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> |
1da177e4 LT |
16 | |
17 | #include <linux/sunrpc/clnt.h> | |
18 | #include <linux/nfs_fs.h> | |
19 | #include <linux/nfs_mount.h> | |
20 | #include <linux/nfs_page.h> | |
3fcfab16 AM |
21 | #include <linux/backing-dev.h> |
22 | ||
1da177e4 LT |
23 | #include <asm/uaccess.h> |
24 | #include <linux/smp_lock.h> | |
25 | ||
26 | #include "delegation.h" | |
49a70f27 | 27 | #include "internal.h" |
91d5b470 | 28 | #include "iostat.h" |
1da177e4 LT |
29 | |
30 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
31 | ||
32 | #define MIN_POOL_WRITE (32) | |
33 | #define MIN_POOL_COMMIT (4) | |
34 | ||
35 | /* | |
36 | * Local function declarations | |
37 | */ | |
38 | static struct nfs_page * nfs_update_request(struct nfs_open_context*, | |
1da177e4 LT |
39 | struct page *, |
40 | unsigned int, unsigned int); | |
c63c7b05 TM |
41 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
42 | struct inode *inode, int ioflags); | |
788e7a89 TM |
43 | static const struct rpc_call_ops nfs_write_partial_ops; |
44 | static const struct rpc_call_ops nfs_write_full_ops; | |
45 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 | 46 | |
e18b890b | 47 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 48 | static mempool_t *nfs_wdata_mempool; |
1da177e4 LT |
49 | static mempool_t *nfs_commit_mempool; |
50 | ||
e9f7bee1 | 51 | struct nfs_write_data *nfs_commit_alloc(void) |
1da177e4 | 52 | { |
e6b4f8da | 53 | struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 54 | |
1da177e4 LT |
55 | if (p) { |
56 | memset(p, 0, sizeof(*p)); | |
57 | INIT_LIST_HEAD(&p->pages); | |
58 | } | |
59 | return p; | |
60 | } | |
61 | ||
8aca67f0 | 62 | void nfs_commit_rcu_free(struct rcu_head *head) |
1da177e4 | 63 | { |
8aca67f0 | 64 | struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); |
40859d7e CL |
65 | if (p && (p->pagevec != &p->page_array[0])) |
66 | kfree(p->pagevec); | |
1da177e4 LT |
67 | mempool_free(p, nfs_commit_mempool); |
68 | } | |
69 | ||
8aca67f0 TM |
70 | void nfs_commit_free(struct nfs_write_data *wdata) |
71 | { | |
72 | call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); | |
73 | } | |
74 | ||
e9f7bee1 | 75 | struct nfs_write_data *nfs_writedata_alloc(size_t len) |
3feb2d49 | 76 | { |
e9f7bee1 | 77 | unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e6b4f8da | 78 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
79 | |
80 | if (p) { | |
81 | memset(p, 0, sizeof(*p)); | |
82 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 83 | p->npages = pagecount; |
0d0b5cb3 CL |
84 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
85 | p->pagevec = p->page_array; | |
3feb2d49 | 86 | else { |
0d0b5cb3 CL |
87 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
88 | if (!p->pagevec) { | |
3feb2d49 TM |
89 | mempool_free(p, nfs_wdata_mempool); |
90 | p = NULL; | |
91 | } | |
92 | } | |
93 | } | |
94 | return p; | |
95 | } | |
96 | ||
8aca67f0 | 97 | static void nfs_writedata_rcu_free(struct rcu_head *head) |
3feb2d49 | 98 | { |
8aca67f0 | 99 | struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); |
3feb2d49 TM |
100 | if (p && (p->pagevec != &p->page_array[0])) |
101 | kfree(p->pagevec); | |
102 | mempool_free(p, nfs_wdata_mempool); | |
103 | } | |
104 | ||
8aca67f0 TM |
105 | static void nfs_writedata_free(struct nfs_write_data *wdata) |
106 | { | |
107 | call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); | |
108 | } | |
109 | ||
963d8fe5 | 110 | void nfs_writedata_release(void *wdata) |
1da177e4 | 111 | { |
1da177e4 LT |
112 | nfs_writedata_free(wdata); |
113 | } | |
114 | ||
277459d2 TM |
115 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
116 | { | |
117 | struct nfs_page *req = NULL; | |
118 | ||
119 | if (PagePrivate(page)) { | |
120 | req = (struct nfs_page *)page_private(page); | |
121 | if (req != NULL) | |
122 | atomic_inc(&req->wb_count); | |
123 | } | |
124 | return req; | |
125 | } | |
126 | ||
127 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
128 | { | |
129 | struct nfs_page *req = NULL; | |
130 | spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; | |
131 | ||
132 | spin_lock(req_lock); | |
133 | req = nfs_page_find_request_locked(page); | |
134 | spin_unlock(req_lock); | |
135 | return req; | |
136 | } | |
137 | ||
1da177e4 LT |
138 | /* Adjust the file length if we're writing beyond the end */ |
139 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
140 | { | |
141 | struct inode *inode = page->mapping->host; | |
142 | loff_t end, i_size = i_size_read(inode); | |
143 | unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
144 | ||
145 | if (i_size > 0 && page->index < end_index) | |
146 | return; | |
147 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); | |
148 | if (i_size >= end) | |
149 | return; | |
91d5b470 | 150 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
1da177e4 LT |
151 | i_size_write(inode, end); |
152 | } | |
153 | ||
a301b777 TM |
154 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
155 | static void nfs_set_pageerror(struct page *page) | |
156 | { | |
157 | SetPageError(page); | |
158 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
159 | } | |
160 | ||
1da177e4 LT |
161 | /* We can set the PG_uptodate flag if we see that a write request |
162 | * covers the full page. | |
163 | */ | |
164 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
165 | { | |
1da177e4 LT |
166 | if (PageUptodate(page)) |
167 | return; | |
168 | if (base != 0) | |
169 | return; | |
49a70f27 | 170 | if (count != nfs_page_length(page)) |
1da177e4 | 171 | return; |
49a70f27 | 172 | if (count != PAGE_CACHE_SIZE) |
1da177e4 | 173 | memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); |
49a70f27 | 174 | SetPageUptodate(page); |
1da177e4 LT |
175 | } |
176 | ||
e21195a7 | 177 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1da177e4 LT |
178 | unsigned int offset, unsigned int count) |
179 | { | |
180 | struct nfs_page *req; | |
e21195a7 | 181 | int ret; |
1da177e4 | 182 | |
e21195a7 TM |
183 | for (;;) { |
184 | req = nfs_update_request(ctx, page, offset, count); | |
185 | if (!IS_ERR(req)) | |
186 | break; | |
187 | ret = PTR_ERR(req); | |
188 | if (ret != -EBUSY) | |
189 | return ret; | |
190 | ret = nfs_wb_page(page->mapping->host, page); | |
191 | if (ret != 0) | |
192 | return ret; | |
193 | } | |
1da177e4 LT |
194 | /* Update file length */ |
195 | nfs_grow_file(page, offset, count); | |
196 | /* Set the PG_uptodate flag? */ | |
197 | nfs_mark_uptodate(page, offset, count); | |
198 | nfs_unlock_request(req); | |
abd3e641 | 199 | return 0; |
1da177e4 LT |
200 | } |
201 | ||
202 | static int wb_priority(struct writeback_control *wbc) | |
203 | { | |
204 | if (wbc->for_reclaim) | |
c63c7b05 | 205 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
1da177e4 LT |
206 | if (wbc->for_kupdate) |
207 | return FLUSH_LOWPRI; | |
208 | return 0; | |
209 | } | |
210 | ||
89a09141 PZ |
211 | /* |
212 | * NFS congestion control | |
213 | */ | |
214 | ||
215 | int nfs_congestion_kb; | |
216 | ||
217 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
218 | #define NFS_CONGESTION_OFF_THRESH \ | |
219 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
220 | ||
5a6d41b3 | 221 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 222 | { |
5a6d41b3 TM |
223 | int ret = test_set_page_writeback(page); |
224 | ||
225 | if (!ret) { | |
89a09141 PZ |
226 | struct inode *inode = page->mapping->host; |
227 | struct nfs_server *nfss = NFS_SERVER(inode); | |
228 | ||
229 | if (atomic_inc_return(&nfss->writeback) > | |
230 | NFS_CONGESTION_ON_THRESH) | |
231 | set_bdi_congested(&nfss->backing_dev_info, WRITE); | |
232 | } | |
5a6d41b3 | 233 | return ret; |
89a09141 PZ |
234 | } |
235 | ||
236 | static void nfs_end_page_writeback(struct page *page) | |
237 | { | |
238 | struct inode *inode = page->mapping->host; | |
239 | struct nfs_server *nfss = NFS_SERVER(inode); | |
240 | ||
241 | end_page_writeback(page); | |
242 | if (atomic_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) { | |
243 | clear_bdi_congested(&nfss->backing_dev_info, WRITE); | |
244 | congestion_end(WRITE); | |
245 | } | |
246 | } | |
247 | ||
e261f51f TM |
248 | /* |
249 | * Find an associated nfs write request, and prepare to flush it out | |
250 | * Returns 1 if there was no write request, or if the request was | |
251 | * already tagged by nfs_set_page_dirty.Returns 0 if the request | |
252 | * was not tagged. | |
253 | * May also return an error if the user signalled nfs_wait_on_request(). | |
254 | */ | |
c63c7b05 TM |
255 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, |
256 | struct page *page) | |
e261f51f TM |
257 | { |
258 | struct nfs_page *req; | |
612c9384 TM |
259 | struct nfs_inode *nfsi = NFS_I(page->mapping->host); |
260 | spinlock_t *req_lock = &nfsi->req_lock; | |
e261f51f TM |
261 | int ret; |
262 | ||
263 | spin_lock(req_lock); | |
264 | for(;;) { | |
265 | req = nfs_page_find_request_locked(page); | |
266 | if (req == NULL) { | |
267 | spin_unlock(req_lock); | |
268 | return 1; | |
269 | } | |
270 | if (nfs_lock_request_dontget(req)) | |
271 | break; | |
272 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
273 | * then the call to nfs_lock_request_dontget() will always | |
274 | * succeed provided that someone hasn't already marked the | |
275 | * request as dirty (in which case we don't care). | |
276 | */ | |
277 | spin_unlock(req_lock); | |
c63c7b05 TM |
278 | /* Prevent deadlock! */ |
279 | nfs_pageio_complete(pgio); | |
e261f51f TM |
280 | ret = nfs_wait_on_request(req); |
281 | nfs_release_request(req); | |
282 | if (ret != 0) | |
283 | return ret; | |
284 | spin_lock(req_lock); | |
285 | } | |
612c9384 TM |
286 | if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
287 | /* This request is marked for commit */ | |
288 | spin_unlock(req_lock); | |
289 | nfs_unlock_request(req); | |
c63c7b05 | 290 | nfs_pageio_complete(pgio); |
612c9384 TM |
291 | return 1; |
292 | } | |
c63c7b05 | 293 | if (nfs_set_page_writeback(page) != 0) { |
612c9384 | 294 | spin_unlock(req_lock); |
c63c7b05 TM |
295 | BUG(); |
296 | } | |
297 | radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, | |
298 | NFS_PAGE_TAG_WRITEBACK); | |
e261f51f | 299 | ret = test_bit(PG_NEED_FLUSH, &req->wb_flags); |
c63c7b05 TM |
300 | spin_unlock(req_lock); |
301 | nfs_pageio_add_request(pgio, req); | |
e261f51f TM |
302 | return ret; |
303 | } | |
304 | ||
1da177e4 LT |
305 | /* |
306 | * Write an mmapped page to the server. | |
307 | */ | |
4d770ccf | 308 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) |
1da177e4 | 309 | { |
c63c7b05 | 310 | struct nfs_pageio_descriptor mypgio, *pgio; |
1da177e4 LT |
311 | struct nfs_open_context *ctx; |
312 | struct inode *inode = page->mapping->host; | |
49a70f27 | 313 | unsigned offset; |
e261f51f | 314 | int err; |
1da177e4 | 315 | |
91d5b470 CL |
316 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
317 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
318 | ||
c63c7b05 TM |
319 | if (wbc->for_writepages) |
320 | pgio = wbc->fs_private; | |
321 | else { | |
322 | nfs_pageio_init_write(&mypgio, inode, wb_priority(wbc)); | |
323 | pgio = &mypgio; | |
324 | } | |
325 | ||
326 | err = nfs_page_async_flush(pgio, page); | |
e261f51f TM |
327 | if (err <= 0) |
328 | goto out; | |
329 | err = 0; | |
49a70f27 TM |
330 | offset = nfs_page_length(page); |
331 | if (!offset) | |
1da177e4 | 332 | goto out; |
49a70f27 | 333 | |
d530838b | 334 | ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); |
1da177e4 LT |
335 | if (ctx == NULL) { |
336 | err = -EBADF; | |
337 | goto out; | |
338 | } | |
200baa21 | 339 | err = nfs_writepage_setup(ctx, page, 0, offset); |
1da177e4 | 340 | put_nfs_open_context(ctx); |
e261f51f TM |
341 | if (err != 0) |
342 | goto out; | |
c63c7b05 | 343 | err = nfs_page_async_flush(pgio, page); |
e261f51f TM |
344 | if (err > 0) |
345 | err = 0; | |
1da177e4 | 346 | out: |
200baa21 | 347 | if (!wbc->for_writepages) |
c63c7b05 | 348 | nfs_pageio_complete(pgio); |
4d770ccf TM |
349 | return err; |
350 | } | |
351 | ||
352 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
353 | { | |
354 | int err; | |
355 | ||
356 | err = nfs_writepage_locked(page, wbc); | |
1da177e4 | 357 | unlock_page(page); |
1da177e4 LT |
358 | return err; |
359 | } | |
360 | ||
1da177e4 LT |
361 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
362 | { | |
1da177e4 | 363 | struct inode *inode = mapping->host; |
c63c7b05 | 364 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
365 | int err; |
366 | ||
91d5b470 CL |
367 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
368 | ||
c63c7b05 TM |
369 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
370 | wbc->fs_private = &pgio; | |
1da177e4 | 371 | err = generic_writepages(mapping, wbc); |
c63c7b05 | 372 | nfs_pageio_complete(&pgio); |
1da177e4 LT |
373 | if (err) |
374 | return err; | |
c63c7b05 TM |
375 | if (pgio.pg_error) |
376 | return pgio.pg_error; | |
377 | return 0; | |
1da177e4 LT |
378 | } |
379 | ||
380 | /* | |
381 | * Insert a write request into an inode | |
382 | */ | |
383 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) | |
384 | { | |
385 | struct nfs_inode *nfsi = NFS_I(inode); | |
386 | int error; | |
387 | ||
388 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); | |
389 | BUG_ON(error == -EEXIST); | |
390 | if (error) | |
391 | return error; | |
392 | if (!nfsi->npages) { | |
393 | igrab(inode); | |
394 | nfs_begin_data_update(inode); | |
395 | if (nfs_have_delegation(inode, FMODE_WRITE)) | |
396 | nfsi->change_attr++; | |
397 | } | |
deb7d638 | 398 | SetPagePrivate(req->wb_page); |
277459d2 | 399 | set_page_private(req->wb_page, (unsigned long)req); |
2b82f190 TM |
400 | if (PageDirty(req->wb_page)) |
401 | set_bit(PG_NEED_FLUSH, &req->wb_flags); | |
1da177e4 LT |
402 | nfsi->npages++; |
403 | atomic_inc(&req->wb_count); | |
404 | return 0; | |
405 | } | |
406 | ||
407 | /* | |
89a09141 | 408 | * Remove a write request from an inode |
1da177e4 LT |
409 | */ |
410 | static void nfs_inode_remove_request(struct nfs_page *req) | |
411 | { | |
412 | struct inode *inode = req->wb_context->dentry->d_inode; | |
413 | struct nfs_inode *nfsi = NFS_I(inode); | |
414 | ||
415 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
416 | ||
417 | spin_lock(&nfsi->req_lock); | |
277459d2 | 418 | set_page_private(req->wb_page, 0); |
deb7d638 | 419 | ClearPagePrivate(req->wb_page); |
1da177e4 | 420 | radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); |
2b82f190 TM |
421 | if (test_and_clear_bit(PG_NEED_FLUSH, &req->wb_flags)) |
422 | __set_page_dirty_nobuffers(req->wb_page); | |
1da177e4 LT |
423 | nfsi->npages--; |
424 | if (!nfsi->npages) { | |
425 | spin_unlock(&nfsi->req_lock); | |
951a143b | 426 | nfs_end_data_update(inode); |
1da177e4 LT |
427 | iput(inode); |
428 | } else | |
429 | spin_unlock(&nfsi->req_lock); | |
430 | nfs_clear_request(req); | |
431 | nfs_release_request(req); | |
432 | } | |
433 | ||
61822ab5 TM |
434 | static void |
435 | nfs_redirty_request(struct nfs_page *req) | |
436 | { | |
61822ab5 TM |
437 | __set_page_dirty_nobuffers(req->wb_page); |
438 | } | |
439 | ||
1da177e4 LT |
440 | /* |
441 | * Check if a request is dirty | |
442 | */ | |
443 | static inline int | |
444 | nfs_dirty_request(struct nfs_page *req) | |
445 | { | |
5a6d41b3 TM |
446 | struct page *page = req->wb_page; |
447 | ||
612c9384 | 448 | if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) |
5a6d41b3 TM |
449 | return 0; |
450 | return !PageWriteback(req->wb_page); | |
1da177e4 LT |
451 | } |
452 | ||
453 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
454 | /* | |
455 | * Add a request to the inode's commit list. | |
456 | */ | |
457 | static void | |
458 | nfs_mark_request_commit(struct nfs_page *req) | |
459 | { | |
460 | struct inode *inode = req->wb_context->dentry->d_inode; | |
461 | struct nfs_inode *nfsi = NFS_I(inode); | |
462 | ||
463 | spin_lock(&nfsi->req_lock); | |
464 | nfs_list_add_request(req, &nfsi->commit); | |
465 | nfsi->ncommit++; | |
612c9384 | 466 | set_bit(PG_NEED_COMMIT, &(req)->wb_flags); |
1da177e4 | 467 | spin_unlock(&nfsi->req_lock); |
fd39fc85 | 468 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
a1803044 | 469 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 470 | } |
8e821cad TM |
471 | |
472 | static inline | |
473 | int nfs_write_need_commit(struct nfs_write_data *data) | |
474 | { | |
475 | return data->verf.committed != NFS_FILE_SYNC; | |
476 | } | |
477 | ||
478 | static inline | |
479 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
480 | { | |
612c9384 | 481 | if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
8e821cad TM |
482 | nfs_mark_request_commit(req); |
483 | return 1; | |
484 | } | |
485 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
486 | nfs_redirty_request(req); | |
487 | return 1; | |
488 | } | |
489 | return 0; | |
490 | } | |
491 | #else | |
492 | static inline void | |
493 | nfs_mark_request_commit(struct nfs_page *req) | |
494 | { | |
495 | } | |
496 | ||
497 | static inline | |
498 | int nfs_write_need_commit(struct nfs_write_data *data) | |
499 | { | |
500 | return 0; | |
501 | } | |
502 | ||
503 | static inline | |
504 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
505 | { | |
506 | return 0; | |
507 | } | |
1da177e4 LT |
508 | #endif |
509 | ||
510 | /* | |
511 | * Wait for a request to complete. | |
512 | * | |
513 | * Interruptible by signals only if mounted with intr flag. | |
514 | */ | |
c42de9dd | 515 | static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) |
1da177e4 LT |
516 | { |
517 | struct nfs_inode *nfsi = NFS_I(inode); | |
518 | struct nfs_page *req; | |
519 | unsigned long idx_end, next; | |
520 | unsigned int res = 0; | |
521 | int error; | |
522 | ||
523 | if (npages == 0) | |
524 | idx_end = ~0; | |
525 | else | |
526 | idx_end = idx_start + npages - 1; | |
527 | ||
1da177e4 | 528 | next = idx_start; |
c6a556b8 | 529 | while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { |
1da177e4 LT |
530 | if (req->wb_index > idx_end) |
531 | break; | |
532 | ||
533 | next = req->wb_index + 1; | |
c6a556b8 | 534 | BUG_ON(!NFS_WBACK_BUSY(req)); |
1da177e4 LT |
535 | |
536 | atomic_inc(&req->wb_count); | |
537 | spin_unlock(&nfsi->req_lock); | |
538 | error = nfs_wait_on_request(req); | |
539 | nfs_release_request(req); | |
c42de9dd | 540 | spin_lock(&nfsi->req_lock); |
1da177e4 LT |
541 | if (error < 0) |
542 | return error; | |
1da177e4 LT |
543 | res++; |
544 | } | |
1da177e4 LT |
545 | return res; |
546 | } | |
547 | ||
83715ad5 TM |
548 | static void nfs_cancel_commit_list(struct list_head *head) |
549 | { | |
550 | struct nfs_page *req; | |
551 | ||
552 | while(!list_empty(head)) { | |
553 | req = nfs_list_entry(head->next); | |
b6dff26a | 554 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
83715ad5 | 555 | nfs_list_remove_request(req); |
612c9384 | 556 | clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); |
83715ad5 | 557 | nfs_inode_remove_request(req); |
b6dff26a | 558 | nfs_unlock_request(req); |
83715ad5 TM |
559 | } |
560 | } | |
561 | ||
1da177e4 LT |
562 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
563 | /* | |
564 | * nfs_scan_commit - Scan an inode for commit requests | |
565 | * @inode: NFS inode to scan | |
566 | * @dst: destination list | |
567 | * @idx_start: lower bound of page->index to scan. | |
568 | * @npages: idx_start + npages sets the upper bound to scan. | |
569 | * | |
570 | * Moves requests from the inode's 'commit' request list. | |
571 | * The requests are *not* checked to ensure that they form a contiguous set. | |
572 | */ | |
573 | static int | |
574 | nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) | |
575 | { | |
576 | struct nfs_inode *nfsi = NFS_I(inode); | |
3da28eb1 TM |
577 | int res = 0; |
578 | ||
579 | if (nfsi->ncommit != 0) { | |
d2ccddf0 | 580 | res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); |
3da28eb1 TM |
581 | nfsi->ncommit -= res; |
582 | if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) | |
583 | printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); | |
584 | } | |
1da177e4 LT |
585 | return res; |
586 | } | |
c42de9dd TM |
587 | #else |
588 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) | |
589 | { | |
590 | return 0; | |
591 | } | |
1da177e4 LT |
592 | #endif |
593 | ||
1da177e4 LT |
594 | /* |
595 | * Try to update any existing write request, or create one if there is none. | |
596 | * In order to match, the request's credentials must match those of | |
597 | * the calling process. | |
598 | * | |
599 | * Note: Should always be called with the Page Lock held! | |
600 | */ | |
601 | static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, | |
e21195a7 | 602 | struct page *page, unsigned int offset, unsigned int bytes) |
1da177e4 | 603 | { |
89a09141 PZ |
604 | struct address_space *mapping = page->mapping; |
605 | struct inode *inode = mapping->host; | |
1da177e4 LT |
606 | struct nfs_inode *nfsi = NFS_I(inode); |
607 | struct nfs_page *req, *new = NULL; | |
608 | unsigned long rqend, end; | |
609 | ||
610 | end = offset + bytes; | |
611 | ||
1da177e4 LT |
612 | for (;;) { |
613 | /* Loop over all inode entries and see if we find | |
614 | * A request for the page we wish to update | |
615 | */ | |
616 | spin_lock(&nfsi->req_lock); | |
277459d2 | 617 | req = nfs_page_find_request_locked(page); |
1da177e4 LT |
618 | if (req) { |
619 | if (!nfs_lock_request_dontget(req)) { | |
620 | int error; | |
277459d2 | 621 | |
1da177e4 LT |
622 | spin_unlock(&nfsi->req_lock); |
623 | error = nfs_wait_on_request(req); | |
624 | nfs_release_request(req); | |
1dd594b2 NB |
625 | if (error < 0) { |
626 | if (new) | |
627 | nfs_release_request(new); | |
1da177e4 | 628 | return ERR_PTR(error); |
1dd594b2 | 629 | } |
1da177e4 LT |
630 | continue; |
631 | } | |
632 | spin_unlock(&nfsi->req_lock); | |
633 | if (new) | |
634 | nfs_release_request(new); | |
635 | break; | |
636 | } | |
637 | ||
638 | if (new) { | |
639 | int error; | |
640 | nfs_lock_request_dontget(new); | |
641 | error = nfs_inode_add_request(inode, new); | |
642 | if (error) { | |
643 | spin_unlock(&nfsi->req_lock); | |
644 | nfs_unlock_request(new); | |
645 | return ERR_PTR(error); | |
646 | } | |
647 | spin_unlock(&nfsi->req_lock); | |
1da177e4 LT |
648 | return new; |
649 | } | |
650 | spin_unlock(&nfsi->req_lock); | |
651 | ||
652 | new = nfs_create_request(ctx, inode, page, offset, bytes); | |
653 | if (IS_ERR(new)) | |
654 | return new; | |
655 | } | |
656 | ||
657 | /* We have a request for our page. | |
658 | * If the creds don't match, or the | |
659 | * page addresses don't match, | |
660 | * tell the caller to wait on the conflicting | |
661 | * request. | |
662 | */ | |
663 | rqend = req->wb_offset + req->wb_bytes; | |
664 | if (req->wb_context != ctx | |
665 | || req->wb_page != page | |
666 | || !nfs_dirty_request(req) | |
667 | || offset > rqend || end < req->wb_offset) { | |
668 | nfs_unlock_request(req); | |
669 | return ERR_PTR(-EBUSY); | |
670 | } | |
671 | ||
672 | /* Okay, the request matches. Update the region */ | |
673 | if (offset < req->wb_offset) { | |
674 | req->wb_offset = offset; | |
675 | req->wb_pgbase = offset; | |
676 | req->wb_bytes = rqend - req->wb_offset; | |
677 | } | |
678 | ||
679 | if (end > rqend) | |
680 | req->wb_bytes = end - req->wb_offset; | |
681 | ||
682 | return req; | |
683 | } | |
684 | ||
685 | int nfs_flush_incompatible(struct file *file, struct page *page) | |
686 | { | |
687 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
1da177e4 | 688 | struct nfs_page *req; |
1a54533e | 689 | int do_flush, status; |
1da177e4 LT |
690 | /* |
691 | * Look for a request corresponding to this page. If there | |
692 | * is one, and it belongs to another file, we flush it out | |
693 | * before we try to copy anything into the page. Do this | |
694 | * due to the lack of an ACCESS-type call in NFSv2. | |
695 | * Also do the same if we find a request from an existing | |
696 | * dropped page. | |
697 | */ | |
1a54533e TM |
698 | do { |
699 | req = nfs_page_find_request(page); | |
700 | if (req == NULL) | |
701 | return 0; | |
702 | do_flush = req->wb_page != page || req->wb_context != ctx | |
e261f51f | 703 | || !nfs_dirty_request(req); |
1da177e4 | 704 | nfs_release_request(req); |
1a54533e TM |
705 | if (!do_flush) |
706 | return 0; | |
707 | status = nfs_wb_page(page->mapping->host, page); | |
708 | } while (status == 0); | |
709 | return status; | |
1da177e4 LT |
710 | } |
711 | ||
712 | /* | |
713 | * Update and possibly write a cached page of an NFS file. | |
714 | * | |
715 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
716 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
717 | */ | |
718 | int nfs_updatepage(struct file *file, struct page *page, | |
719 | unsigned int offset, unsigned int count) | |
720 | { | |
721 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
1da177e4 | 722 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
723 | int status = 0; |
724 | ||
91d5b470 CL |
725 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
726 | ||
1da177e4 | 727 | dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", |
01cce933 JJS |
728 | file->f_path.dentry->d_parent->d_name.name, |
729 | file->f_path.dentry->d_name.name, count, | |
0bbacc40 | 730 | (long long)(page_offset(page) +offset)); |
1da177e4 | 731 | |
1da177e4 LT |
732 | /* If we're not using byte range locks, and we know the page |
733 | * is entirely in cache, it may be more efficient to avoid | |
734 | * fragmenting write requests. | |
735 | */ | |
ab0a3dbe | 736 | if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { |
49a70f27 | 737 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 738 | offset = 0; |
1da177e4 LT |
739 | } |
740 | ||
e21195a7 | 741 | status = nfs_writepage_setup(ctx, page, offset, count); |
e261f51f | 742 | __set_page_dirty_nobuffers(page); |
1da177e4 | 743 | |
1da177e4 LT |
744 | dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", |
745 | status, (long long)i_size_read(inode)); | |
746 | if (status < 0) | |
a301b777 | 747 | nfs_set_pageerror(page); |
1da177e4 LT |
748 | return status; |
749 | } | |
750 | ||
751 | static void nfs_writepage_release(struct nfs_page *req) | |
752 | { | |
1da177e4 | 753 | |
8e821cad TM |
754 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { |
755 | nfs_end_page_writeback(req->wb_page); | |
756 | nfs_inode_remove_request(req); | |
757 | } else | |
758 | nfs_end_page_writeback(req->wb_page); | |
c6a556b8 | 759 | nfs_clear_page_writeback(req); |
1da177e4 LT |
760 | } |
761 | ||
762 | static inline int flush_task_priority(int how) | |
763 | { | |
764 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
765 | case FLUSH_HIGHPRI: | |
766 | return RPC_PRIORITY_HIGH; | |
767 | case FLUSH_LOWPRI: | |
768 | return RPC_PRIORITY_LOW; | |
769 | } | |
770 | return RPC_PRIORITY_NORMAL; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Set up the argument/result storage required for the RPC call. | |
775 | */ | |
776 | static void nfs_write_rpcsetup(struct nfs_page *req, | |
777 | struct nfs_write_data *data, | |
788e7a89 | 778 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
779 | unsigned int count, unsigned int offset, |
780 | int how) | |
781 | { | |
1da177e4 | 782 | struct inode *inode; |
788e7a89 | 783 | int flags; |
1da177e4 LT |
784 | |
785 | /* Set up the RPC argument and reply structs | |
786 | * NB: take care not to mess about with data->commit et al. */ | |
787 | ||
788 | data->req = req; | |
789 | data->inode = inode = req->wb_context->dentry->d_inode; | |
790 | data->cred = req->wb_context->cred; | |
791 | ||
792 | data->args.fh = NFS_FH(inode); | |
793 | data->args.offset = req_offset(req) + offset; | |
794 | data->args.pgbase = req->wb_pgbase + offset; | |
795 | data->args.pages = data->pagevec; | |
796 | data->args.count = count; | |
797 | data->args.context = req->wb_context; | |
798 | ||
799 | data->res.fattr = &data->fattr; | |
800 | data->res.count = count; | |
801 | data->res.verf = &data->verf; | |
0e574af1 | 802 | nfs_fattr_init(&data->fattr); |
1da177e4 | 803 | |
788e7a89 TM |
804 | /* Set up the initial task struct. */ |
805 | flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
806 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); | |
1da177e4 LT |
807 | NFS_PROTO(inode)->write_setup(data, how); |
808 | ||
809 | data->task.tk_priority = flush_task_priority(how); | |
810 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 | 811 | |
a3f565b1 CL |
812 | dprintk("NFS: %5u initiated write call " |
813 | "(req %s/%Ld, %u bytes @ offset %Lu)\n", | |
0bbacc40 | 814 | data->task.tk_pid, |
1da177e4 LT |
815 | inode->i_sb->s_id, |
816 | (long long)NFS_FILEID(inode), | |
817 | count, | |
818 | (unsigned long long)data->args.offset); | |
819 | } | |
820 | ||
821 | static void nfs_execute_write(struct nfs_write_data *data) | |
822 | { | |
823 | struct rpc_clnt *clnt = NFS_CLIENT(data->inode); | |
824 | sigset_t oldset; | |
825 | ||
826 | rpc_clnt_sigmask(clnt, &oldset); | |
1da177e4 | 827 | rpc_execute(&data->task); |
1da177e4 LT |
828 | rpc_clnt_sigunmask(clnt, &oldset); |
829 | } | |
830 | ||
831 | /* | |
832 | * Generate multiple small requests to write out a single | |
833 | * contiguous dirty area on one page. | |
834 | */ | |
bcb71bba | 835 | static int nfs_flush_multi(struct inode *inode, struct list_head *head, size_t count, int how) |
1da177e4 LT |
836 | { |
837 | struct nfs_page *req = nfs_list_entry(head->next); | |
838 | struct page *page = req->wb_page; | |
839 | struct nfs_write_data *data; | |
e9f7bee1 TM |
840 | size_t wsize = NFS_SERVER(inode)->wsize, nbytes; |
841 | unsigned int offset; | |
1da177e4 LT |
842 | int requests = 0; |
843 | LIST_HEAD(list); | |
844 | ||
845 | nfs_list_remove_request(req); | |
846 | ||
bcb71bba | 847 | nbytes = count; |
e9f7bee1 TM |
848 | do { |
849 | size_t len = min(nbytes, wsize); | |
850 | ||
851 | data = nfs_writedata_alloc(len); | |
1da177e4 LT |
852 | if (!data) |
853 | goto out_bad; | |
854 | list_add(&data->pages, &list); | |
855 | requests++; | |
e9f7bee1 TM |
856 | nbytes -= len; |
857 | } while (nbytes != 0); | |
1da177e4 LT |
858 | atomic_set(&req->wb_complete, requests); |
859 | ||
860 | ClearPageError(page); | |
1da177e4 | 861 | offset = 0; |
bcb71bba | 862 | nbytes = count; |
1da177e4 LT |
863 | do { |
864 | data = list_entry(list.next, struct nfs_write_data, pages); | |
865 | list_del_init(&data->pages); | |
866 | ||
867 | data->pagevec[0] = page; | |
1da177e4 | 868 | |
bcb71bba TM |
869 | if (nbytes < wsize) |
870 | wsize = nbytes; | |
871 | nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, | |
872 | wsize, offset, how); | |
873 | offset += wsize; | |
874 | nbytes -= wsize; | |
1da177e4 LT |
875 | nfs_execute_write(data); |
876 | } while (nbytes != 0); | |
877 | ||
878 | return 0; | |
879 | ||
880 | out_bad: | |
881 | while (!list_empty(&list)) { | |
882 | data = list_entry(list.next, struct nfs_write_data, pages); | |
883 | list_del(&data->pages); | |
8aca67f0 | 884 | nfs_writedata_release(data); |
1da177e4 | 885 | } |
61822ab5 | 886 | nfs_redirty_request(req); |
6d677e35 | 887 | nfs_end_page_writeback(req->wb_page); |
c6a556b8 | 888 | nfs_clear_page_writeback(req); |
1da177e4 LT |
889 | return -ENOMEM; |
890 | } | |
891 | ||
892 | /* | |
893 | * Create an RPC task for the given write request and kick it. | |
894 | * The page must have been locked by the caller. | |
895 | * | |
896 | * It may happen that the page we're passed is not marked dirty. | |
897 | * This is the case if nfs_updatepage detects a conflicting request | |
898 | * that has been written but not committed. | |
899 | */ | |
bcb71bba | 900 | static int nfs_flush_one(struct inode *inode, struct list_head *head, size_t count, int how) |
1da177e4 LT |
901 | { |
902 | struct nfs_page *req; | |
903 | struct page **pages; | |
904 | struct nfs_write_data *data; | |
1da177e4 | 905 | |
bcb71bba | 906 | data = nfs_writedata_alloc(count); |
1da177e4 LT |
907 | if (!data) |
908 | goto out_bad; | |
909 | ||
910 | pages = data->pagevec; | |
1da177e4 LT |
911 | while (!list_empty(head)) { |
912 | req = nfs_list_entry(head->next); | |
913 | nfs_list_remove_request(req); | |
914 | nfs_list_add_request(req, &data->pages); | |
915 | ClearPageError(req->wb_page); | |
1da177e4 | 916 | *pages++ = req->wb_page; |
1da177e4 LT |
917 | } |
918 | req = nfs_list_entry(data->pages.next); | |
919 | ||
1da177e4 | 920 | /* Set up the argument struct */ |
788e7a89 | 921 | nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); |
1da177e4 LT |
922 | |
923 | nfs_execute_write(data); | |
924 | return 0; | |
925 | out_bad: | |
926 | while (!list_empty(head)) { | |
927 | struct nfs_page *req = nfs_list_entry(head->next); | |
928 | nfs_list_remove_request(req); | |
61822ab5 | 929 | nfs_redirty_request(req); |
6d677e35 | 930 | nfs_end_page_writeback(req->wb_page); |
c6a556b8 | 931 | nfs_clear_page_writeback(req); |
1da177e4 LT |
932 | } |
933 | return -ENOMEM; | |
934 | } | |
935 | ||
c63c7b05 TM |
936 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
937 | struct inode *inode, int ioflags) | |
1da177e4 | 938 | { |
7d46a49f | 939 | int wsize = NFS_SERVER(inode)->wsize; |
1da177e4 | 940 | |
bcb71bba | 941 | if (wsize < PAGE_CACHE_SIZE) |
c63c7b05 | 942 | nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); |
bcb71bba | 943 | else |
c63c7b05 | 944 | nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); |
1da177e4 LT |
945 | } |
946 | ||
947 | /* | |
948 | * Handle a write reply that flushed part of a page. | |
949 | */ | |
788e7a89 | 950 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 951 | { |
788e7a89 | 952 | struct nfs_write_data *data = calldata; |
1da177e4 LT |
953 | struct nfs_page *req = data->req; |
954 | struct page *page = req->wb_page; | |
955 | ||
956 | dprintk("NFS: write (%s/%Ld %d@%Ld)", | |
957 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
958 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
959 | req->wb_bytes, | |
960 | (long long)req_offset(req)); | |
961 | ||
788e7a89 TM |
962 | if (nfs_writeback_done(task, data) != 0) |
963 | return; | |
964 | ||
965 | if (task->tk_status < 0) { | |
a301b777 | 966 | nfs_set_pageerror(page); |
788e7a89 TM |
967 | req->wb_context->error = task->tk_status; |
968 | dprintk(", error = %d\n", task->tk_status); | |
8e821cad | 969 | goto out; |
1da177e4 LT |
970 | } |
971 | ||
8e821cad TM |
972 | if (nfs_write_need_commit(data)) { |
973 | spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; | |
974 | ||
975 | spin_lock(req_lock); | |
976 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
977 | /* Do nothing we need to resend the writes */ | |
978 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
979 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
980 | dprintk(" defer commit\n"); | |
981 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
982 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
983 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
984 | dprintk(" server reboot detected\n"); | |
985 | } | |
986 | spin_unlock(req_lock); | |
987 | } else | |
988 | dprintk(" OK\n"); | |
989 | ||
990 | out: | |
1da177e4 LT |
991 | if (atomic_dec_and_test(&req->wb_complete)) |
992 | nfs_writepage_release(req); | |
993 | } | |
994 | ||
788e7a89 TM |
995 | static const struct rpc_call_ops nfs_write_partial_ops = { |
996 | .rpc_call_done = nfs_writeback_done_partial, | |
997 | .rpc_release = nfs_writedata_release, | |
998 | }; | |
999 | ||
1da177e4 LT |
1000 | /* |
1001 | * Handle a write reply that flushes a whole page. | |
1002 | * | |
1003 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1004 | * writebacks since the page->count is kept > 1 for as long | |
1005 | * as the page has a write request pending. | |
1006 | */ | |
788e7a89 | 1007 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1008 | { |
788e7a89 | 1009 | struct nfs_write_data *data = calldata; |
1da177e4 LT |
1010 | struct nfs_page *req; |
1011 | struct page *page; | |
1012 | ||
788e7a89 TM |
1013 | if (nfs_writeback_done(task, data) != 0) |
1014 | return; | |
1015 | ||
1da177e4 LT |
1016 | /* Update attributes as result of writeback. */ |
1017 | while (!list_empty(&data->pages)) { | |
1018 | req = nfs_list_entry(data->pages.next); | |
1019 | nfs_list_remove_request(req); | |
1020 | page = req->wb_page; | |
1021 | ||
1022 | dprintk("NFS: write (%s/%Ld %d@%Ld)", | |
1023 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
1024 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1025 | req->wb_bytes, | |
1026 | (long long)req_offset(req)); | |
1027 | ||
788e7a89 | 1028 | if (task->tk_status < 0) { |
a301b777 | 1029 | nfs_set_pageerror(page); |
788e7a89 | 1030 | req->wb_context->error = task->tk_status; |
788e7a89 | 1031 | dprintk(", error = %d\n", task->tk_status); |
8e821cad | 1032 | goto remove_request; |
1da177e4 | 1033 | } |
1da177e4 | 1034 | |
8e821cad TM |
1035 | if (nfs_write_need_commit(data)) { |
1036 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1037 | nfs_mark_request_commit(req); | |
1038 | nfs_end_page_writeback(page); | |
1039 | dprintk(" marked for commit\n"); | |
1da177e4 LT |
1040 | goto next; |
1041 | } | |
8e821cad TM |
1042 | dprintk(" OK\n"); |
1043 | remove_request: | |
1044 | nfs_end_page_writeback(page); | |
1da177e4 | 1045 | nfs_inode_remove_request(req); |
1da177e4 | 1046 | next: |
c6a556b8 | 1047 | nfs_clear_page_writeback(req); |
1da177e4 LT |
1048 | } |
1049 | } | |
1050 | ||
788e7a89 TM |
1051 | static const struct rpc_call_ops nfs_write_full_ops = { |
1052 | .rpc_call_done = nfs_writeback_done_full, | |
1053 | .rpc_release = nfs_writedata_release, | |
1054 | }; | |
1055 | ||
1056 | ||
1da177e4 LT |
1057 | /* |
1058 | * This function is called when the WRITE call is complete. | |
1059 | */ | |
462d5b32 | 1060 | int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1061 | { |
1da177e4 LT |
1062 | struct nfs_writeargs *argp = &data->args; |
1063 | struct nfs_writeres *resp = &data->res; | |
788e7a89 | 1064 | int status; |
1da177e4 | 1065 | |
a3f565b1 | 1066 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1067 | task->tk_pid, task->tk_status); |
1068 | ||
f551e44f CL |
1069 | /* |
1070 | * ->write_done will attempt to use post-op attributes to detect | |
1071 | * conflicting writes by other clients. A strict interpretation | |
1072 | * of close-to-open would allow us to continue caching even if | |
1073 | * another writer had changed the file, but some applications | |
1074 | * depend on tighter cache coherency when writing. | |
1075 | */ | |
788e7a89 TM |
1076 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1077 | if (status != 0) | |
1078 | return status; | |
91d5b470 CL |
1079 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1080 | ||
1da177e4 LT |
1081 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1082 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1083 | /* We tried a write call, but the server did not | |
1084 | * commit data to stable storage even though we | |
1085 | * requested it. | |
1086 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1087 | * the server reports NFS_DATA_SYNC, but performs | |
1088 | * NFS_FILE_SYNC. We therefore implement this checking | |
1089 | * as a dprintk() in order to avoid filling syslog. | |
1090 | */ | |
1091 | static unsigned long complain; | |
1092 | ||
1093 | if (time_before(complain, jiffies)) { | |
1094 | dprintk("NFS: faulty NFS server %s:" | |
1095 | " (committed = %d) != (stable = %d)\n", | |
54ceac45 | 1096 | NFS_SERVER(data->inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1097 | resp->verf->committed, argp->stable); |
1098 | complain = jiffies + 300 * HZ; | |
1099 | } | |
1100 | } | |
1101 | #endif | |
1102 | /* Is this a short write? */ | |
1103 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1104 | static unsigned long complain; | |
1105 | ||
91d5b470 CL |
1106 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1107 | ||
1da177e4 LT |
1108 | /* Has the server at least made some progress? */ |
1109 | if (resp->count != 0) { | |
1110 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1111 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1112 | /* Resend from where the server left off */ | |
1113 | argp->offset += resp->count; | |
1114 | argp->pgbase += resp->count; | |
1115 | argp->count -= resp->count; | |
1116 | } else { | |
1117 | /* Resend as a stable write in order to avoid | |
1118 | * headaches in the case of a server crash. | |
1119 | */ | |
1120 | argp->stable = NFS_FILE_SYNC; | |
1121 | } | |
1122 | rpc_restart_call(task); | |
788e7a89 | 1123 | return -EAGAIN; |
1da177e4 LT |
1124 | } |
1125 | if (time_before(complain, jiffies)) { | |
1126 | printk(KERN_WARNING | |
1127 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1128 | argp->count); | |
1129 | complain = jiffies + 300 * HZ; | |
1130 | } | |
1131 | /* Can't do anything about it except throw an error. */ | |
1132 | task->tk_status = -EIO; | |
1133 | } | |
788e7a89 | 1134 | return 0; |
1da177e4 LT |
1135 | } |
1136 | ||
1137 | ||
1138 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
963d8fe5 | 1139 | void nfs_commit_release(void *wdata) |
1da177e4 | 1140 | { |
1da177e4 LT |
1141 | nfs_commit_free(wdata); |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * Set up the argument/result storage required for the RPC call. | |
1146 | */ | |
1147 | static void nfs_commit_rpcsetup(struct list_head *head, | |
788e7a89 TM |
1148 | struct nfs_write_data *data, |
1149 | int how) | |
1da177e4 | 1150 | { |
3da28eb1 | 1151 | struct nfs_page *first; |
1da177e4 | 1152 | struct inode *inode; |
788e7a89 | 1153 | int flags; |
1da177e4 LT |
1154 | |
1155 | /* Set up the RPC argument and reply structs | |
1156 | * NB: take care not to mess about with data->commit et al. */ | |
1157 | ||
1158 | list_splice_init(head, &data->pages); | |
1159 | first = nfs_list_entry(data->pages.next); | |
1da177e4 LT |
1160 | inode = first->wb_context->dentry->d_inode; |
1161 | ||
1da177e4 LT |
1162 | data->inode = inode; |
1163 | data->cred = first->wb_context->cred; | |
1164 | ||
1165 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1166 | /* Note: we always request a commit of the entire inode */ |
1167 | data->args.offset = 0; | |
1168 | data->args.count = 0; | |
1169 | data->res.count = 0; | |
1da177e4 LT |
1170 | data->res.fattr = &data->fattr; |
1171 | data->res.verf = &data->verf; | |
0e574af1 | 1172 | nfs_fattr_init(&data->fattr); |
788e7a89 TM |
1173 | |
1174 | /* Set up the initial task struct. */ | |
1175 | flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
1176 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); | |
1da177e4 LT |
1177 | NFS_PROTO(inode)->commit_setup(data, how); |
1178 | ||
1179 | data->task.tk_priority = flush_task_priority(how); | |
1180 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 | 1181 | |
a3f565b1 | 1182 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); |
1da177e4 LT |
1183 | } |
1184 | ||
1185 | /* | |
1186 | * Commit dirty pages | |
1187 | */ | |
1188 | static int | |
40859d7e | 1189 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1190 | { |
1191 | struct nfs_write_data *data; | |
1192 | struct nfs_page *req; | |
1193 | ||
e9f7bee1 | 1194 | data = nfs_commit_alloc(); |
1da177e4 LT |
1195 | |
1196 | if (!data) | |
1197 | goto out_bad; | |
1198 | ||
1199 | /* Set up the argument struct */ | |
1200 | nfs_commit_rpcsetup(head, data, how); | |
1201 | ||
1202 | nfs_execute_write(data); | |
1203 | return 0; | |
1204 | out_bad: | |
1205 | while (!list_empty(head)) { | |
1206 | req = nfs_list_entry(head->next); | |
1207 | nfs_list_remove_request(req); | |
1208 | nfs_mark_request_commit(req); | |
83715ad5 | 1209 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
5c2d97cb | 1210 | nfs_clear_page_writeback(req); |
1da177e4 LT |
1211 | } |
1212 | return -ENOMEM; | |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * COMMIT call returned | |
1217 | */ | |
788e7a89 | 1218 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1219 | { |
963d8fe5 | 1220 | struct nfs_write_data *data = calldata; |
1da177e4 | 1221 | struct nfs_page *req; |
1da177e4 | 1222 | |
a3f565b1 | 1223 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1224 | task->tk_pid, task->tk_status); |
1225 | ||
788e7a89 TM |
1226 | /* Call the NFS version-specific code */ |
1227 | if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) | |
1228 | return; | |
1229 | ||
1da177e4 LT |
1230 | while (!list_empty(&data->pages)) { |
1231 | req = nfs_list_entry(data->pages.next); | |
1232 | nfs_list_remove_request(req); | |
612c9384 | 1233 | clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); |
fd39fc85 | 1234 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1da177e4 LT |
1235 | |
1236 | dprintk("NFS: commit (%s/%Ld %d@%Ld)", | |
1237 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
1238 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1239 | req->wb_bytes, | |
1240 | (long long)req_offset(req)); | |
1241 | if (task->tk_status < 0) { | |
1242 | req->wb_context->error = task->tk_status; | |
1243 | nfs_inode_remove_request(req); | |
1244 | dprintk(", error = %d\n", task->tk_status); | |
1245 | goto next; | |
1246 | } | |
1247 | ||
1248 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1249 | * returned by the server against all stored verfs. */ | |
1250 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1251 | /* We have a match */ | |
1252 | nfs_inode_remove_request(req); | |
1253 | dprintk(" OK\n"); | |
1254 | goto next; | |
1255 | } | |
1256 | /* We have a mismatch. Write the page again */ | |
1257 | dprintk(" mismatch\n"); | |
61822ab5 | 1258 | nfs_redirty_request(req); |
1da177e4 | 1259 | next: |
c6a556b8 | 1260 | nfs_clear_page_writeback(req); |
1da177e4 | 1261 | } |
1da177e4 | 1262 | } |
788e7a89 TM |
1263 | |
1264 | static const struct rpc_call_ops nfs_commit_ops = { | |
1265 | .rpc_call_done = nfs_commit_done, | |
1266 | .rpc_release = nfs_commit_release, | |
1267 | }; | |
1da177e4 | 1268 | |
3da28eb1 | 1269 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 LT |
1270 | { |
1271 | struct nfs_inode *nfsi = NFS_I(inode); | |
1272 | LIST_HEAD(head); | |
7d46a49f | 1273 | int res; |
1da177e4 LT |
1274 | |
1275 | spin_lock(&nfsi->req_lock); | |
3da28eb1 TM |
1276 | res = nfs_scan_commit(inode, &head, 0, 0); |
1277 | spin_unlock(&nfsi->req_lock); | |
1da177e4 | 1278 | if (res) { |
7d46a49f | 1279 | int error = nfs_commit_list(inode, &head, how); |
3da28eb1 TM |
1280 | if (error < 0) |
1281 | return error; | |
1282 | } | |
1da177e4 LT |
1283 | return res; |
1284 | } | |
c63c7b05 TM |
1285 | #else |
1286 | static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) | |
1287 | { | |
1288 | return 0; | |
1289 | } | |
1da177e4 LT |
1290 | #endif |
1291 | ||
1c75950b | 1292 | long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) |
1da177e4 | 1293 | { |
1c75950b | 1294 | struct inode *inode = mapping->host; |
c42de9dd | 1295 | struct nfs_inode *nfsi = NFS_I(inode); |
1c75950b TM |
1296 | unsigned long idx_start, idx_end; |
1297 | unsigned int npages = 0; | |
c42de9dd | 1298 | LIST_HEAD(head); |
70b9ecbd | 1299 | int nocommit = how & FLUSH_NOCOMMIT; |
3f442547 | 1300 | long pages, ret; |
1da177e4 | 1301 | |
1c75950b TM |
1302 | /* FIXME */ |
1303 | if (wbc->range_cyclic) | |
1304 | idx_start = 0; | |
1305 | else { | |
1306 | idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
1307 | idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
1308 | if (idx_end > idx_start) { | |
1309 | unsigned long l_npages = 1 + idx_end - idx_start; | |
1310 | npages = l_npages; | |
1311 | if (sizeof(npages) != sizeof(l_npages) && | |
1312 | (unsigned long)npages != l_npages) | |
1313 | npages = 0; | |
1314 | } | |
1315 | } | |
c42de9dd TM |
1316 | how &= ~FLUSH_NOCOMMIT; |
1317 | spin_lock(&nfsi->req_lock); | |
1da177e4 | 1318 | do { |
1c75950b | 1319 | wbc->pages_skipped = 0; |
c42de9dd TM |
1320 | ret = nfs_wait_on_requests_locked(inode, idx_start, npages); |
1321 | if (ret != 0) | |
70b9ecbd | 1322 | continue; |
c42de9dd TM |
1323 | if (nocommit) |
1324 | break; | |
d2ccddf0 | 1325 | pages = nfs_scan_commit(inode, &head, idx_start, npages); |
1c75950b TM |
1326 | if (pages == 0) { |
1327 | if (wbc->pages_skipped != 0) | |
1328 | continue; | |
c42de9dd | 1329 | break; |
1c75950b | 1330 | } |
d2ccddf0 TM |
1331 | if (how & FLUSH_INVALIDATE) { |
1332 | spin_unlock(&nfsi->req_lock); | |
83715ad5 | 1333 | nfs_cancel_commit_list(&head); |
e8e058e8 | 1334 | ret = pages; |
d2ccddf0 TM |
1335 | spin_lock(&nfsi->req_lock); |
1336 | continue; | |
1337 | } | |
1338 | pages += nfs_scan_commit(inode, &head, 0, 0); | |
c42de9dd TM |
1339 | spin_unlock(&nfsi->req_lock); |
1340 | ret = nfs_commit_list(inode, &head, how); | |
1341 | spin_lock(&nfsi->req_lock); | |
1342 | } while (ret >= 0); | |
1343 | spin_unlock(&nfsi->req_lock); | |
1344 | return ret; | |
1da177e4 LT |
1345 | } |
1346 | ||
1c75950b TM |
1347 | /* |
1348 | * flush the inode to disk. | |
1349 | */ | |
1350 | int nfs_wb_all(struct inode *inode) | |
1351 | { | |
1352 | struct address_space *mapping = inode->i_mapping; | |
1353 | struct writeback_control wbc = { | |
1354 | .bdi = mapping->backing_dev_info, | |
1355 | .sync_mode = WB_SYNC_ALL, | |
1356 | .nr_to_write = LONG_MAX, | |
61822ab5 | 1357 | .for_writepages = 1, |
1c75950b TM |
1358 | .range_cyclic = 1, |
1359 | }; | |
1360 | int ret; | |
1361 | ||
c63c7b05 | 1362 | ret = nfs_writepages(mapping, &wbc); |
61822ab5 TM |
1363 | if (ret < 0) |
1364 | goto out; | |
1c75950b TM |
1365 | ret = nfs_sync_mapping_wait(mapping, &wbc, 0); |
1366 | if (ret >= 0) | |
1367 | return 0; | |
61822ab5 | 1368 | out: |
e507d9eb | 1369 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
1c75950b TM |
1370 | return ret; |
1371 | } | |
1372 | ||
1373 | int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how) | |
1374 | { | |
1375 | struct writeback_control wbc = { | |
1376 | .bdi = mapping->backing_dev_info, | |
1377 | .sync_mode = WB_SYNC_ALL, | |
1378 | .nr_to_write = LONG_MAX, | |
1379 | .range_start = range_start, | |
1380 | .range_end = range_end, | |
61822ab5 | 1381 | .for_writepages = 1, |
1c75950b TM |
1382 | }; |
1383 | int ret; | |
1384 | ||
c63c7b05 TM |
1385 | ret = nfs_writepages(mapping, &wbc); |
1386 | if (ret < 0) | |
1387 | goto out; | |
1c75950b TM |
1388 | ret = nfs_sync_mapping_wait(mapping, &wbc, how); |
1389 | if (ret >= 0) | |
1390 | return 0; | |
61822ab5 | 1391 | out: |
e507d9eb | 1392 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
1c75950b TM |
1393 | return ret; |
1394 | } | |
1395 | ||
61822ab5 | 1396 | int nfs_wb_page_priority(struct inode *inode, struct page *page, int how) |
1c75950b TM |
1397 | { |
1398 | loff_t range_start = page_offset(page); | |
1399 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf TM |
1400 | struct writeback_control wbc = { |
1401 | .bdi = page->mapping->backing_dev_info, | |
1402 | .sync_mode = WB_SYNC_ALL, | |
1403 | .nr_to_write = LONG_MAX, | |
1404 | .range_start = range_start, | |
1405 | .range_end = range_end, | |
1406 | }; | |
1407 | int ret; | |
1c75950b | 1408 | |
4d770ccf | 1409 | BUG_ON(!PageLocked(page)); |
c63c7b05 | 1410 | if (clear_page_dirty_for_io(page)) { |
4d770ccf TM |
1411 | ret = nfs_writepage_locked(page, &wbc); |
1412 | if (ret < 0) | |
1413 | goto out; | |
1414 | } | |
f40313ac TM |
1415 | if (!PagePrivate(page)) |
1416 | return 0; | |
4d770ccf TM |
1417 | ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); |
1418 | if (ret >= 0) | |
1419 | return 0; | |
1420 | out: | |
e507d9eb | 1421 | __mark_inode_dirty(inode, I_DIRTY_PAGES); |
4d770ccf | 1422 | return ret; |
1c75950b TM |
1423 | } |
1424 | ||
1425 | /* | |
1426 | * Write back all requests on one page - we do this before reading it. | |
1427 | */ | |
1428 | int nfs_wb_page(struct inode *inode, struct page* page) | |
1429 | { | |
4d770ccf | 1430 | return nfs_wb_page_priority(inode, page, FLUSH_STABLE); |
1c75950b TM |
1431 | } |
1432 | ||
1a54533e TM |
1433 | int nfs_set_page_dirty(struct page *page) |
1434 | { | |
d585158b TM |
1435 | struct address_space *mapping = page->mapping; |
1436 | struct inode *inode; | |
1437 | spinlock_t *req_lock; | |
1a54533e | 1438 | struct nfs_page *req; |
2b82f190 | 1439 | int ret; |
1a54533e | 1440 | |
d585158b TM |
1441 | if (!mapping) |
1442 | goto out_raced; | |
1443 | inode = mapping->host; | |
1444 | if (!inode) | |
1445 | goto out_raced; | |
1446 | req_lock = &NFS_I(inode)->req_lock; | |
2b82f190 TM |
1447 | spin_lock(req_lock); |
1448 | req = nfs_page_find_request_locked(page); | |
1a54533e TM |
1449 | if (req != NULL) { |
1450 | /* Mark any existing write requests for flushing */ | |
2b82f190 TM |
1451 | ret = !test_and_set_bit(PG_NEED_FLUSH, &req->wb_flags); |
1452 | spin_unlock(req_lock); | |
1a54533e | 1453 | nfs_release_request(req); |
2b82f190 | 1454 | return ret; |
1a54533e | 1455 | } |
2b82f190 TM |
1456 | ret = __set_page_dirty_nobuffers(page); |
1457 | spin_unlock(req_lock); | |
1458 | return ret; | |
d585158b TM |
1459 | out_raced: |
1460 | return !TestSetPageDirty(page); | |
1a54533e TM |
1461 | } |
1462 | ||
1c75950b | 1463 | |
f7b422b1 | 1464 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1465 | { |
1466 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1467 | sizeof(struct nfs_write_data), | |
1468 | 0, SLAB_HWCACHE_ALIGN, | |
1469 | NULL, NULL); | |
1470 | if (nfs_wdata_cachep == NULL) | |
1471 | return -ENOMEM; | |
1472 | ||
93d2341c MD |
1473 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1474 | nfs_wdata_cachep); | |
1da177e4 LT |
1475 | if (nfs_wdata_mempool == NULL) |
1476 | return -ENOMEM; | |
1477 | ||
93d2341c MD |
1478 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1479 | nfs_wdata_cachep); | |
1da177e4 LT |
1480 | if (nfs_commit_mempool == NULL) |
1481 | return -ENOMEM; | |
1482 | ||
89a09141 PZ |
1483 | /* |
1484 | * NFS congestion size, scale with available memory. | |
1485 | * | |
1486 | * 64MB: 8192k | |
1487 | * 128MB: 11585k | |
1488 | * 256MB: 16384k | |
1489 | * 512MB: 23170k | |
1490 | * 1GB: 32768k | |
1491 | * 2GB: 46340k | |
1492 | * 4GB: 65536k | |
1493 | * 8GB: 92681k | |
1494 | * 16GB: 131072k | |
1495 | * | |
1496 | * This allows larger machines to have larger/more transfers. | |
1497 | * Limit the default to 256M | |
1498 | */ | |
1499 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1500 | if (nfs_congestion_kb > 256*1024) | |
1501 | nfs_congestion_kb = 256*1024; | |
1502 | ||
1da177e4 LT |
1503 | return 0; |
1504 | } | |
1505 | ||
266bee88 | 1506 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1507 | { |
1508 | mempool_destroy(nfs_commit_mempool); | |
1509 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1510 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1511 | } |
1512 |