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 AM |
22 | #include <linux/backing-dev.h> |
23 | ||
1da177e4 | 24 | #include <asm/uaccess.h> |
1da177e4 LT |
25 | |
26 | #include "delegation.h" | |
49a70f27 | 27 | #include "internal.h" |
91d5b470 | 28 | #include "iostat.h" |
def6ed7e | 29 | #include "nfs4_fs.h" |
074cc1de | 30 | #include "fscache.h" |
94ad1c80 | 31 | #include "pnfs.h" |
1da177e4 LT |
32 | |
33 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
34 | ||
35 | #define MIN_POOL_WRITE (32) | |
36 | #define MIN_POOL_COMMIT (4) | |
37 | ||
38 | /* | |
39 | * Local function declarations | |
40 | */ | |
c63c7b05 TM |
41 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
42 | struct inode *inode, int ioflags); | |
f8512ad0 | 43 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 TM |
44 | static const struct rpc_call_ops nfs_write_partial_ops; |
45 | static const struct rpc_call_ops nfs_write_full_ops; | |
46 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 | 47 | |
e18b890b | 48 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 49 | static mempool_t *nfs_wdata_mempool; |
1da177e4 LT |
50 | static mempool_t *nfs_commit_mempool; |
51 | ||
c9d8f89d | 52 | struct nfs_write_data *nfs_commitdata_alloc(void) |
1da177e4 | 53 | { |
e6b4f8da | 54 | struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 55 | |
1da177e4 LT |
56 | if (p) { |
57 | memset(p, 0, sizeof(*p)); | |
58 | INIT_LIST_HEAD(&p->pages); | |
59 | } | |
60 | return p; | |
61 | } | |
62 | ||
5e4424af | 63 | void nfs_commit_free(struct nfs_write_data *p) |
1da177e4 | 64 | { |
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 | ||
8d5658c9 | 70 | struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) |
3feb2d49 | 71 | { |
e6b4f8da | 72 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
73 | |
74 | if (p) { | |
75 | memset(p, 0, sizeof(*p)); | |
76 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 77 | p->npages = pagecount; |
0d0b5cb3 CL |
78 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
79 | p->pagevec = p->page_array; | |
3feb2d49 | 80 | else { |
0d0b5cb3 CL |
81 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
82 | if (!p->pagevec) { | |
3feb2d49 TM |
83 | mempool_free(p, nfs_wdata_mempool); |
84 | p = NULL; | |
85 | } | |
86 | } | |
87 | } | |
88 | return p; | |
89 | } | |
90 | ||
1ae88b2e | 91 | void nfs_writedata_free(struct nfs_write_data *p) |
3feb2d49 TM |
92 | { |
93 | if (p && (p->pagevec != &p->page_array[0])) | |
94 | kfree(p->pagevec); | |
95 | mempool_free(p, nfs_wdata_mempool); | |
96 | } | |
97 | ||
1ae88b2e | 98 | static void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 99 | { |
383ba719 | 100 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
101 | nfs_writedata_free(wdata); |
102 | } | |
103 | ||
7b159fc1 TM |
104 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
105 | { | |
106 | ctx->error = error; | |
107 | smp_wmb(); | |
108 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
109 | } | |
110 | ||
277459d2 TM |
111 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
112 | { | |
113 | struct nfs_page *req = NULL; | |
114 | ||
115 | if (PagePrivate(page)) { | |
116 | req = (struct nfs_page *)page_private(page); | |
117 | if (req != NULL) | |
c03b4024 | 118 | kref_get(&req->wb_kref); |
277459d2 TM |
119 | } |
120 | return req; | |
121 | } | |
122 | ||
123 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
124 | { | |
587142f8 | 125 | struct inode *inode = page->mapping->host; |
277459d2 | 126 | struct nfs_page *req = NULL; |
277459d2 | 127 | |
587142f8 | 128 | spin_lock(&inode->i_lock); |
277459d2 | 129 | req = nfs_page_find_request_locked(page); |
587142f8 | 130 | spin_unlock(&inode->i_lock); |
277459d2 TM |
131 | return req; |
132 | } | |
133 | ||
1da177e4 LT |
134 | /* Adjust the file length if we're writing beyond the end */ |
135 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
136 | { | |
137 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
138 | loff_t end, i_size; |
139 | pgoff_t end_index; | |
1da177e4 | 140 | |
a3d01454 TM |
141 | spin_lock(&inode->i_lock); |
142 | i_size = i_size_read(inode); | |
143 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 144 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 145 | goto out; |
1da177e4 LT |
146 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
147 | if (i_size >= end) | |
a3d01454 | 148 | goto out; |
1da177e4 | 149 | i_size_write(inode, end); |
a3d01454 TM |
150 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
151 | out: | |
152 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
153 | } |
154 | ||
a301b777 TM |
155 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
156 | static void nfs_set_pageerror(struct page *page) | |
157 | { | |
158 | SetPageError(page); | |
159 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
160 | } | |
161 | ||
1da177e4 LT |
162 | /* We can set the PG_uptodate flag if we see that a write request |
163 | * covers the full page. | |
164 | */ | |
165 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
166 | { | |
1da177e4 LT |
167 | if (PageUptodate(page)) |
168 | return; | |
169 | if (base != 0) | |
170 | return; | |
49a70f27 | 171 | if (count != nfs_page_length(page)) |
1da177e4 | 172 | return; |
49a70f27 | 173 | SetPageUptodate(page); |
1da177e4 LT |
174 | } |
175 | ||
1da177e4 LT |
176 | static int wb_priority(struct writeback_control *wbc) |
177 | { | |
178 | if (wbc->for_reclaim) | |
c63c7b05 | 179 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 180 | if (wbc->for_kupdate || wbc->for_background) |
1da177e4 LT |
181 | return FLUSH_LOWPRI; |
182 | return 0; | |
183 | } | |
184 | ||
89a09141 PZ |
185 | /* |
186 | * NFS congestion control | |
187 | */ | |
188 | ||
189 | int nfs_congestion_kb; | |
190 | ||
191 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
192 | #define NFS_CONGESTION_OFF_THRESH \ | |
193 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
194 | ||
5a6d41b3 | 195 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 196 | { |
5a6d41b3 TM |
197 | int ret = test_set_page_writeback(page); |
198 | ||
199 | if (!ret) { | |
89a09141 PZ |
200 | struct inode *inode = page->mapping->host; |
201 | struct nfs_server *nfss = NFS_SERVER(inode); | |
202 | ||
a6305ddb | 203 | page_cache_get(page); |
277866a0 | 204 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
205 | NFS_CONGESTION_ON_THRESH) { |
206 | set_bdi_congested(&nfss->backing_dev_info, | |
207 | BLK_RW_ASYNC); | |
208 | } | |
89a09141 | 209 | } |
5a6d41b3 | 210 | return ret; |
89a09141 PZ |
211 | } |
212 | ||
213 | static void nfs_end_page_writeback(struct page *page) | |
214 | { | |
215 | struct inode *inode = page->mapping->host; | |
216 | struct nfs_server *nfss = NFS_SERVER(inode); | |
217 | ||
218 | end_page_writeback(page); | |
a6305ddb | 219 | page_cache_release(page); |
c4dc4bee | 220 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 221 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
222 | } |
223 | ||
cfb506e1 | 224 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 225 | { |
587142f8 | 226 | struct inode *inode = page->mapping->host; |
e261f51f | 227 | struct nfs_page *req; |
e261f51f TM |
228 | int ret; |
229 | ||
587142f8 | 230 | spin_lock(&inode->i_lock); |
074cc1de | 231 | for (;;) { |
e261f51f | 232 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
233 | if (req == NULL) |
234 | break; | |
acee478a | 235 | if (nfs_set_page_tag_locked(req)) |
e261f51f TM |
236 | break; |
237 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
acee478a | 238 | * then the call to nfs_set_page_tag_locked() will always |
e261f51f TM |
239 | * succeed provided that someone hasn't already marked the |
240 | * request as dirty (in which case we don't care). | |
241 | */ | |
587142f8 | 242 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
243 | if (!nonblock) |
244 | ret = nfs_wait_on_request(req); | |
245 | else | |
246 | ret = -EAGAIN; | |
e261f51f TM |
247 | nfs_release_request(req); |
248 | if (ret != 0) | |
074cc1de | 249 | return ERR_PTR(ret); |
587142f8 | 250 | spin_lock(&inode->i_lock); |
e261f51f | 251 | } |
587142f8 | 252 | spin_unlock(&inode->i_lock); |
074cc1de TM |
253 | return req; |
254 | } | |
255 | ||
256 | /* | |
257 | * Find an associated nfs write request, and prepare to flush it out | |
258 | * May return an error if the user signalled nfs_wait_on_request(). | |
259 | */ | |
260 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 261 | struct page *page, bool nonblock) |
074cc1de TM |
262 | { |
263 | struct nfs_page *req; | |
264 | int ret = 0; | |
265 | ||
cfb506e1 | 266 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
267 | if (!req) |
268 | goto out; | |
269 | ret = PTR_ERR(req); | |
270 | if (IS_ERR(req)) | |
271 | goto out; | |
272 | ||
273 | ret = nfs_set_page_writeback(page); | |
274 | BUG_ON(ret != 0); | |
275 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
276 | ||
f8512ad0 FI |
277 | if (!nfs_pageio_add_request(pgio, req)) { |
278 | nfs_redirty_request(req); | |
074cc1de | 279 | ret = pgio->pg_error; |
f8512ad0 | 280 | } |
074cc1de TM |
281 | out: |
282 | return ret; | |
e261f51f TM |
283 | } |
284 | ||
f758c885 | 285 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 286 | { |
1da177e4 | 287 | struct inode *inode = page->mapping->host; |
cfb506e1 | 288 | int ret; |
1da177e4 | 289 | |
91d5b470 CL |
290 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
291 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
292 | ||
7fe7f848 | 293 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 294 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
295 | if (ret == -EAGAIN) { |
296 | redirty_page_for_writepage(wbc, page); | |
297 | ret = 0; | |
298 | } | |
299 | return ret; | |
f758c885 | 300 | } |
7fe7f848 | 301 | |
f758c885 TM |
302 | /* |
303 | * Write an mmapped page to the server. | |
304 | */ | |
305 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
306 | { | |
307 | struct nfs_pageio_descriptor pgio; | |
308 | int err; | |
49a70f27 | 309 | |
f758c885 TM |
310 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
311 | err = nfs_do_writepage(page, wbc, &pgio); | |
312 | nfs_pageio_complete(&pgio); | |
313 | if (err < 0) | |
314 | return err; | |
315 | if (pgio.pg_error < 0) | |
316 | return pgio.pg_error; | |
317 | return 0; | |
4d770ccf TM |
318 | } |
319 | ||
320 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
321 | { | |
f758c885 | 322 | int ret; |
4d770ccf | 323 | |
f758c885 | 324 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 325 | unlock_page(page); |
f758c885 TM |
326 | return ret; |
327 | } | |
328 | ||
329 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
330 | { | |
331 | int ret; | |
332 | ||
333 | ret = nfs_do_writepage(page, wbc, data); | |
334 | unlock_page(page); | |
335 | return ret; | |
1da177e4 LT |
336 | } |
337 | ||
1da177e4 LT |
338 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
339 | { | |
1da177e4 | 340 | struct inode *inode = mapping->host; |
72cb77f4 | 341 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 342 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
343 | int err; |
344 | ||
72cb77f4 TM |
345 | /* Stop dirtying of new pages while we sync */ |
346 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
347 | nfs_wait_bit_killable, TASK_KILLABLE); | |
348 | if (err) | |
349 | goto out_err; | |
350 | ||
91d5b470 CL |
351 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
352 | ||
c63c7b05 | 353 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 354 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 355 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
356 | |
357 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
358 | smp_mb__after_clear_bit(); | |
359 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
360 | ||
f758c885 | 361 | if (err < 0) |
72cb77f4 TM |
362 | goto out_err; |
363 | err = pgio.pg_error; | |
364 | if (err < 0) | |
365 | goto out_err; | |
c63c7b05 | 366 | return 0; |
72cb77f4 TM |
367 | out_err: |
368 | return err; | |
1da177e4 LT |
369 | } |
370 | ||
371 | /* | |
372 | * Insert a write request into an inode | |
373 | */ | |
e7d39069 | 374 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
375 | { |
376 | struct nfs_inode *nfsi = NFS_I(inode); | |
377 | int error; | |
378 | ||
e7d39069 TM |
379 | error = radix_tree_preload(GFP_NOFS); |
380 | if (error != 0) | |
381 | goto out; | |
382 | ||
383 | /* Lock the request! */ | |
384 | nfs_lock_request_dontget(req); | |
385 | ||
386 | spin_lock(&inode->i_lock); | |
1da177e4 | 387 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); |
27852596 | 388 | BUG_ON(error); |
1da177e4 LT |
389 | if (!nfsi->npages) { |
390 | igrab(inode); | |
1da177e4 LT |
391 | if (nfs_have_delegation(inode, FMODE_WRITE)) |
392 | nfsi->change_attr++; | |
393 | } | |
2df485a7 | 394 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 395 | SetPagePrivate(req->wb_page); |
277459d2 | 396 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 397 | nfsi->npages++; |
c03b4024 | 398 | kref_get(&req->wb_kref); |
27852596 NP |
399 | radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, |
400 | NFS_PAGE_TAG_LOCKED); | |
e7d39069 TM |
401 | spin_unlock(&inode->i_lock); |
402 | radix_tree_preload_end(); | |
403 | out: | |
404 | return error; | |
1da177e4 LT |
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 | { | |
88be9f99 | 412 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
413 | struct nfs_inode *nfsi = NFS_I(inode); |
414 | ||
415 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
416 | ||
587142f8 | 417 | spin_lock(&inode->i_lock); |
277459d2 | 418 | set_page_private(req->wb_page, 0); |
deb7d638 | 419 | ClearPagePrivate(req->wb_page); |
2df485a7 | 420 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 LT |
421 | radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); |
422 | nfsi->npages--; | |
423 | if (!nfsi->npages) { | |
587142f8 | 424 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
425 | iput(inode); |
426 | } else | |
587142f8 | 427 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
428 | nfs_release_request(req); |
429 | } | |
430 | ||
61822ab5 | 431 | static void |
6d884e8f | 432 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 433 | { |
61822ab5 | 434 | __set_page_dirty_nobuffers(req->wb_page); |
b80c3cb6 | 435 | __mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC); |
61822ab5 TM |
436 | } |
437 | ||
1da177e4 LT |
438 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
439 | /* | |
440 | * Add a request to the inode's commit list. | |
441 | */ | |
442 | static void | |
443 | nfs_mark_request_commit(struct nfs_page *req) | |
444 | { | |
88be9f99 | 445 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
446 | struct nfs_inode *nfsi = NFS_I(inode); |
447 | ||
587142f8 | 448 | spin_lock(&inode->i_lock); |
e468bae9 | 449 | set_bit(PG_CLEAN, &(req)->wb_flags); |
5c369683 TM |
450 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
451 | req->wb_index, | |
452 | NFS_PAGE_TAG_COMMIT); | |
ff778d02 | 453 | nfsi->ncommit++; |
587142f8 | 454 | spin_unlock(&inode->i_lock); |
fd39fc85 | 455 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 456 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 457 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 458 | } |
8e821cad | 459 | |
e468bae9 TM |
460 | static int |
461 | nfs_clear_request_commit(struct nfs_page *req) | |
462 | { | |
463 | struct page *page = req->wb_page; | |
464 | ||
465 | if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { | |
466 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
467 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
468 | return 1; | |
469 | } | |
470 | return 0; | |
471 | } | |
472 | ||
8e821cad TM |
473 | static inline |
474 | int nfs_write_need_commit(struct nfs_write_data *data) | |
475 | { | |
476 | return data->verf.committed != NFS_FILE_SYNC; | |
477 | } | |
478 | ||
479 | static inline | |
480 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
481 | { | |
e468bae9 | 482 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
8e821cad TM |
483 | nfs_mark_request_commit(req); |
484 | return 1; | |
485 | } | |
486 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
6d884e8f | 487 | nfs_mark_request_dirty(req); |
8e821cad TM |
488 | return 1; |
489 | } | |
490 | return 0; | |
491 | } | |
492 | #else | |
493 | static inline void | |
494 | nfs_mark_request_commit(struct nfs_page *req) | |
495 | { | |
496 | } | |
497 | ||
e468bae9 TM |
498 | static inline int |
499 | nfs_clear_request_commit(struct nfs_page *req) | |
500 | { | |
501 | return 0; | |
502 | } | |
503 | ||
8e821cad TM |
504 | static inline |
505 | int nfs_write_need_commit(struct nfs_write_data *data) | |
506 | { | |
507 | return 0; | |
508 | } | |
509 | ||
510 | static inline | |
511 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
512 | { | |
513 | return 0; | |
514 | } | |
1da177e4 LT |
515 | #endif |
516 | ||
47c62564 | 517 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
518 | static int |
519 | nfs_need_commit(struct nfs_inode *nfsi) | |
520 | { | |
521 | return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); | |
522 | } | |
523 | ||
1da177e4 LT |
524 | /* |
525 | * nfs_scan_commit - Scan an inode for commit requests | |
526 | * @inode: NFS inode to scan | |
527 | * @dst: destination list | |
528 | * @idx_start: lower bound of page->index to scan. | |
529 | * @npages: idx_start + npages sets the upper bound to scan. | |
530 | * | |
531 | * Moves requests from the inode's 'commit' request list. | |
532 | * The requests are *not* checked to ensure that they form a contiguous set. | |
533 | */ | |
534 | static int | |
ca52fec1 | 535 | nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
536 | { |
537 | struct nfs_inode *nfsi = NFS_I(inode); | |
ff778d02 | 538 | int ret; |
3da28eb1 | 539 | |
fb8a1f11 TM |
540 | if (!nfs_need_commit(nfsi)) |
541 | return 0; | |
542 | ||
ff778d02 TM |
543 | ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); |
544 | if (ret > 0) | |
545 | nfsi->ncommit -= ret; | |
2928db1f TM |
546 | if (nfs_need_commit(NFS_I(inode))) |
547 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
ff778d02 | 548 | return ret; |
1da177e4 | 549 | } |
c42de9dd | 550 | #else |
fb8a1f11 TM |
551 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
552 | { | |
553 | return 0; | |
554 | } | |
555 | ||
ca52fec1 | 556 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
c42de9dd TM |
557 | { |
558 | return 0; | |
559 | } | |
1da177e4 LT |
560 | #endif |
561 | ||
1da177e4 | 562 | /* |
e7d39069 TM |
563 | * Search for an existing write request, and attempt to update |
564 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 565 | * |
e7d39069 TM |
566 | * If the attempt fails, then the existing request is flushed out |
567 | * to disk. | |
1da177e4 | 568 | */ |
e7d39069 TM |
569 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
570 | struct page *page, | |
571 | unsigned int offset, | |
572 | unsigned int bytes) | |
1da177e4 | 573 | { |
e7d39069 TM |
574 | struct nfs_page *req; |
575 | unsigned int rqend; | |
576 | unsigned int end; | |
577 | int error; | |
578 | ||
579 | if (!PagePrivate(page)) | |
580 | return NULL; | |
1da177e4 LT |
581 | |
582 | end = offset + bytes; | |
e7d39069 | 583 | spin_lock(&inode->i_lock); |
1da177e4 | 584 | |
1da177e4 | 585 | for (;;) { |
277459d2 | 586 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
587 | if (req == NULL) |
588 | goto out_unlock; | |
589 | ||
590 | rqend = req->wb_offset + req->wb_bytes; | |
591 | /* | |
592 | * Tell the caller to flush out the request if | |
593 | * the offsets are non-contiguous. | |
594 | * Note: nfs_flush_incompatible() will already | |
595 | * have flushed out requests having wrong owners. | |
596 | */ | |
e468bae9 | 597 | if (offset > rqend |
e7d39069 TM |
598 | || end < req->wb_offset) |
599 | goto out_flushme; | |
600 | ||
601 | if (nfs_set_page_tag_locked(req)) | |
1da177e4 | 602 | break; |
1da177e4 | 603 | |
e7d39069 | 604 | /* The request is locked, so wait and then retry */ |
587142f8 | 605 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
606 | error = nfs_wait_on_request(req); |
607 | nfs_release_request(req); | |
608 | if (error != 0) | |
609 | goto out_err; | |
610 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
611 | } |
612 | ||
ff778d02 TM |
613 | if (nfs_clear_request_commit(req) && |
614 | radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, | |
615 | req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) | |
616 | NFS_I(inode)->ncommit--; | |
e468bae9 | 617 | |
1da177e4 LT |
618 | /* Okay, the request matches. Update the region */ |
619 | if (offset < req->wb_offset) { | |
620 | req->wb_offset = offset; | |
621 | req->wb_pgbase = offset; | |
1da177e4 | 622 | } |
1da177e4 LT |
623 | if (end > rqend) |
624 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
625 | else |
626 | req->wb_bytes = rqend - req->wb_offset; | |
627 | out_unlock: | |
628 | spin_unlock(&inode->i_lock); | |
629 | return req; | |
630 | out_flushme: | |
631 | spin_unlock(&inode->i_lock); | |
632 | nfs_release_request(req); | |
633 | error = nfs_wb_page(inode, page); | |
634 | out_err: | |
635 | return ERR_PTR(error); | |
636 | } | |
637 | ||
638 | /* | |
639 | * Try to update an existing write request, or create one if there is none. | |
640 | * | |
641 | * Note: Should always be called with the Page Lock held to prevent races | |
642 | * if we have to add a new request. Also assumes that the caller has | |
643 | * already called nfs_flush_incompatible() if necessary. | |
644 | */ | |
645 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
646 | struct page *page, unsigned int offset, unsigned int bytes) | |
647 | { | |
648 | struct inode *inode = page->mapping->host; | |
649 | struct nfs_page *req; | |
650 | int error; | |
1da177e4 | 651 | |
e7d39069 TM |
652 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
653 | if (req != NULL) | |
654 | goto out; | |
655 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
656 | if (IS_ERR(req)) | |
657 | goto out; | |
658 | error = nfs_inode_add_request(inode, req); | |
659 | if (error != 0) { | |
660 | nfs_release_request(req); | |
661 | req = ERR_PTR(error); | |
662 | } | |
efc91ed0 | 663 | out: |
61e930a9 | 664 | return req; |
1da177e4 LT |
665 | } |
666 | ||
e7d39069 TM |
667 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
668 | unsigned int offset, unsigned int count) | |
669 | { | |
670 | struct nfs_page *req; | |
671 | ||
672 | req = nfs_setup_write_request(ctx, page, offset, count); | |
673 | if (IS_ERR(req)) | |
674 | return PTR_ERR(req); | |
b80c3cb6 | 675 | nfs_mark_request_dirty(req); |
e7d39069 TM |
676 | /* Update file length */ |
677 | nfs_grow_file(page, offset, count); | |
678 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 679 | nfs_mark_request_dirty(req); |
e7d39069 TM |
680 | nfs_clear_page_tag_locked(req); |
681 | return 0; | |
682 | } | |
683 | ||
1da177e4 LT |
684 | int nfs_flush_incompatible(struct file *file, struct page *page) |
685 | { | |
cd3758e3 | 686 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 687 | struct nfs_page *req; |
1a54533e | 688 | int do_flush, status; |
1da177e4 LT |
689 | /* |
690 | * Look for a request corresponding to this page. If there | |
691 | * is one, and it belongs to another file, we flush it out | |
692 | * before we try to copy anything into the page. Do this | |
693 | * due to the lack of an ACCESS-type call in NFSv2. | |
694 | * Also do the same if we find a request from an existing | |
695 | * dropped page. | |
696 | */ | |
1a54533e TM |
697 | do { |
698 | req = nfs_page_find_request(page); | |
699 | if (req == NULL) | |
700 | return 0; | |
f11ac8db TM |
701 | do_flush = req->wb_page != page || req->wb_context != ctx || |
702 | req->wb_lock_context->lockowner != current->files || | |
703 | req->wb_lock_context->pid != current->tgid; | |
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 | ||
5d47a356 TM |
712 | /* |
713 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
714 | * the PageUptodate() flag. In this case, we will need to turn off | |
715 | * write optimisations that depend on the page contents being correct. | |
716 | */ | |
717 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
718 | { | |
719 | return PageUptodate(page) && | |
720 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
721 | } | |
722 | ||
1da177e4 LT |
723 | /* |
724 | * Update and possibly write a cached page of an NFS file. | |
725 | * | |
726 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
727 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
728 | */ | |
729 | int nfs_updatepage(struct file *file, struct page *page, | |
730 | unsigned int offset, unsigned int count) | |
731 | { | |
cd3758e3 | 732 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 733 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
734 | int status = 0; |
735 | ||
91d5b470 CL |
736 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
737 | ||
48186c7d | 738 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
739 | file->f_path.dentry->d_parent->d_name.name, |
740 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 741 | (long long)(page_offset(page) + offset)); |
1da177e4 | 742 | |
1da177e4 | 743 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
744 | * is up to date, it may be more efficient to extend the write |
745 | * to cover the entire page in order to avoid fragmentation | |
746 | * inefficiencies. | |
1da177e4 | 747 | */ |
5d47a356 TM |
748 | if (nfs_write_pageuptodate(page, inode) && |
749 | inode->i_flock == NULL && | |
6b2f3d1f | 750 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 751 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 752 | offset = 0; |
1da177e4 LT |
753 | } |
754 | ||
e21195a7 | 755 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
756 | if (status < 0) |
757 | nfs_set_pageerror(page); | |
1da177e4 | 758 | |
48186c7d | 759 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 760 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
761 | return status; |
762 | } | |
763 | ||
764 | static void nfs_writepage_release(struct nfs_page *req) | |
765 | { | |
a6305ddb | 766 | struct page *page = req->wb_page; |
1da177e4 | 767 | |
a6305ddb | 768 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) |
8e821cad | 769 | nfs_inode_remove_request(req); |
9fd367f0 | 770 | nfs_clear_page_tag_locked(req); |
a6305ddb | 771 | nfs_end_page_writeback(page); |
1da177e4 LT |
772 | } |
773 | ||
3ff7576d | 774 | static int flush_task_priority(int how) |
1da177e4 LT |
775 | { |
776 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
777 | case FLUSH_HIGHPRI: | |
778 | return RPC_PRIORITY_HIGH; | |
779 | case FLUSH_LOWPRI: | |
780 | return RPC_PRIORITY_LOW; | |
781 | } | |
782 | return RPC_PRIORITY_NORMAL; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Set up the argument/result storage required for the RPC call. | |
787 | */ | |
dbae4c73 | 788 | static int nfs_write_rpcsetup(struct nfs_page *req, |
1da177e4 | 789 | struct nfs_write_data *data, |
788e7a89 | 790 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
791 | unsigned int count, unsigned int offset, |
792 | int how) | |
793 | { | |
84115e1c | 794 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
3ff7576d | 795 | int priority = flush_task_priority(how); |
07737691 | 796 | struct rpc_task *task; |
bdc7f021 TM |
797 | struct rpc_message msg = { |
798 | .rpc_argp = &data->args, | |
799 | .rpc_resp = &data->res, | |
800 | .rpc_cred = req->wb_context->cred, | |
801 | }; | |
84115e1c TM |
802 | struct rpc_task_setup task_setup_data = { |
803 | .rpc_client = NFS_CLIENT(inode), | |
07737691 | 804 | .task = &data->task, |
bdc7f021 | 805 | .rpc_message = &msg, |
84115e1c TM |
806 | .callback_ops = call_ops, |
807 | .callback_data = data, | |
101070ca | 808 | .workqueue = nfsiod_workqueue, |
2c61be0a | 809 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 810 | .priority = priority, |
84115e1c | 811 | }; |
2c61be0a | 812 | int ret = 0; |
1da177e4 LT |
813 | |
814 | /* Set up the RPC argument and reply structs | |
815 | * NB: take care not to mess about with data->commit et al. */ | |
816 | ||
817 | data->req = req; | |
88be9f99 | 818 | data->inode = inode = req->wb_context->path.dentry->d_inode; |
bdc7f021 | 819 | data->cred = msg.rpc_cred; |
1da177e4 LT |
820 | |
821 | data->args.fh = NFS_FH(inode); | |
822 | data->args.offset = req_offset(req) + offset; | |
823 | data->args.pgbase = req->wb_pgbase + offset; | |
824 | data->args.pages = data->pagevec; | |
825 | data->args.count = count; | |
383ba719 | 826 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 827 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 TM |
828 | data->args.stable = NFS_UNSTABLE; |
829 | if (how & FLUSH_STABLE) { | |
830 | data->args.stable = NFS_DATA_SYNC; | |
fb8a1f11 | 831 | if (!nfs_need_commit(NFS_I(inode))) |
bdc7f021 TM |
832 | data->args.stable = NFS_FILE_SYNC; |
833 | } | |
1da177e4 LT |
834 | |
835 | data->res.fattr = &data->fattr; | |
836 | data->res.count = count; | |
837 | data->res.verf = &data->verf; | |
0e574af1 | 838 | nfs_fattr_init(&data->fattr); |
1da177e4 | 839 | |
788e7a89 | 840 | /* Set up the initial task struct. */ |
bdc7f021 | 841 | NFS_PROTO(inode)->write_setup(data, &msg); |
1da177e4 | 842 | |
a3f565b1 | 843 | dprintk("NFS: %5u initiated write call " |
48186c7d | 844 | "(req %s/%lld, %u bytes @ offset %llu)\n", |
0bbacc40 | 845 | data->task.tk_pid, |
1da177e4 LT |
846 | inode->i_sb->s_id, |
847 | (long long)NFS_FILEID(inode), | |
848 | count, | |
849 | (unsigned long long)data->args.offset); | |
1da177e4 | 850 | |
07737691 | 851 | task = rpc_run_task(&task_setup_data); |
2c61be0a TM |
852 | if (IS_ERR(task)) { |
853 | ret = PTR_ERR(task); | |
854 | goto out; | |
855 | } | |
856 | if (how & FLUSH_SYNC) { | |
857 | ret = rpc_wait_for_completion_task(task); | |
858 | if (ret == 0) | |
859 | ret = task->tk_status; | |
860 | } | |
dbae4c73 | 861 | rpc_put_task(task); |
2c61be0a TM |
862 | out: |
863 | return ret; | |
1da177e4 LT |
864 | } |
865 | ||
6d884e8f F |
866 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
867 | * call this on each, which will prepare them to be retried on next | |
868 | * writeback using standard nfs. | |
869 | */ | |
870 | static void nfs_redirty_request(struct nfs_page *req) | |
871 | { | |
a6305ddb TM |
872 | struct page *page = req->wb_page; |
873 | ||
6d884e8f | 874 | nfs_mark_request_dirty(req); |
6d884e8f | 875 | nfs_clear_page_tag_locked(req); |
a6305ddb | 876 | nfs_end_page_writeback(page); |
6d884e8f F |
877 | } |
878 | ||
1da177e4 LT |
879 | /* |
880 | * Generate multiple small requests to write out a single | |
881 | * contiguous dirty area on one page. | |
882 | */ | |
bae724ef | 883 | static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how, struct pnfs_layout_segment *lseg) |
1da177e4 LT |
884 | { |
885 | struct nfs_page *req = nfs_list_entry(head->next); | |
886 | struct page *page = req->wb_page; | |
887 | struct nfs_write_data *data; | |
e9f7bee1 TM |
888 | size_t wsize = NFS_SERVER(inode)->wsize, nbytes; |
889 | unsigned int offset; | |
1da177e4 | 890 | int requests = 0; |
dbae4c73 | 891 | int ret = 0; |
1da177e4 LT |
892 | LIST_HEAD(list); |
893 | ||
894 | nfs_list_remove_request(req); | |
895 | ||
bcb71bba | 896 | nbytes = count; |
e9f7bee1 TM |
897 | do { |
898 | size_t len = min(nbytes, wsize); | |
899 | ||
8d5658c9 | 900 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
901 | if (!data) |
902 | goto out_bad; | |
903 | list_add(&data->pages, &list); | |
904 | requests++; | |
e9f7bee1 TM |
905 | nbytes -= len; |
906 | } while (nbytes != 0); | |
1da177e4 LT |
907 | atomic_set(&req->wb_complete, requests); |
908 | ||
909 | ClearPageError(page); | |
1da177e4 | 910 | offset = 0; |
bcb71bba | 911 | nbytes = count; |
1da177e4 | 912 | do { |
dbae4c73 TM |
913 | int ret2; |
914 | ||
1da177e4 LT |
915 | data = list_entry(list.next, struct nfs_write_data, pages); |
916 | list_del_init(&data->pages); | |
917 | ||
918 | data->pagevec[0] = page; | |
1da177e4 | 919 | |
bcb71bba TM |
920 | if (nbytes < wsize) |
921 | wsize = nbytes; | |
dbae4c73 | 922 | ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, |
bcb71bba | 923 | wsize, offset, how); |
dbae4c73 TM |
924 | if (ret == 0) |
925 | ret = ret2; | |
bcb71bba TM |
926 | offset += wsize; |
927 | nbytes -= wsize; | |
1da177e4 LT |
928 | } while (nbytes != 0); |
929 | ||
dbae4c73 | 930 | return ret; |
1da177e4 LT |
931 | |
932 | out_bad: | |
933 | while (!list_empty(&list)) { | |
934 | data = list_entry(list.next, struct nfs_write_data, pages); | |
935 | list_del(&data->pages); | |
0da2a4ac | 936 | nfs_writedata_free(data); |
1da177e4 | 937 | } |
61822ab5 | 938 | nfs_redirty_request(req); |
1da177e4 LT |
939 | return -ENOMEM; |
940 | } | |
941 | ||
942 | /* | |
943 | * Create an RPC task for the given write request and kick it. | |
944 | * The page must have been locked by the caller. | |
945 | * | |
946 | * It may happen that the page we're passed is not marked dirty. | |
947 | * This is the case if nfs_updatepage detects a conflicting request | |
948 | * that has been written but not committed. | |
949 | */ | |
bae724ef | 950 | static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how, struct pnfs_layout_segment *lseg) |
1da177e4 LT |
951 | { |
952 | struct nfs_page *req; | |
953 | struct page **pages; | |
954 | struct nfs_write_data *data; | |
1da177e4 | 955 | |
8d5658c9 | 956 | data = nfs_writedata_alloc(npages); |
1da177e4 LT |
957 | if (!data) |
958 | goto out_bad; | |
959 | ||
960 | pages = data->pagevec; | |
1da177e4 LT |
961 | while (!list_empty(head)) { |
962 | req = nfs_list_entry(head->next); | |
963 | nfs_list_remove_request(req); | |
964 | nfs_list_add_request(req, &data->pages); | |
965 | ClearPageError(req->wb_page); | |
1da177e4 | 966 | *pages++ = req->wb_page; |
1da177e4 LT |
967 | } |
968 | req = nfs_list_entry(data->pages.next); | |
969 | ||
1da177e4 | 970 | /* Set up the argument struct */ |
dbae4c73 | 971 | return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); |
1da177e4 LT |
972 | out_bad: |
973 | while (!list_empty(head)) { | |
10afec90 | 974 | req = nfs_list_entry(head->next); |
1da177e4 | 975 | nfs_list_remove_request(req); |
61822ab5 | 976 | nfs_redirty_request(req); |
1da177e4 LT |
977 | } |
978 | return -ENOMEM; | |
979 | } | |
980 | ||
c63c7b05 TM |
981 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
982 | struct inode *inode, int ioflags) | |
1da177e4 | 983 | { |
bf4285e7 | 984 | size_t wsize = NFS_SERVER(inode)->wsize; |
1da177e4 | 985 | |
94ad1c80 FI |
986 | pgio->pg_test = NULL; |
987 | ||
bcb71bba | 988 | if (wsize < PAGE_CACHE_SIZE) |
c63c7b05 | 989 | nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); |
bcb71bba | 990 | else |
c63c7b05 | 991 | nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); |
1da177e4 LT |
992 | } |
993 | ||
994 | /* | |
995 | * Handle a write reply that flushed part of a page. | |
996 | */ | |
788e7a89 | 997 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 998 | { |
788e7a89 | 999 | struct nfs_write_data *data = calldata; |
1da177e4 | 1000 | |
48186c7d CL |
1001 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1002 | task->tk_pid, | |
1003 | data->req->wb_context->path.dentry->d_inode->i_sb->s_id, | |
1004 | (long long) | |
1005 | NFS_FILEID(data->req->wb_context->path.dentry->d_inode), | |
1006 | data->req->wb_bytes, (long long)req_offset(data->req)); | |
1da177e4 | 1007 | |
c9d8f89d TM |
1008 | nfs_writeback_done(task, data); |
1009 | } | |
788e7a89 | 1010 | |
c9d8f89d TM |
1011 | static void nfs_writeback_release_partial(void *calldata) |
1012 | { | |
1013 | struct nfs_write_data *data = calldata; | |
1014 | struct nfs_page *req = data->req; | |
1015 | struct page *page = req->wb_page; | |
1016 | int status = data->task.tk_status; | |
1017 | ||
1018 | if (status < 0) { | |
a301b777 | 1019 | nfs_set_pageerror(page); |
c9d8f89d TM |
1020 | nfs_context_set_write_error(req->wb_context, status); |
1021 | dprintk(", error = %d\n", status); | |
8e821cad | 1022 | goto out; |
1da177e4 LT |
1023 | } |
1024 | ||
8e821cad | 1025 | if (nfs_write_need_commit(data)) { |
587142f8 | 1026 | struct inode *inode = page->mapping->host; |
8e821cad | 1027 | |
587142f8 | 1028 | spin_lock(&inode->i_lock); |
8e821cad TM |
1029 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1030 | /* Do nothing we need to resend the writes */ | |
1031 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1032 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1033 | dprintk(" defer commit\n"); | |
1034 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1035 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1036 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1037 | dprintk(" server reboot detected\n"); | |
1038 | } | |
587142f8 | 1039 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1040 | } else |
1041 | dprintk(" OK\n"); | |
1042 | ||
1043 | out: | |
1da177e4 LT |
1044 | if (atomic_dec_and_test(&req->wb_complete)) |
1045 | nfs_writepage_release(req); | |
c9d8f89d | 1046 | nfs_writedata_release(calldata); |
1da177e4 LT |
1047 | } |
1048 | ||
def6ed7e AA |
1049 | #if defined(CONFIG_NFS_V4_1) |
1050 | void nfs_write_prepare(struct rpc_task *task, void *calldata) | |
1051 | { | |
1052 | struct nfs_write_data *data = calldata; | |
def6ed7e | 1053 | |
035168ab TM |
1054 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
1055 | &data->args.seq_args, | |
def6ed7e AA |
1056 | &data->res.seq_res, 1, task)) |
1057 | return; | |
1058 | rpc_call_start(task); | |
1059 | } | |
1060 | #endif /* CONFIG_NFS_V4_1 */ | |
1061 | ||
788e7a89 | 1062 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e AA |
1063 | #if defined(CONFIG_NFS_V4_1) |
1064 | .rpc_call_prepare = nfs_write_prepare, | |
1065 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1066 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1067 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1068 | }; |
1069 | ||
1da177e4 LT |
1070 | /* |
1071 | * Handle a write reply that flushes a whole page. | |
1072 | * | |
1073 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1074 | * writebacks since the page->count is kept > 1 for as long | |
1075 | * as the page has a write request pending. | |
1076 | */ | |
788e7a89 | 1077 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1078 | { |
788e7a89 | 1079 | struct nfs_write_data *data = calldata; |
1da177e4 | 1080 | |
c9d8f89d TM |
1081 | nfs_writeback_done(task, data); |
1082 | } | |
1083 | ||
1084 | static void nfs_writeback_release_full(void *calldata) | |
1085 | { | |
1086 | struct nfs_write_data *data = calldata; | |
1087 | int status = data->task.tk_status; | |
788e7a89 | 1088 | |
1da177e4 LT |
1089 | /* Update attributes as result of writeback. */ |
1090 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1091 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1092 | struct page *page = req->wb_page; | |
1093 | ||
1da177e4 | 1094 | nfs_list_remove_request(req); |
1da177e4 | 1095 | |
48186c7d CL |
1096 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1097 | data->task.tk_pid, | |
88be9f99 TM |
1098 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1099 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1100 | req->wb_bytes, |
1101 | (long long)req_offset(req)); | |
1102 | ||
c9d8f89d | 1103 | if (status < 0) { |
a301b777 | 1104 | nfs_set_pageerror(page); |
c9d8f89d TM |
1105 | nfs_context_set_write_error(req->wb_context, status); |
1106 | dprintk(", error = %d\n", status); | |
8e821cad | 1107 | goto remove_request; |
1da177e4 | 1108 | } |
1da177e4 | 1109 | |
8e821cad TM |
1110 | if (nfs_write_need_commit(data)) { |
1111 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1112 | nfs_mark_request_commit(req); | |
8e821cad | 1113 | dprintk(" marked for commit\n"); |
1da177e4 LT |
1114 | goto next; |
1115 | } | |
8e821cad TM |
1116 | dprintk(" OK\n"); |
1117 | remove_request: | |
1da177e4 | 1118 | nfs_inode_remove_request(req); |
1da177e4 | 1119 | next: |
9fd367f0 | 1120 | nfs_clear_page_tag_locked(req); |
a6305ddb | 1121 | nfs_end_page_writeback(page); |
1da177e4 | 1122 | } |
c9d8f89d | 1123 | nfs_writedata_release(calldata); |
1da177e4 LT |
1124 | } |
1125 | ||
788e7a89 | 1126 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e AA |
1127 | #if defined(CONFIG_NFS_V4_1) |
1128 | .rpc_call_prepare = nfs_write_prepare, | |
1129 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1130 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1131 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1132 | }; |
1133 | ||
1134 | ||
1da177e4 LT |
1135 | /* |
1136 | * This function is called when the WRITE call is complete. | |
1137 | */ | |
13602896 | 1138 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1139 | { |
1da177e4 LT |
1140 | struct nfs_writeargs *argp = &data->args; |
1141 | struct nfs_writeres *resp = &data->res; | |
eedc020e | 1142 | struct nfs_server *server = NFS_SERVER(data->inode); |
788e7a89 | 1143 | int status; |
1da177e4 | 1144 | |
a3f565b1 | 1145 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1146 | task->tk_pid, task->tk_status); |
1147 | ||
f551e44f CL |
1148 | /* |
1149 | * ->write_done will attempt to use post-op attributes to detect | |
1150 | * conflicting writes by other clients. A strict interpretation | |
1151 | * of close-to-open would allow us to continue caching even if | |
1152 | * another writer had changed the file, but some applications | |
1153 | * depend on tighter cache coherency when writing. | |
1154 | */ | |
788e7a89 TM |
1155 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1156 | if (status != 0) | |
13602896 | 1157 | return; |
91d5b470 CL |
1158 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1159 | ||
1da177e4 LT |
1160 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1161 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1162 | /* We tried a write call, but the server did not | |
1163 | * commit data to stable storage even though we | |
1164 | * requested it. | |
1165 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1166 | * the server reports NFS_DATA_SYNC, but performs | |
1167 | * NFS_FILE_SYNC. We therefore implement this checking | |
1168 | * as a dprintk() in order to avoid filling syslog. | |
1169 | */ | |
1170 | static unsigned long complain; | |
1171 | ||
1172 | if (time_before(complain, jiffies)) { | |
48186c7d | 1173 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1174 | " (committed = %d) != (stable = %d)\n", |
eedc020e | 1175 | server->nfs_client->cl_hostname, |
1da177e4 LT |
1176 | resp->verf->committed, argp->stable); |
1177 | complain = jiffies + 300 * HZ; | |
1178 | } | |
1179 | } | |
1180 | #endif | |
1181 | /* Is this a short write? */ | |
1182 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1183 | static unsigned long complain; | |
1184 | ||
91d5b470 CL |
1185 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1186 | ||
1da177e4 LT |
1187 | /* Has the server at least made some progress? */ |
1188 | if (resp->count != 0) { | |
1189 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1190 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1191 | /* Resend from where the server left off */ | |
1192 | argp->offset += resp->count; | |
1193 | argp->pgbase += resp->count; | |
1194 | argp->count -= resp->count; | |
1195 | } else { | |
1196 | /* Resend as a stable write in order to avoid | |
1197 | * headaches in the case of a server crash. | |
1198 | */ | |
1199 | argp->stable = NFS_FILE_SYNC; | |
1200 | } | |
0110ee15 | 1201 | nfs_restart_rpc(task, server->nfs_client); |
13602896 | 1202 | return; |
1da177e4 LT |
1203 | } |
1204 | if (time_before(complain, jiffies)) { | |
1205 | printk(KERN_WARNING | |
1206 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1207 | argp->count); | |
1208 | complain = jiffies + 300 * HZ; | |
1209 | } | |
1210 | /* Can't do anything about it except throw an error. */ | |
1211 | task->tk_status = -EIO; | |
1212 | } | |
13602896 | 1213 | return; |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | ||
1217 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1218 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1219 | { | |
1220 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) | |
1221 | return 1; | |
1222 | if (may_wait && !out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1223 | NFS_INO_COMMIT, nfs_wait_bit_killable, | |
1224 | TASK_KILLABLE)) | |
1225 | return 1; | |
1226 | return 0; | |
1227 | } | |
1228 | ||
1229 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) | |
1230 | { | |
1231 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1232 | smp_mb__after_clear_bit(); | |
1233 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1234 | } | |
1235 | ||
1236 | ||
0aa05887 | 1237 | static void nfs_commitdata_release(void *data) |
1da177e4 | 1238 | { |
383ba719 TM |
1239 | struct nfs_write_data *wdata = data; |
1240 | ||
1241 | put_nfs_open_context(wdata->args.context); | |
1da177e4 LT |
1242 | nfs_commit_free(wdata); |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Set up the argument/result storage required for the RPC call. | |
1247 | */ | |
dbae4c73 | 1248 | static int nfs_commit_rpcsetup(struct list_head *head, |
788e7a89 TM |
1249 | struct nfs_write_data *data, |
1250 | int how) | |
1da177e4 | 1251 | { |
84115e1c TM |
1252 | struct nfs_page *first = nfs_list_entry(head->next); |
1253 | struct inode *inode = first->wb_context->path.dentry->d_inode; | |
3ff7576d | 1254 | int priority = flush_task_priority(how); |
07737691 | 1255 | struct rpc_task *task; |
bdc7f021 TM |
1256 | struct rpc_message msg = { |
1257 | .rpc_argp = &data->args, | |
1258 | .rpc_resp = &data->res, | |
1259 | .rpc_cred = first->wb_context->cred, | |
1260 | }; | |
84115e1c | 1261 | struct rpc_task_setup task_setup_data = { |
07737691 | 1262 | .task = &data->task, |
84115e1c | 1263 | .rpc_client = NFS_CLIENT(inode), |
bdc7f021 | 1264 | .rpc_message = &msg, |
84115e1c TM |
1265 | .callback_ops = &nfs_commit_ops, |
1266 | .callback_data = data, | |
101070ca | 1267 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1268 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1269 | .priority = priority, |
84115e1c | 1270 | }; |
1da177e4 LT |
1271 | |
1272 | /* Set up the RPC argument and reply structs | |
1273 | * NB: take care not to mess about with data->commit et al. */ | |
1274 | ||
1275 | list_splice_init(head, &data->pages); | |
1da177e4 | 1276 | |
1da177e4 | 1277 | data->inode = inode; |
bdc7f021 | 1278 | data->cred = msg.rpc_cred; |
1da177e4 LT |
1279 | |
1280 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1281 | /* Note: we always request a commit of the entire inode */ |
1282 | data->args.offset = 0; | |
1283 | data->args.count = 0; | |
383ba719 | 1284 | data->args.context = get_nfs_open_context(first->wb_context); |
3da28eb1 | 1285 | data->res.count = 0; |
1da177e4 LT |
1286 | data->res.fattr = &data->fattr; |
1287 | data->res.verf = &data->verf; | |
0e574af1 | 1288 | nfs_fattr_init(&data->fattr); |
788e7a89 TM |
1289 | |
1290 | /* Set up the initial task struct. */ | |
bdc7f021 | 1291 | NFS_PROTO(inode)->commit_setup(data, &msg); |
1da177e4 | 1292 | |
a3f565b1 | 1293 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); |
bdc7f021 | 1294 | |
07737691 | 1295 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
1296 | if (IS_ERR(task)) |
1297 | return PTR_ERR(task); | |
d2224e7a JL |
1298 | if (how & FLUSH_SYNC) |
1299 | rpc_wait_for_completion_task(task); | |
dbae4c73 TM |
1300 | rpc_put_task(task); |
1301 | return 0; | |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Commit dirty pages | |
1306 | */ | |
1307 | static int | |
40859d7e | 1308 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1309 | { |
1310 | struct nfs_write_data *data; | |
1311 | struct nfs_page *req; | |
1312 | ||
c9d8f89d | 1313 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1314 | |
1315 | if (!data) | |
1316 | goto out_bad; | |
1317 | ||
1318 | /* Set up the argument struct */ | |
dbae4c73 | 1319 | return nfs_commit_rpcsetup(head, data, how); |
1da177e4 LT |
1320 | out_bad: |
1321 | while (!list_empty(head)) { | |
1322 | req = nfs_list_entry(head->next); | |
1323 | nfs_list_remove_request(req); | |
1324 | nfs_mark_request_commit(req); | |
83715ad5 | 1325 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 PZ |
1326 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, |
1327 | BDI_RECLAIMABLE); | |
9fd367f0 | 1328 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1329 | } |
71d0a611 | 1330 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1331 | return -ENOMEM; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * COMMIT call returned | |
1336 | */ | |
788e7a89 | 1337 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1338 | { |
963d8fe5 | 1339 | struct nfs_write_data *data = calldata; |
1da177e4 | 1340 | |
a3f565b1 | 1341 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1342 | task->tk_pid, task->tk_status); |
1343 | ||
788e7a89 TM |
1344 | /* Call the NFS version-specific code */ |
1345 | if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) | |
1346 | return; | |
c9d8f89d TM |
1347 | } |
1348 | ||
1349 | static void nfs_commit_release(void *calldata) | |
1350 | { | |
1351 | struct nfs_write_data *data = calldata; | |
1352 | struct nfs_page *req; | |
1353 | int status = data->task.tk_status; | |
788e7a89 | 1354 | |
1da177e4 LT |
1355 | while (!list_empty(&data->pages)) { |
1356 | req = nfs_list_entry(data->pages.next); | |
1357 | nfs_list_remove_request(req); | |
e468bae9 | 1358 | nfs_clear_request_commit(req); |
1da177e4 | 1359 | |
48186c7d | 1360 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
88be9f99 TM |
1361 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1362 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1363 | req->wb_bytes, |
1364 | (long long)req_offset(req)); | |
c9d8f89d TM |
1365 | if (status < 0) { |
1366 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1367 | nfs_inode_remove_request(req); |
c9d8f89d | 1368 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1369 | goto next; |
1370 | } | |
1371 | ||
1372 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1373 | * returned by the server against all stored verfs. */ | |
1374 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1375 | /* We have a match */ | |
1376 | nfs_inode_remove_request(req); | |
1377 | dprintk(" OK\n"); | |
1378 | goto next; | |
1379 | } | |
1380 | /* We have a mismatch. Write the page again */ | |
1381 | dprintk(" mismatch\n"); | |
6d884e8f | 1382 | nfs_mark_request_dirty(req); |
1da177e4 | 1383 | next: |
9fd367f0 | 1384 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1385 | } |
71d0a611 | 1386 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1387 | nfs_commitdata_release(calldata); |
1da177e4 | 1388 | } |
788e7a89 TM |
1389 | |
1390 | static const struct rpc_call_ops nfs_commit_ops = { | |
21d9a851 AA |
1391 | #if defined(CONFIG_NFS_V4_1) |
1392 | .rpc_call_prepare = nfs_write_prepare, | |
1393 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 TM |
1394 | .rpc_call_done = nfs_commit_done, |
1395 | .rpc_release = nfs_commit_release, | |
1396 | }; | |
1da177e4 | 1397 | |
b608b283 | 1398 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1399 | { |
1da177e4 | 1400 | LIST_HEAD(head); |
71d0a611 TM |
1401 | int may_wait = how & FLUSH_SYNC; |
1402 | int res = 0; | |
1da177e4 | 1403 | |
71d0a611 | 1404 | if (!nfs_commit_set_lock(NFS_I(inode), may_wait)) |
c5efa5fc | 1405 | goto out_mark_dirty; |
587142f8 | 1406 | spin_lock(&inode->i_lock); |
3da28eb1 | 1407 | res = nfs_scan_commit(inode, &head, 0, 0); |
587142f8 | 1408 | spin_unlock(&inode->i_lock); |
1da177e4 | 1409 | if (res) { |
7d46a49f | 1410 | int error = nfs_commit_list(inode, &head, how); |
3da28eb1 TM |
1411 | if (error < 0) |
1412 | return error; | |
71d0a611 TM |
1413 | if (may_wait) |
1414 | wait_on_bit(&NFS_I(inode)->flags, NFS_INO_COMMIT, | |
1415 | nfs_wait_bit_killable, | |
1416 | TASK_KILLABLE); | |
c5efa5fc TM |
1417 | else |
1418 | goto out_mark_dirty; | |
71d0a611 TM |
1419 | } else |
1420 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1421 | return res; |
1422 | /* Note: If we exit without ensuring that the commit is complete, | |
1423 | * we must mark the inode as dirty. Otherwise, future calls to | |
1424 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1425 | * that the data is on the disk. | |
1426 | */ | |
1427 | out_mark_dirty: | |
1428 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1429 | return res; |
1430 | } | |
8fc795f7 TM |
1431 | |
1432 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1433 | { | |
420e3646 TM |
1434 | struct nfs_inode *nfsi = NFS_I(inode); |
1435 | int flags = FLUSH_SYNC; | |
1436 | int ret = 0; | |
8fc795f7 | 1437 | |
a00dd6c0 JL |
1438 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1439 | /* Don't commit yet if this is a non-blocking flush and there | |
1440 | * are a lot of outstanding writes for this mapping. | |
1441 | */ | |
1442 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1443 | goto out_mark_dirty; | |
420e3646 | 1444 | |
a00dd6c0 | 1445 | /* don't wait for the COMMIT response */ |
420e3646 | 1446 | flags = 0; |
a00dd6c0 JL |
1447 | } |
1448 | ||
420e3646 TM |
1449 | ret = nfs_commit_inode(inode, flags); |
1450 | if (ret >= 0) { | |
1451 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1452 | if (ret < wbc->nr_to_write) | |
1453 | wbc->nr_to_write -= ret; | |
1454 | else | |
1455 | wbc->nr_to_write = 0; | |
1456 | } | |
8fc795f7 | 1457 | return 0; |
420e3646 TM |
1458 | } |
1459 | out_mark_dirty: | |
8fc795f7 TM |
1460 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1461 | return ret; | |
1462 | } | |
c63c7b05 | 1463 | #else |
8fc795f7 TM |
1464 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1465 | { | |
1466 | return 0; | |
1467 | } | |
1da177e4 LT |
1468 | #endif |
1469 | ||
8fc795f7 TM |
1470 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1471 | { | |
1472 | return nfs_commit_unstable_pages(inode, wbc); | |
1473 | } | |
1474 | ||
acdc53b2 TM |
1475 | /* |
1476 | * flush the inode to disk. | |
1477 | */ | |
1478 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1479 | { |
1480 | struct writeback_control wbc = { | |
72cb77f4 | 1481 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1482 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1483 | .range_start = 0, |
1484 | .range_end = LLONG_MAX, | |
34901f70 | 1485 | }; |
34901f70 | 1486 | |
acdc53b2 | 1487 | return sync_inode(inode, &wbc); |
1c75950b TM |
1488 | } |
1489 | ||
1b3b4a1a TM |
1490 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1491 | { | |
1492 | struct nfs_page *req; | |
1b3b4a1a TM |
1493 | int ret = 0; |
1494 | ||
1495 | BUG_ON(!PageLocked(page)); | |
1496 | for (;;) { | |
ba8b06e6 | 1497 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1498 | req = nfs_page_find_request(page); |
1499 | if (req == NULL) | |
1b3b4a1a | 1500 | break; |
1b3b4a1a TM |
1501 | if (nfs_lock_request_dontget(req)) { |
1502 | nfs_inode_remove_request(req); | |
1503 | /* | |
1504 | * In case nfs_inode_remove_request has marked the | |
1505 | * page as being dirty | |
1506 | */ | |
1507 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1508 | nfs_unlock_request(req); | |
1509 | break; | |
1510 | } | |
1511 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1512 | nfs_release_request(req); |
1b3b4a1a | 1513 | if (ret < 0) |
c988950e | 1514 | break; |
1b3b4a1a | 1515 | } |
1b3b4a1a TM |
1516 | return ret; |
1517 | } | |
1518 | ||
7f2f12d9 TM |
1519 | /* |
1520 | * Write back all requests on one page - we do this before reading it. | |
1521 | */ | |
1522 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1523 | { |
1524 | loff_t range_start = page_offset(page); | |
1525 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1526 | struct writeback_control wbc = { |
4d770ccf | 1527 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1528 | .nr_to_write = 0, |
4d770ccf TM |
1529 | .range_start = range_start, |
1530 | .range_end = range_end, | |
1531 | }; | |
1532 | int ret; | |
1c75950b | 1533 | |
0522f6ad | 1534 | for (;;) { |
ba8b06e6 | 1535 | wait_on_page_writeback(page); |
73e3302f TM |
1536 | if (clear_page_dirty_for_io(page)) { |
1537 | ret = nfs_writepage_locked(page, &wbc); | |
1538 | if (ret < 0) | |
1539 | goto out_error; | |
0522f6ad | 1540 | continue; |
7f2f12d9 | 1541 | } |
0522f6ad TM |
1542 | if (!PagePrivate(page)) |
1543 | break; | |
1544 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1545 | if (ret < 0) |
73e3302f | 1546 | goto out_error; |
7f2f12d9 | 1547 | } |
73e3302f TM |
1548 | return 0; |
1549 | out_error: | |
4d770ccf | 1550 | return ret; |
1c75950b TM |
1551 | } |
1552 | ||
074cc1de TM |
1553 | #ifdef CONFIG_MIGRATION |
1554 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
1555 | struct page *page) | |
1556 | { | |
1557 | struct nfs_page *req; | |
1558 | int ret; | |
1559 | ||
7549ad5f | 1560 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1561 | |
cfb506e1 | 1562 | req = nfs_find_and_lock_request(page, false); |
074cc1de TM |
1563 | ret = PTR_ERR(req); |
1564 | if (IS_ERR(req)) | |
1565 | goto out; | |
1566 | ||
1567 | ret = migrate_page(mapping, newpage, page); | |
1568 | if (!req) | |
1569 | goto out; | |
1570 | if (ret) | |
1571 | goto out_unlock; | |
1572 | page_cache_get(newpage); | |
190f38e5 | 1573 | spin_lock(&mapping->host->i_lock); |
074cc1de TM |
1574 | req->wb_page = newpage; |
1575 | SetPagePrivate(newpage); | |
190f38e5 | 1576 | set_page_private(newpage, (unsigned long)req); |
074cc1de TM |
1577 | ClearPagePrivate(page); |
1578 | set_page_private(page, 0); | |
190f38e5 | 1579 | spin_unlock(&mapping->host->i_lock); |
074cc1de TM |
1580 | page_cache_release(page); |
1581 | out_unlock: | |
1582 | nfs_clear_page_tag_locked(req); | |
074cc1de TM |
1583 | out: |
1584 | return ret; | |
1585 | } | |
1586 | #endif | |
1587 | ||
f7b422b1 | 1588 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1589 | { |
1590 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1591 | sizeof(struct nfs_write_data), | |
1592 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1593 | NULL); |
1da177e4 LT |
1594 | if (nfs_wdata_cachep == NULL) |
1595 | return -ENOMEM; | |
1596 | ||
93d2341c MD |
1597 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1598 | nfs_wdata_cachep); | |
1da177e4 LT |
1599 | if (nfs_wdata_mempool == NULL) |
1600 | return -ENOMEM; | |
1601 | ||
93d2341c MD |
1602 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1603 | nfs_wdata_cachep); | |
1da177e4 LT |
1604 | if (nfs_commit_mempool == NULL) |
1605 | return -ENOMEM; | |
1606 | ||
89a09141 PZ |
1607 | /* |
1608 | * NFS congestion size, scale with available memory. | |
1609 | * | |
1610 | * 64MB: 8192k | |
1611 | * 128MB: 11585k | |
1612 | * 256MB: 16384k | |
1613 | * 512MB: 23170k | |
1614 | * 1GB: 32768k | |
1615 | * 2GB: 46340k | |
1616 | * 4GB: 65536k | |
1617 | * 8GB: 92681k | |
1618 | * 16GB: 131072k | |
1619 | * | |
1620 | * This allows larger machines to have larger/more transfers. | |
1621 | * Limit the default to 256M | |
1622 | */ | |
1623 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1624 | if (nfs_congestion_kb > 256*1024) | |
1625 | nfs_congestion_kb = 256*1024; | |
1626 | ||
1da177e4 LT |
1627 | return 0; |
1628 | } | |
1629 | ||
266bee88 | 1630 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1631 | { |
1632 | mempool_destroy(nfs_commit_mempool); | |
1633 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1634 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1635 | } |
1636 |