]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
3fcfab16 | 24 | |
1da177e4 | 25 | #include <asm/uaccess.h> |
1da177e4 LT |
26 | |
27 | #include "delegation.h" | |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
def6ed7e | 30 | #include "nfs4_fs.h" |
074cc1de | 31 | #include "fscache.h" |
94ad1c80 | 32 | #include "pnfs.h" |
1da177e4 | 33 | |
f4ce1299 TM |
34 | #include "nfstrace.h" |
35 | ||
1da177e4 LT |
36 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
37 | ||
38 | #define MIN_POOL_WRITE (32) | |
39 | #define MIN_POOL_COMMIT (4) | |
40 | ||
41 | /* | |
42 | * Local function declarations | |
43 | */ | |
f8512ad0 | 44 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 45 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 46 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 47 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 48 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
1da177e4 | 49 | |
e18b890b | 50 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 51 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 52 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
53 | static mempool_t *nfs_commit_mempool; |
54 | ||
0b7c0153 | 55 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 56 | { |
192e501b | 57 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
40859d7e | 58 | |
1da177e4 LT |
59 | if (p) { |
60 | memset(p, 0, sizeof(*p)); | |
61 | INIT_LIST_HEAD(&p->pages); | |
62 | } | |
63 | return p; | |
64 | } | |
e0c2b380 | 65 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 66 | |
0b7c0153 | 67 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
68 | { |
69 | mempool_free(p, nfs_commit_mempool); | |
70 | } | |
e0c2b380 | 71 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 72 | |
6c75dc0d | 73 | struct nfs_write_header *nfs_writehdr_alloc(void) |
3feb2d49 | 74 | { |
192e501b | 75 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
3feb2d49 TM |
76 | |
77 | if (p) { | |
cd841605 | 78 | struct nfs_pgio_header *hdr = &p->header; |
cd841605 | 79 | |
3feb2d49 | 80 | memset(p, 0, sizeof(*p)); |
cd841605 | 81 | INIT_LIST_HEAD(&hdr->pages); |
6c75dc0d FI |
82 | INIT_LIST_HEAD(&hdr->rpc_list); |
83 | spin_lock_init(&hdr->lock); | |
84 | atomic_set(&hdr->refcnt, 0); | |
9bce008b | 85 | hdr->verf = &p->verf; |
3feb2d49 TM |
86 | } |
87 | return p; | |
88 | } | |
89d77c8f | 89 | EXPORT_SYMBOL_GPL(nfs_writehdr_alloc); |
3feb2d49 | 90 | |
1763da12 FI |
91 | static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
92 | unsigned int pagecount) | |
6c75dc0d FI |
93 | { |
94 | struct nfs_write_data *data, *prealloc; | |
95 | ||
96 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
97 | if (prealloc->header == NULL) | |
98 | data = prealloc; | |
99 | else | |
100 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
101 | if (!data) | |
102 | goto out; | |
103 | ||
104 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
105 | data->header = hdr; | |
106 | atomic_inc(&hdr->refcnt); | |
107 | } else { | |
108 | if (data != prealloc) | |
109 | kfree(data); | |
110 | data = NULL; | |
111 | } | |
112 | out: | |
113 | return data; | |
114 | } | |
115 | ||
cd841605 | 116 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 117 | { |
cd841605 FI |
118 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
119 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 | 120 | } |
89d77c8f | 121 | EXPORT_SYMBOL_GPL(nfs_writehdr_free); |
3feb2d49 | 122 | |
dce81290 | 123 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 124 | { |
6c75dc0d FI |
125 | struct nfs_pgio_header *hdr = wdata->header; |
126 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
127 | ||
383ba719 | 128 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
129 | if (wdata->pages.pagevec != wdata->pages.page_array) |
130 | kfree(wdata->pages.pagevec); | |
6db6dd7d | 131 | if (wdata == &write_header->rpc_data) { |
6c75dc0d | 132 | wdata->header = NULL; |
6db6dd7d TM |
133 | wdata = NULL; |
134 | } | |
6c75dc0d | 135 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 136 | hdr->completion_ops->completion(hdr); |
6db6dd7d TM |
137 | /* Note: we only free the rpc_task after callbacks are done. |
138 | * See the comment in rpc_free_task() for why | |
139 | */ | |
140 | kfree(wdata); | |
1da177e4 | 141 | } |
89d77c8f | 142 | EXPORT_SYMBOL_GPL(nfs_writedata_release); |
1da177e4 | 143 | |
7b159fc1 TM |
144 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
145 | { | |
146 | ctx->error = error; | |
147 | smp_wmb(); | |
148 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
149 | } | |
150 | ||
29418aa4 MG |
151 | static struct nfs_page * |
152 | nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) | |
277459d2 TM |
153 | { |
154 | struct nfs_page *req = NULL; | |
155 | ||
29418aa4 | 156 | if (PagePrivate(page)) |
277459d2 | 157 | req = (struct nfs_page *)page_private(page); |
29418aa4 MG |
158 | else if (unlikely(PageSwapCache(page))) { |
159 | struct nfs_page *freq, *t; | |
160 | ||
161 | /* Linearly search the commit list for the correct req */ | |
162 | list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { | |
163 | if (freq->wb_page == page) { | |
164 | req = freq; | |
165 | break; | |
166 | } | |
167 | } | |
277459d2 | 168 | } |
29418aa4 MG |
169 | |
170 | if (req) | |
171 | kref_get(&req->wb_kref); | |
172 | ||
277459d2 TM |
173 | return req; |
174 | } | |
175 | ||
176 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
177 | { | |
d56b4ddf | 178 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 179 | struct nfs_page *req = NULL; |
277459d2 | 180 | |
587142f8 | 181 | spin_lock(&inode->i_lock); |
29418aa4 | 182 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
587142f8 | 183 | spin_unlock(&inode->i_lock); |
277459d2 TM |
184 | return req; |
185 | } | |
186 | ||
1da177e4 LT |
187 | /* Adjust the file length if we're writing beyond the end */ |
188 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
189 | { | |
d56b4ddf | 190 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
191 | loff_t end, i_size; |
192 | pgoff_t end_index; | |
1da177e4 | 193 | |
a3d01454 TM |
194 | spin_lock(&inode->i_lock); |
195 | i_size = i_size_read(inode); | |
196 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 197 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 198 | goto out; |
d56b4ddf | 199 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 200 | if (i_size >= end) |
a3d01454 | 201 | goto out; |
1da177e4 | 202 | i_size_write(inode, end); |
a3d01454 TM |
203 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
204 | out: | |
205 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
206 | } |
207 | ||
a301b777 TM |
208 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
209 | static void nfs_set_pageerror(struct page *page) | |
210 | { | |
d56b4ddf | 211 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
212 | } |
213 | ||
1da177e4 LT |
214 | /* We can set the PG_uptodate flag if we see that a write request |
215 | * covers the full page. | |
216 | */ | |
217 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
218 | { | |
1da177e4 LT |
219 | if (PageUptodate(page)) |
220 | return; | |
221 | if (base != 0) | |
222 | return; | |
49a70f27 | 223 | if (count != nfs_page_length(page)) |
1da177e4 | 224 | return; |
49a70f27 | 225 | SetPageUptodate(page); |
1da177e4 LT |
226 | } |
227 | ||
1da177e4 LT |
228 | static int wb_priority(struct writeback_control *wbc) |
229 | { | |
230 | if (wbc->for_reclaim) | |
c63c7b05 | 231 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 232 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
233 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
234 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
235 | } |
236 | ||
89a09141 PZ |
237 | /* |
238 | * NFS congestion control | |
239 | */ | |
240 | ||
241 | int nfs_congestion_kb; | |
242 | ||
243 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
244 | #define NFS_CONGESTION_OFF_THRESH \ | |
245 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
246 | ||
deed85e7 | 247 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 248 | { |
deed85e7 | 249 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
250 | int ret = test_set_page_writeback(page); |
251 | ||
deed85e7 | 252 | WARN_ON_ONCE(ret != 0); |
89a09141 | 253 | |
deed85e7 TM |
254 | if (atomic_long_inc_return(&nfss->writeback) > |
255 | NFS_CONGESTION_ON_THRESH) { | |
256 | set_bdi_congested(&nfss->backing_dev_info, | |
257 | BLK_RW_ASYNC); | |
89a09141 PZ |
258 | } |
259 | } | |
260 | ||
261 | static void nfs_end_page_writeback(struct page *page) | |
262 | { | |
d56b4ddf | 263 | struct inode *inode = page_file_mapping(page)->host; |
89a09141 PZ |
264 | struct nfs_server *nfss = NFS_SERVER(inode); |
265 | ||
266 | end_page_writeback(page); | |
c4dc4bee | 267 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 268 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
269 | } |
270 | ||
cfb506e1 | 271 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 272 | { |
d56b4ddf | 273 | struct inode *inode = page_file_mapping(page)->host; |
e261f51f | 274 | struct nfs_page *req; |
e261f51f TM |
275 | int ret; |
276 | ||
587142f8 | 277 | spin_lock(&inode->i_lock); |
074cc1de | 278 | for (;;) { |
29418aa4 | 279 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
074cc1de TM |
280 | if (req == NULL) |
281 | break; | |
7ad84aa9 | 282 | if (nfs_lock_request(req)) |
e261f51f TM |
283 | break; |
284 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 285 | * then the call to nfs_lock_request() will always |
e261f51f TM |
286 | * succeed provided that someone hasn't already marked the |
287 | * request as dirty (in which case we don't care). | |
288 | */ | |
587142f8 | 289 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
290 | if (!nonblock) |
291 | ret = nfs_wait_on_request(req); | |
292 | else | |
293 | ret = -EAGAIN; | |
e261f51f TM |
294 | nfs_release_request(req); |
295 | if (ret != 0) | |
074cc1de | 296 | return ERR_PTR(ret); |
587142f8 | 297 | spin_lock(&inode->i_lock); |
e261f51f | 298 | } |
587142f8 | 299 | spin_unlock(&inode->i_lock); |
074cc1de TM |
300 | return req; |
301 | } | |
302 | ||
303 | /* | |
304 | * Find an associated nfs write request, and prepare to flush it out | |
305 | * May return an error if the user signalled nfs_wait_on_request(). | |
306 | */ | |
307 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 308 | struct page *page, bool nonblock) |
074cc1de TM |
309 | { |
310 | struct nfs_page *req; | |
311 | int ret = 0; | |
312 | ||
cfb506e1 | 313 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
314 | if (!req) |
315 | goto out; | |
316 | ret = PTR_ERR(req); | |
317 | if (IS_ERR(req)) | |
318 | goto out; | |
319 | ||
deed85e7 TM |
320 | nfs_set_page_writeback(page); |
321 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 322 | |
deed85e7 | 323 | ret = 0; |
f8512ad0 FI |
324 | if (!nfs_pageio_add_request(pgio, req)) { |
325 | nfs_redirty_request(req); | |
074cc1de | 326 | ret = pgio->pg_error; |
f8512ad0 | 327 | } |
074cc1de TM |
328 | out: |
329 | return ret; | |
e261f51f TM |
330 | } |
331 | ||
f758c885 | 332 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 333 | { |
d56b4ddf | 334 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 335 | int ret; |
1da177e4 | 336 | |
91d5b470 CL |
337 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
338 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
339 | ||
d56b4ddf | 340 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 341 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
342 | if (ret == -EAGAIN) { |
343 | redirty_page_for_writepage(wbc, page); | |
344 | ret = 0; | |
345 | } | |
346 | return ret; | |
f758c885 | 347 | } |
7fe7f848 | 348 | |
f758c885 TM |
349 | /* |
350 | * Write an mmapped page to the server. | |
351 | */ | |
352 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
353 | { | |
354 | struct nfs_pageio_descriptor pgio; | |
355 | int err; | |
49a70f27 | 356 | |
a20c93e3 CH |
357 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
358 | false, &nfs_async_write_completion_ops); | |
f758c885 TM |
359 | err = nfs_do_writepage(page, wbc, &pgio); |
360 | nfs_pageio_complete(&pgio); | |
361 | if (err < 0) | |
362 | return err; | |
363 | if (pgio.pg_error < 0) | |
364 | return pgio.pg_error; | |
365 | return 0; | |
4d770ccf TM |
366 | } |
367 | ||
368 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
369 | { | |
f758c885 | 370 | int ret; |
4d770ccf | 371 | |
f758c885 | 372 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 373 | unlock_page(page); |
f758c885 TM |
374 | return ret; |
375 | } | |
376 | ||
377 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
378 | { | |
379 | int ret; | |
380 | ||
381 | ret = nfs_do_writepage(page, wbc, data); | |
382 | unlock_page(page); | |
383 | return ret; | |
1da177e4 LT |
384 | } |
385 | ||
1da177e4 LT |
386 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
387 | { | |
1da177e4 | 388 | struct inode *inode = mapping->host; |
72cb77f4 | 389 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 390 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
391 | int err; |
392 | ||
72cb77f4 TM |
393 | /* Stop dirtying of new pages while we sync */ |
394 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
395 | nfs_wait_bit_killable, TASK_KILLABLE); | |
396 | if (err) | |
397 | goto out_err; | |
398 | ||
91d5b470 CL |
399 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
400 | ||
a20c93e3 CH |
401 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
402 | &nfs_async_write_completion_ops); | |
f758c885 | 403 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 404 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
405 | |
406 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
407 | smp_mb__after_clear_bit(); | |
408 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
409 | ||
f758c885 | 410 | if (err < 0) |
72cb77f4 TM |
411 | goto out_err; |
412 | err = pgio.pg_error; | |
413 | if (err < 0) | |
414 | goto out_err; | |
c63c7b05 | 415 | return 0; |
72cb77f4 TM |
416 | out_err: |
417 | return err; | |
1da177e4 LT |
418 | } |
419 | ||
420 | /* | |
421 | * Insert a write request into an inode | |
422 | */ | |
d6d6dc7c | 423 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
424 | { |
425 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
426 | |
427 | /* Lock the request! */ | |
7ad84aa9 | 428 | nfs_lock_request(req); |
e7d39069 TM |
429 | |
430 | spin_lock(&inode->i_lock); | |
011e2a7f | 431 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 432 | inode->i_version++; |
29418aa4 MG |
433 | /* |
434 | * Swap-space should not get truncated. Hence no need to plug the race | |
435 | * with invalidate/truncate. | |
436 | */ | |
437 | if (likely(!PageSwapCache(req->wb_page))) { | |
438 | set_bit(PG_MAPPED, &req->wb_flags); | |
439 | SetPagePrivate(req->wb_page); | |
440 | set_page_private(req->wb_page, (unsigned long)req); | |
441 | } | |
1da177e4 | 442 | nfsi->npages++; |
c03b4024 | 443 | kref_get(&req->wb_kref); |
e7d39069 | 444 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
445 | } |
446 | ||
447 | /* | |
89a09141 | 448 | * Remove a write request from an inode |
1da177e4 LT |
449 | */ |
450 | static void nfs_inode_remove_request(struct nfs_page *req) | |
451 | { | |
3d4ff43d | 452 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
453 | struct nfs_inode *nfsi = NFS_I(inode); |
454 | ||
587142f8 | 455 | spin_lock(&inode->i_lock); |
29418aa4 MG |
456 | if (likely(!PageSwapCache(req->wb_page))) { |
457 | set_page_private(req->wb_page, 0); | |
458 | ClearPagePrivate(req->wb_page); | |
459 | clear_bit(PG_MAPPED, &req->wb_flags); | |
460 | } | |
1da177e4 | 461 | nfsi->npages--; |
4d65c520 | 462 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
463 | nfs_release_request(req); |
464 | } | |
465 | ||
61822ab5 | 466 | static void |
6d884e8f | 467 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 468 | { |
61822ab5 TM |
469 | __set_page_dirty_nobuffers(req->wb_page); |
470 | } | |
471 | ||
89d77c8f | 472 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
8dd37758 TM |
473 | /** |
474 | * nfs_request_add_commit_list - add request to a commit list | |
475 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
476 | * @dst: commit list head |
477 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 478 | * |
ea2cf228 | 479 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
480 | * number of outstanding requests requiring a commit as well as |
481 | * the MM page stats. | |
482 | * | |
ea2cf228 | 483 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 484 | * holding the nfs_page lock. |
1da177e4 | 485 | */ |
8dd37758 | 486 | void |
ea2cf228 FI |
487 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
488 | struct nfs_commit_info *cinfo) | |
1da177e4 | 489 | { |
e468bae9 | 490 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
491 | spin_lock(cinfo->lock); |
492 | nfs_list_add_request(req, dst); | |
493 | cinfo->mds->ncommit++; | |
494 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
495 | if (!cinfo->dreq) { |
496 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 497 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
498 | BDI_RECLAIMABLE); |
499 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
500 | I_DIRTY_DATASYNC); | |
501 | } | |
1da177e4 | 502 | } |
8dd37758 TM |
503 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
504 | ||
505 | /** | |
506 | * nfs_request_remove_commit_list - Remove request from a commit list | |
507 | * @req: pointer to a nfs_page | |
ea2cf228 | 508 | * @cinfo: holds list lock and accounting info |
8dd37758 | 509 | * |
ea2cf228 | 510 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
511 | * number of outstanding requests requiring a commit |
512 | * It does not update the MM page stats. | |
513 | * | |
ea2cf228 | 514 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
515 | */ |
516 | void | |
ea2cf228 FI |
517 | nfs_request_remove_commit_list(struct nfs_page *req, |
518 | struct nfs_commit_info *cinfo) | |
8dd37758 | 519 | { |
8dd37758 TM |
520 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
521 | return; | |
522 | nfs_list_remove_request(req); | |
ea2cf228 | 523 | cinfo->mds->ncommit--; |
8dd37758 TM |
524 | } |
525 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
526 | ||
ea2cf228 FI |
527 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
528 | struct inode *inode) | |
529 | { | |
530 | cinfo->lock = &inode->i_lock; | |
531 | cinfo->mds = &NFS_I(inode)->commit_info; | |
532 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 533 | cinfo->dreq = NULL; |
f453a54a | 534 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
535 | } |
536 | ||
537 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
538 | struct inode *inode, | |
539 | struct nfs_direct_req *dreq) | |
540 | { | |
1763da12 FI |
541 | if (dreq) |
542 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
543 | else | |
544 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
545 | } |
546 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
547 | |
548 | /* | |
549 | * Add a request to the inode's commit list. | |
550 | */ | |
1763da12 | 551 | void |
ea2cf228 FI |
552 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
553 | struct nfs_commit_info *cinfo) | |
8dd37758 | 554 | { |
ea2cf228 | 555 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 556 | return; |
ea2cf228 | 557 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 558 | } |
8e821cad | 559 | |
d6d6dc7c FI |
560 | static void |
561 | nfs_clear_page_commit(struct page *page) | |
562 | { | |
563 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 564 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
565 | } |
566 | ||
8dd37758 | 567 | static void |
e468bae9 TM |
568 | nfs_clear_request_commit(struct nfs_page *req) |
569 | { | |
8dd37758 TM |
570 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
571 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 572 | struct nfs_commit_info cinfo; |
e468bae9 | 573 | |
ea2cf228 FI |
574 | nfs_init_cinfo_from_inode(&cinfo, inode); |
575 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
576 | spin_lock(cinfo.lock); | |
577 | nfs_request_remove_commit_list(req, &cinfo); | |
578 | spin_unlock(cinfo.lock); | |
8dd37758 | 579 | } |
d6d6dc7c | 580 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 581 | } |
e468bae9 TM |
582 | } |
583 | ||
8e821cad TM |
584 | static inline |
585 | int nfs_write_need_commit(struct nfs_write_data *data) | |
586 | { | |
465d5243 | 587 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
588 | return data->header->lseg == NULL; |
589 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
590 | } |
591 | ||
8e821cad | 592 | #else |
68cd6fa4 BS |
593 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
594 | struct inode *inode) | |
595 | { | |
596 | } | |
597 | ||
598 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
599 | struct inode *inode, | |
600 | struct nfs_direct_req *dreq) | |
601 | { | |
602 | } | |
603 | ||
1763da12 | 604 | void |
ea2cf228 FI |
605 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
606 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
607 | { |
608 | } | |
609 | ||
8dd37758 | 610 | static void |
e468bae9 TM |
611 | nfs_clear_request_commit(struct nfs_page *req) |
612 | { | |
e468bae9 TM |
613 | } |
614 | ||
8e821cad TM |
615 | static inline |
616 | int nfs_write_need_commit(struct nfs_write_data *data) | |
617 | { | |
618 | return 0; | |
619 | } | |
620 | ||
6c75dc0d FI |
621 | #endif |
622 | ||
061ae2ed | 623 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 624 | { |
ea2cf228 | 625 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
626 | unsigned long bytes = 0; |
627 | ||
628 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
629 | goto out; | |
ea2cf228 | 630 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
631 | while (!list_empty(&hdr->pages)) { |
632 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
633 | |
634 | bytes += req->wb_bytes; | |
635 | nfs_list_remove_request(req); | |
636 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
637 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 638 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
639 | nfs_context_set_write_error(req->wb_context, hdr->error); |
640 | goto remove_req; | |
641 | } | |
642 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
643 | nfs_mark_request_dirty(req); | |
644 | goto next; | |
645 | } | |
646 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
2f2c63bc | 647 | memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf)); |
ea2cf228 | 648 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
649 | goto next; |
650 | } | |
651 | remove_req: | |
652 | nfs_inode_remove_request(req); | |
653 | next: | |
1d1afcbc | 654 | nfs_unlock_request(req); |
d1182b33 | 655 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 656 | nfs_release_request(req); |
6c75dc0d FI |
657 | } |
658 | out: | |
659 | hdr->release(hdr); | |
8e821cad | 660 | } |
1da177e4 | 661 | |
89d77c8f | 662 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
ea2cf228 FI |
663 | static unsigned long |
664 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) | |
fb8a1f11 | 665 | { |
ea2cf228 | 666 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
667 | } |
668 | ||
ea2cf228 | 669 | /* cinfo->lock held by caller */ |
1763da12 | 670 | int |
ea2cf228 FI |
671 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
672 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
673 | { |
674 | struct nfs_page *req, *tmp; | |
675 | int ret = 0; | |
676 | ||
677 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
678 | if (!nfs_lock_request(req)) |
679 | continue; | |
7ad84aa9 | 680 | kref_get(&req->wb_kref); |
ea2cf228 | 681 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 682 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 683 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
684 | nfs_list_add_request(req, dst); |
685 | ret++; | |
1763da12 | 686 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 687 | break; |
d6d6dc7c FI |
688 | } |
689 | return ret; | |
fb8a1f11 TM |
690 | } |
691 | ||
1da177e4 LT |
692 | /* |
693 | * nfs_scan_commit - Scan an inode for commit requests | |
694 | * @inode: NFS inode to scan | |
ea2cf228 FI |
695 | * @dst: mds destination list |
696 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
697 | * |
698 | * Moves requests from the inode's 'commit' request list. | |
699 | * The requests are *not* checked to ensure that they form a contiguous set. | |
700 | */ | |
1763da12 | 701 | int |
ea2cf228 FI |
702 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
703 | struct nfs_commit_info *cinfo) | |
1da177e4 | 704 | { |
d6d6dc7c | 705 | int ret = 0; |
fb8a1f11 | 706 | |
ea2cf228 FI |
707 | spin_lock(cinfo->lock); |
708 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 709 | const int max = INT_MAX; |
d6d6dc7c | 710 | |
ea2cf228 FI |
711 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
712 | cinfo, max); | |
713 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 714 | } |
ea2cf228 | 715 | spin_unlock(cinfo->lock); |
ff778d02 | 716 | return ret; |
1da177e4 | 717 | } |
d6d6dc7c | 718 | |
c42de9dd | 719 | #else |
ea2cf228 | 720 | static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
721 | { |
722 | return 0; | |
723 | } | |
724 | ||
1763da12 FI |
725 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
726 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
727 | { |
728 | return 0; | |
729 | } | |
1da177e4 LT |
730 | #endif |
731 | ||
1da177e4 | 732 | /* |
e7d39069 TM |
733 | * Search for an existing write request, and attempt to update |
734 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 735 | * |
e7d39069 TM |
736 | * If the attempt fails, then the existing request is flushed out |
737 | * to disk. | |
1da177e4 | 738 | */ |
e7d39069 TM |
739 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
740 | struct page *page, | |
741 | unsigned int offset, | |
742 | unsigned int bytes) | |
1da177e4 | 743 | { |
e7d39069 TM |
744 | struct nfs_page *req; |
745 | unsigned int rqend; | |
746 | unsigned int end; | |
747 | int error; | |
748 | ||
749 | if (!PagePrivate(page)) | |
750 | return NULL; | |
1da177e4 LT |
751 | |
752 | end = offset + bytes; | |
e7d39069 | 753 | spin_lock(&inode->i_lock); |
1da177e4 | 754 | |
1da177e4 | 755 | for (;;) { |
29418aa4 | 756 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
e7d39069 TM |
757 | if (req == NULL) |
758 | goto out_unlock; | |
759 | ||
760 | rqend = req->wb_offset + req->wb_bytes; | |
761 | /* | |
762 | * Tell the caller to flush out the request if | |
763 | * the offsets are non-contiguous. | |
764 | * Note: nfs_flush_incompatible() will already | |
765 | * have flushed out requests having wrong owners. | |
766 | */ | |
e468bae9 | 767 | if (offset > rqend |
e7d39069 TM |
768 | || end < req->wb_offset) |
769 | goto out_flushme; | |
770 | ||
7ad84aa9 | 771 | if (nfs_lock_request(req)) |
1da177e4 | 772 | break; |
1da177e4 | 773 | |
e7d39069 | 774 | /* The request is locked, so wait and then retry */ |
587142f8 | 775 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
776 | error = nfs_wait_on_request(req); |
777 | nfs_release_request(req); | |
778 | if (error != 0) | |
779 | goto out_err; | |
780 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
781 | } |
782 | ||
783 | /* Okay, the request matches. Update the region */ | |
784 | if (offset < req->wb_offset) { | |
785 | req->wb_offset = offset; | |
786 | req->wb_pgbase = offset; | |
1da177e4 | 787 | } |
1da177e4 LT |
788 | if (end > rqend) |
789 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
790 | else |
791 | req->wb_bytes = rqend - req->wb_offset; | |
792 | out_unlock: | |
793 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
794 | if (req) |
795 | nfs_clear_request_commit(req); | |
e7d39069 TM |
796 | return req; |
797 | out_flushme: | |
798 | spin_unlock(&inode->i_lock); | |
799 | nfs_release_request(req); | |
800 | error = nfs_wb_page(inode, page); | |
801 | out_err: | |
802 | return ERR_PTR(error); | |
803 | } | |
804 | ||
805 | /* | |
806 | * Try to update an existing write request, or create one if there is none. | |
807 | * | |
808 | * Note: Should always be called with the Page Lock held to prevent races | |
809 | * if we have to add a new request. Also assumes that the caller has | |
810 | * already called nfs_flush_incompatible() if necessary. | |
811 | */ | |
812 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
813 | struct page *page, unsigned int offset, unsigned int bytes) | |
814 | { | |
d56b4ddf | 815 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 816 | struct nfs_page *req; |
1da177e4 | 817 | |
e7d39069 TM |
818 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
819 | if (req != NULL) | |
820 | goto out; | |
821 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
822 | if (IS_ERR(req)) | |
823 | goto out; | |
d6d6dc7c | 824 | nfs_inode_add_request(inode, req); |
efc91ed0 | 825 | out: |
61e930a9 | 826 | return req; |
1da177e4 LT |
827 | } |
828 | ||
e7d39069 TM |
829 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
830 | unsigned int offset, unsigned int count) | |
831 | { | |
832 | struct nfs_page *req; | |
833 | ||
834 | req = nfs_setup_write_request(ctx, page, offset, count); | |
835 | if (IS_ERR(req)) | |
836 | return PTR_ERR(req); | |
837 | /* Update file length */ | |
838 | nfs_grow_file(page, offset, count); | |
839 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 840 | nfs_mark_request_dirty(req); |
1d1afcbc | 841 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
842 | return 0; |
843 | } | |
844 | ||
1da177e4 LT |
845 | int nfs_flush_incompatible(struct file *file, struct page *page) |
846 | { | |
cd3758e3 | 847 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 848 | struct nfs_lock_context *l_ctx; |
1da177e4 | 849 | struct nfs_page *req; |
1a54533e | 850 | int do_flush, status; |
1da177e4 LT |
851 | /* |
852 | * Look for a request corresponding to this page. If there | |
853 | * is one, and it belongs to another file, we flush it out | |
854 | * before we try to copy anything into the page. Do this | |
855 | * due to the lack of an ACCESS-type call in NFSv2. | |
856 | * Also do the same if we find a request from an existing | |
857 | * dropped page. | |
858 | */ | |
1a54533e TM |
859 | do { |
860 | req = nfs_page_find_request(page); | |
861 | if (req == NULL) | |
862 | return 0; | |
2a369153 TM |
863 | l_ctx = req->wb_lock_context; |
864 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
0f1d2605 | 865 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
866 | do_flush |= l_ctx->lockowner.l_owner != current->files |
867 | || l_ctx->lockowner.l_pid != current->tgid; | |
868 | } | |
1da177e4 | 869 | nfs_release_request(req); |
1a54533e TM |
870 | if (!do_flush) |
871 | return 0; | |
d56b4ddf | 872 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
873 | } while (status == 0); |
874 | return status; | |
1da177e4 LT |
875 | } |
876 | ||
dc24826b AA |
877 | /* |
878 | * Avoid buffered writes when a open context credential's key would | |
879 | * expire soon. | |
880 | * | |
881 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
882 | * | |
883 | * Return 0 and set a credential flag which triggers the inode to flush | |
884 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
885 | * RPC_KEY_EXPIRE_TIMEO. | |
886 | */ | |
887 | int | |
888 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
889 | { | |
890 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
891 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
892 | ||
893 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
894 | } | |
895 | ||
896 | /* | |
897 | * Test if the open context credential key is marked to expire soon. | |
898 | */ | |
899 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
900 | { | |
901 | return rpcauth_cred_key_to_expire(ctx->cred); | |
902 | } | |
903 | ||
5d47a356 TM |
904 | /* |
905 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
906 | * the PageUptodate() flag. In this case, we will need to turn off | |
907 | * write optimisations that depend on the page contents being correct. | |
908 | */ | |
8d197a56 | 909 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 910 | { |
d529ef83 JL |
911 | struct nfs_inode *nfsi = NFS_I(inode); |
912 | ||
8d197a56 TM |
913 | if (nfs_have_delegated_attributes(inode)) |
914 | goto out; | |
d529ef83 JL |
915 | if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) |
916 | return false; | |
4db72b40 | 917 | smp_rmb(); |
d529ef83 | 918 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
919 | return false; |
920 | out: | |
921 | return PageUptodate(page) != 0; | |
5d47a356 TM |
922 | } |
923 | ||
c7559663 SM |
924 | /* If we know the page is up to date, and we're not using byte range locks (or |
925 | * if we have the whole file locked for writing), it may be more efficient to | |
926 | * extend the write to cover the entire page in order to avoid fragmentation | |
927 | * inefficiencies. | |
928 | * | |
263b4509 SM |
929 | * If the file is opened for synchronous writes then we can just skip the rest |
930 | * of the checks. | |
c7559663 SM |
931 | */ |
932 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
933 | { | |
934 | if (file->f_flags & O_DSYNC) | |
935 | return 0; | |
263b4509 SM |
936 | if (!nfs_write_pageuptodate(page, inode)) |
937 | return 0; | |
c7559663 SM |
938 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
939 | return 1; | |
263b4509 | 940 | if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && |
c7559663 | 941 | inode->i_flock->fl_end == OFFSET_MAX && |
263b4509 | 942 | inode->i_flock->fl_type != F_RDLCK)) |
c7559663 SM |
943 | return 1; |
944 | return 0; | |
945 | } | |
946 | ||
1da177e4 LT |
947 | /* |
948 | * Update and possibly write a cached page of an NFS file. | |
949 | * | |
950 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
951 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
952 | */ | |
953 | int nfs_updatepage(struct file *file, struct page *page, | |
954 | unsigned int offset, unsigned int count) | |
955 | { | |
cd3758e3 | 956 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 957 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
958 | int status = 0; |
959 | ||
91d5b470 CL |
960 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
961 | ||
6de1472f AV |
962 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
963 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 964 | |
c7559663 | 965 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 966 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 967 | offset = 0; |
1da177e4 LT |
968 | } |
969 | ||
e21195a7 | 970 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
971 | if (status < 0) |
972 | nfs_set_pageerror(page); | |
59b7c05f TM |
973 | else |
974 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 975 | |
48186c7d | 976 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 977 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
978 | return status; |
979 | } | |
980 | ||
3ff7576d | 981 | static int flush_task_priority(int how) |
1da177e4 LT |
982 | { |
983 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
984 | case FLUSH_HIGHPRI: | |
985 | return RPC_PRIORITY_HIGH; | |
986 | case FLUSH_LOWPRI: | |
987 | return RPC_PRIORITY_LOW; | |
988 | } | |
989 | return RPC_PRIORITY_NORMAL; | |
990 | } | |
991 | ||
c5996c4e FI |
992 | int nfs_initiate_write(struct rpc_clnt *clnt, |
993 | struct nfs_write_data *data, | |
d138d5d1 | 994 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 995 | int how, int flags) |
1da177e4 | 996 | { |
cd841605 | 997 | struct inode *inode = data->header->inode; |
3ff7576d | 998 | int priority = flush_task_priority(how); |
07737691 | 999 | struct rpc_task *task; |
bdc7f021 TM |
1000 | struct rpc_message msg = { |
1001 | .rpc_argp = &data->args, | |
1002 | .rpc_resp = &data->res, | |
cd841605 | 1003 | .rpc_cred = data->header->cred, |
bdc7f021 | 1004 | }; |
84115e1c | 1005 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 1006 | .rpc_client = clnt, |
07737691 | 1007 | .task = &data->task, |
bdc7f021 | 1008 | .rpc_message = &msg, |
84115e1c TM |
1009 | .callback_ops = call_ops, |
1010 | .callback_data = data, | |
101070ca | 1011 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1012 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1013 | .priority = priority, |
84115e1c | 1014 | }; |
2c61be0a | 1015 | int ret = 0; |
1da177e4 | 1016 | |
d138d5d1 AA |
1017 | /* Set up the initial task struct. */ |
1018 | NFS_PROTO(inode)->write_setup(data, &msg); | |
1019 | ||
1020 | dprintk("NFS: %5u initiated write call " | |
1e8968c5 | 1021 | "(req %s/%llu, %u bytes @ offset %llu)\n", |
d138d5d1 AA |
1022 | data->task.tk_pid, |
1023 | inode->i_sb->s_id, | |
1e8968c5 | 1024 | (unsigned long long)NFS_FILEID(inode), |
d138d5d1 AA |
1025 | data->args.count, |
1026 | (unsigned long long)data->args.offset); | |
1027 | ||
8c21c62c WAA |
1028 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
1029 | &task_setup_data.rpc_client, &msg, data); | |
1030 | ||
d138d5d1 AA |
1031 | task = rpc_run_task(&task_setup_data); |
1032 | if (IS_ERR(task)) { | |
1033 | ret = PTR_ERR(task); | |
1034 | goto out; | |
1035 | } | |
1036 | if (how & FLUSH_SYNC) { | |
1037 | ret = rpc_wait_for_completion_task(task); | |
1038 | if (ret == 0) | |
1039 | ret = task->tk_status; | |
1040 | } | |
1041 | rpc_put_task(task); | |
1042 | out: | |
1043 | return ret; | |
1044 | } | |
a69aef14 | 1045 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
1046 | |
1047 | /* | |
1048 | * Set up the argument/result storage required for the RPC call. | |
1049 | */ | |
6c75dc0d | 1050 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 | 1051 | unsigned int count, unsigned int offset, |
ea2cf228 | 1052 | int how, struct nfs_commit_info *cinfo) |
d138d5d1 | 1053 | { |
6c75dc0d | 1054 | struct nfs_page *req = data->header->req; |
d138d5d1 | 1055 | |
1da177e4 LT |
1056 | /* Set up the RPC argument and reply structs |
1057 | * NB: take care not to mess about with data->commit et al. */ | |
1058 | ||
6c75dc0d | 1059 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 1060 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
1061 | /* pnfs_set_layoutcommit needs this */ |
1062 | data->mds_offset = data->args.offset; | |
1da177e4 | 1063 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 1064 | data->args.pages = data->pages.pagevec; |
1da177e4 | 1065 | data->args.count = count; |
383ba719 | 1066 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 1067 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 1068 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
1069 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
1070 | case 0: | |
1071 | break; | |
1072 | case FLUSH_COND_STABLE: | |
ea2cf228 | 1073 | if (nfs_reqs_to_commit(cinfo)) |
87ed5eb4 TM |
1074 | break; |
1075 | default: | |
1076 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 1077 | } |
1da177e4 LT |
1078 | |
1079 | data->res.fattr = &data->fattr; | |
1080 | data->res.count = count; | |
1081 | data->res.verf = &data->verf; | |
0e574af1 | 1082 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 1083 | } |
1da177e4 | 1084 | |
6e4efd56 TM |
1085 | static int nfs_do_write(struct nfs_write_data *data, |
1086 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
1087 | int how) |
1088 | { | |
cd841605 | 1089 | struct inode *inode = data->header->inode; |
0382b744 | 1090 | |
9f0ec176 | 1091 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0); |
1da177e4 LT |
1092 | } |
1093 | ||
275acaaf TM |
1094 | static int nfs_do_multiple_writes(struct list_head *head, |
1095 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
1096 | int how) |
1097 | { | |
1098 | struct nfs_write_data *data; | |
1099 | int ret = 0; | |
1100 | ||
1101 | while (!list_empty(head)) { | |
1102 | int ret2; | |
1103 | ||
6c75dc0d | 1104 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
1105 | list_del_init(&data->list); |
1106 | ||
dce81290 | 1107 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
1108 | if (ret == 0) |
1109 | ret = ret2; | |
1110 | } | |
1111 | return ret; | |
1112 | } | |
1113 | ||
6d884e8f F |
1114 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1115 | * call this on each, which will prepare them to be retried on next | |
1116 | * writeback using standard nfs. | |
1117 | */ | |
1118 | static void nfs_redirty_request(struct nfs_page *req) | |
1119 | { | |
1120 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1121 | nfs_unlock_request(req); |
d1182b33 | 1122 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 1123 | nfs_release_request(req); |
6d884e8f F |
1124 | } |
1125 | ||
061ae2ed | 1126 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1127 | { |
1128 | struct nfs_page *req; | |
1129 | ||
1130 | while (!list_empty(head)) { | |
1131 | req = nfs_list_entry(head->next); | |
1132 | nfs_list_remove_request(req); | |
1133 | nfs_redirty_request(req); | |
1134 | } | |
1135 | } | |
1136 | ||
061ae2ed FI |
1137 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1138 | .error_cleanup = nfs_async_write_error, | |
1139 | .completion = nfs_write_completion, | |
1140 | }; | |
1141 | ||
25b11dcd TM |
1142 | static void nfs_flush_error(struct nfs_pageio_descriptor *desc, |
1143 | struct nfs_pgio_header *hdr) | |
1144 | { | |
1145 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1146 | while (!list_empty(&hdr->rpc_list)) { | |
1147 | struct nfs_write_data *data = list_first_entry(&hdr->rpc_list, | |
1148 | struct nfs_write_data, list); | |
1149 | list_del(&data->list); | |
1150 | nfs_writedata_release(data); | |
1151 | } | |
1152 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
1153 | } | |
1154 | ||
1da177e4 LT |
1155 | /* |
1156 | * Generate multiple small requests to write out a single | |
1157 | * contiguous dirty area on one page. | |
1158 | */ | |
6c75dc0d FI |
1159 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1160 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1161 | { |
6c75dc0d | 1162 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1163 | struct page *page = req->wb_page; |
1164 | struct nfs_write_data *data; | |
d097971d | 1165 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1166 | unsigned int offset; |
1da177e4 | 1167 | int requests = 0; |
ea2cf228 | 1168 | struct nfs_commit_info cinfo; |
1da177e4 | 1169 | |
ea2cf228 | 1170 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
1da177e4 | 1171 | |
b31268ac | 1172 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1173 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || |
b31268ac TM |
1174 | desc->pg_count > wsize)) |
1175 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1176 | ||
1177 | ||
275acaaf | 1178 | offset = 0; |
c76069bd | 1179 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1180 | do { |
1181 | size_t len = min(nbytes, wsize); | |
1182 | ||
6c75dc0d | 1183 | data = nfs_writedata_alloc(hdr, 1); |
25b11dcd TM |
1184 | if (!data) { |
1185 | nfs_flush_error(desc, hdr); | |
1186 | return -ENOMEM; | |
1187 | } | |
30dd374f | 1188 | data->pages.pagevec[0] = page; |
ea2cf228 | 1189 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); |
6c75dc0d | 1190 | list_add(&data->list, &hdr->rpc_list); |
1da177e4 | 1191 | requests++; |
e9f7bee1 | 1192 | nbytes -= len; |
275acaaf | 1193 | offset += len; |
e9f7bee1 | 1194 | } while (nbytes != 0); |
25b11dcd TM |
1195 | nfs_list_remove_request(req); |
1196 | nfs_list_add_request(req, &hdr->pages); | |
6c75dc0d | 1197 | desc->pg_rpc_callops = &nfs_write_common_ops; |
25b11dcd | 1198 | return 0; |
1da177e4 LT |
1199 | } |
1200 | ||
1201 | /* | |
1202 | * Create an RPC task for the given write request and kick it. | |
1203 | * The page must have been locked by the caller. | |
1204 | * | |
1205 | * It may happen that the page we're passed is not marked dirty. | |
1206 | * This is the case if nfs_updatepage detects a conflicting request | |
1207 | * that has been written but not committed. | |
1208 | */ | |
6c75dc0d FI |
1209 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1210 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1211 | { |
1212 | struct nfs_page *req; | |
1213 | struct page **pages; | |
1214 | struct nfs_write_data *data; | |
c76069bd | 1215 | struct list_head *head = &desc->pg_list; |
ea2cf228 | 1216 | struct nfs_commit_info cinfo; |
1da177e4 | 1217 | |
6c75dc0d FI |
1218 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1219 | desc->pg_count)); | |
1220 | if (!data) { | |
25b11dcd TM |
1221 | nfs_flush_error(desc, hdr); |
1222 | return -ENOMEM; | |
44b83799 | 1223 | } |
6c75dc0d | 1224 | |
ea2cf228 | 1225 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
30dd374f | 1226 | pages = data->pages.pagevec; |
1da177e4 LT |
1227 | while (!list_empty(head)) { |
1228 | req = nfs_list_entry(head->next); | |
1229 | nfs_list_remove_request(req); | |
6c75dc0d | 1230 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1231 | *pages++ = req->wb_page; |
1da177e4 | 1232 | } |
1da177e4 | 1233 | |
b31268ac | 1234 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1235 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) |
b31268ac TM |
1236 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; |
1237 | ||
1da177e4 | 1238 | /* Set up the argument struct */ |
ea2cf228 | 1239 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); |
6c75dc0d FI |
1240 | list_add(&data->list, &hdr->rpc_list); |
1241 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
25b11dcd | 1242 | return 0; |
1da177e4 LT |
1243 | } |
1244 | ||
6c75dc0d FI |
1245 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1246 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1247 | { |
1248 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1249 | return nfs_flush_multi(desc, hdr); |
1250 | return nfs_flush_one(desc, hdr); | |
dce81290 | 1251 | } |
89d77c8f | 1252 | EXPORT_SYMBOL_GPL(nfs_generic_flush); |
dce81290 TM |
1253 | |
1254 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1255 | { |
6c75dc0d FI |
1256 | struct nfs_write_header *whdr; |
1257 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1258 | int ret; |
1259 | ||
6c75dc0d FI |
1260 | whdr = nfs_writehdr_alloc(); |
1261 | if (!whdr) { | |
9b5415b5 | 1262 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
6c75dc0d FI |
1263 | return -ENOMEM; |
1264 | } | |
1265 | hdr = &whdr->header; | |
1266 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1267 | atomic_inc(&hdr->refcnt); | |
1268 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1269 | if (ret == 0) |
6c75dc0d FI |
1270 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1271 | desc->pg_rpc_callops, | |
1272 | desc->pg_ioflags); | |
6c75dc0d | 1273 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 1274 | hdr->completion_ops->completion(hdr); |
275acaaf | 1275 | return ret; |
1751c363 | 1276 | } |
1751c363 TM |
1277 | |
1278 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1279 | .pg_test = nfs_generic_pg_test, | |
1280 | .pg_doio = nfs_generic_pg_writepages, | |
1281 | }; | |
1282 | ||
57208fa7 | 1283 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1284 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1285 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1286 | { |
a20c93e3 CH |
1287 | struct nfs_server *server = NFS_SERVER(inode); |
1288 | const struct nfs_pageio_ops *pg_ops = &nfs_pageio_write_ops; | |
1289 | ||
1290 | #ifdef CONFIG_NFS_V4_1 | |
1291 | if (server->pnfs_curr_ld && !force_mds) | |
1292 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1293 | #endif | |
1294 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->wsize, ioflags); | |
1751c363 | 1295 | } |
ddda8e0a | 1296 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1297 | |
dce81290 TM |
1298 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1299 | { | |
1300 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1301 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1302 | } | |
1f945357 | 1303 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1304 | |
1da177e4 | 1305 | |
def6ed7e AA |
1306 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1307 | { | |
1308 | struct nfs_write_data *data = calldata; | |
ef1820f9 N |
1309 | int err; |
1310 | err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); | |
1311 | if (err) | |
1312 | rpc_exit(task, err); | |
def6ed7e | 1313 | } |
def6ed7e | 1314 | |
0b7c0153 FI |
1315 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1316 | { | |
1317 | struct nfs_commit_data *data = calldata; | |
1318 | ||
1319 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1320 | } | |
1321 | ||
1da177e4 LT |
1322 | /* |
1323 | * Handle a write reply that flushes a whole page. | |
1324 | * | |
1325 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1326 | * writebacks since the page->count is kept > 1 for as long | |
1327 | * as the page has a write request pending. | |
1328 | */ | |
6c75dc0d | 1329 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1330 | { |
788e7a89 | 1331 | struct nfs_write_data *data = calldata; |
1da177e4 | 1332 | |
c9d8f89d TM |
1333 | nfs_writeback_done(task, data); |
1334 | } | |
1335 | ||
6c75dc0d | 1336 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1337 | { |
1338 | struct nfs_write_data *data = calldata; | |
cd841605 | 1339 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1340 | int status = data->task.tk_status; |
788e7a89 | 1341 | |
6c75dc0d FI |
1342 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1343 | spin_lock(&hdr->lock); | |
1344 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1345 | ; /* Do nothing */ | |
1346 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
9bce008b TM |
1347 | memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf)); |
1348 | else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf))) | |
6c75dc0d FI |
1349 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); |
1350 | spin_unlock(&hdr->lock); | |
1da177e4 | 1351 | } |
cd841605 | 1352 | nfs_writedata_release(data); |
1da177e4 LT |
1353 | } |
1354 | ||
6c75dc0d | 1355 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1356 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1357 | .rpc_call_done = nfs_writeback_done_common, |
1358 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1359 | }; |
1360 | ||
1f2edbe3 TM |
1361 | /* |
1362 | * Special version of should_remove_suid() that ignores capabilities. | |
1363 | */ | |
1364 | static int nfs_should_remove_suid(const struct inode *inode) | |
1365 | { | |
1366 | umode_t mode = inode->i_mode; | |
1367 | int kill = 0; | |
1368 | ||
1369 | /* suid always must be killed */ | |
1370 | if (unlikely(mode & S_ISUID)) | |
1371 | kill = ATTR_KILL_SUID; | |
1372 | ||
1373 | /* | |
1374 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1375 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1376 | */ | |
1377 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1378 | kill |= ATTR_KILL_SGID; | |
1379 | ||
1380 | if (unlikely(kill && S_ISREG(mode))) | |
1381 | return kill; | |
1382 | ||
1383 | return 0; | |
1384 | } | |
788e7a89 | 1385 | |
1da177e4 LT |
1386 | /* |
1387 | * This function is called when the WRITE call is complete. | |
1388 | */ | |
13602896 | 1389 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1390 | { |
3c6b899c | 1391 | struct nfs_pgio_args *argp = &data->args; |
1da177e4 | 1392 | struct nfs_writeres *resp = &data->res; |
cd841605 | 1393 | struct inode *inode = data->header->inode; |
788e7a89 | 1394 | int status; |
1da177e4 | 1395 | |
a3f565b1 | 1396 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1397 | task->tk_pid, task->tk_status); |
1398 | ||
f551e44f CL |
1399 | /* |
1400 | * ->write_done will attempt to use post-op attributes to detect | |
1401 | * conflicting writes by other clients. A strict interpretation | |
1402 | * of close-to-open would allow us to continue caching even if | |
1403 | * another writer had changed the file, but some applications | |
1404 | * depend on tighter cache coherency when writing. | |
1405 | */ | |
cd841605 | 1406 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1407 | if (status != 0) |
13602896 | 1408 | return; |
cd841605 | 1409 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1410 | |
89d77c8f | 1411 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
1da177e4 LT |
1412 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { |
1413 | /* We tried a write call, but the server did not | |
1414 | * commit data to stable storage even though we | |
1415 | * requested it. | |
1416 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1417 | * the server reports NFS_DATA_SYNC, but performs | |
1418 | * NFS_FILE_SYNC. We therefore implement this checking | |
1419 | * as a dprintk() in order to avoid filling syslog. | |
1420 | */ | |
1421 | static unsigned long complain; | |
1422 | ||
a69aef14 | 1423 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1424 | if (time_before(complain, jiffies)) { |
48186c7d | 1425 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1426 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1427 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1428 | resp->verf->committed, argp->stable); |
1429 | complain = jiffies + 300 * HZ; | |
1430 | } | |
1431 | } | |
1432 | #endif | |
1f2edbe3 | 1433 | if (task->tk_status < 0) { |
6c75dc0d | 1434 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); |
1f2edbe3 TM |
1435 | return; |
1436 | } | |
1437 | ||
1438 | /* Deal with the suid/sgid bit corner case */ | |
1439 | if (nfs_should_remove_suid(inode)) | |
1440 | nfs_mark_for_revalidate(inode); | |
1441 | ||
1442 | if (resp->count < argp->count) { | |
1da177e4 LT |
1443 | static unsigned long complain; |
1444 | ||
6c75dc0d | 1445 | /* This a short write! */ |
cd841605 | 1446 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1447 | |
1da177e4 | 1448 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1449 | if (resp->count == 0) { |
1450 | if (time_before(complain, jiffies)) { | |
1451 | printk(KERN_WARNING | |
1452 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1453 | argp->count); | |
1454 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1455 | } |
6c75dc0d FI |
1456 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1457 | task->tk_status = -EIO; | |
13602896 | 1458 | return; |
1da177e4 | 1459 | } |
6c75dc0d FI |
1460 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1461 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1462 | /* Resend from where the server left off */ | |
1463 | data->mds_offset += resp->count; | |
1464 | argp->offset += resp->count; | |
1465 | argp->pgbase += resp->count; | |
1466 | argp->count -= resp->count; | |
1467 | } else { | |
1468 | /* Resend as a stable write in order to avoid | |
1469 | * headaches in the case of a server crash. | |
1470 | */ | |
1471 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1472 | } |
6c75dc0d | 1473 | rpc_restart_call_prepare(task); |
1da177e4 | 1474 | } |
1da177e4 LT |
1475 | } |
1476 | ||
1477 | ||
89d77c8f | 1478 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
71d0a611 TM |
1479 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1480 | { | |
b8413f98 TM |
1481 | int ret; |
1482 | ||
71d0a611 TM |
1483 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1484 | return 1; | |
b8413f98 TM |
1485 | if (!may_wait) |
1486 | return 0; | |
1487 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1488 | NFS_INO_COMMIT, | |
1489 | nfs_wait_bit_killable, | |
1490 | TASK_KILLABLE); | |
1491 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1492 | } |
1493 | ||
f453a54a | 1494 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1495 | { |
1496 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1497 | smp_mb__after_clear_bit(); | |
1498 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1499 | } | |
1500 | ||
0b7c0153 | 1501 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1502 | { |
0b7c0153 FI |
1503 | put_nfs_open_context(data->context); |
1504 | nfs_commit_free(data); | |
1da177e4 | 1505 | } |
e0c2b380 | 1506 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1507 | |
0b7c0153 | 1508 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1509 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1510 | int how, int flags) |
1da177e4 | 1511 | { |
07737691 | 1512 | struct rpc_task *task; |
9ace33cd | 1513 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1514 | struct rpc_message msg = { |
1515 | .rpc_argp = &data->args, | |
1516 | .rpc_resp = &data->res, | |
9ace33cd | 1517 | .rpc_cred = data->cred, |
bdc7f021 | 1518 | }; |
84115e1c | 1519 | struct rpc_task_setup task_setup_data = { |
07737691 | 1520 | .task = &data->task, |
9ace33cd | 1521 | .rpc_client = clnt, |
bdc7f021 | 1522 | .rpc_message = &msg, |
9ace33cd | 1523 | .callback_ops = call_ops, |
84115e1c | 1524 | .callback_data = data, |
101070ca | 1525 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1526 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1527 | .priority = priority, |
84115e1c | 1528 | }; |
9ace33cd FI |
1529 | /* Set up the initial task struct. */ |
1530 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1531 | ||
1532 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1533 | ||
8c21c62c WAA |
1534 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1535 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1536 | ||
9ace33cd FI |
1537 | task = rpc_run_task(&task_setup_data); |
1538 | if (IS_ERR(task)) | |
1539 | return PTR_ERR(task); | |
1540 | if (how & FLUSH_SYNC) | |
1541 | rpc_wait_for_completion_task(task); | |
1542 | rpc_put_task(task); | |
1543 | return 0; | |
1544 | } | |
e0c2b380 | 1545 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1546 | |
1547 | /* | |
1548 | * Set up the argument/result storage required for the RPC call. | |
1549 | */ | |
0b7c0153 | 1550 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1551 | struct list_head *head, |
1552 | struct pnfs_layout_segment *lseg, | |
1553 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1554 | { |
1555 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1556 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1557 | |
1558 | /* Set up the RPC argument and reply structs | |
1559 | * NB: take care not to mess about with data->commit et al. */ | |
1560 | ||
1561 | list_splice_init(head, &data->pages); | |
1da177e4 | 1562 | |
1da177e4 | 1563 | data->inode = inode; |
9ace33cd | 1564 | data->cred = first->wb_context->cred; |
988b6dce | 1565 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1566 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1567 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1568 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1569 | |
1570 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1571 | /* Note: we always request a commit of the entire inode */ |
1572 | data->args.offset = 0; | |
1573 | data->args.count = 0; | |
0b7c0153 | 1574 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1575 | data->res.fattr = &data->fattr; |
1576 | data->res.verf = &data->verf; | |
0e574af1 | 1577 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1578 | } |
e0c2b380 | 1579 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1580 | |
e0c2b380 | 1581 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1582 | struct pnfs_layout_segment *lseg, |
1583 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1584 | { |
1585 | struct nfs_page *req; | |
1586 | ||
1587 | while (!list_empty(page_list)) { | |
1588 | req = nfs_list_entry(page_list->next); | |
1589 | nfs_list_remove_request(req); | |
ea2cf228 | 1590 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1591 | if (!cinfo->dreq) { |
1592 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1593 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1594 | BDI_RECLAIMABLE); |
1595 | } | |
1d1afcbc | 1596 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1597 | } |
1598 | } | |
e0c2b380 | 1599 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1600 | |
1da177e4 LT |
1601 | /* |
1602 | * Commit dirty pages | |
1603 | */ | |
1604 | static int | |
ea2cf228 FI |
1605 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1606 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1607 | { |
0b7c0153 | 1608 | struct nfs_commit_data *data; |
1da177e4 | 1609 | |
c9d8f89d | 1610 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1611 | |
1612 | if (!data) | |
1613 | goto out_bad; | |
1614 | ||
1615 | /* Set up the argument struct */ | |
f453a54a FI |
1616 | nfs_init_commit(data, head, NULL, cinfo); |
1617 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1618 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1619 | how, 0); | |
1da177e4 | 1620 | out_bad: |
ea2cf228 | 1621 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1622 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1623 | return -ENOMEM; |
1624 | } | |
1625 | ||
1626 | /* | |
1627 | * COMMIT call returned | |
1628 | */ | |
788e7a89 | 1629 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1630 | { |
0b7c0153 | 1631 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1632 | |
a3f565b1 | 1633 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1634 | task->tk_pid, task->tk_status); |
1635 | ||
788e7a89 | 1636 | /* Call the NFS version-specific code */ |
c0d0e96b | 1637 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1638 | } |
1639 | ||
f453a54a | 1640 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1641 | { |
5917ce84 | 1642 | struct nfs_page *req; |
c9d8f89d | 1643 | int status = data->task.tk_status; |
f453a54a | 1644 | struct nfs_commit_info cinfo; |
788e7a89 | 1645 | |
1da177e4 LT |
1646 | while (!list_empty(&data->pages)) { |
1647 | req = nfs_list_entry(data->pages.next); | |
1648 | nfs_list_remove_request(req); | |
d6d6dc7c | 1649 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1650 | |
1e8968c5 | 1651 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1652 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1653 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1654 | req->wb_bytes, |
1655 | (long long)req_offset(req)); | |
c9d8f89d TM |
1656 | if (status < 0) { |
1657 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1658 | nfs_inode_remove_request(req); |
c9d8f89d | 1659 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1660 | goto next; |
1661 | } | |
1662 | ||
1663 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1664 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1665 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1666 | /* We have a match */ |
1667 | nfs_inode_remove_request(req); | |
1668 | dprintk(" OK\n"); | |
1669 | goto next; | |
1670 | } | |
1671 | /* We have a mismatch. Write the page again */ | |
1672 | dprintk(" mismatch\n"); | |
6d884e8f | 1673 | nfs_mark_request_dirty(req); |
05990d1b | 1674 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1675 | next: |
1d1afcbc | 1676 | nfs_unlock_and_release_request(req); |
1da177e4 | 1677 | } |
f453a54a FI |
1678 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1679 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1680 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1681 | } |
1682 | ||
1683 | static void nfs_commit_release(void *calldata) | |
1684 | { | |
0b7c0153 | 1685 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1686 | |
f453a54a | 1687 | data->completion_ops->completion(data); |
c9d8f89d | 1688 | nfs_commitdata_release(calldata); |
1da177e4 | 1689 | } |
788e7a89 TM |
1690 | |
1691 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1692 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1693 | .rpc_call_done = nfs_commit_done, |
1694 | .rpc_release = nfs_commit_release, | |
1695 | }; | |
1da177e4 | 1696 | |
f453a54a FI |
1697 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1698 | .completion = nfs_commit_release_pages, | |
1699 | .error_cleanup = nfs_commit_clear_lock, | |
1700 | }; | |
1701 | ||
1763da12 FI |
1702 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1703 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1704 | { |
1705 | int status; | |
1706 | ||
ea2cf228 | 1707 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1708 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1709 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1710 | return status; |
1711 | } | |
1712 | ||
b608b283 | 1713 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1714 | { |
1da177e4 | 1715 | LIST_HEAD(head); |
ea2cf228 | 1716 | struct nfs_commit_info cinfo; |
71d0a611 | 1717 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1718 | int res; |
1da177e4 | 1719 | |
b8413f98 TM |
1720 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1721 | if (res <= 0) | |
c5efa5fc | 1722 | goto out_mark_dirty; |
ea2cf228 FI |
1723 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1724 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1725 | if (res) { |
a861a1e1 FI |
1726 | int error; |
1727 | ||
ea2cf228 | 1728 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1729 | if (error < 0) |
1730 | return error; | |
b8413f98 | 1731 | if (!may_wait) |
c5efa5fc | 1732 | goto out_mark_dirty; |
b8413f98 TM |
1733 | error = wait_on_bit(&NFS_I(inode)->flags, |
1734 | NFS_INO_COMMIT, | |
1735 | nfs_wait_bit_killable, | |
1736 | TASK_KILLABLE); | |
1737 | if (error < 0) | |
1738 | return error; | |
71d0a611 TM |
1739 | } else |
1740 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1741 | return res; |
1742 | /* Note: If we exit without ensuring that the commit is complete, | |
1743 | * we must mark the inode as dirty. Otherwise, future calls to | |
1744 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1745 | * that the data is on the disk. | |
1746 | */ | |
1747 | out_mark_dirty: | |
1748 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1749 | return res; |
1750 | } | |
8fc795f7 TM |
1751 | |
1752 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1753 | { | |
420e3646 TM |
1754 | struct nfs_inode *nfsi = NFS_I(inode); |
1755 | int flags = FLUSH_SYNC; | |
1756 | int ret = 0; | |
8fc795f7 | 1757 | |
3236c3e1 | 1758 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1759 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1760 | return ret; |
1761 | ||
a00dd6c0 JL |
1762 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1763 | /* Don't commit yet if this is a non-blocking flush and there | |
1764 | * are a lot of outstanding writes for this mapping. | |
1765 | */ | |
ea2cf228 | 1766 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1767 | goto out_mark_dirty; |
420e3646 | 1768 | |
a00dd6c0 | 1769 | /* don't wait for the COMMIT response */ |
420e3646 | 1770 | flags = 0; |
a00dd6c0 JL |
1771 | } |
1772 | ||
420e3646 TM |
1773 | ret = nfs_commit_inode(inode, flags); |
1774 | if (ret >= 0) { | |
1775 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1776 | if (ret < wbc->nr_to_write) | |
1777 | wbc->nr_to_write -= ret; | |
1778 | else | |
1779 | wbc->nr_to_write = 0; | |
1780 | } | |
8fc795f7 | 1781 | return 0; |
420e3646 TM |
1782 | } |
1783 | out_mark_dirty: | |
8fc795f7 TM |
1784 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1785 | return ret; | |
1786 | } | |
c63c7b05 | 1787 | #else |
8fc795f7 TM |
1788 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1789 | { | |
1790 | return 0; | |
1791 | } | |
1da177e4 LT |
1792 | #endif |
1793 | ||
8fc795f7 TM |
1794 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1795 | { | |
a8d8f02c BS |
1796 | return nfs_commit_unstable_pages(inode, wbc); |
1797 | } | |
89d77c8f | 1798 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1799 | |
acdc53b2 TM |
1800 | /* |
1801 | * flush the inode to disk. | |
1802 | */ | |
1803 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1804 | { |
1805 | struct writeback_control wbc = { | |
72cb77f4 | 1806 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1807 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1808 | .range_start = 0, |
1809 | .range_end = LLONG_MAX, | |
34901f70 | 1810 | }; |
f4ce1299 TM |
1811 | int ret; |
1812 | ||
1813 | trace_nfs_writeback_inode_enter(inode); | |
1814 | ||
1815 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1816 | |
f4ce1299 TM |
1817 | trace_nfs_writeback_inode_exit(inode, ret); |
1818 | return ret; | |
1c75950b | 1819 | } |
ddda8e0a | 1820 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1821 | |
1b3b4a1a TM |
1822 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1823 | { | |
1824 | struct nfs_page *req; | |
1b3b4a1a TM |
1825 | int ret = 0; |
1826 | ||
1b3b4a1a | 1827 | for (;;) { |
ba8b06e6 | 1828 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1829 | req = nfs_page_find_request(page); |
1830 | if (req == NULL) | |
1b3b4a1a | 1831 | break; |
7ad84aa9 | 1832 | if (nfs_lock_request(req)) { |
8dd37758 | 1833 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1834 | nfs_inode_remove_request(req); |
1835 | /* | |
1836 | * In case nfs_inode_remove_request has marked the | |
1837 | * page as being dirty | |
1838 | */ | |
1839 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1d1afcbc | 1840 | nfs_unlock_and_release_request(req); |
1b3b4a1a TM |
1841 | break; |
1842 | } | |
1843 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1844 | nfs_release_request(req); |
1b3b4a1a | 1845 | if (ret < 0) |
c988950e | 1846 | break; |
1b3b4a1a | 1847 | } |
1b3b4a1a TM |
1848 | return ret; |
1849 | } | |
1850 | ||
7f2f12d9 TM |
1851 | /* |
1852 | * Write back all requests on one page - we do this before reading it. | |
1853 | */ | |
1854 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1855 | { |
29418aa4 | 1856 | loff_t range_start = page_file_offset(page); |
1c75950b | 1857 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1858 | struct writeback_control wbc = { |
4d770ccf | 1859 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1860 | .nr_to_write = 0, |
4d770ccf TM |
1861 | .range_start = range_start, |
1862 | .range_end = range_end, | |
1863 | }; | |
1864 | int ret; | |
1c75950b | 1865 | |
f4ce1299 TM |
1866 | trace_nfs_writeback_page_enter(inode); |
1867 | ||
0522f6ad | 1868 | for (;;) { |
ba8b06e6 | 1869 | wait_on_page_writeback(page); |
73e3302f TM |
1870 | if (clear_page_dirty_for_io(page)) { |
1871 | ret = nfs_writepage_locked(page, &wbc); | |
1872 | if (ret < 0) | |
1873 | goto out_error; | |
0522f6ad | 1874 | continue; |
7f2f12d9 | 1875 | } |
f4ce1299 | 1876 | ret = 0; |
0522f6ad TM |
1877 | if (!PagePrivate(page)) |
1878 | break; | |
1879 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1880 | if (ret < 0) |
73e3302f | 1881 | goto out_error; |
7f2f12d9 | 1882 | } |
73e3302f | 1883 | out_error: |
f4ce1299 | 1884 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1885 | return ret; |
1c75950b TM |
1886 | } |
1887 | ||
074cc1de TM |
1888 | #ifdef CONFIG_MIGRATION |
1889 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1890 | struct page *page, enum migrate_mode mode) |
074cc1de | 1891 | { |
2da95652 JL |
1892 | /* |
1893 | * If PagePrivate is set, then the page is currently associated with | |
1894 | * an in-progress read or write request. Don't try to migrate it. | |
1895 | * | |
1896 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1897 | * that we can safely release the inode reference while holding | |
1898 | * the page lock. | |
1899 | */ | |
1900 | if (PagePrivate(page)) | |
1901 | return -EBUSY; | |
074cc1de | 1902 | |
8c209ce7 DH |
1903 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1904 | return -EBUSY; | |
074cc1de | 1905 | |
a6bc32b8 | 1906 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1907 | } |
1908 | #endif | |
1909 | ||
f7b422b1 | 1910 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1911 | { |
1912 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1913 | sizeof(struct nfs_write_header), |
1da177e4 | 1914 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1915 | NULL); |
1da177e4 LT |
1916 | if (nfs_wdata_cachep == NULL) |
1917 | return -ENOMEM; | |
1918 | ||
93d2341c MD |
1919 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1920 | nfs_wdata_cachep); | |
1da177e4 | 1921 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1922 | goto out_destroy_write_cache; |
1da177e4 | 1923 | |
0b7c0153 FI |
1924 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1925 | sizeof(struct nfs_commit_data), | |
1926 | 0, SLAB_HWCACHE_ALIGN, | |
1927 | NULL); | |
1928 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1929 | goto out_destroy_write_mempool; |
0b7c0153 | 1930 | |
93d2341c | 1931 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1932 | nfs_cdata_cachep); |
1da177e4 | 1933 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1934 | goto out_destroy_commit_cache; |
1da177e4 | 1935 | |
89a09141 PZ |
1936 | /* |
1937 | * NFS congestion size, scale with available memory. | |
1938 | * | |
1939 | * 64MB: 8192k | |
1940 | * 128MB: 11585k | |
1941 | * 256MB: 16384k | |
1942 | * 512MB: 23170k | |
1943 | * 1GB: 32768k | |
1944 | * 2GB: 46340k | |
1945 | * 4GB: 65536k | |
1946 | * 8GB: 92681k | |
1947 | * 16GB: 131072k | |
1948 | * | |
1949 | * This allows larger machines to have larger/more transfers. | |
1950 | * Limit the default to 256M | |
1951 | */ | |
1952 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1953 | if (nfs_congestion_kb > 256*1024) | |
1954 | nfs_congestion_kb = 256*1024; | |
1955 | ||
1da177e4 | 1956 | return 0; |
3dd4765f JL |
1957 | |
1958 | out_destroy_commit_cache: | |
1959 | kmem_cache_destroy(nfs_cdata_cachep); | |
1960 | out_destroy_write_mempool: | |
1961 | mempool_destroy(nfs_wdata_mempool); | |
1962 | out_destroy_write_cache: | |
1963 | kmem_cache_destroy(nfs_wdata_cachep); | |
1964 | return -ENOMEM; | |
1da177e4 LT |
1965 | } |
1966 | ||
266bee88 | 1967 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1968 | { |
1969 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1970 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1971 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1972 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1973 | } |
1974 |