | |
d0b9875d | 459 | |
85128b2b | 460 | if (requested > 0) { |
607f31e8 | 461 | result = nfs_direct_wait(dreq); |
85128b2b AV |
462 | if (result > 0) { |
463 | requested -= result; | |
c8b8e32d | 464 | iocb->ki_pos += result; |
85128b2b AV |
465 | } |
466 | iov_iter_revert(iter, requested); | |
467 | } else { | |
468 | result = requested; | |
14a3ec79 | 469 | } |
d0b9875d | 470 | |
f11ac8db | 471 | out_release: |
b4946ffb | 472 | nfs_direct_req_release(dreq); |
f11ac8db | 473 | out: |
1da177e4 LT |
474 | return result; |
475 | } | |
476 | ||
ed5d588f TM |
477 | static void |
478 | nfs_direct_join_group(struct list_head *list, struct inode *inode) | |
479 | { | |
480 | struct nfs_page *req, *next; | |
481 | ||
482 | list_for_each_entry(req, list, wb_list) { | |
483 | if (req->wb_head != req || req->wb_this_page == req) | |
484 | continue; | |
485 | for (next = req->wb_this_page; | |
486 | next != req->wb_head; | |
487 | next = next->wb_this_page) { | |
488 | nfs_list_remove_request(next); | |
489 | nfs_release_request(next); | |
490 | } | |
491 | nfs_join_page_group(req, inode); | |
492 | } | |
493 | } | |
494 | ||
085d1e33 TH |
495 | static void |
496 | nfs_direct_write_scan_commit_list(struct inode *inode, | |
497 | struct list_head *list, | |
498 | struct nfs_commit_info *cinfo) | |
499 | { | |
e824f99a | 500 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
9c455a8c | 501 | pnfs_recover_commit_reqs(list, cinfo); |
085d1e33 | 502 | nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0); |
e824f99a | 503 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
085d1e33 TH |
504 | } |
505 | ||
fad61490 TM |
506 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) |
507 | { | |
1763da12 FI |
508 | struct nfs_pageio_descriptor desc; |
509 | struct nfs_page *req, *tmp; | |
510 | LIST_HEAD(reqs); | |
511 | struct nfs_commit_info cinfo; | |
512 | LIST_HEAD(failed); | |
513 | ||
514 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
085d1e33 | 515 | nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); |
1da177e4 | 516 | |
ed5d588f TM |
517 | nfs_direct_join_group(&reqs, dreq->inode); |
518 | ||
fad61490 | 519 | dreq->count = 0; |
031d73ed TM |
520 | dreq->max_count = 0; |
521 | list_for_each_entry(req, &reqs, wb_list) | |
522 | dreq->max_count += req->wb_bytes; | |
a5314a74 | 523 | nfs_clear_pnfs_ds_commit_verifiers(&dreq->ds_cinfo); |
607f31e8 TM |
524 | get_dreq(dreq); |
525 | ||
a20c93e3 | 526 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false, |
1763da12 FI |
527 | &nfs_direct_write_completion_ops); |
528 | desc.pg_dreq = dreq; | |
fedb595c | 529 | |
1763da12 | 530 | list_for_each_entry_safe(req, tmp, &reqs, wb_list) { |
33344e0f TM |
531 | /* Bump the transmission count */ |
532 | req->wb_nio++; | |
1763da12 | 533 | if (!nfs_pageio_add_request(&desc, req)) { |
078b5fd9 | 534 | nfs_list_move_request(req, &failed); |
fe238e60 | 535 | spin_lock(&cinfo.inode->i_lock); |
1763da12 | 536 | dreq->flags = 0; |
d600ad1f PT |
537 | if (desc.pg_error < 0) |
538 | dreq->error = desc.pg_error; | |
539 | else | |
540 | dreq->error = -EIO; | |
fe238e60 | 541 | spin_unlock(&cinfo.inode->i_lock); |
1763da12 | 542 | } |
5a695da2 | 543 | nfs_release_request(req); |
1763da12 FI |
544 | } |
545 | nfs_pageio_complete(&desc); | |
fad61490 | 546 | |
4035c248 TM |
547 | while (!list_empty(&failed)) { |
548 | req = nfs_list_entry(failed.next); | |
549 | nfs_list_remove_request(req); | |
1d1afcbc | 550 | nfs_unlock_and_release_request(req); |
4035c248 | 551 | } |
fad61490 | 552 | |
1763da12 | 553 | if (put_dreq(dreq)) |
4d3b55d3 | 554 | nfs_direct_write_complete(dreq); |
c9d8f89d TM |
555 | } |
556 | ||
1763da12 | 557 | static void nfs_direct_commit_complete(struct nfs_commit_data *data) |
c9d8f89d | 558 | { |
1f28476d | 559 | const struct nfs_writeverf *verf = data->res.verf; |
0b7c0153 | 560 | struct nfs_direct_req *dreq = data->dreq; |
1763da12 FI |
561 | struct nfs_commit_info cinfo; |
562 | struct nfs_page *req; | |
c9d8f89d TM |
563 | int status = data->task.tk_status; |
564 | ||
8efc4bbe JL |
565 | trace_nfs_direct_commit_complete(dreq); |
566 | ||
fb5f7f20 TM |
567 | if (status < 0) { |
568 | /* Errors in commit are fatal */ | |
569 | dreq->error = status; | |
570 | dreq->max_count = 0; | |
571 | dreq->count = 0; | |
572 | dreq->flags = NFS_ODIRECT_DONE; | |
55051c0c | 573 | } else { |
fb5f7f20 | 574 | status = dreq->error; |
55051c0c | 575 | } |
fb5f7f20 | 576 | |
1763da12 | 577 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
1da177e4 | 578 | |
1763da12 FI |
579 | while (!list_empty(&data->pages)) { |
580 | req = nfs_list_entry(data->pages.next); | |
581 | nfs_list_remove_request(req); | |
1f28476d TM |
582 | if (status >= 0 && !nfs_write_match_verf(verf, req)) { |
583 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
33344e0f TM |
584 | /* |
585 | * Despite the reboot, the write was successful, | |
586 | * so reset wb_nio. | |
587 | */ | |
588 | req->wb_nio = 0; | |
b57ff130 | 589 | nfs_mark_request_commit(req, NULL, &cinfo, 0); |
1f28476d | 590 | } else /* Error or match */ |
906369e4 | 591 | nfs_release_request(req); |
1d1afcbc | 592 | nfs_unlock_and_release_request(req); |
1763da12 FI |
593 | } |
594 | ||
133a48ab | 595 | if (nfs_commit_end(cinfo.mds)) |
4d3b55d3 | 596 | nfs_direct_write_complete(dreq); |
1da177e4 LT |
597 | } |
598 | ||
b20135d0 TM |
599 | static void nfs_direct_resched_write(struct nfs_commit_info *cinfo, |
600 | struct nfs_page *req) | |
1763da12 | 601 | { |
b20135d0 TM |
602 | struct nfs_direct_req *dreq = cinfo->dreq; |
603 | ||
8efc4bbe JL |
604 | trace_nfs_direct_resched_write(dreq); |
605 | ||
b20135d0 | 606 | spin_lock(&dreq->lock); |
fb5f7f20 TM |
607 | if (dreq->flags != NFS_ODIRECT_DONE) |
608 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
b20135d0 TM |
609 | spin_unlock(&dreq->lock); |
610 | nfs_mark_request_commit(req, NULL, cinfo, 0); | |
1763da12 FI |
611 | } |
612 | ||
613 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { | |
614 | .completion = nfs_direct_commit_complete, | |
b20135d0 | 615 | .resched_write = nfs_direct_resched_write, |
fad61490 TM |
616 | }; |
617 | ||
618 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 619 | { |
1763da12 FI |
620 | int res; |
621 | struct nfs_commit_info cinfo; | |
622 | LIST_HEAD(mds_list); | |
623 | ||
624 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
625 | nfs_scan_commit(dreq->inode, &mds_list, &cinfo); | |
626 | res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); | |
627 | if (res < 0) /* res == -ENOMEM */ | |
628 | nfs_direct_write_reschedule(dreq); | |
fad61490 | 629 | } |
1da177e4 | 630 | |
fb5f7f20 TM |
631 | static void nfs_direct_write_clear_reqs(struct nfs_direct_req *dreq) |
632 | { | |
633 | struct nfs_commit_info cinfo; | |
634 | struct nfs_page *req; | |
635 | LIST_HEAD(reqs); | |
636 | ||
637 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
638 | nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); | |
639 | ||
640 | while (!list_empty(&reqs)) { | |
641 | req = nfs_list_entry(reqs.next); | |
642 | nfs_list_remove_request(req); | |
f02cec9d | 643 | nfs_release_request(req); |
fb5f7f20 TM |
644 | nfs_unlock_and_release_request(req); |
645 | } | |
646 | } | |
647 | ||
1763da12 | 648 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
fad61490 | 649 | { |
1763da12 | 650 | struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); |
fad61490 | 651 | int flags = dreq->flags; |
1da177e4 | 652 | |
fad61490 TM |
653 | dreq->flags = 0; |
654 | switch (flags) { | |
655 | case NFS_ODIRECT_DO_COMMIT: | |
656 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 657 | break; |
fad61490 TM |
658 | case NFS_ODIRECT_RESCHED_WRITES: |
659 | nfs_direct_write_reschedule(dreq); | |
660 | break; | |
661 | default: | |
fb5f7f20 | 662 | nfs_direct_write_clear_reqs(dreq); |
f7b5c340 TM |
663 | nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping); |
664 | nfs_direct_complete(dreq); | |
fad61490 TM |
665 | } |
666 | } | |
1da177e4 | 667 | |
4d3b55d3 | 668 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq) |
fad61490 | 669 | { |
8efc4bbe | 670 | trace_nfs_direct_write_complete(dreq); |
46483c2e | 671 | queue_work(nfsiod_workqueue, &dreq->work); /* Calls nfs_direct_write_schedule_work */ |
fad61490 | 672 | } |
1763da12 | 673 | |
1763da12 FI |
674 | static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) |
675 | { | |
676 | struct nfs_direct_req *dreq = hdr->dreq; | |
677 | struct nfs_commit_info cinfo; | |
1763da12 | 678 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); |
3731d44b | 679 | int flags = NFS_ODIRECT_DONE; |
1763da12 | 680 | |
8efc4bbe JL |
681 | trace_nfs_direct_write_completion(dreq); |
682 | ||
1763da12 FI |
683 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
684 | ||
685 | spin_lock(&dreq->lock); | |
eb2c50da TM |
686 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
687 | spin_unlock(&dreq->lock); | |
688 | goto out_put; | |
689 | } | |
690 | ||
031d73ed | 691 | nfs_direct_count_bytes(dreq, hdr); |
69d96651 | 692 | if (test_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags)) { |
3731d44b | 693 | if (!dreq->flags) |
1f28476d | 694 | dreq->flags = NFS_ODIRECT_DO_COMMIT; |
3731d44b | 695 | flags = dreq->flags; |
1763da12 FI |
696 | } |
697 | spin_unlock(&dreq->lock); | |
698 | ||
699 | while (!list_empty(&hdr->pages)) { | |
2bfc6e56 | 700 | |
1763da12 FI |
701 | req = nfs_list_entry(hdr->pages.next); |
702 | nfs_list_remove_request(req); | |
3731d44b | 703 | if (flags == NFS_ODIRECT_DO_COMMIT) { |
04277086 | 704 | kref_get(&req->wb_kref); |
ba838a75 CL |
705 | memcpy(&req->wb_verf, &hdr->verf.verifier, |
706 | sizeof(req->wb_verf)); | |
b57ff130 WAA |
707 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
708 | hdr->ds_commit_idx); | |
3731d44b TM |
709 | } else if (flags == NFS_ODIRECT_RESCHED_WRITES) { |
710 | kref_get(&req->wb_kref); | |
711 | nfs_mark_request_commit(req, NULL, &cinfo, 0); | |
1763da12 | 712 | } |
1d1afcbc | 713 | nfs_unlock_and_release_request(req); |
1763da12 FI |
714 | } |
715 | ||
716 | out_put: | |
717 | if (put_dreq(dreq)) | |
4d3b55d3 | 718 | nfs_direct_write_complete(dreq); |
1763da12 FI |
719 | hdr->release(hdr); |
720 | } | |
721 | ||
df3accb8 | 722 | static void nfs_write_sync_pgio_error(struct list_head *head, int error) |
3e9e0ca3 TM |
723 | { |
724 | struct nfs_page *req; | |
725 | ||
726 | while (!list_empty(head)) { | |
727 | req = nfs_list_entry(head->next); | |
728 | nfs_list_remove_request(req); | |
1d1afcbc | 729 | nfs_unlock_and_release_request(req); |
3e9e0ca3 TM |
730 | } |
731 | } | |
732 | ||
dc602dd7 TM |
733 | static void nfs_direct_write_reschedule_io(struct nfs_pgio_header *hdr) |
734 | { | |
735 | struct nfs_direct_req *dreq = hdr->dreq; | |
736 | ||
8efc4bbe JL |
737 | trace_nfs_direct_write_reschedule_io(dreq); |
738 | ||
dc602dd7 TM |
739 | spin_lock(&dreq->lock); |
740 | if (dreq->error == 0) { | |
741 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
742 | /* fake unstable write to let common nfs resend pages */ | |
743 | hdr->verf.committed = NFS_UNSTABLE; | |
4daaeba9 TM |
744 | hdr->good_bytes = hdr->args.offset + hdr->args.count - |
745 | hdr->io_start; | |
dc602dd7 TM |
746 | } |
747 | spin_unlock(&dreq->lock); | |
748 | } | |
749 | ||
1763da12 | 750 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { |
3e9e0ca3 | 751 | .error_cleanup = nfs_write_sync_pgio_error, |
1763da12 FI |
752 | .init_hdr = nfs_direct_pgio_init, |
753 | .completion = nfs_direct_write_completion, | |
dc602dd7 | 754 | .reschedule_io = nfs_direct_write_reschedule_io, |
1763da12 FI |
755 | }; |
756 | ||
91f79c43 AV |
757 | |
758 | /* | |
759 | * NB: Return the value of the first error return code. Subsequent | |
760 | * errors after the first one are ignored. | |
761 | */ | |
762 | /* | |
763 | * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE | |
764 | * operation. If nfs_writedata_alloc() or get_user_pages() fails, | |
765 | * bail and stop sending more writes. Write length accounting is | |
766 | * handled automatically by nfs_direct_write_result(). Otherwise, if | |
767 | * no requests have been sent, just return an error. | |
768 | */ | |
19f73787 | 769 | static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, |
619d30b4 | 770 | struct iov_iter *iter, |
c265de25 | 771 | loff_t pos, int ioflags) |
19f73787 | 772 | { |
1763da12 | 773 | struct nfs_pageio_descriptor desc; |
1d59d61f | 774 | struct inode *inode = dreq->inode; |
19f73787 CL |
775 | ssize_t result = 0; |
776 | size_t requested_bytes = 0; | |
91f79c43 | 777 | size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE); |
19f73787 | 778 | |
8efc4bbe JL |
779 | trace_nfs_direct_write_schedule_iovec(dreq); |
780 | ||
c265de25 | 781 | nfs_pageio_init_write(&desc, inode, ioflags, false, |
1763da12 FI |
782 | &nfs_direct_write_completion_ops); |
783 | desc.pg_dreq = dreq; | |
19f73787 | 784 | get_dreq(dreq); |
fe0f07d0 | 785 | inode_dio_begin(inode); |
19f73787 | 786 | |
91f79c43 AV |
787 | NFS_I(inode)->write_io += iov_iter_count(iter); |
788 | while (iov_iter_count(iter)) { | |
789 | struct page **pagevec; | |
790 | size_t bytes; | |
791 | size_t pgbase; | |
792 | unsigned npages, i; | |
793 | ||
1ef255e2 | 794 | result = iov_iter_get_pages_alloc2(iter, &pagevec, |
91f79c43 | 795 | wsize, &pgbase); |
19f73787 CL |
796 | if (result < 0) |
797 | break; | |
91f79c43 AV |
798 | |
799 | bytes = result; | |
91f79c43 AV |
800 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; |
801 | for (i = 0; i < npages; i++) { | |
802 | struct nfs_page *req; | |
803 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); | |
804 | ||
28b1d3f5 | 805 | req = nfs_create_request(dreq->ctx, pagevec[i], |
91f79c43 AV |
806 | pgbase, req_len); |
807 | if (IS_ERR(req)) { | |
808 | result = PTR_ERR(req); | |
809 | break; | |
810 | } | |
0a00b77b | 811 | |
d600ad1f PT |
812 | if (desc.pg_error < 0) { |
813 | nfs_free_request(req); | |
814 | result = desc.pg_error; | |
815 | break; | |
816 | } | |
0a00b77b | 817 | |
91f79c43 AV |
818 | nfs_lock_request(req); |
819 | req->wb_index = pos >> PAGE_SHIFT; | |
820 | req->wb_offset = pos & ~PAGE_MASK; | |
821 | if (!nfs_pageio_add_request(&desc, req)) { | |
822 | result = desc.pg_error; | |
823 | nfs_unlock_and_release_request(req); | |
824 | break; | |
825 | } | |
826 | pgbase = 0; | |
827 | bytes -= req_len; | |
828 | requested_bytes += req_len; | |
829 | pos += req_len; | |
830 | dreq->bytes_left -= req_len; | |
831 | } | |
832 | nfs_direct_release_pages(pagevec, npages); | |
833 | kvfree(pagevec); | |
834 | if (result < 0) | |
19f73787 | 835 | break; |
19f73787 | 836 | } |
1763da12 | 837 | nfs_pageio_complete(&desc); |
19f73787 | 838 | |
839f7ad6 CL |
839 | /* |
840 | * If no bytes were started, return the error, and let the | |
841 | * generic layer handle the completion. | |
842 | */ | |
843 | if (requested_bytes == 0) { | |
d03727b2 | 844 | inode_dio_end(inode); |
65caafd0 | 845 | nfs_direct_req_release(dreq); |
839f7ad6 CL |
846 | return result < 0 ? result : -EIO; |
847 | } | |
848 | ||
19f73787 | 849 | if (put_dreq(dreq)) |
4d3b55d3 | 850 | nfs_direct_write_complete(dreq); |
85128b2b | 851 | return requested_bytes; |
19f73787 CL |
852 | } |
853 | ||
1da177e4 LT |
854 | /** |
855 | * nfs_file_direct_write - file direct write operation for NFS files | |
856 | * @iocb: target I/O control block | |
619d30b4 | 857 | * @iter: vector of user buffers from which to write data |
64158668 | 858 | * @swap: flag indicating this is swap IO, not O_DIRECT IO |
1da177e4 LT |
859 | * |
860 | * We use this function for direct writes instead of calling | |
861 | * generic_file_aio_write() in order to avoid taking the inode | |
862 | * semaphore and updating the i_size. The NFS server will set | |
863 | * the new i_size and this client must read the updated size | |
864 | * back into its cache. We let the server do generic write | |
865 | * parameter checking and report problems. | |
866 | * | |
1da177e4 LT |
867 | * We eliminate local atime updates, see direct read above. |
868 | * | |
869 | * We avoid unnecessary page cache invalidations for normal cached | |
870 | * readers of this file. | |
871 | * | |
872 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
873 | * is no atomic O_APPEND write facility in the NFS protocol. | |
874 | */ | |
64158668 N |
875 | ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter, |
876 | bool swap) | |
1da177e4 | 877 | { |
9a74a2b8 | 878 | ssize_t result, requested; |
89698b24 | 879 | size_t count; |
1da177e4 | 880 | struct file *file = iocb->ki_filp; |
1da177e4 | 881 | struct address_space *mapping = file->f_mapping; |
22cd1bf1 CH |
882 | struct inode *inode = mapping->host; |
883 | struct nfs_direct_req *dreq; | |
884 | struct nfs_lock_context *l_ctx; | |
65a4a1ca | 885 | loff_t pos, end; |
c216fd70 | 886 | |
6de1472f | 887 | dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n", |
3309dd04 | 888 | file, iov_iter_count(iter), (long long) iocb->ki_pos); |
027445c3 | 889 | |
64158668 N |
890 | if (swap) |
891 | /* bypass generic checks */ | |
892 | result = iov_iter_count(iter); | |
893 | else | |
894 | result = generic_write_checks(iocb, iter); | |
89698b24 TM |
895 | if (result <= 0) |
896 | return result; | |
897 | count = result; | |
898 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); | |
3309dd04 AV |
899 | |
900 | pos = iocb->ki_pos; | |
09cbfeaf | 901 | end = (pos + iov_iter_count(iter) - 1) >> PAGE_SHIFT; |
ce1a8e67 | 902 | |
89698b24 | 903 | task_io_account_write(count); |
7ec10f26 | 904 | |
22cd1bf1 CH |
905 | result = -ENOMEM; |
906 | dreq = nfs_direct_req_alloc(); | |
907 | if (!dreq) | |
a5864c99 | 908 | goto out; |
9eafa8cc | 909 | |
22cd1bf1 | 910 | dreq->inode = inode; |
89698b24 | 911 | dreq->bytes_left = dreq->max_count = count; |
5fadeb47 | 912 | dreq->io_start = pos; |
22cd1bf1 CH |
913 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
914 | l_ctx = nfs_get_lock_context(dreq->ctx); | |
915 | if (IS_ERR(l_ctx)) { | |
916 | result = PTR_ERR(l_ctx); | |
8605cf0e | 917 | nfs_direct_req_release(dreq); |
22cd1bf1 CH |
918 | goto out_release; |
919 | } | |
920 | dreq->l_ctx = l_ctx; | |
921 | if (!is_sync_kiocb(iocb)) | |
922 | dreq->iocb = iocb; | |
9c455a8c | 923 | pnfs_init_ds_commit_info_ops(&dreq->ds_cinfo, inode); |
22cd1bf1 | 924 | |
64158668 | 925 | if (swap) { |
c265de25 N |
926 | requested = nfs_direct_write_schedule_iovec(dreq, iter, pos, |
927 | FLUSH_STABLE); | |
64158668 N |
928 | } else { |
929 | nfs_start_io_direct(inode); | |
a5864c99 | 930 | |
c265de25 N |
931 | requested = nfs_direct_write_schedule_iovec(dreq, iter, pos, |
932 | FLUSH_COND_STABLE); | |
a9ab5e84 | 933 | |
64158668 N |
934 | if (mapping->nrpages) { |
935 | invalidate_inode_pages2_range(mapping, | |
936 | pos >> PAGE_SHIFT, end); | |
937 | } | |
a9ab5e84 | 938 | |
64158668 N |
939 | nfs_end_io_direct(inode); |
940 | } | |
a9ab5e84 | 941 | |
85128b2b | 942 | if (requested > 0) { |
22cd1bf1 CH |
943 | result = nfs_direct_wait(dreq); |
944 | if (result > 0) { | |
85128b2b | 945 | requested -= result; |
22cd1bf1 | 946 | iocb->ki_pos = pos + result; |
e2592217 CH |
947 | /* XXX: should check the generic_write_sync retval */ |
948 | generic_write_sync(iocb, result); | |
22cd1bf1 | 949 | } |
85128b2b AV |
950 | iov_iter_revert(iter, requested); |
951 | } else { | |
952 | result = requested; | |
1763da12 | 953 | } |
a6b5a28e | 954 | nfs_fscache_invalidate(inode, FSCACHE_INVAL_DIO_WRITE); |
22cd1bf1 CH |
955 | out_release: |
956 | nfs_direct_req_release(dreq); | |
a5864c99 | 957 | out: |
22cd1bf1 | 958 | return result; |
1da177e4 LT |
959 | } |
960 | ||
88467055 CL |
961 | /** |
962 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
963 | * | |
964 | */ | |
f7b422b1 | 965 | int __init nfs_init_directcache(void) |
1da177e4 LT |
966 | { |
967 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
968 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
969 | 0, (SLAB_RECLAIM_ACCOUNT| |
970 | SLAB_MEM_SPREAD), | |
20c2df83 | 971 | NULL); |
1da177e4 LT |
972 | if (nfs_direct_cachep == NULL) |
973 | return -ENOMEM; | |
974 | ||
975 | return 0; | |
976 | } | |
977 | ||
88467055 | 978 | /** |
f7b422b1 | 979 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
980 | * |
981 | */ | |
266bee88 | 982 | void nfs_destroy_directcache(void) |
1da177e4 | 983 | { |
1a1d92c1 | 984 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 985 | } |