]>
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" |
91d5b470 | 62 | |
1da177e4 | 63 | #define NFSDBG_FACILITY NFSDBG_VFS |
1da177e4 | 64 | |
e18b890b | 65 | static struct kmem_cache *nfs_direct_cachep; |
1da177e4 | 66 | |
1da177e4 LT |
67 | struct nfs_direct_req { |
68 | struct kref kref; /* release manager */ | |
15ce4a0c CL |
69 | |
70 | /* I/O parameters */ | |
a8881f5a | 71 | struct nfs_open_context *ctx; /* file open context info */ |
f11ac8db | 72 | struct nfs_lock_context *l_ctx; /* Lock context info */ |
99514f8f | 73 | struct kiocb * iocb; /* controlling i/o request */ |
88467055 | 74 | struct inode * inode; /* target file of i/o */ |
15ce4a0c CL |
75 | |
76 | /* completion state */ | |
607f31e8 | 77 | atomic_t io_count; /* i/os we're waiting for */ |
15ce4a0c | 78 | spinlock_t lock; /* protect completion state */ |
0a00b77b | 79 | |
d9ee6553 | 80 | loff_t io_start; /* Start offset for I/O */ |
15ce4a0c | 81 | ssize_t count, /* bytes actually processed */ |
ed3743a6 | 82 | max_count, /* max expected count */ |
35754bc0 | 83 | bytes_left, /* bytes left to be sent */ |
1da177e4 | 84 | error; /* any reported error */ |
d72b7a6b | 85 | struct completion completion; /* wait for i/o completion */ |
fad61490 TM |
86 | |
87 | /* commit state */ | |
1763da12 FI |
88 | struct nfs_mds_commit_info mds_cinfo; /* Storage for cinfo */ |
89 | struct pnfs_ds_commit_info ds_cinfo; /* Storage for cinfo */ | |
90 | struct work_struct work; | |
fad61490 | 91 | int flags; |
ad3cba22 | 92 | /* for write */ |
fad61490 TM |
93 | #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */ |
94 | #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */ | |
ad3cba22 DK |
95 | /* for read */ |
96 | #define NFS_ODIRECT_SHOULD_DIRTY (3) /* dirty user-space page after read */ | |
fb5f7f20 | 97 | #define NFS_ODIRECT_DONE INT_MAX /* write verification failed */ |
1da177e4 LT |
98 | }; |
99 | ||
1763da12 FI |
100 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops; |
101 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops; | |
4d3b55d3 | 102 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq); |
1763da12 | 103 | static void nfs_direct_write_schedule_work(struct work_struct *work); |
607f31e8 TM |
104 | |
105 | static inline void get_dreq(struct nfs_direct_req *dreq) | |
106 | { | |
107 | atomic_inc(&dreq->io_count); | |
108 | } | |
109 | ||
110 | static inline int put_dreq(struct nfs_direct_req *dreq) | |
111 | { | |
112 | return atomic_dec_and_test(&dreq->io_count); | |
113 | } | |
114 | ||
0a00b77b | 115 | static void |
031d73ed TM |
116 | nfs_direct_handle_truncated(struct nfs_direct_req *dreq, |
117 | const struct nfs_pgio_header *hdr, | |
118 | ssize_t dreq_len) | |
0a00b77b | 119 | { |
031d73ed TM |
120 | if (!(test_bit(NFS_IOHDR_ERROR, &hdr->flags) || |
121 | test_bit(NFS_IOHDR_EOF, &hdr->flags))) | |
122 | return; | |
123 | if (dreq->max_count >= dreq_len) { | |
124 | dreq->max_count = dreq_len; | |
125 | if (dreq->count > dreq_len) | |
126 | dreq->count = dreq_len; | |
127 | ||
128 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) | |
129 | dreq->error = hdr->error; | |
130 | else /* Clear outstanding error if this is EOF */ | |
131 | dreq->error = 0; | |
132 | } | |
031d73ed | 133 | } |
0a00b77b | 134 | |
031d73ed TM |
135 | static void |
136 | nfs_direct_count_bytes(struct nfs_direct_req *dreq, | |
137 | const struct nfs_pgio_header *hdr) | |
138 | { | |
031d73ed TM |
139 | loff_t hdr_end = hdr->io_start + hdr->good_bytes; |
140 | ssize_t dreq_len = 0; | |
ed3743a6 | 141 | |
031d73ed TM |
142 | if (hdr_end > dreq->io_start) |
143 | dreq_len = hdr_end - dreq->io_start; | |
0a00b77b | 144 | |
031d73ed | 145 | nfs_direct_handle_truncated(dreq, hdr, dreq_len); |
0a00b77b | 146 | |
031d73ed TM |
147 | if (dreq_len > dreq->max_count) |
148 | dreq_len = dreq->max_count; | |
149 | ||
031d73ed TM |
150 | if (dreq->count < dreq_len) |
151 | dreq->count = dreq_len; | |
0a00b77b WAA |
152 | } |
153 | ||
1da177e4 | 154 | /** |
b8a32e2b | 155 | * nfs_direct_IO - NFS address space operation for direct I/O |
b8a32e2b | 156 | * @iocb: target I/O control block |
90090ae6 | 157 | * @iter: I/O buffer |
b8a32e2b CL |
158 | * |
159 | * The presence of this routine in the address space ops vector means | |
a564b8f0 MG |
160 | * the NFS client supports direct I/O. However, for most direct IO, we |
161 | * shunt off direct read and write requests before the VFS gets them, | |
162 | * so this method is only ever called for swap. | |
1da177e4 | 163 | */ |
c8b8e32d | 164 | ssize_t nfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
b8a32e2b | 165 | { |
ee8a1a8b PT |
166 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
167 | ||
168 | /* we only support swap file calling nfs_direct_IO */ | |
169 | if (!IS_SWAPFILE(inode)) | |
170 | return 0; | |
171 | ||
66ee59af | 172 | VM_BUG_ON(iov_iter_count(iter) != PAGE_SIZE); |
a564b8f0 | 173 | |
6f673763 | 174 | if (iov_iter_rw(iter) == READ) |
c8b8e32d | 175 | return nfs_file_direct_read(iocb, iter); |
65a4a1ca | 176 | return nfs_file_direct_write(iocb, iter); |
b8a32e2b CL |
177 | } |
178 | ||
749e146e | 179 | static void nfs_direct_release_pages(struct page **pages, unsigned int npages) |
9c93ab7d | 180 | { |
749e146e | 181 | unsigned int i; |
607f31e8 | 182 | for (i = 0; i < npages; i++) |
09cbfeaf | 183 | put_page(pages[i]); |
6b45d858 TM |
184 | } |
185 | ||
1763da12 FI |
186 | void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, |
187 | struct nfs_direct_req *dreq) | |
188 | { | |
fe238e60 | 189 | cinfo->inode = dreq->inode; |
1763da12 FI |
190 | cinfo->mds = &dreq->mds_cinfo; |
191 | cinfo->ds = &dreq->ds_cinfo; | |
192 | cinfo->dreq = dreq; | |
193 | cinfo->completion_ops = &nfs_direct_commit_completion_ops; | |
194 | } | |
195 | ||
93619e59 | 196 | static inline struct nfs_direct_req *nfs_direct_req_alloc(void) |
1da177e4 | 197 | { |
93619e59 CL |
198 | struct nfs_direct_req *dreq; |
199 | ||
292f3eee | 200 | dreq = kmem_cache_zalloc(nfs_direct_cachep, GFP_KERNEL); |
93619e59 CL |
201 | if (!dreq) |
202 | return NULL; | |
203 | ||
204 | kref_init(&dreq->kref); | |
607f31e8 | 205 | kref_get(&dreq->kref); |
d72b7a6b | 206 | init_completion(&dreq->completion); |
1763da12 | 207 | INIT_LIST_HEAD(&dreq->mds_cinfo.list); |
c21e7168 | 208 | pnfs_init_ds_commit_info(&dreq->ds_cinfo); |
1763da12 | 209 | INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); |
15ce4a0c | 210 | spin_lock_init(&dreq->lock); |
93619e59 CL |
211 | |
212 | return dreq; | |
1da177e4 LT |
213 | } |
214 | ||
b4946ffb | 215 | static void nfs_direct_req_free(struct kref *kref) |
1da177e4 LT |
216 | { |
217 | struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); | |
a8881f5a | 218 | |
18f41296 | 219 | pnfs_release_ds_info(&dreq->ds_cinfo, dreq->inode); |
f11ac8db TM |
220 | if (dreq->l_ctx != NULL) |
221 | nfs_put_lock_context(dreq->l_ctx); | |
a8881f5a TM |
222 | if (dreq->ctx != NULL) |
223 | put_nfs_open_context(dreq->ctx); | |
1da177e4 LT |
224 | kmem_cache_free(nfs_direct_cachep, dreq); |
225 | } | |
226 | ||
b4946ffb TM |
227 | static void nfs_direct_req_release(struct nfs_direct_req *dreq) |
228 | { | |
229 | kref_put(&dreq->kref, nfs_direct_req_free); | |
230 | } | |
231 | ||
6296556f PT |
232 | ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq) |
233 | { | |
234 | return dreq->bytes_left; | |
235 | } | |
236 | EXPORT_SYMBOL_GPL(nfs_dreq_bytes_left); | |
237 | ||
bc0fb201 CL |
238 | /* |
239 | * Collects and returns the final error value/byte-count. | |
240 | */ | |
241 | static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) | |
242 | { | |
15ce4a0c | 243 | ssize_t result = -EIOCBQUEUED; |
bc0fb201 CL |
244 | |
245 | /* Async requests don't wait here */ | |
246 | if (dreq->iocb) | |
247 | goto out; | |
248 | ||
150030b7 | 249 | result = wait_for_completion_killable(&dreq->completion); |
bc0fb201 | 250 | |
d2a7de0b TM |
251 | if (!result) { |
252 | result = dreq->count; | |
253 | WARN_ON_ONCE(dreq->count < 0); | |
254 | } | |
bc0fb201 | 255 | if (!result) |
15ce4a0c | 256 | result = dreq->error; |
bc0fb201 CL |
257 | |
258 | out: | |
bc0fb201 CL |
259 | return (ssize_t) result; |
260 | } | |
261 | ||
63ab46ab | 262 | /* |
607f31e8 TM |
263 | * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust |
264 | * the iocb is still valid here if this is a synchronous request. | |
63ab46ab | 265 | */ |
f7b5c340 | 266 | static void nfs_direct_complete(struct nfs_direct_req *dreq) |
63ab46ab | 267 | { |
9811cd57 CH |
268 | struct inode *inode = dreq->inode; |
269 | ||
65caafd0 OK |
270 | inode_dio_end(inode); |
271 | ||
2a009ec9 | 272 | if (dreq->iocb) { |
15ce4a0c | 273 | long res = (long) dreq->error; |
d2a7de0b | 274 | if (dreq->count != 0) { |
15ce4a0c | 275 | res = (long) dreq->count; |
d2a7de0b TM |
276 | WARN_ON_ONCE(dreq->count < 0); |
277 | } | |
04b2fa9f | 278 | dreq->iocb->ki_complete(dreq->iocb, res, 0); |
d72b7a6b | 279 | } |
2a009ec9 | 280 | |
024de8f1 | 281 | complete(&dreq->completion); |
63ab46ab | 282 | |
b4946ffb | 283 | nfs_direct_req_release(dreq); |
63ab46ab CL |
284 | } |
285 | ||
584aa810 | 286 | static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) |
fdd1e74c | 287 | { |
584aa810 FI |
288 | unsigned long bytes = 0; |
289 | struct nfs_direct_req *dreq = hdr->dreq; | |
fdd1e74c | 290 | |
15ce4a0c | 291 | spin_lock(&dreq->lock); |
eb2c50da TM |
292 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
293 | spin_unlock(&dreq->lock); | |
294 | goto out_put; | |
295 | } | |
296 | ||
031d73ed | 297 | nfs_direct_count_bytes(dreq, hdr); |
584aa810 FI |
298 | spin_unlock(&dreq->lock); |
299 | ||
4bd8b010 TM |
300 | while (!list_empty(&hdr->pages)) { |
301 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
302 | struct page *page = req->wb_page; | |
303 | ||
ad3cba22 DK |
304 | if (!PageCompound(page) && bytes < hdr->good_bytes && |
305 | (dreq->flags == NFS_ODIRECT_SHOULD_DIRTY)) | |
be7e9858 | 306 | set_page_dirty(page); |
4bd8b010 TM |
307 | bytes += req->wb_bytes; |
308 | nfs_list_remove_request(req); | |
beeb5338 | 309 | nfs_release_request(req); |
d4a8f367 | 310 | } |
584aa810 | 311 | out_put: |
607f31e8 | 312 | if (put_dreq(dreq)) |
f7b5c340 | 313 | nfs_direct_complete(dreq); |
584aa810 | 314 | hdr->release(hdr); |
1da177e4 LT |
315 | } |
316 | ||
df3accb8 | 317 | static void nfs_read_sync_pgio_error(struct list_head *head, int error) |
cd841605 | 318 | { |
584aa810 | 319 | struct nfs_page *req; |
cd841605 | 320 | |
584aa810 FI |
321 | while (!list_empty(head)) { |
322 | req = nfs_list_entry(head->next); | |
323 | nfs_list_remove_request(req); | |
324 | nfs_release_request(req); | |
325 | } | |
cd841605 FI |
326 | } |
327 | ||
584aa810 FI |
328 | static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) |
329 | { | |
330 | get_dreq(hdr->dreq); | |
331 | } | |
332 | ||
333 | static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { | |
3e9e0ca3 | 334 | .error_cleanup = nfs_read_sync_pgio_error, |
584aa810 FI |
335 | .init_hdr = nfs_direct_pgio_init, |
336 | .completion = nfs_direct_read_completion, | |
337 | }; | |
338 | ||
d4cc948b | 339 | /* |
607f31e8 TM |
340 | * For each rsize'd chunk of the user's buffer, dispatch an NFS READ |
341 | * operation. If nfs_readdata_alloc() or get_user_pages() fails, | |
342 | * bail and stop sending more reads. Read length accounting is | |
343 | * handled automatically by nfs_direct_read_result(). Otherwise, if | |
344 | * no requests have been sent, just return an error. | |
1da177e4 | 345 | */ |
1da177e4 | 346 | |
91f79c43 AV |
347 | static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, |
348 | struct iov_iter *iter, | |
349 | loff_t pos) | |
350 | { | |
351 | struct nfs_pageio_descriptor desc; | |
352 | struct inode *inode = dreq->inode; | |
353 | ssize_t result = -EINVAL; | |
354 | size_t requested_bytes = 0; | |
355 | size_t rsize = max_t(size_t, NFS_SERVER(inode)->rsize, PAGE_SIZE); | |
e9f7bee1 | 356 | |
16b90578 | 357 | nfs_pageio_init_read(&desc, dreq->inode, false, |
91f79c43 AV |
358 | &nfs_direct_read_completion_ops); |
359 | get_dreq(dreq); | |
360 | desc.pg_dreq = dreq; | |
fe0f07d0 | 361 | inode_dio_begin(inode); |
a564b8f0 | 362 | |
91f79c43 AV |
363 | while (iov_iter_count(iter)) { |
364 | struct page **pagevec; | |
365 | size_t bytes; | |
366 | size_t pgbase; | |
367 | unsigned npages, i; | |
607f31e8 | 368 | |
91f79c43 AV |
369 | result = iov_iter_get_pages_alloc(iter, &pagevec, |
370 | rsize, &pgbase); | |
371 | if (result < 0) | |
372 | break; | |
373 | ||
374 | bytes = result; | |
375 | iov_iter_advance(iter, bytes); | |
376 | npages = (result + pgbase + PAGE_SIZE - 1) / PAGE_SIZE; | |
584aa810 FI |
377 | for (i = 0; i < npages; i++) { |
378 | struct nfs_page *req; | |
bf5fc402 | 379 | unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase); |
584aa810 | 380 | /* XXX do we need to do the eof zeroing found in async_filler? */ |
28b1d3f5 | 381 | req = nfs_create_request(dreq->ctx, pagevec[i], |
584aa810 FI |
382 | pgbase, req_len); |
383 | if (IS_ERR(req)) { | |
584aa810 FI |
384 | result = PTR_ERR(req); |
385 | break; | |
386 | } | |
387 | req->wb_index = pos >> PAGE_SHIFT; | |
388 | req->wb_offset = pos & ~PAGE_MASK; | |
91f79c43 AV |
389 | if (!nfs_pageio_add_request(&desc, req)) { |
390 | result = desc.pg_error; | |
584aa810 | 391 | nfs_release_request(req); |
584aa810 FI |
392 | break; |
393 | } | |
394 | pgbase = 0; | |
395 | bytes -= req_len; | |
91f79c43 | 396 | requested_bytes += req_len; |
584aa810 | 397 | pos += req_len; |
35754bc0 | 398 | dreq->bytes_left -= req_len; |
584aa810 | 399 | } |
6d74743b | 400 | nfs_direct_release_pages(pagevec, npages); |
91f79c43 | 401 | kvfree(pagevec); |
19f73787 CL |
402 | if (result < 0) |
403 | break; | |
19f73787 CL |
404 | } |
405 | ||
584aa810 FI |
406 | nfs_pageio_complete(&desc); |
407 | ||
839f7ad6 CL |
408 | /* |
409 | * If no bytes were started, return the error, and let the | |
410 | * generic layer handle the completion. | |
411 | */ | |
412 | if (requested_bytes == 0) { | |
d03727b2 | 413 | inode_dio_end(inode); |
65caafd0 | 414 | nfs_direct_req_release(dreq); |
839f7ad6 CL |
415 | return result < 0 ? result : -EIO; |
416 | } | |
417 | ||
19f73787 | 418 | if (put_dreq(dreq)) |
f7b5c340 | 419 | nfs_direct_complete(dreq); |
85128b2b | 420 | return requested_bytes; |
19f73787 CL |
421 | } |
422 | ||
14a3ec79 CH |
423 | /** |
424 | * nfs_file_direct_read - file direct read operation for NFS files | |
425 | * @iocb: target I/O control block | |
619d30b4 | 426 | * @iter: vector of user buffers into which to read data |
14a3ec79 CH |
427 | * |
428 | * We use this function for direct reads instead of calling | |
429 | * generic_file_aio_read() in order to avoid gfar's check to see if | |
430 | * the request starts before the end of the file. For that check | |
431 | * to work, we must generate a GETATTR before each direct read, and | |
432 | * even then there is a window between the GETATTR and the subsequent | |
433 | * READ where the file size could change. Our preference is simply | |
434 | * to do all reads the application wants, and the server will take | |
435 | * care of managing the end of file boundary. | |
436 | * | |
437 | * This function also eliminates unnecessarily updating the file's | |
438 | * atime locally, as the NFS server sets the file's atime, and this | |
439 | * client must read the updated atime from the server back into its | |
440 | * cache. | |
441 | */ | |
c8b8e32d | 442 | ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 443 | { |
14a3ec79 CH |
444 | struct file *file = iocb->ki_filp; |
445 | struct address_space *mapping = file->f_mapping; | |
446 | struct inode *inode = mapping->host; | |
1da177e4 | 447 | struct nfs_direct_req *dreq; |
b3c54de6 | 448 | struct nfs_lock_context *l_ctx; |
86b93667 | 449 | ssize_t result, requested; |
a6cbcd4a | 450 | size_t count = iov_iter_count(iter); |
14a3ec79 CH |
451 | nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); |
452 | ||
453 | dfprintk(FILE, "NFS: direct read(%pD2, %zd@%Ld)\n", | |
c8b8e32d | 454 | file, count, (long long) iocb->ki_pos); |
14a3ec79 CH |
455 | |
456 | result = 0; | |
457 | if (!count) | |
458 | goto out; | |
459 | ||
14a3ec79 CH |
460 | task_io_account_read(count); |
461 | ||
462 | result = -ENOMEM; | |
607f31e8 | 463 | dreq = nfs_direct_req_alloc(); |
f11ac8db | 464 | if (dreq == NULL) |
a5864c99 | 465 | goto out; |
1da177e4 | 466 | |
91d5b470 | 467 | dreq->inode = inode; |
ed3743a6 | 468 | dreq->bytes_left = dreq->max_count = count; |
c8b8e32d | 469 | dreq->io_start = iocb->ki_pos; |
cd3758e3 | 470 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
b3c54de6 TM |
471 | l_ctx = nfs_get_lock_context(dreq->ctx); |
472 | if (IS_ERR(l_ctx)) { | |
473 | result = PTR_ERR(l_ctx); | |
8605cf0e | 474 | nfs_direct_req_release(dreq); |
f11ac8db | 475 | goto out_release; |
b3c54de6 TM |
476 | } |
477 | dreq->l_ctx = l_ctx; | |
487b8372 CL |
478 | if (!is_sync_kiocb(iocb)) |
479 | dreq->iocb = iocb; | |
1da177e4 | 480 | |
ad3cba22 DK |
481 | if (iter_is_iovec(iter)) |
482 | dreq->flags = NFS_ODIRECT_SHOULD_DIRTY; | |
483 | ||
a5864c99 TM |
484 | nfs_start_io_direct(inode); |
485 | ||
619d30b4 | 486 | NFS_I(inode)->read_io += count; |
85128b2b | 487 | requested = nfs_direct_read_schedule_iovec(dreq, iter, iocb->ki_pos); |
d0b9875d | 488 | |
a5864c99 | 489 | nfs_end_io_direct(inode); |
d0b9875d | 490 | |
85128b2b | 491 | if (requested > 0) { |
607f31e8 | 492 | result = nfs_direct_wait(dreq); |
85128b2b AV |
493 | if (result > 0) { |
494 | requested -= result; | |
c8b8e32d | 495 | iocb->ki_pos += result; |
85128b2b AV |
496 | } |
497 | iov_iter_revert(iter, requested); | |
498 | } else { | |
499 | result = requested; | |
14a3ec79 | 500 | } |
d0b9875d | 501 | |
f11ac8db | 502 | out_release: |
b4946ffb | 503 | nfs_direct_req_release(dreq); |
f11ac8db | 504 | out: |
1da177e4 LT |
505 | return result; |
506 | } | |
507 | ||
ed5d588f TM |
508 | static void |
509 | nfs_direct_join_group(struct list_head *list, struct inode *inode) | |
510 | { | |
511 | struct nfs_page *req, *next; | |
512 | ||
513 | list_for_each_entry(req, list, wb_list) { | |
514 | if (req->wb_head != req || req->wb_this_page == req) | |
515 | continue; | |
516 | for (next = req->wb_this_page; | |
517 | next != req->wb_head; | |
518 | next = next->wb_this_page) { | |
519 | nfs_list_remove_request(next); | |
520 | nfs_release_request(next); | |
521 | } | |
522 | nfs_join_page_group(req, inode); | |
523 | } | |
524 | } | |
525 | ||
085d1e33 TH |
526 | static void |
527 | nfs_direct_write_scan_commit_list(struct inode *inode, | |
528 | struct list_head *list, | |
529 | struct nfs_commit_info *cinfo) | |
530 | { | |
e824f99a | 531 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
9c455a8c | 532 | pnfs_recover_commit_reqs(list, cinfo); |
085d1e33 | 533 | nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0); |
e824f99a | 534 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
085d1e33 TH |
535 | } |
536 | ||
fad61490 TM |
537 | static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) |
538 | { | |
1763da12 FI |
539 | struct nfs_pageio_descriptor desc; |
540 | struct nfs_page *req, *tmp; | |
541 | LIST_HEAD(reqs); | |
542 | struct nfs_commit_info cinfo; | |
543 | LIST_HEAD(failed); | |
544 | ||
545 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
085d1e33 | 546 | nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); |
1da177e4 | 547 | |
ed5d588f TM |
548 | nfs_direct_join_group(&reqs, dreq->inode); |
549 | ||
fad61490 | 550 | dreq->count = 0; |
031d73ed TM |
551 | dreq->max_count = 0; |
552 | list_for_each_entry(req, &reqs, wb_list) | |
553 | dreq->max_count += req->wb_bytes; | |
a5314a74 | 554 | nfs_clear_pnfs_ds_commit_verifiers(&dreq->ds_cinfo); |
607f31e8 TM |
555 | get_dreq(dreq); |
556 | ||
a20c93e3 | 557 | nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false, |
1763da12 FI |
558 | &nfs_direct_write_completion_ops); |
559 | desc.pg_dreq = dreq; | |
fedb595c | 560 | |
1763da12 | 561 | list_for_each_entry_safe(req, tmp, &reqs, wb_list) { |
33344e0f TM |
562 | /* Bump the transmission count */ |
563 | req->wb_nio++; | |
1763da12 | 564 | if (!nfs_pageio_add_request(&desc, req)) { |
078b5fd9 | 565 | nfs_list_move_request(req, &failed); |
fe238e60 | 566 | spin_lock(&cinfo.inode->i_lock); |
1763da12 | 567 | dreq->flags = 0; |
d600ad1f PT |
568 | if (desc.pg_error < 0) |
569 | dreq->error = desc.pg_error; | |
570 | else | |
571 | dreq->error = -EIO; | |
fe238e60 | 572 | spin_unlock(&cinfo.inode->i_lock); |
1763da12 | 573 | } |
5a695da2 | 574 | nfs_release_request(req); |
1763da12 FI |
575 | } |
576 | nfs_pageio_complete(&desc); | |
fad61490 | 577 | |
4035c248 TM |
578 | while (!list_empty(&failed)) { |
579 | req = nfs_list_entry(failed.next); | |
580 | nfs_list_remove_request(req); | |
1d1afcbc | 581 | nfs_unlock_and_release_request(req); |
4035c248 | 582 | } |
fad61490 | 583 | |
1763da12 | 584 | if (put_dreq(dreq)) |
4d3b55d3 | 585 | nfs_direct_write_complete(dreq); |
c9d8f89d TM |
586 | } |
587 | ||
1763da12 | 588 | static void nfs_direct_commit_complete(struct nfs_commit_data *data) |
c9d8f89d | 589 | { |
1f28476d | 590 | const struct nfs_writeverf *verf = data->res.verf; |
0b7c0153 | 591 | struct nfs_direct_req *dreq = data->dreq; |
1763da12 FI |
592 | struct nfs_commit_info cinfo; |
593 | struct nfs_page *req; | |
c9d8f89d TM |
594 | int status = data->task.tk_status; |
595 | ||
fb5f7f20 TM |
596 | if (status < 0) { |
597 | /* Errors in commit are fatal */ | |
598 | dreq->error = status; | |
599 | dreq->max_count = 0; | |
600 | dreq->count = 0; | |
601 | dreq->flags = NFS_ODIRECT_DONE; | |
602 | } else if (dreq->flags == NFS_ODIRECT_DONE) | |
603 | status = dreq->error; | |
604 | ||
1763da12 | 605 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
1da177e4 | 606 | |
1763da12 FI |
607 | while (!list_empty(&data->pages)) { |
608 | req = nfs_list_entry(data->pages.next); | |
609 | nfs_list_remove_request(req); | |
1f28476d TM |
610 | if (status >= 0 && !nfs_write_match_verf(verf, req)) { |
611 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
33344e0f TM |
612 | /* |
613 | * Despite the reboot, the write was successful, | |
614 | * so reset wb_nio. | |
615 | */ | |
616 | req->wb_nio = 0; | |
b57ff130 | 617 | nfs_mark_request_commit(req, NULL, &cinfo, 0); |
1f28476d | 618 | } else /* Error or match */ |
906369e4 | 619 | nfs_release_request(req); |
1d1afcbc | 620 | nfs_unlock_and_release_request(req); |
1763da12 FI |
621 | } |
622 | ||
623 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
4d3b55d3 | 624 | nfs_direct_write_complete(dreq); |
1da177e4 LT |
625 | } |
626 | ||
b20135d0 TM |
627 | static void nfs_direct_resched_write(struct nfs_commit_info *cinfo, |
628 | struct nfs_page *req) | |
1763da12 | 629 | { |
b20135d0 TM |
630 | struct nfs_direct_req *dreq = cinfo->dreq; |
631 | ||
632 | spin_lock(&dreq->lock); | |
fb5f7f20 TM |
633 | if (dreq->flags != NFS_ODIRECT_DONE) |
634 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
b20135d0 TM |
635 | spin_unlock(&dreq->lock); |
636 | nfs_mark_request_commit(req, NULL, cinfo, 0); | |
1763da12 FI |
637 | } |
638 | ||
639 | static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { | |
640 | .completion = nfs_direct_commit_complete, | |
b20135d0 | 641 | .resched_write = nfs_direct_resched_write, |
fad61490 TM |
642 | }; |
643 | ||
644 | static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) | |
1da177e4 | 645 | { |
1763da12 FI |
646 | int res; |
647 | struct nfs_commit_info cinfo; | |
648 | LIST_HEAD(mds_list); | |
649 | ||
650 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
651 | nfs_scan_commit(dreq->inode, &mds_list, &cinfo); | |
652 | res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); | |
653 | if (res < 0) /* res == -ENOMEM */ | |
654 | nfs_direct_write_reschedule(dreq); | |
fad61490 | 655 | } |
1da177e4 | 656 | |
fb5f7f20 TM |
657 | static void nfs_direct_write_clear_reqs(struct nfs_direct_req *dreq) |
658 | { | |
659 | struct nfs_commit_info cinfo; | |
660 | struct nfs_page *req; | |
661 | LIST_HEAD(reqs); | |
662 | ||
663 | nfs_init_cinfo_from_dreq(&cinfo, dreq); | |
664 | nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); | |
665 | ||
666 | while (!list_empty(&reqs)) { | |
667 | req = nfs_list_entry(reqs.next); | |
668 | nfs_list_remove_request(req); | |
f02cec9d | 669 | nfs_release_request(req); |
fb5f7f20 TM |
670 | nfs_unlock_and_release_request(req); |
671 | } | |
672 | } | |
673 | ||
1763da12 | 674 | static void nfs_direct_write_schedule_work(struct work_struct *work) |
fad61490 | 675 | { |
1763da12 | 676 | struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); |
fad61490 | 677 | int flags = dreq->flags; |
1da177e4 | 678 | |
fad61490 TM |
679 | dreq->flags = 0; |
680 | switch (flags) { | |
681 | case NFS_ODIRECT_DO_COMMIT: | |
682 | nfs_direct_commit_schedule(dreq); | |
1da177e4 | 683 | break; |
fad61490 TM |
684 | case NFS_ODIRECT_RESCHED_WRITES: |
685 | nfs_direct_write_reschedule(dreq); | |
686 | break; | |
687 | default: | |
fb5f7f20 | 688 | nfs_direct_write_clear_reqs(dreq); |
f7b5c340 TM |
689 | nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping); |
690 | nfs_direct_complete(dreq); | |
fad61490 TM |
691 | } |
692 | } | |
1da177e4 | 693 | |
4d3b55d3 | 694 | static void nfs_direct_write_complete(struct nfs_direct_req *dreq) |
fad61490 | 695 | { |
46483c2e | 696 | queue_work(nfsiod_workqueue, &dreq->work); /* Calls nfs_direct_write_schedule_work */ |
fad61490 | 697 | } |
1763da12 | 698 | |
1763da12 FI |
699 | static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) |
700 | { | |
701 | struct nfs_direct_req *dreq = hdr->dreq; | |
702 | struct nfs_commit_info cinfo; | |
c65e6254 | 703 | bool request_commit = false; |
1763da12 FI |
704 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); |
705 | ||
1763da12 FI |
706 | nfs_init_cinfo_from_dreq(&cinfo, dreq); |
707 | ||
708 | spin_lock(&dreq->lock); | |
eb2c50da TM |
709 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
710 | spin_unlock(&dreq->lock); | |
711 | goto out_put; | |
712 | } | |
713 | ||
031d73ed | 714 | nfs_direct_count_bytes(dreq, hdr); |
1f28476d TM |
715 | if (hdr->good_bytes != 0 && nfs_write_need_commit(hdr)) { |
716 | switch (dreq->flags) { | |
717 | case 0: | |
718 | dreq->flags = NFS_ODIRECT_DO_COMMIT; | |
719 | request_commit = true; | |
720 | break; | |
721 | case NFS_ODIRECT_RESCHED_WRITES: | |
722 | case NFS_ODIRECT_DO_COMMIT: | |
723 | request_commit = true; | |
1763da12 FI |
724 | } |
725 | } | |
726 | spin_unlock(&dreq->lock); | |
727 | ||
728 | while (!list_empty(&hdr->pages)) { | |
2bfc6e56 | 729 | |
1763da12 FI |
730 | req = nfs_list_entry(hdr->pages.next); |
731 | nfs_list_remove_request(req); | |
c65e6254 | 732 | if (request_commit) { |
04277086 | 733 | kref_get(&req->wb_kref); |
ba838a75 CL |
734 | memcpy(&req->wb_verf, &hdr->verf.verifier, |
735 | sizeof(req->wb_verf)); | |
b57ff130 WAA |
736 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
737 | hdr->ds_commit_idx); | |
1763da12 | 738 | } |
1d1afcbc | 739 | nfs_unlock_and_release_request(req); |
1763da12 FI |
740 | } |
741 | ||
742 | out_put: | |
743 | if (put_dreq(dreq)) | |
4d3b55d3 | 744 | nfs_direct_write_complete(dreq); |
1763da12 FI |
745 | hdr->release(hdr); |
746 | } | |
747 | ||
df3accb8 | 748 | static void nfs_write_sync_pgio_error(struct list_head *head, int error) |
3e9e0ca3 TM |
749 | { |
750 | struct nfs_page *req; | |
751 | ||
752 | while (!list_empty(head)) { | |
753 | req = nfs_list_entry(head->next); | |
754 | nfs_list_remove_request(req); | |
1d1afcbc | 755 | nfs_unlock_and_release_request(req); |
3e9e0ca3 TM |
756 | } |
757 | } | |
758 | ||
dc602dd7 TM |
759 | static void nfs_direct_write_reschedule_io(struct nfs_pgio_header *hdr) |
760 | { | |
761 | struct nfs_direct_req *dreq = hdr->dreq; | |
762 | ||
763 | spin_lock(&dreq->lock); | |
764 | if (dreq->error == 0) { | |
765 | dreq->flags = NFS_ODIRECT_RESCHED_WRITES; | |
766 | /* fake unstable write to let common nfs resend pages */ | |
767 | hdr->verf.committed = NFS_UNSTABLE; | |
4daaeba9 TM |
768 | hdr->good_bytes = hdr->args.offset + hdr->args.count - |
769 | hdr->io_start; | |
dc602dd7 TM |
770 | } |
771 | spin_unlock(&dreq->lock); | |
772 | } | |
773 | ||
1763da12 | 774 | static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { |
3e9e0ca3 | 775 | .error_cleanup = nfs_write_sync_pgio_error, |
1763da12 FI |
776 | .init_hdr = nfs_direct_pgio_init, |
777 | .completion = nfs_direct_write_completion, | |
dc602dd7 | 778 | .reschedule_io = nfs_direct_write_reschedule_io, |
1763da12 FI |
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); |
fe0f07d0 | 807 | inode_dio_begin(inode); |
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 | ||
28b1d3f5 | 828 | req = nfs_create_request(dreq->ctx, pagevec[i], |
91f79c43 AV |
829 | pgbase, req_len); |
830 | if (IS_ERR(req)) { | |
831 | result = PTR_ERR(req); | |
832 | break; | |
833 | } | |
0a00b77b | 834 | |
d600ad1f PT |
835 | if (desc.pg_error < 0) { |
836 | nfs_free_request(req); | |
837 | result = desc.pg_error; | |
838 | break; | |
839 | } | |
0a00b77b | 840 | |
91f79c43 AV |
841 | nfs_lock_request(req); |
842 | req->wb_index = pos >> PAGE_SHIFT; | |
843 | req->wb_offset = pos & ~PAGE_MASK; | |
844 | if (!nfs_pageio_add_request(&desc, req)) { | |
845 | result = desc.pg_error; | |
846 | nfs_unlock_and_release_request(req); | |
847 | break; | |
848 | } | |
849 | pgbase = 0; | |
850 | bytes -= req_len; | |
851 | requested_bytes += req_len; | |
852 | pos += req_len; | |
853 | dreq->bytes_left -= req_len; | |
854 | } | |
855 | nfs_direct_release_pages(pagevec, npages); | |
856 | kvfree(pagevec); | |
857 | if (result < 0) | |
19f73787 | 858 | break; |
19f73787 | 859 | } |
1763da12 | 860 | nfs_pageio_complete(&desc); |
19f73787 | 861 | |
839f7ad6 CL |
862 | /* |
863 | * If no bytes were started, return the error, and let the | |
864 | * generic layer handle the completion. | |
865 | */ | |
866 | if (requested_bytes == 0) { | |
d03727b2 | 867 | inode_dio_end(inode); |
65caafd0 | 868 | nfs_direct_req_release(dreq); |
839f7ad6 CL |
869 | return result < 0 ? result : -EIO; |
870 | } | |
871 | ||
19f73787 | 872 | if (put_dreq(dreq)) |
4d3b55d3 | 873 | nfs_direct_write_complete(dreq); |
85128b2b | 874 | return requested_bytes; |
19f73787 CL |
875 | } |
876 | ||
1da177e4 LT |
877 | /** |
878 | * nfs_file_direct_write - file direct write operation for NFS files | |
879 | * @iocb: target I/O control block | |
619d30b4 | 880 | * @iter: vector of user buffers from which to write data |
1da177e4 LT |
881 | * |
882 | * We use this function for direct writes instead of calling | |
883 | * generic_file_aio_write() in order to avoid taking the inode | |
884 | * semaphore and updating the i_size. The NFS server will set | |
885 | * the new i_size and this client must read the updated size | |
886 | * back into its cache. We let the server do generic write | |
887 | * parameter checking and report problems. | |
888 | * | |
1da177e4 LT |
889 | * We eliminate local atime updates, see direct read above. |
890 | * | |
891 | * We avoid unnecessary page cache invalidations for normal cached | |
892 | * readers of this file. | |
893 | * | |
894 | * Note that O_APPEND is not supported for NFS direct writes, as there | |
895 | * is no atomic O_APPEND write facility in the NFS protocol. | |
896 | */ | |
65a4a1ca | 897 | ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 898 | { |
9a74a2b8 | 899 | ssize_t result, requested; |
89698b24 | 900 | size_t count; |
1da177e4 | 901 | struct file *file = iocb->ki_filp; |
1da177e4 | 902 | struct address_space *mapping = file->f_mapping; |
22cd1bf1 CH |
903 | struct inode *inode = mapping->host; |
904 | struct nfs_direct_req *dreq; | |
905 | struct nfs_lock_context *l_ctx; | |
65a4a1ca | 906 | loff_t pos, end; |
c216fd70 | 907 | |
6de1472f | 908 | dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n", |
3309dd04 | 909 | file, iov_iter_count(iter), (long long) iocb->ki_pos); |
027445c3 | 910 | |
89698b24 TM |
911 | result = generic_write_checks(iocb, iter); |
912 | if (result <= 0) | |
913 | return result; | |
914 | count = result; | |
915 | nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); | |
3309dd04 AV |
916 | |
917 | pos = iocb->ki_pos; | |
09cbfeaf | 918 | end = (pos + iov_iter_count(iter) - 1) >> PAGE_SHIFT; |
ce1a8e67 | 919 | |
89698b24 | 920 | task_io_account_write(count); |
7ec10f26 | 921 | |
22cd1bf1 CH |
922 | result = -ENOMEM; |
923 | dreq = nfs_direct_req_alloc(); | |
924 | if (!dreq) | |
a5864c99 | 925 | goto out; |
9eafa8cc | 926 | |
22cd1bf1 | 927 | dreq->inode = inode; |
89698b24 | 928 | dreq->bytes_left = dreq->max_count = count; |
5fadeb47 | 929 | dreq->io_start = pos; |
22cd1bf1 CH |
930 | dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); |
931 | l_ctx = nfs_get_lock_context(dreq->ctx); | |
932 | if (IS_ERR(l_ctx)) { | |
933 | result = PTR_ERR(l_ctx); | |
8605cf0e | 934 | nfs_direct_req_release(dreq); |
22cd1bf1 CH |
935 | goto out_release; |
936 | } | |
937 | dreq->l_ctx = l_ctx; | |
938 | if (!is_sync_kiocb(iocb)) | |
939 | dreq->iocb = iocb; | |
9c455a8c | 940 | pnfs_init_ds_commit_info_ops(&dreq->ds_cinfo, inode); |
22cd1bf1 | 941 | |
a5864c99 TM |
942 | nfs_start_io_direct(inode); |
943 | ||
85128b2b | 944 | requested = nfs_direct_write_schedule_iovec(dreq, iter, pos); |
a9ab5e84 CH |
945 | |
946 | if (mapping->nrpages) { | |
947 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 948 | pos >> PAGE_SHIFT, end); |
a9ab5e84 CH |
949 | } |
950 | ||
a5864c99 | 951 | nfs_end_io_direct(inode); |
a9ab5e84 | 952 | |
85128b2b | 953 | if (requested > 0) { |
22cd1bf1 CH |
954 | result = nfs_direct_wait(dreq); |
955 | if (result > 0) { | |
85128b2b | 956 | requested -= result; |
22cd1bf1 | 957 | iocb->ki_pos = pos + result; |
e2592217 CH |
958 | /* XXX: should check the generic_write_sync retval */ |
959 | generic_write_sync(iocb, result); | |
22cd1bf1 | 960 | } |
85128b2b AV |
961 | iov_iter_revert(iter, requested); |
962 | } else { | |
963 | result = requested; | |
1763da12 | 964 | } |
22cd1bf1 CH |
965 | out_release: |
966 | nfs_direct_req_release(dreq); | |
a5864c99 | 967 | out: |
22cd1bf1 | 968 | return result; |
1da177e4 LT |
969 | } |
970 | ||
88467055 CL |
971 | /** |
972 | * nfs_init_directcache - create a slab cache for nfs_direct_req structures | |
973 | * | |
974 | */ | |
f7b422b1 | 975 | int __init nfs_init_directcache(void) |
1da177e4 LT |
976 | { |
977 | nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", | |
978 | sizeof(struct nfs_direct_req), | |
fffb60f9 PJ |
979 | 0, (SLAB_RECLAIM_ACCOUNT| |
980 | SLAB_MEM_SPREAD), | |
20c2df83 | 981 | NULL); |
1da177e4 LT |
982 | if (nfs_direct_cachep == NULL) |
983 | return -ENOMEM; | |
984 | ||
985 | return 0; | |
986 | } | |
987 | ||
88467055 | 988 | /** |
f7b422b1 | 989 | * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures |
88467055 CL |
990 | * |
991 | */ | |
266bee88 | 992 | void nfs_destroy_directcache(void) |
1da177e4 | 993 | { |
1a1d92c1 | 994 | kmem_cache_destroy(nfs_direct_cachep); |
1da177e4 | 995 | } |