]>
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); |
788e7a89 | 45 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 46 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 47 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 48 | static const struct nfs_rw_ops nfs_rw_write_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 | |
4a0de55c | 73 | static struct nfs_rw_header *nfs_writehdr_alloc(void) |
3feb2d49 | 74 | { |
c0752cdf | 75 | struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 76 | |
4a0de55c | 77 | if (p) |
3feb2d49 | 78 | memset(p, 0, sizeof(*p)); |
3feb2d49 TM |
79 | return p; |
80 | } | |
6c75dc0d | 81 | |
4a0de55c | 82 | static void nfs_writehdr_free(struct nfs_rw_header *whdr) |
3feb2d49 | 83 | { |
cd841605 | 84 | mempool_free(whdr, nfs_wdata_mempool); |
3feb2d49 | 85 | } |
1da177e4 | 86 | |
7b159fc1 TM |
87 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
88 | { | |
89 | ctx->error = error; | |
90 | smp_wmb(); | |
91 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
92 | } | |
93 | ||
84d3a9a9 WAA |
94 | /* |
95 | * nfs_page_find_head_request_locked - find head request associated with @page | |
96 | * | |
97 | * must be called while holding the inode lock. | |
98 | * | |
99 | * returns matching head request with reference held, or NULL if not found. | |
100 | */ | |
29418aa4 | 101 | static struct nfs_page * |
84d3a9a9 | 102 | nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) |
277459d2 TM |
103 | { |
104 | struct nfs_page *req = NULL; | |
105 | ||
29418aa4 | 106 | if (PagePrivate(page)) |
277459d2 | 107 | req = (struct nfs_page *)page_private(page); |
29418aa4 MG |
108 | else if (unlikely(PageSwapCache(page))) { |
109 | struct nfs_page *freq, *t; | |
110 | ||
111 | /* Linearly search the commit list for the correct req */ | |
112 | list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { | |
113 | if (freq->wb_page == page) { | |
84d3a9a9 | 114 | req = freq->wb_head; |
29418aa4 MG |
115 | break; |
116 | } | |
117 | } | |
277459d2 | 118 | } |
29418aa4 | 119 | |
84d3a9a9 WAA |
120 | if (req) { |
121 | WARN_ON_ONCE(req->wb_head != req); | |
122 | ||
29418aa4 | 123 | kref_get(&req->wb_kref); |
84d3a9a9 | 124 | } |
29418aa4 | 125 | |
277459d2 TM |
126 | return req; |
127 | } | |
128 | ||
84d3a9a9 WAA |
129 | /* |
130 | * nfs_page_find_head_request - find head request associated with @page | |
131 | * | |
132 | * returns matching head request with reference held, or NULL if not found. | |
133 | */ | |
134 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 135 | { |
d56b4ddf | 136 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 137 | struct nfs_page *req = NULL; |
277459d2 | 138 | |
587142f8 | 139 | spin_lock(&inode->i_lock); |
84d3a9a9 | 140 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
587142f8 | 141 | spin_unlock(&inode->i_lock); |
277459d2 TM |
142 | return req; |
143 | } | |
144 | ||
1da177e4 LT |
145 | /* Adjust the file length if we're writing beyond the end */ |
146 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
147 | { | |
d56b4ddf | 148 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
149 | loff_t end, i_size; |
150 | pgoff_t end_index; | |
1da177e4 | 151 | |
a3d01454 TM |
152 | spin_lock(&inode->i_lock); |
153 | i_size = i_size_read(inode); | |
154 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 155 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 156 | goto out; |
d56b4ddf | 157 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 158 | if (i_size >= end) |
a3d01454 | 159 | goto out; |
1da177e4 | 160 | i_size_write(inode, end); |
a3d01454 TM |
161 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
162 | out: | |
163 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
164 | } |
165 | ||
a301b777 TM |
166 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
167 | static void nfs_set_pageerror(struct page *page) | |
168 | { | |
d56b4ddf | 169 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
170 | } |
171 | ||
d72ddcba WAA |
172 | /* |
173 | * nfs_page_group_search_locked | |
174 | * @head - head request of page group | |
175 | * @page_offset - offset into page | |
176 | * | |
177 | * Search page group with head @head to find a request that contains the | |
178 | * page offset @page_offset. | |
179 | * | |
180 | * Returns a pointer to the first matching nfs request, or NULL if no | |
181 | * match is found. | |
182 | * | |
183 | * Must be called with the page group lock held | |
184 | */ | |
185 | static struct nfs_page * | |
186 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
187 | { | |
188 | struct nfs_page *req; | |
189 | ||
190 | WARN_ON_ONCE(head != head->wb_head); | |
191 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); | |
192 | ||
193 | req = head; | |
194 | do { | |
195 | if (page_offset >= req->wb_pgbase && | |
196 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
197 | return req; | |
198 | ||
199 | req = req->wb_this_page; | |
200 | } while (req != head); | |
201 | ||
202 | return NULL; | |
203 | } | |
204 | ||
205 | /* | |
206 | * nfs_page_group_covers_page | |
207 | * @head - head request of page group | |
208 | * | |
209 | * Return true if the page group with head @head covers the whole page, | |
210 | * returns false otherwise | |
211 | */ | |
212 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
213 | { | |
214 | struct nfs_page *tmp; | |
215 | unsigned int pos = 0; | |
216 | unsigned int len = nfs_page_length(req->wb_page); | |
217 | ||
218 | nfs_page_group_lock(req); | |
219 | ||
220 | do { | |
221 | tmp = nfs_page_group_search_locked(req->wb_head, pos); | |
222 | if (tmp) { | |
223 | /* no way this should happen */ | |
224 | WARN_ON_ONCE(tmp->wb_pgbase != pos); | |
225 | pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); | |
226 | } | |
227 | } while (tmp && pos < len); | |
228 | ||
229 | nfs_page_group_unlock(req); | |
230 | WARN_ON_ONCE(pos > len); | |
231 | return pos == len; | |
232 | } | |
233 | ||
1da177e4 LT |
234 | /* We can set the PG_uptodate flag if we see that a write request |
235 | * covers the full page. | |
236 | */ | |
d72ddcba | 237 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 238 | { |
d72ddcba | 239 | if (PageUptodate(req->wb_page)) |
1da177e4 | 240 | return; |
d72ddcba | 241 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 242 | return; |
d72ddcba | 243 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
244 | } |
245 | ||
1da177e4 LT |
246 | static int wb_priority(struct writeback_control *wbc) |
247 | { | |
248 | if (wbc->for_reclaim) | |
c63c7b05 | 249 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 250 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
251 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
252 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
253 | } |
254 | ||
89a09141 PZ |
255 | /* |
256 | * NFS congestion control | |
257 | */ | |
258 | ||
259 | int nfs_congestion_kb; | |
260 | ||
261 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
262 | #define NFS_CONGESTION_OFF_THRESH \ | |
263 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
264 | ||
deed85e7 | 265 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 266 | { |
deed85e7 | 267 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
268 | int ret = test_set_page_writeback(page); |
269 | ||
deed85e7 | 270 | WARN_ON_ONCE(ret != 0); |
89a09141 | 271 | |
deed85e7 TM |
272 | if (atomic_long_inc_return(&nfss->writeback) > |
273 | NFS_CONGESTION_ON_THRESH) { | |
274 | set_bdi_congested(&nfss->backing_dev_info, | |
275 | BLK_RW_ASYNC); | |
89a09141 PZ |
276 | } |
277 | } | |
278 | ||
20633f04 | 279 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 280 | { |
20633f04 | 281 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 PZ |
282 | struct nfs_server *nfss = NFS_SERVER(inode); |
283 | ||
20633f04 WAA |
284 | if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) |
285 | return; | |
286 | ||
287 | end_page_writeback(req->wb_page); | |
c4dc4bee | 288 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 289 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
290 | } |
291 | ||
cfb506e1 | 292 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 293 | { |
d56b4ddf | 294 | struct inode *inode = page_file_mapping(page)->host; |
e261f51f | 295 | struct nfs_page *req; |
e261f51f TM |
296 | int ret; |
297 | ||
587142f8 | 298 | spin_lock(&inode->i_lock); |
074cc1de | 299 | for (;;) { |
84d3a9a9 | 300 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
074cc1de TM |
301 | if (req == NULL) |
302 | break; | |
7ad84aa9 | 303 | if (nfs_lock_request(req)) |
e261f51f TM |
304 | break; |
305 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 306 | * then the call to nfs_lock_request() will always |
e261f51f TM |
307 | * succeed provided that someone hasn't already marked the |
308 | * request as dirty (in which case we don't care). | |
309 | */ | |
587142f8 | 310 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
311 | if (!nonblock) |
312 | ret = nfs_wait_on_request(req); | |
313 | else | |
314 | ret = -EAGAIN; | |
e261f51f TM |
315 | nfs_release_request(req); |
316 | if (ret != 0) | |
074cc1de | 317 | return ERR_PTR(ret); |
587142f8 | 318 | spin_lock(&inode->i_lock); |
e261f51f | 319 | } |
587142f8 | 320 | spin_unlock(&inode->i_lock); |
074cc1de TM |
321 | return req; |
322 | } | |
323 | ||
324 | /* | |
325 | * Find an associated nfs write request, and prepare to flush it out | |
326 | * May return an error if the user signalled nfs_wait_on_request(). | |
327 | */ | |
328 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 329 | struct page *page, bool nonblock) |
074cc1de TM |
330 | { |
331 | struct nfs_page *req; | |
332 | int ret = 0; | |
333 | ||
cfb506e1 | 334 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
335 | if (!req) |
336 | goto out; | |
337 | ret = PTR_ERR(req); | |
338 | if (IS_ERR(req)) | |
339 | goto out; | |
340 | ||
deed85e7 TM |
341 | nfs_set_page_writeback(page); |
342 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 343 | |
deed85e7 | 344 | ret = 0; |
f8512ad0 FI |
345 | if (!nfs_pageio_add_request(pgio, req)) { |
346 | nfs_redirty_request(req); | |
074cc1de | 347 | ret = pgio->pg_error; |
f8512ad0 | 348 | } |
074cc1de TM |
349 | out: |
350 | return ret; | |
e261f51f TM |
351 | } |
352 | ||
f758c885 | 353 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 354 | { |
d56b4ddf | 355 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 356 | int ret; |
1da177e4 | 357 | |
91d5b470 CL |
358 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
359 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
360 | ||
d56b4ddf | 361 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 362 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
363 | if (ret == -EAGAIN) { |
364 | redirty_page_for_writepage(wbc, page); | |
365 | ret = 0; | |
366 | } | |
367 | return ret; | |
f758c885 | 368 | } |
7fe7f848 | 369 | |
f758c885 TM |
370 | /* |
371 | * Write an mmapped page to the server. | |
372 | */ | |
373 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
374 | { | |
375 | struct nfs_pageio_descriptor pgio; | |
376 | int err; | |
49a70f27 | 377 | |
a20c93e3 CH |
378 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
379 | false, &nfs_async_write_completion_ops); | |
f758c885 TM |
380 | err = nfs_do_writepage(page, wbc, &pgio); |
381 | nfs_pageio_complete(&pgio); | |
382 | if (err < 0) | |
383 | return err; | |
384 | if (pgio.pg_error < 0) | |
385 | return pgio.pg_error; | |
386 | return 0; | |
4d770ccf TM |
387 | } |
388 | ||
389 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
390 | { | |
f758c885 | 391 | int ret; |
4d770ccf | 392 | |
f758c885 | 393 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 394 | unlock_page(page); |
f758c885 TM |
395 | return ret; |
396 | } | |
397 | ||
398 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
399 | { | |
400 | int ret; | |
401 | ||
402 | ret = nfs_do_writepage(page, wbc, data); | |
403 | unlock_page(page); | |
404 | return ret; | |
1da177e4 LT |
405 | } |
406 | ||
1da177e4 LT |
407 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
408 | { | |
1da177e4 | 409 | struct inode *inode = mapping->host; |
72cb77f4 | 410 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 411 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
412 | int err; |
413 | ||
72cb77f4 TM |
414 | /* Stop dirtying of new pages while we sync */ |
415 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
416 | nfs_wait_bit_killable, TASK_KILLABLE); | |
417 | if (err) | |
418 | goto out_err; | |
419 | ||
91d5b470 CL |
420 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
421 | ||
a20c93e3 CH |
422 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
423 | &nfs_async_write_completion_ops); | |
f758c885 | 424 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 425 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
426 | |
427 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
4e857c58 | 428 | smp_mb__after_atomic(); |
72cb77f4 TM |
429 | wake_up_bit(bitlock, NFS_INO_FLUSHING); |
430 | ||
f758c885 | 431 | if (err < 0) |
72cb77f4 TM |
432 | goto out_err; |
433 | err = pgio.pg_error; | |
434 | if (err < 0) | |
435 | goto out_err; | |
c63c7b05 | 436 | return 0; |
72cb77f4 TM |
437 | out_err: |
438 | return err; | |
1da177e4 LT |
439 | } |
440 | ||
441 | /* | |
442 | * Insert a write request into an inode | |
443 | */ | |
d6d6dc7c | 444 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
445 | { |
446 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 | 447 | |
2bfc6e56 WAA |
448 | WARN_ON_ONCE(req->wb_this_page != req); |
449 | ||
e7d39069 | 450 | /* Lock the request! */ |
7ad84aa9 | 451 | nfs_lock_request(req); |
e7d39069 TM |
452 | |
453 | spin_lock(&inode->i_lock); | |
011e2a7f | 454 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 455 | inode->i_version++; |
29418aa4 MG |
456 | /* |
457 | * Swap-space should not get truncated. Hence no need to plug the race | |
458 | * with invalidate/truncate. | |
459 | */ | |
460 | if (likely(!PageSwapCache(req->wb_page))) { | |
461 | set_bit(PG_MAPPED, &req->wb_flags); | |
462 | SetPagePrivate(req->wb_page); | |
463 | set_page_private(req->wb_page, (unsigned long)req); | |
464 | } | |
1da177e4 | 465 | nfsi->npages++; |
17089a29 WAA |
466 | /* this a head request for a page group - mark it as having an |
467 | * extra reference so sub groups can follow suit */ | |
468 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); | |
c03b4024 | 469 | kref_get(&req->wb_kref); |
e7d39069 | 470 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
89a09141 | 474 | * Remove a write request from an inode |
1da177e4 LT |
475 | */ |
476 | static void nfs_inode_remove_request(struct nfs_page *req) | |
477 | { | |
3d4ff43d | 478 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 | 479 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 480 | struct nfs_page *head; |
1da177e4 | 481 | |
20633f04 WAA |
482 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
483 | head = req->wb_head; | |
484 | ||
485 | spin_lock(&inode->i_lock); | |
486 | if (likely(!PageSwapCache(head->wb_page))) { | |
487 | set_page_private(head->wb_page, 0); | |
488 | ClearPagePrivate(head->wb_page); | |
489 | clear_bit(PG_MAPPED, &head->wb_flags); | |
490 | } | |
491 | nfsi->npages--; | |
492 | spin_unlock(&inode->i_lock); | |
29418aa4 | 493 | } |
17089a29 WAA |
494 | |
495 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
496 | nfs_release_request(req); | |
1da177e4 LT |
497 | } |
498 | ||
61822ab5 | 499 | static void |
6d884e8f | 500 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 501 | { |
61822ab5 TM |
502 | __set_page_dirty_nobuffers(req->wb_page); |
503 | } | |
504 | ||
89d77c8f | 505 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
8dd37758 TM |
506 | /** |
507 | * nfs_request_add_commit_list - add request to a commit list | |
508 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
509 | * @dst: commit list head |
510 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 511 | * |
ea2cf228 | 512 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
513 | * number of outstanding requests requiring a commit as well as |
514 | * the MM page stats. | |
515 | * | |
ea2cf228 | 516 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 517 | * holding the nfs_page lock. |
1da177e4 | 518 | */ |
8dd37758 | 519 | void |
ea2cf228 FI |
520 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
521 | struct nfs_commit_info *cinfo) | |
1da177e4 | 522 | { |
e468bae9 | 523 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
524 | spin_lock(cinfo->lock); |
525 | nfs_list_add_request(req, dst); | |
526 | cinfo->mds->ncommit++; | |
527 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
528 | if (!cinfo->dreq) { |
529 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 530 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
531 | BDI_RECLAIMABLE); |
532 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
533 | I_DIRTY_DATASYNC); | |
534 | } | |
1da177e4 | 535 | } |
8dd37758 TM |
536 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
537 | ||
538 | /** | |
539 | * nfs_request_remove_commit_list - Remove request from a commit list | |
540 | * @req: pointer to a nfs_page | |
ea2cf228 | 541 | * @cinfo: holds list lock and accounting info |
8dd37758 | 542 | * |
ea2cf228 | 543 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
544 | * number of outstanding requests requiring a commit |
545 | * It does not update the MM page stats. | |
546 | * | |
ea2cf228 | 547 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
548 | */ |
549 | void | |
ea2cf228 FI |
550 | nfs_request_remove_commit_list(struct nfs_page *req, |
551 | struct nfs_commit_info *cinfo) | |
8dd37758 | 552 | { |
8dd37758 TM |
553 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
554 | return; | |
555 | nfs_list_remove_request(req); | |
ea2cf228 | 556 | cinfo->mds->ncommit--; |
8dd37758 TM |
557 | } |
558 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
559 | ||
ea2cf228 FI |
560 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
561 | struct inode *inode) | |
562 | { | |
563 | cinfo->lock = &inode->i_lock; | |
564 | cinfo->mds = &NFS_I(inode)->commit_info; | |
565 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 566 | cinfo->dreq = NULL; |
f453a54a | 567 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
568 | } |
569 | ||
570 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
571 | struct inode *inode, | |
572 | struct nfs_direct_req *dreq) | |
573 | { | |
1763da12 FI |
574 | if (dreq) |
575 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
576 | else | |
577 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
578 | } |
579 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
580 | |
581 | /* | |
582 | * Add a request to the inode's commit list. | |
583 | */ | |
1763da12 | 584 | void |
ea2cf228 FI |
585 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
586 | struct nfs_commit_info *cinfo) | |
8dd37758 | 587 | { |
ea2cf228 | 588 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 589 | return; |
ea2cf228 | 590 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 591 | } |
8e821cad | 592 | |
d6d6dc7c FI |
593 | static void |
594 | nfs_clear_page_commit(struct page *page) | |
595 | { | |
596 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 597 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
598 | } |
599 | ||
8dd37758 | 600 | static void |
e468bae9 TM |
601 | nfs_clear_request_commit(struct nfs_page *req) |
602 | { | |
8dd37758 TM |
603 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
604 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 605 | struct nfs_commit_info cinfo; |
e468bae9 | 606 | |
ea2cf228 FI |
607 | nfs_init_cinfo_from_inode(&cinfo, inode); |
608 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
609 | spin_lock(cinfo.lock); | |
610 | nfs_request_remove_commit_list(req, &cinfo); | |
611 | spin_unlock(cinfo.lock); | |
8dd37758 | 612 | } |
d6d6dc7c | 613 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 614 | } |
e468bae9 TM |
615 | } |
616 | ||
8e821cad | 617 | static inline |
9c7e1b3d | 618 | int nfs_write_need_commit(struct nfs_pgio_data *data) |
8e821cad | 619 | { |
465d5243 | 620 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
621 | return data->header->lseg == NULL; |
622 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
623 | } |
624 | ||
8e821cad | 625 | #else |
68cd6fa4 BS |
626 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
627 | struct inode *inode) | |
628 | { | |
629 | } | |
630 | ||
631 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
632 | struct inode *inode, | |
633 | struct nfs_direct_req *dreq) | |
634 | { | |
635 | } | |
636 | ||
1763da12 | 637 | void |
ea2cf228 FI |
638 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
639 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
640 | { |
641 | } | |
642 | ||
8dd37758 | 643 | static void |
e468bae9 TM |
644 | nfs_clear_request_commit(struct nfs_page *req) |
645 | { | |
e468bae9 TM |
646 | } |
647 | ||
8e821cad | 648 | static inline |
9c7e1b3d | 649 | int nfs_write_need_commit(struct nfs_pgio_data *data) |
8e821cad TM |
650 | { |
651 | return 0; | |
652 | } | |
653 | ||
6c75dc0d FI |
654 | #endif |
655 | ||
061ae2ed | 656 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 657 | { |
ea2cf228 | 658 | struct nfs_commit_info cinfo; |
6c75dc0d | 659 | unsigned long bytes = 0; |
2bfc6e56 | 660 | bool do_destroy; |
6c75dc0d FI |
661 | |
662 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
663 | goto out; | |
ea2cf228 | 664 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
665 | while (!list_empty(&hdr->pages)) { |
666 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
667 | |
668 | bytes += req->wb_bytes; | |
669 | nfs_list_remove_request(req); | |
670 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
671 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 672 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
673 | nfs_context_set_write_error(req->wb_context, hdr->error); |
674 | goto remove_req; | |
675 | } | |
676 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
677 | nfs_mark_request_dirty(req); | |
678 | goto next; | |
679 | } | |
680 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
f79d06f5 | 681 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
ea2cf228 | 682 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
683 | goto next; |
684 | } | |
685 | remove_req: | |
686 | nfs_inode_remove_request(req); | |
687 | next: | |
1d1afcbc | 688 | nfs_unlock_request(req); |
20633f04 | 689 | nfs_end_page_writeback(req); |
2bfc6e56 | 690 | do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags); |
3aff4ebb | 691 | nfs_release_request(req); |
6c75dc0d FI |
692 | } |
693 | out: | |
694 | hdr->release(hdr); | |
8e821cad | 695 | } |
1da177e4 | 696 | |
89d77c8f | 697 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
ce59515c | 698 | unsigned long |
ea2cf228 | 699 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 700 | { |
ea2cf228 | 701 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
702 | } |
703 | ||
ea2cf228 | 704 | /* cinfo->lock held by caller */ |
1763da12 | 705 | int |
ea2cf228 FI |
706 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
707 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
708 | { |
709 | struct nfs_page *req, *tmp; | |
710 | int ret = 0; | |
711 | ||
712 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
713 | if (!nfs_lock_request(req)) |
714 | continue; | |
7ad84aa9 | 715 | kref_get(&req->wb_kref); |
ea2cf228 | 716 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 717 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 718 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
719 | nfs_list_add_request(req, dst); |
720 | ret++; | |
1763da12 | 721 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 722 | break; |
d6d6dc7c FI |
723 | } |
724 | return ret; | |
fb8a1f11 TM |
725 | } |
726 | ||
1da177e4 LT |
727 | /* |
728 | * nfs_scan_commit - Scan an inode for commit requests | |
729 | * @inode: NFS inode to scan | |
ea2cf228 FI |
730 | * @dst: mds destination list |
731 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
732 | * |
733 | * Moves requests from the inode's 'commit' request list. | |
734 | * The requests are *not* checked to ensure that they form a contiguous set. | |
735 | */ | |
1763da12 | 736 | int |
ea2cf228 FI |
737 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
738 | struct nfs_commit_info *cinfo) | |
1da177e4 | 739 | { |
d6d6dc7c | 740 | int ret = 0; |
fb8a1f11 | 741 | |
ea2cf228 FI |
742 | spin_lock(cinfo->lock); |
743 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 744 | const int max = INT_MAX; |
d6d6dc7c | 745 | |
ea2cf228 FI |
746 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
747 | cinfo, max); | |
748 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 749 | } |
ea2cf228 | 750 | spin_unlock(cinfo->lock); |
ff778d02 | 751 | return ret; |
1da177e4 | 752 | } |
d6d6dc7c | 753 | |
c42de9dd | 754 | #else |
ce59515c | 755 | unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
756 | { |
757 | return 0; | |
758 | } | |
759 | ||
1763da12 FI |
760 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
761 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
762 | { |
763 | return 0; | |
764 | } | |
1da177e4 LT |
765 | #endif |
766 | ||
1da177e4 | 767 | /* |
e7d39069 TM |
768 | * Search for an existing write request, and attempt to update |
769 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 770 | * |
e7d39069 TM |
771 | * If the attempt fails, then the existing request is flushed out |
772 | * to disk. | |
1da177e4 | 773 | */ |
e7d39069 TM |
774 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
775 | struct page *page, | |
776 | unsigned int offset, | |
777 | unsigned int bytes) | |
1da177e4 | 778 | { |
e7d39069 TM |
779 | struct nfs_page *req; |
780 | unsigned int rqend; | |
781 | unsigned int end; | |
782 | int error; | |
783 | ||
784 | if (!PagePrivate(page)) | |
785 | return NULL; | |
1da177e4 LT |
786 | |
787 | end = offset + bytes; | |
e7d39069 | 788 | spin_lock(&inode->i_lock); |
1da177e4 | 789 | |
1da177e4 | 790 | for (;;) { |
84d3a9a9 | 791 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
e7d39069 TM |
792 | if (req == NULL) |
793 | goto out_unlock; | |
794 | ||
2bfc6e56 WAA |
795 | /* should be handled by nfs_flush_incompatible */ |
796 | WARN_ON_ONCE(req->wb_head != req); | |
797 | WARN_ON_ONCE(req->wb_this_page != req); | |
798 | ||
e7d39069 TM |
799 | rqend = req->wb_offset + req->wb_bytes; |
800 | /* | |
801 | * Tell the caller to flush out the request if | |
802 | * the offsets are non-contiguous. | |
803 | * Note: nfs_flush_incompatible() will already | |
804 | * have flushed out requests having wrong owners. | |
805 | */ | |
e468bae9 | 806 | if (offset > rqend |
e7d39069 TM |
807 | || end < req->wb_offset) |
808 | goto out_flushme; | |
809 | ||
7ad84aa9 | 810 | if (nfs_lock_request(req)) |
1da177e4 | 811 | break; |
1da177e4 | 812 | |
e7d39069 | 813 | /* The request is locked, so wait and then retry */ |
587142f8 | 814 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
815 | error = nfs_wait_on_request(req); |
816 | nfs_release_request(req); | |
817 | if (error != 0) | |
818 | goto out_err; | |
819 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
820 | } |
821 | ||
822 | /* Okay, the request matches. Update the region */ | |
823 | if (offset < req->wb_offset) { | |
824 | req->wb_offset = offset; | |
825 | req->wb_pgbase = offset; | |
1da177e4 | 826 | } |
1da177e4 LT |
827 | if (end > rqend) |
828 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
829 | else |
830 | req->wb_bytes = rqend - req->wb_offset; | |
831 | out_unlock: | |
832 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
833 | if (req) |
834 | nfs_clear_request_commit(req); | |
e7d39069 TM |
835 | return req; |
836 | out_flushme: | |
837 | spin_unlock(&inode->i_lock); | |
838 | nfs_release_request(req); | |
839 | error = nfs_wb_page(inode, page); | |
840 | out_err: | |
841 | return ERR_PTR(error); | |
842 | } | |
843 | ||
844 | /* | |
845 | * Try to update an existing write request, or create one if there is none. | |
846 | * | |
847 | * Note: Should always be called with the Page Lock held to prevent races | |
848 | * if we have to add a new request. Also assumes that the caller has | |
849 | * already called nfs_flush_incompatible() if necessary. | |
850 | */ | |
851 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
852 | struct page *page, unsigned int offset, unsigned int bytes) | |
853 | { | |
d56b4ddf | 854 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 855 | struct nfs_page *req; |
1da177e4 | 856 | |
e7d39069 TM |
857 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
858 | if (req != NULL) | |
859 | goto out; | |
2bfc6e56 | 860 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
861 | if (IS_ERR(req)) |
862 | goto out; | |
d6d6dc7c | 863 | nfs_inode_add_request(inode, req); |
efc91ed0 | 864 | out: |
61e930a9 | 865 | return req; |
1da177e4 LT |
866 | } |
867 | ||
e7d39069 TM |
868 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
869 | unsigned int offset, unsigned int count) | |
870 | { | |
871 | struct nfs_page *req; | |
872 | ||
873 | req = nfs_setup_write_request(ctx, page, offset, count); | |
874 | if (IS_ERR(req)) | |
875 | return PTR_ERR(req); | |
876 | /* Update file length */ | |
877 | nfs_grow_file(page, offset, count); | |
d72ddcba | 878 | nfs_mark_uptodate(req); |
a6305ddb | 879 | nfs_mark_request_dirty(req); |
1d1afcbc | 880 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
881 | return 0; |
882 | } | |
883 | ||
1da177e4 LT |
884 | int nfs_flush_incompatible(struct file *file, struct page *page) |
885 | { | |
cd3758e3 | 886 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 887 | struct nfs_lock_context *l_ctx; |
1da177e4 | 888 | struct nfs_page *req; |
1a54533e | 889 | int do_flush, status; |
1da177e4 LT |
890 | /* |
891 | * Look for a request corresponding to this page. If there | |
892 | * is one, and it belongs to another file, we flush it out | |
893 | * before we try to copy anything into the page. Do this | |
894 | * due to the lack of an ACCESS-type call in NFSv2. | |
895 | * Also do the same if we find a request from an existing | |
896 | * dropped page. | |
897 | */ | |
1a54533e | 898 | do { |
84d3a9a9 | 899 | req = nfs_page_find_head_request(page); |
1a54533e TM |
900 | if (req == NULL) |
901 | return 0; | |
2a369153 TM |
902 | l_ctx = req->wb_lock_context; |
903 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
2bfc6e56 WAA |
904 | /* for now, flush if more than 1 request in page_group */ |
905 | do_flush |= req->wb_this_page != req; | |
0f1d2605 | 906 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
907 | do_flush |= l_ctx->lockowner.l_owner != current->files |
908 | || l_ctx->lockowner.l_pid != current->tgid; | |
909 | } | |
1da177e4 | 910 | nfs_release_request(req); |
1a54533e TM |
911 | if (!do_flush) |
912 | return 0; | |
d56b4ddf | 913 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
914 | } while (status == 0); |
915 | return status; | |
1da177e4 LT |
916 | } |
917 | ||
dc24826b AA |
918 | /* |
919 | * Avoid buffered writes when a open context credential's key would | |
920 | * expire soon. | |
921 | * | |
922 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
923 | * | |
924 | * Return 0 and set a credential flag which triggers the inode to flush | |
925 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
926 | * RPC_KEY_EXPIRE_TIMEO. | |
927 | */ | |
928 | int | |
929 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
930 | { | |
931 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
932 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
933 | ||
934 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
935 | } | |
936 | ||
937 | /* | |
938 | * Test if the open context credential key is marked to expire soon. | |
939 | */ | |
940 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
941 | { | |
942 | return rpcauth_cred_key_to_expire(ctx->cred); | |
943 | } | |
944 | ||
5d47a356 TM |
945 | /* |
946 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
947 | * the PageUptodate() flag. In this case, we will need to turn off | |
948 | * write optimisations that depend on the page contents being correct. | |
949 | */ | |
8d197a56 | 950 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 951 | { |
d529ef83 JL |
952 | struct nfs_inode *nfsi = NFS_I(inode); |
953 | ||
8d197a56 TM |
954 | if (nfs_have_delegated_attributes(inode)) |
955 | goto out; | |
18dd78c4 | 956 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 957 | return false; |
4db72b40 | 958 | smp_rmb(); |
d529ef83 | 959 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
960 | return false; |
961 | out: | |
18dd78c4 SM |
962 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
963 | return false; | |
8d197a56 | 964 | return PageUptodate(page) != 0; |
5d47a356 TM |
965 | } |
966 | ||
c7559663 SM |
967 | /* If we know the page is up to date, and we're not using byte range locks (or |
968 | * if we have the whole file locked for writing), it may be more efficient to | |
969 | * extend the write to cover the entire page in order to avoid fragmentation | |
970 | * inefficiencies. | |
971 | * | |
263b4509 SM |
972 | * If the file is opened for synchronous writes then we can just skip the rest |
973 | * of the checks. | |
c7559663 SM |
974 | */ |
975 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
976 | { | |
977 | if (file->f_flags & O_DSYNC) | |
978 | return 0; | |
263b4509 SM |
979 | if (!nfs_write_pageuptodate(page, inode)) |
980 | return 0; | |
c7559663 SM |
981 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
982 | return 1; | |
263b4509 | 983 | if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && |
c7559663 | 984 | inode->i_flock->fl_end == OFFSET_MAX && |
263b4509 | 985 | inode->i_flock->fl_type != F_RDLCK)) |
c7559663 SM |
986 | return 1; |
987 | return 0; | |
988 | } | |
989 | ||
1da177e4 LT |
990 | /* |
991 | * Update and possibly write a cached page of an NFS file. | |
992 | * | |
993 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
994 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
995 | */ | |
996 | int nfs_updatepage(struct file *file, struct page *page, | |
997 | unsigned int offset, unsigned int count) | |
998 | { | |
cd3758e3 | 999 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1000 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1001 | int status = 0; |
1002 | ||
91d5b470 CL |
1003 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1004 | ||
6de1472f AV |
1005 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1006 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1007 | |
c7559663 | 1008 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1009 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1010 | offset = 0; |
1da177e4 LT |
1011 | } |
1012 | ||
e21195a7 | 1013 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1014 | if (status < 0) |
1015 | nfs_set_pageerror(page); | |
59b7c05f TM |
1016 | else |
1017 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 1018 | |
48186c7d | 1019 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1020 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1021 | return status; |
1022 | } | |
1023 | ||
3ff7576d | 1024 | static int flush_task_priority(int how) |
1da177e4 LT |
1025 | { |
1026 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1027 | case FLUSH_HIGHPRI: | |
1028 | return RPC_PRIORITY_HIGH; | |
1029 | case FLUSH_LOWPRI: | |
1030 | return RPC_PRIORITY_LOW; | |
1031 | } | |
1032 | return RPC_PRIORITY_NORMAL; | |
1033 | } | |
1034 | ||
1ed26f33 AS |
1035 | static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg, |
1036 | struct rpc_task_setup *task_setup_data, int how) | |
1da177e4 | 1037 | { |
cd841605 | 1038 | struct inode *inode = data->header->inode; |
3ff7576d | 1039 | int priority = flush_task_priority(how); |
d138d5d1 | 1040 | |
1ed26f33 AS |
1041 | task_setup_data->priority = priority; |
1042 | NFS_PROTO(inode)->write_setup(data, msg); | |
d138d5d1 | 1043 | |
8c21c62c | 1044 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
1ed26f33 | 1045 | &task_setup_data->rpc_client, msg, data); |
275acaaf TM |
1046 | } |
1047 | ||
6d884e8f F |
1048 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1049 | * call this on each, which will prepare them to be retried on next | |
1050 | * writeback using standard nfs. | |
1051 | */ | |
1052 | static void nfs_redirty_request(struct nfs_page *req) | |
1053 | { | |
1054 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1055 | nfs_unlock_request(req); |
20633f04 | 1056 | nfs_end_page_writeback(req); |
3aff4ebb | 1057 | nfs_release_request(req); |
6d884e8f F |
1058 | } |
1059 | ||
061ae2ed | 1060 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1061 | { |
1062 | struct nfs_page *req; | |
1063 | ||
1064 | while (!list_empty(head)) { | |
1065 | req = nfs_list_entry(head->next); | |
1066 | nfs_list_remove_request(req); | |
1067 | nfs_redirty_request(req); | |
1068 | } | |
1069 | } | |
1070 | ||
061ae2ed FI |
1071 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1072 | .error_cleanup = nfs_async_write_error, | |
1073 | .completion = nfs_write_completion, | |
1074 | }; | |
1075 | ||
57208fa7 | 1076 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1077 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1078 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1079 | { |
a20c93e3 | 1080 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1081 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1082 | |
1083 | #ifdef CONFIG_NFS_V4_1 | |
1084 | if (server->pnfs_curr_ld && !force_mds) | |
1085 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1086 | #endif | |
4a0de55c AS |
1087 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
1088 | server->wsize, ioflags); | |
1751c363 | 1089 | } |
ddda8e0a | 1090 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1091 | |
dce81290 TM |
1092 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1093 | { | |
41d8d5b7 | 1094 | pgio->pg_ops = &nfs_pgio_rw_ops; |
dce81290 TM |
1095 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; |
1096 | } | |
1f945357 | 1097 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1098 | |
1da177e4 | 1099 | |
0b7c0153 FI |
1100 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1101 | { | |
1102 | struct nfs_commit_data *data = calldata; | |
1103 | ||
1104 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1105 | } | |
1106 | ||
a4cdda59 | 1107 | static void nfs_writeback_release_common(struct nfs_pgio_data *data) |
c9d8f89d | 1108 | { |
cd841605 | 1109 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1110 | int status = data->task.tk_status; |
788e7a89 | 1111 | |
6c75dc0d FI |
1112 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1113 | spin_lock(&hdr->lock); | |
1114 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1115 | ; /* Do nothing */ | |
1116 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
f79d06f5 AS |
1117 | memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); |
1118 | else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) | |
6c75dc0d FI |
1119 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); |
1120 | spin_unlock(&hdr->lock); | |
1da177e4 LT |
1121 | } |
1122 | } | |
1123 | ||
1f2edbe3 TM |
1124 | /* |
1125 | * Special version of should_remove_suid() that ignores capabilities. | |
1126 | */ | |
1127 | static int nfs_should_remove_suid(const struct inode *inode) | |
1128 | { | |
1129 | umode_t mode = inode->i_mode; | |
1130 | int kill = 0; | |
1131 | ||
1132 | /* suid always must be killed */ | |
1133 | if (unlikely(mode & S_ISUID)) | |
1134 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1135 | |
1f2edbe3 TM |
1136 | /* |
1137 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1138 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1139 | */ | |
1140 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1141 | kill |= ATTR_KILL_SGID; | |
1142 | ||
1143 | if (unlikely(kill && S_ISREG(mode))) | |
1144 | return kill; | |
1145 | ||
1146 | return 0; | |
1147 | } | |
788e7a89 | 1148 | |
1da177e4 LT |
1149 | /* |
1150 | * This function is called when the WRITE call is complete. | |
1151 | */ | |
0eecb214 AS |
1152 | static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data, |
1153 | struct inode *inode) | |
1da177e4 | 1154 | { |
788e7a89 | 1155 | int status; |
1da177e4 | 1156 | |
f551e44f CL |
1157 | /* |
1158 | * ->write_done will attempt to use post-op attributes to detect | |
1159 | * conflicting writes by other clients. A strict interpretation | |
1160 | * of close-to-open would allow us to continue caching even if | |
1161 | * another writer had changed the file, but some applications | |
1162 | * depend on tighter cache coherency when writing. | |
1163 | */ | |
cd841605 | 1164 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1165 | if (status != 0) |
0eecb214 AS |
1166 | return status; |
1167 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count); | |
91d5b470 | 1168 | |
89d77c8f | 1169 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
0eecb214 | 1170 | if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) { |
1da177e4 LT |
1171 | /* We tried a write call, but the server did not |
1172 | * commit data to stable storage even though we | |
1173 | * requested it. | |
1174 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1175 | * the server reports NFS_DATA_SYNC, but performs | |
1176 | * NFS_FILE_SYNC. We therefore implement this checking | |
1177 | * as a dprintk() in order to avoid filling syslog. | |
1178 | */ | |
1179 | static unsigned long complain; | |
1180 | ||
a69aef14 | 1181 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1182 | if (time_before(complain, jiffies)) { |
48186c7d | 1183 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1184 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1185 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
0eecb214 | 1186 | data->res.verf->committed, data->args.stable); |
1da177e4 LT |
1187 | complain = jiffies + 300 * HZ; |
1188 | } | |
1189 | } | |
1190 | #endif | |
1f2edbe3 TM |
1191 | |
1192 | /* Deal with the suid/sgid bit corner case */ | |
1193 | if (nfs_should_remove_suid(inode)) | |
1194 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1195 | return 0; |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * This function is called when the WRITE call is complete. | |
1200 | */ | |
1201 | static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data) | |
1202 | { | |
1203 | struct nfs_pgio_args *argp = &data->args; | |
1204 | struct nfs_pgio_res *resp = &data->res; | |
1f2edbe3 TM |
1205 | |
1206 | if (resp->count < argp->count) { | |
1da177e4 LT |
1207 | static unsigned long complain; |
1208 | ||
6c75dc0d | 1209 | /* This a short write! */ |
0eecb214 | 1210 | nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1211 | |
1da177e4 | 1212 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1213 | if (resp->count == 0) { |
1214 | if (time_before(complain, jiffies)) { | |
1215 | printk(KERN_WARNING | |
1216 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1217 | argp->count); | |
1218 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1219 | } |
6c75dc0d FI |
1220 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1221 | task->tk_status = -EIO; | |
13602896 | 1222 | return; |
1da177e4 | 1223 | } |
6c75dc0d FI |
1224 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1225 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1226 | /* Resend from where the server left off */ | |
1227 | data->mds_offset += resp->count; | |
1228 | argp->offset += resp->count; | |
1229 | argp->pgbase += resp->count; | |
1230 | argp->count -= resp->count; | |
1231 | } else { | |
1232 | /* Resend as a stable write in order to avoid | |
1233 | * headaches in the case of a server crash. | |
1234 | */ | |
1235 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1236 | } |
6c75dc0d | 1237 | rpc_restart_call_prepare(task); |
1da177e4 | 1238 | } |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | ||
89d77c8f | 1242 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
71d0a611 TM |
1243 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1244 | { | |
b8413f98 TM |
1245 | int ret; |
1246 | ||
71d0a611 TM |
1247 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1248 | return 1; | |
b8413f98 TM |
1249 | if (!may_wait) |
1250 | return 0; | |
1251 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1252 | NFS_INO_COMMIT, | |
1253 | nfs_wait_bit_killable, | |
1254 | TASK_KILLABLE); | |
1255 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1256 | } |
1257 | ||
f453a54a | 1258 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1259 | { |
1260 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
4e857c58 | 1261 | smp_mb__after_atomic(); |
71d0a611 TM |
1262 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); |
1263 | } | |
1264 | ||
0b7c0153 | 1265 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1266 | { |
0b7c0153 FI |
1267 | put_nfs_open_context(data->context); |
1268 | nfs_commit_free(data); | |
1da177e4 | 1269 | } |
e0c2b380 | 1270 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1271 | |
0b7c0153 | 1272 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1273 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1274 | int how, int flags) |
1da177e4 | 1275 | { |
07737691 | 1276 | struct rpc_task *task; |
9ace33cd | 1277 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1278 | struct rpc_message msg = { |
1279 | .rpc_argp = &data->args, | |
1280 | .rpc_resp = &data->res, | |
9ace33cd | 1281 | .rpc_cred = data->cred, |
bdc7f021 | 1282 | }; |
84115e1c | 1283 | struct rpc_task_setup task_setup_data = { |
07737691 | 1284 | .task = &data->task, |
9ace33cd | 1285 | .rpc_client = clnt, |
bdc7f021 | 1286 | .rpc_message = &msg, |
9ace33cd | 1287 | .callback_ops = call_ops, |
84115e1c | 1288 | .callback_data = data, |
101070ca | 1289 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1290 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1291 | .priority = priority, |
84115e1c | 1292 | }; |
9ace33cd FI |
1293 | /* Set up the initial task struct. */ |
1294 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1295 | ||
1296 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1297 | ||
8c21c62c WAA |
1298 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1299 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1300 | ||
9ace33cd FI |
1301 | task = rpc_run_task(&task_setup_data); |
1302 | if (IS_ERR(task)) | |
1303 | return PTR_ERR(task); | |
1304 | if (how & FLUSH_SYNC) | |
1305 | rpc_wait_for_completion_task(task); | |
1306 | rpc_put_task(task); | |
1307 | return 0; | |
1308 | } | |
e0c2b380 | 1309 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1310 | |
1311 | /* | |
1312 | * Set up the argument/result storage required for the RPC call. | |
1313 | */ | |
0b7c0153 | 1314 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1315 | struct list_head *head, |
1316 | struct pnfs_layout_segment *lseg, | |
1317 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1318 | { |
1319 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1320 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1321 | |
1322 | /* Set up the RPC argument and reply structs | |
1323 | * NB: take care not to mess about with data->commit et al. */ | |
1324 | ||
1325 | list_splice_init(head, &data->pages); | |
1da177e4 | 1326 | |
1da177e4 | 1327 | data->inode = inode; |
9ace33cd | 1328 | data->cred = first->wb_context->cred; |
988b6dce | 1329 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1330 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1331 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1332 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1333 | |
1334 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1335 | /* Note: we always request a commit of the entire inode */ |
1336 | data->args.offset = 0; | |
1337 | data->args.count = 0; | |
0b7c0153 | 1338 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1339 | data->res.fattr = &data->fattr; |
1340 | data->res.verf = &data->verf; | |
0e574af1 | 1341 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1342 | } |
e0c2b380 | 1343 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1344 | |
e0c2b380 | 1345 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1346 | struct pnfs_layout_segment *lseg, |
1347 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1348 | { |
1349 | struct nfs_page *req; | |
1350 | ||
1351 | while (!list_empty(page_list)) { | |
1352 | req = nfs_list_entry(page_list->next); | |
1353 | nfs_list_remove_request(req); | |
ea2cf228 | 1354 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1355 | if (!cinfo->dreq) { |
1356 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1357 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1358 | BDI_RECLAIMABLE); |
1359 | } | |
1d1afcbc | 1360 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1361 | } |
1362 | } | |
e0c2b380 | 1363 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1364 | |
1da177e4 LT |
1365 | /* |
1366 | * Commit dirty pages | |
1367 | */ | |
1368 | static int | |
ea2cf228 FI |
1369 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1370 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1371 | { |
0b7c0153 | 1372 | struct nfs_commit_data *data; |
1da177e4 | 1373 | |
c9d8f89d | 1374 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1375 | |
1376 | if (!data) | |
1377 | goto out_bad; | |
1378 | ||
1379 | /* Set up the argument struct */ | |
f453a54a FI |
1380 | nfs_init_commit(data, head, NULL, cinfo); |
1381 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1382 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1383 | how, 0); | |
1da177e4 | 1384 | out_bad: |
ea2cf228 | 1385 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1386 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1387 | return -ENOMEM; |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * COMMIT call returned | |
1392 | */ | |
788e7a89 | 1393 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1394 | { |
0b7c0153 | 1395 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1396 | |
a3f565b1 | 1397 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1398 | task->tk_pid, task->tk_status); |
1399 | ||
788e7a89 | 1400 | /* Call the NFS version-specific code */ |
c0d0e96b | 1401 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1402 | } |
1403 | ||
f453a54a | 1404 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1405 | { |
5917ce84 | 1406 | struct nfs_page *req; |
c9d8f89d | 1407 | int status = data->task.tk_status; |
f453a54a | 1408 | struct nfs_commit_info cinfo; |
788e7a89 | 1409 | |
1da177e4 LT |
1410 | while (!list_empty(&data->pages)) { |
1411 | req = nfs_list_entry(data->pages.next); | |
1412 | nfs_list_remove_request(req); | |
d6d6dc7c | 1413 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1414 | |
1e8968c5 | 1415 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1416 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1417 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1418 | req->wb_bytes, |
1419 | (long long)req_offset(req)); | |
c9d8f89d TM |
1420 | if (status < 0) { |
1421 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1422 | nfs_inode_remove_request(req); |
c9d8f89d | 1423 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1424 | goto next; |
1425 | } | |
1426 | ||
1427 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1428 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1429 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1430 | /* We have a match */ |
1431 | nfs_inode_remove_request(req); | |
1432 | dprintk(" OK\n"); | |
1433 | goto next; | |
1434 | } | |
1435 | /* We have a mismatch. Write the page again */ | |
1436 | dprintk(" mismatch\n"); | |
6d884e8f | 1437 | nfs_mark_request_dirty(req); |
05990d1b | 1438 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1439 | next: |
1d1afcbc | 1440 | nfs_unlock_and_release_request(req); |
1da177e4 | 1441 | } |
f453a54a FI |
1442 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1443 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1444 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1445 | } |
1446 | ||
1447 | static void nfs_commit_release(void *calldata) | |
1448 | { | |
0b7c0153 | 1449 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1450 | |
f453a54a | 1451 | data->completion_ops->completion(data); |
c9d8f89d | 1452 | nfs_commitdata_release(calldata); |
1da177e4 | 1453 | } |
788e7a89 TM |
1454 | |
1455 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1456 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1457 | .rpc_call_done = nfs_commit_done, |
1458 | .rpc_release = nfs_commit_release, | |
1459 | }; | |
1da177e4 | 1460 | |
f453a54a FI |
1461 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1462 | .completion = nfs_commit_release_pages, | |
1463 | .error_cleanup = nfs_commit_clear_lock, | |
1464 | }; | |
1465 | ||
1763da12 FI |
1466 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1467 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1468 | { |
1469 | int status; | |
1470 | ||
ea2cf228 | 1471 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1472 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1473 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1474 | return status; |
1475 | } | |
1476 | ||
b608b283 | 1477 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1478 | { |
1da177e4 | 1479 | LIST_HEAD(head); |
ea2cf228 | 1480 | struct nfs_commit_info cinfo; |
71d0a611 | 1481 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1482 | int res; |
1da177e4 | 1483 | |
b8413f98 TM |
1484 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1485 | if (res <= 0) | |
c5efa5fc | 1486 | goto out_mark_dirty; |
ea2cf228 FI |
1487 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1488 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1489 | if (res) { |
a861a1e1 FI |
1490 | int error; |
1491 | ||
ea2cf228 | 1492 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1493 | if (error < 0) |
1494 | return error; | |
b8413f98 | 1495 | if (!may_wait) |
c5efa5fc | 1496 | goto out_mark_dirty; |
b8413f98 TM |
1497 | error = wait_on_bit(&NFS_I(inode)->flags, |
1498 | NFS_INO_COMMIT, | |
1499 | nfs_wait_bit_killable, | |
1500 | TASK_KILLABLE); | |
1501 | if (error < 0) | |
1502 | return error; | |
71d0a611 TM |
1503 | } else |
1504 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1505 | return res; |
1506 | /* Note: If we exit without ensuring that the commit is complete, | |
1507 | * we must mark the inode as dirty. Otherwise, future calls to | |
1508 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1509 | * that the data is on the disk. | |
1510 | */ | |
1511 | out_mark_dirty: | |
1512 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1513 | return res; |
1514 | } | |
8fc795f7 TM |
1515 | |
1516 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1517 | { | |
420e3646 TM |
1518 | struct nfs_inode *nfsi = NFS_I(inode); |
1519 | int flags = FLUSH_SYNC; | |
1520 | int ret = 0; | |
8fc795f7 | 1521 | |
3236c3e1 | 1522 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1523 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1524 | return ret; |
1525 | ||
a00dd6c0 JL |
1526 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1527 | /* Don't commit yet if this is a non-blocking flush and there | |
1528 | * are a lot of outstanding writes for this mapping. | |
1529 | */ | |
ea2cf228 | 1530 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1531 | goto out_mark_dirty; |
420e3646 | 1532 | |
a00dd6c0 | 1533 | /* don't wait for the COMMIT response */ |
420e3646 | 1534 | flags = 0; |
a00dd6c0 JL |
1535 | } |
1536 | ||
420e3646 TM |
1537 | ret = nfs_commit_inode(inode, flags); |
1538 | if (ret >= 0) { | |
1539 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1540 | if (ret < wbc->nr_to_write) | |
1541 | wbc->nr_to_write -= ret; | |
1542 | else | |
1543 | wbc->nr_to_write = 0; | |
1544 | } | |
8fc795f7 | 1545 | return 0; |
420e3646 TM |
1546 | } |
1547 | out_mark_dirty: | |
8fc795f7 TM |
1548 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1549 | return ret; | |
1550 | } | |
c63c7b05 | 1551 | #else |
8fc795f7 TM |
1552 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1553 | { | |
1554 | return 0; | |
1555 | } | |
1da177e4 LT |
1556 | #endif |
1557 | ||
8fc795f7 TM |
1558 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1559 | { | |
a8d8f02c BS |
1560 | return nfs_commit_unstable_pages(inode, wbc); |
1561 | } | |
89d77c8f | 1562 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1563 | |
acdc53b2 TM |
1564 | /* |
1565 | * flush the inode to disk. | |
1566 | */ | |
1567 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1568 | { |
1569 | struct writeback_control wbc = { | |
72cb77f4 | 1570 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1571 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1572 | .range_start = 0, |
1573 | .range_end = LLONG_MAX, | |
34901f70 | 1574 | }; |
f4ce1299 TM |
1575 | int ret; |
1576 | ||
1577 | trace_nfs_writeback_inode_enter(inode); | |
1578 | ||
1579 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1580 | |
f4ce1299 TM |
1581 | trace_nfs_writeback_inode_exit(inode, ret); |
1582 | return ret; | |
1c75950b | 1583 | } |
ddda8e0a | 1584 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1585 | |
1b3b4a1a TM |
1586 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1587 | { | |
1588 | struct nfs_page *req; | |
1b3b4a1a TM |
1589 | int ret = 0; |
1590 | ||
1b3b4a1a | 1591 | for (;;) { |
ba8b06e6 | 1592 | wait_on_page_writeback(page); |
84d3a9a9 | 1593 | req = nfs_page_find_head_request(page); |
1b3b4a1a | 1594 | if (req == NULL) |
1b3b4a1a | 1595 | break; |
7ad84aa9 | 1596 | if (nfs_lock_request(req)) { |
8dd37758 | 1597 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1598 | nfs_inode_remove_request(req); |
1599 | /* | |
1600 | * In case nfs_inode_remove_request has marked the | |
1601 | * page as being dirty | |
1602 | */ | |
1603 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1d1afcbc | 1604 | nfs_unlock_and_release_request(req); |
1b3b4a1a TM |
1605 | break; |
1606 | } | |
1607 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1608 | nfs_release_request(req); |
1b3b4a1a | 1609 | if (ret < 0) |
c988950e | 1610 | break; |
1b3b4a1a | 1611 | } |
1b3b4a1a TM |
1612 | return ret; |
1613 | } | |
1614 | ||
7f2f12d9 TM |
1615 | /* |
1616 | * Write back all requests on one page - we do this before reading it. | |
1617 | */ | |
1618 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1619 | { |
29418aa4 | 1620 | loff_t range_start = page_file_offset(page); |
1c75950b | 1621 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1622 | struct writeback_control wbc = { |
4d770ccf | 1623 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1624 | .nr_to_write = 0, |
4d770ccf TM |
1625 | .range_start = range_start, |
1626 | .range_end = range_end, | |
1627 | }; | |
1628 | int ret; | |
1c75950b | 1629 | |
f4ce1299 TM |
1630 | trace_nfs_writeback_page_enter(inode); |
1631 | ||
0522f6ad | 1632 | for (;;) { |
ba8b06e6 | 1633 | wait_on_page_writeback(page); |
73e3302f TM |
1634 | if (clear_page_dirty_for_io(page)) { |
1635 | ret = nfs_writepage_locked(page, &wbc); | |
1636 | if (ret < 0) | |
1637 | goto out_error; | |
0522f6ad | 1638 | continue; |
7f2f12d9 | 1639 | } |
f4ce1299 | 1640 | ret = 0; |
0522f6ad TM |
1641 | if (!PagePrivate(page)) |
1642 | break; | |
1643 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1644 | if (ret < 0) |
73e3302f | 1645 | goto out_error; |
7f2f12d9 | 1646 | } |
73e3302f | 1647 | out_error: |
f4ce1299 | 1648 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1649 | return ret; |
1c75950b TM |
1650 | } |
1651 | ||
074cc1de TM |
1652 | #ifdef CONFIG_MIGRATION |
1653 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1654 | struct page *page, enum migrate_mode mode) |
074cc1de | 1655 | { |
2da95652 JL |
1656 | /* |
1657 | * If PagePrivate is set, then the page is currently associated with | |
1658 | * an in-progress read or write request. Don't try to migrate it. | |
1659 | * | |
1660 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1661 | * that we can safely release the inode reference while holding | |
1662 | * the page lock. | |
1663 | */ | |
1664 | if (PagePrivate(page)) | |
1665 | return -EBUSY; | |
074cc1de | 1666 | |
8c209ce7 DH |
1667 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1668 | return -EBUSY; | |
074cc1de | 1669 | |
a6bc32b8 | 1670 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1671 | } |
1672 | #endif | |
1673 | ||
f7b422b1 | 1674 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1675 | { |
1676 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
c0752cdf | 1677 | sizeof(struct nfs_rw_header), |
1da177e4 | 1678 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1679 | NULL); |
1da177e4 LT |
1680 | if (nfs_wdata_cachep == NULL) |
1681 | return -ENOMEM; | |
1682 | ||
93d2341c MD |
1683 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1684 | nfs_wdata_cachep); | |
1da177e4 | 1685 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1686 | goto out_destroy_write_cache; |
1da177e4 | 1687 | |
0b7c0153 FI |
1688 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1689 | sizeof(struct nfs_commit_data), | |
1690 | 0, SLAB_HWCACHE_ALIGN, | |
1691 | NULL); | |
1692 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1693 | goto out_destroy_write_mempool; |
0b7c0153 | 1694 | |
93d2341c | 1695 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1696 | nfs_cdata_cachep); |
1da177e4 | 1697 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1698 | goto out_destroy_commit_cache; |
1da177e4 | 1699 | |
89a09141 PZ |
1700 | /* |
1701 | * NFS congestion size, scale with available memory. | |
1702 | * | |
1703 | * 64MB: 8192k | |
1704 | * 128MB: 11585k | |
1705 | * 256MB: 16384k | |
1706 | * 512MB: 23170k | |
1707 | * 1GB: 32768k | |
1708 | * 2GB: 46340k | |
1709 | * 4GB: 65536k | |
1710 | * 8GB: 92681k | |
1711 | * 16GB: 131072k | |
1712 | * | |
1713 | * This allows larger machines to have larger/more transfers. | |
1714 | * Limit the default to 256M | |
1715 | */ | |
1716 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1717 | if (nfs_congestion_kb > 256*1024) | |
1718 | nfs_congestion_kb = 256*1024; | |
1719 | ||
1da177e4 | 1720 | return 0; |
3dd4765f JL |
1721 | |
1722 | out_destroy_commit_cache: | |
1723 | kmem_cache_destroy(nfs_cdata_cachep); | |
1724 | out_destroy_write_mempool: | |
1725 | mempool_destroy(nfs_wdata_mempool); | |
1726 | out_destroy_write_cache: | |
1727 | kmem_cache_destroy(nfs_wdata_cachep); | |
1728 | return -ENOMEM; | |
1da177e4 LT |
1729 | } |
1730 | ||
266bee88 | 1731 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1732 | { |
1733 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1734 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1735 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1736 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1737 | } |
1738 | ||
4a0de55c | 1739 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
a4cdda59 | 1740 | .rw_mode = FMODE_WRITE, |
4a0de55c AS |
1741 | .rw_alloc_header = nfs_writehdr_alloc, |
1742 | .rw_free_header = nfs_writehdr_free, | |
a4cdda59 | 1743 | .rw_release = nfs_writeback_release_common, |
0eecb214 AS |
1744 | .rw_done = nfs_writeback_done, |
1745 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 1746 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 1747 | }; |