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 | } | |
e0c2b380 | 62 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 63 | |
5e4424af | 64 | void nfs_commit_free(struct nfs_write_data *p) |
1da177e4 | 65 | { |
40859d7e CL |
66 | if (p && (p->pagevec != &p->page_array[0])) |
67 | kfree(p->pagevec); | |
1da177e4 LT |
68 | mempool_free(p, nfs_commit_mempool); |
69 | } | |
e0c2b380 | 70 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 71 | |
8d5658c9 | 72 | struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) |
3feb2d49 | 73 | { |
e6b4f8da | 74 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
75 | |
76 | if (p) { | |
77 | memset(p, 0, sizeof(*p)); | |
78 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 79 | p->npages = pagecount; |
0d0b5cb3 CL |
80 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
81 | p->pagevec = p->page_array; | |
3feb2d49 | 82 | else { |
0d0b5cb3 CL |
83 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
84 | if (!p->pagevec) { | |
3feb2d49 TM |
85 | mempool_free(p, nfs_wdata_mempool); |
86 | p = NULL; | |
87 | } | |
88 | } | |
89 | } | |
90 | return p; | |
91 | } | |
92 | ||
1ae88b2e | 93 | void nfs_writedata_free(struct nfs_write_data *p) |
3feb2d49 TM |
94 | { |
95 | if (p && (p->pagevec != &p->page_array[0])) | |
96 | kfree(p->pagevec); | |
97 | mempool_free(p, nfs_wdata_mempool); | |
98 | } | |
99 | ||
dce81290 | 100 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 101 | { |
5053aa56 | 102 | put_lseg(wdata->lseg); |
383ba719 | 103 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
104 | nfs_writedata_free(wdata); |
105 | } | |
106 | ||
7b159fc1 TM |
107 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
108 | { | |
109 | ctx->error = error; | |
110 | smp_wmb(); | |
111 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
112 | } | |
113 | ||
277459d2 TM |
114 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
115 | { | |
116 | struct nfs_page *req = NULL; | |
117 | ||
118 | if (PagePrivate(page)) { | |
119 | req = (struct nfs_page *)page_private(page); | |
120 | if (req != NULL) | |
c03b4024 | 121 | kref_get(&req->wb_kref); |
277459d2 TM |
122 | } |
123 | return req; | |
124 | } | |
125 | ||
126 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
127 | { | |
587142f8 | 128 | struct inode *inode = page->mapping->host; |
277459d2 | 129 | struct nfs_page *req = NULL; |
277459d2 | 130 | |
587142f8 | 131 | spin_lock(&inode->i_lock); |
277459d2 | 132 | req = nfs_page_find_request_locked(page); |
587142f8 | 133 | spin_unlock(&inode->i_lock); |
277459d2 TM |
134 | return req; |
135 | } | |
136 | ||
1da177e4 LT |
137 | /* Adjust the file length if we're writing beyond the end */ |
138 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
139 | { | |
140 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
141 | loff_t end, i_size; |
142 | pgoff_t end_index; | |
1da177e4 | 143 | |
a3d01454 TM |
144 | spin_lock(&inode->i_lock); |
145 | i_size = i_size_read(inode); | |
146 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 147 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 148 | goto out; |
1da177e4 LT |
149 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
150 | if (i_size >= end) | |
a3d01454 | 151 | goto out; |
1da177e4 | 152 | i_size_write(inode, end); |
a3d01454 TM |
153 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
154 | out: | |
155 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
156 | } |
157 | ||
a301b777 TM |
158 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
159 | static void nfs_set_pageerror(struct page *page) | |
160 | { | |
161 | SetPageError(page); | |
162 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
163 | } | |
164 | ||
1da177e4 LT |
165 | /* We can set the PG_uptodate flag if we see that a write request |
166 | * covers the full page. | |
167 | */ | |
168 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
169 | { | |
1da177e4 LT |
170 | if (PageUptodate(page)) |
171 | return; | |
172 | if (base != 0) | |
173 | return; | |
49a70f27 | 174 | if (count != nfs_page_length(page)) |
1da177e4 | 175 | return; |
49a70f27 | 176 | SetPageUptodate(page); |
1da177e4 LT |
177 | } |
178 | ||
1da177e4 LT |
179 | static int wb_priority(struct writeback_control *wbc) |
180 | { | |
181 | if (wbc->for_reclaim) | |
c63c7b05 | 182 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 183 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
184 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
185 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
186 | } |
187 | ||
89a09141 PZ |
188 | /* |
189 | * NFS congestion control | |
190 | */ | |
191 | ||
192 | int nfs_congestion_kb; | |
193 | ||
194 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
195 | #define NFS_CONGESTION_OFF_THRESH \ | |
196 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
197 | ||
5a6d41b3 | 198 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 199 | { |
5a6d41b3 TM |
200 | int ret = test_set_page_writeback(page); |
201 | ||
202 | if (!ret) { | |
89a09141 PZ |
203 | struct inode *inode = page->mapping->host; |
204 | struct nfs_server *nfss = NFS_SERVER(inode); | |
205 | ||
a6305ddb | 206 | page_cache_get(page); |
277866a0 | 207 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
208 | NFS_CONGESTION_ON_THRESH) { |
209 | set_bdi_congested(&nfss->backing_dev_info, | |
210 | BLK_RW_ASYNC); | |
211 | } | |
89a09141 | 212 | } |
5a6d41b3 | 213 | return ret; |
89a09141 PZ |
214 | } |
215 | ||
216 | static void nfs_end_page_writeback(struct page *page) | |
217 | { | |
218 | struct inode *inode = page->mapping->host; | |
219 | struct nfs_server *nfss = NFS_SERVER(inode); | |
220 | ||
221 | end_page_writeback(page); | |
a6305ddb | 222 | page_cache_release(page); |
c4dc4bee | 223 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 224 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
225 | } |
226 | ||
cfb506e1 | 227 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 228 | { |
587142f8 | 229 | struct inode *inode = page->mapping->host; |
e261f51f | 230 | struct nfs_page *req; |
e261f51f TM |
231 | int ret; |
232 | ||
587142f8 | 233 | spin_lock(&inode->i_lock); |
074cc1de | 234 | for (;;) { |
e261f51f | 235 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
236 | if (req == NULL) |
237 | break; | |
acee478a | 238 | if (nfs_set_page_tag_locked(req)) |
e261f51f TM |
239 | break; |
240 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
acee478a | 241 | * then the call to nfs_set_page_tag_locked() will always |
e261f51f TM |
242 | * succeed provided that someone hasn't already marked the |
243 | * request as dirty (in which case we don't care). | |
244 | */ | |
587142f8 | 245 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
246 | if (!nonblock) |
247 | ret = nfs_wait_on_request(req); | |
248 | else | |
249 | ret = -EAGAIN; | |
e261f51f TM |
250 | nfs_release_request(req); |
251 | if (ret != 0) | |
074cc1de | 252 | return ERR_PTR(ret); |
587142f8 | 253 | spin_lock(&inode->i_lock); |
e261f51f | 254 | } |
587142f8 | 255 | spin_unlock(&inode->i_lock); |
074cc1de TM |
256 | return req; |
257 | } | |
258 | ||
259 | /* | |
260 | * Find an associated nfs write request, and prepare to flush it out | |
261 | * May return an error if the user signalled nfs_wait_on_request(). | |
262 | */ | |
263 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 264 | struct page *page, bool nonblock) |
074cc1de TM |
265 | { |
266 | struct nfs_page *req; | |
267 | int ret = 0; | |
268 | ||
cfb506e1 | 269 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
270 | if (!req) |
271 | goto out; | |
272 | ret = PTR_ERR(req); | |
273 | if (IS_ERR(req)) | |
274 | goto out; | |
275 | ||
276 | ret = nfs_set_page_writeback(page); | |
277 | BUG_ON(ret != 0); | |
278 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
279 | ||
f8512ad0 FI |
280 | if (!nfs_pageio_add_request(pgio, req)) { |
281 | nfs_redirty_request(req); | |
074cc1de | 282 | ret = pgio->pg_error; |
f8512ad0 | 283 | } |
074cc1de TM |
284 | out: |
285 | return ret; | |
e261f51f TM |
286 | } |
287 | ||
f758c885 | 288 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 289 | { |
1da177e4 | 290 | struct inode *inode = page->mapping->host; |
cfb506e1 | 291 | int ret; |
1da177e4 | 292 | |
91d5b470 CL |
293 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
294 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
295 | ||
7fe7f848 | 296 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 297 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
298 | if (ret == -EAGAIN) { |
299 | redirty_page_for_writepage(wbc, page); | |
300 | ret = 0; | |
301 | } | |
302 | return ret; | |
f758c885 | 303 | } |
7fe7f848 | 304 | |
f758c885 TM |
305 | /* |
306 | * Write an mmapped page to the server. | |
307 | */ | |
308 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
309 | { | |
310 | struct nfs_pageio_descriptor pgio; | |
311 | int err; | |
49a70f27 | 312 | |
f758c885 TM |
313 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
314 | err = nfs_do_writepage(page, wbc, &pgio); | |
315 | nfs_pageio_complete(&pgio); | |
316 | if (err < 0) | |
317 | return err; | |
318 | if (pgio.pg_error < 0) | |
319 | return pgio.pg_error; | |
320 | return 0; | |
4d770ccf TM |
321 | } |
322 | ||
323 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
324 | { | |
f758c885 | 325 | int ret; |
4d770ccf | 326 | |
f758c885 | 327 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 328 | unlock_page(page); |
f758c885 TM |
329 | return ret; |
330 | } | |
331 | ||
332 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
333 | { | |
334 | int ret; | |
335 | ||
336 | ret = nfs_do_writepage(page, wbc, data); | |
337 | unlock_page(page); | |
338 | return ret; | |
1da177e4 LT |
339 | } |
340 | ||
1da177e4 LT |
341 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
342 | { | |
1da177e4 | 343 | struct inode *inode = mapping->host; |
72cb77f4 | 344 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 345 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
346 | int err; |
347 | ||
72cb77f4 TM |
348 | /* Stop dirtying of new pages while we sync */ |
349 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
350 | nfs_wait_bit_killable, TASK_KILLABLE); | |
351 | if (err) | |
352 | goto out_err; | |
353 | ||
91d5b470 CL |
354 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
355 | ||
c63c7b05 | 356 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 357 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 358 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
359 | |
360 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
361 | smp_mb__after_clear_bit(); | |
362 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
363 | ||
f758c885 | 364 | if (err < 0) |
72cb77f4 TM |
365 | goto out_err; |
366 | err = pgio.pg_error; | |
367 | if (err < 0) | |
368 | goto out_err; | |
c63c7b05 | 369 | return 0; |
72cb77f4 TM |
370 | out_err: |
371 | return err; | |
1da177e4 LT |
372 | } |
373 | ||
374 | /* | |
375 | * Insert a write request into an inode | |
376 | */ | |
e7d39069 | 377 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
378 | { |
379 | struct nfs_inode *nfsi = NFS_I(inode); | |
380 | int error; | |
381 | ||
e7d39069 TM |
382 | error = radix_tree_preload(GFP_NOFS); |
383 | if (error != 0) | |
384 | goto out; | |
385 | ||
386 | /* Lock the request! */ | |
387 | nfs_lock_request_dontget(req); | |
388 | ||
389 | spin_lock(&inode->i_lock); | |
1da177e4 | 390 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); |
27852596 | 391 | BUG_ON(error); |
4d65c520 TM |
392 | if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) |
393 | nfsi->change_attr++; | |
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 | { | |
3d4ff43d | 412 | struct inode *inode = req->wb_context->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--; | |
4d65c520 | 423 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
424 | nfs_release_request(req); |
425 | } | |
426 | ||
61822ab5 | 427 | static void |
6d884e8f | 428 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 429 | { |
61822ab5 | 430 | __set_page_dirty_nobuffers(req->wb_page); |
b80c3cb6 | 431 | __mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC); |
61822ab5 TM |
432 | } |
433 | ||
1da177e4 LT |
434 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
435 | /* | |
436 | * Add a request to the inode's commit list. | |
437 | */ | |
438 | static void | |
a861a1e1 | 439 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
1da177e4 | 440 | { |
3d4ff43d | 441 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
442 | struct nfs_inode *nfsi = NFS_I(inode); |
443 | ||
587142f8 | 444 | spin_lock(&inode->i_lock); |
e468bae9 | 445 | set_bit(PG_CLEAN, &(req)->wb_flags); |
5c369683 TM |
446 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
447 | req->wb_index, | |
448 | NFS_PAGE_TAG_COMMIT); | |
ff778d02 | 449 | nfsi->ncommit++; |
587142f8 | 450 | spin_unlock(&inode->i_lock); |
a861a1e1 | 451 | pnfs_mark_request_commit(req, lseg); |
fd39fc85 | 452 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 453 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 454 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 455 | } |
8e821cad | 456 | |
e468bae9 TM |
457 | static int |
458 | nfs_clear_request_commit(struct nfs_page *req) | |
459 | { | |
460 | struct page *page = req->wb_page; | |
461 | ||
462 | if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { | |
463 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
464 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
465 | return 1; | |
466 | } | |
467 | return 0; | |
468 | } | |
469 | ||
8e821cad TM |
470 | static inline |
471 | int nfs_write_need_commit(struct nfs_write_data *data) | |
472 | { | |
465d5243 FI |
473 | if (data->verf.committed == NFS_DATA_SYNC) |
474 | return data->lseg == NULL; | |
475 | else | |
476 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
477 | } |
478 | ||
479 | static inline | |
a861a1e1 FI |
480 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
481 | struct nfs_write_data *data) | |
8e821cad | 482 | { |
e468bae9 | 483 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
a861a1e1 | 484 | nfs_mark_request_commit(req, data->lseg); |
8e821cad TM |
485 | return 1; |
486 | } | |
487 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
6d884e8f | 488 | nfs_mark_request_dirty(req); |
8e821cad TM |
489 | return 1; |
490 | } | |
491 | return 0; | |
492 | } | |
493 | #else | |
494 | static inline void | |
a861a1e1 | 495 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
8e821cad TM |
496 | { |
497 | } | |
498 | ||
e468bae9 TM |
499 | static inline int |
500 | nfs_clear_request_commit(struct nfs_page *req) | |
501 | { | |
502 | return 0; | |
503 | } | |
504 | ||
8e821cad TM |
505 | static inline |
506 | int nfs_write_need_commit(struct nfs_write_data *data) | |
507 | { | |
508 | return 0; | |
509 | } | |
510 | ||
511 | static inline | |
a861a1e1 FI |
512 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
513 | struct nfs_write_data *data) | |
8e821cad TM |
514 | { |
515 | return 0; | |
516 | } | |
1da177e4 LT |
517 | #endif |
518 | ||
47c62564 | 519 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
520 | static int |
521 | nfs_need_commit(struct nfs_inode *nfsi) | |
522 | { | |
523 | return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); | |
524 | } | |
525 | ||
1da177e4 LT |
526 | /* |
527 | * nfs_scan_commit - Scan an inode for commit requests | |
528 | * @inode: NFS inode to scan | |
529 | * @dst: destination list | |
530 | * @idx_start: lower bound of page->index to scan. | |
531 | * @npages: idx_start + npages sets the upper bound to scan. | |
532 | * | |
533 | * Moves requests from the inode's 'commit' request list. | |
534 | * The requests are *not* checked to ensure that they form a contiguous set. | |
535 | */ | |
536 | static int | |
ca52fec1 | 537 | nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
538 | { |
539 | struct nfs_inode *nfsi = NFS_I(inode); | |
ff778d02 | 540 | int ret; |
3da28eb1 | 541 | |
fb8a1f11 TM |
542 | if (!nfs_need_commit(nfsi)) |
543 | return 0; | |
544 | ||
0d88f6e8 | 545 | spin_lock(&inode->i_lock); |
ff778d02 TM |
546 | ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); |
547 | if (ret > 0) | |
548 | nfsi->ncommit -= ret; | |
0d88f6e8 DC |
549 | spin_unlock(&inode->i_lock); |
550 | ||
2928db1f TM |
551 | if (nfs_need_commit(NFS_I(inode))) |
552 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
0d88f6e8 | 553 | |
ff778d02 | 554 | return ret; |
1da177e4 | 555 | } |
c42de9dd | 556 | #else |
fb8a1f11 TM |
557 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
558 | { | |
559 | return 0; | |
560 | } | |
561 | ||
ca52fec1 | 562 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
c42de9dd TM |
563 | { |
564 | return 0; | |
565 | } | |
1da177e4 LT |
566 | #endif |
567 | ||
1da177e4 | 568 | /* |
e7d39069 TM |
569 | * Search for an existing write request, and attempt to update |
570 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 571 | * |
e7d39069 TM |
572 | * If the attempt fails, then the existing request is flushed out |
573 | * to disk. | |
1da177e4 | 574 | */ |
e7d39069 TM |
575 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
576 | struct page *page, | |
577 | unsigned int offset, | |
578 | unsigned int bytes) | |
1da177e4 | 579 | { |
e7d39069 TM |
580 | struct nfs_page *req; |
581 | unsigned int rqend; | |
582 | unsigned int end; | |
583 | int error; | |
584 | ||
585 | if (!PagePrivate(page)) | |
586 | return NULL; | |
1da177e4 LT |
587 | |
588 | end = offset + bytes; | |
e7d39069 | 589 | spin_lock(&inode->i_lock); |
1da177e4 | 590 | |
1da177e4 | 591 | for (;;) { |
277459d2 | 592 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
593 | if (req == NULL) |
594 | goto out_unlock; | |
595 | ||
596 | rqend = req->wb_offset + req->wb_bytes; | |
597 | /* | |
598 | * Tell the caller to flush out the request if | |
599 | * the offsets are non-contiguous. | |
600 | * Note: nfs_flush_incompatible() will already | |
601 | * have flushed out requests having wrong owners. | |
602 | */ | |
e468bae9 | 603 | if (offset > rqend |
e7d39069 TM |
604 | || end < req->wb_offset) |
605 | goto out_flushme; | |
606 | ||
607 | if (nfs_set_page_tag_locked(req)) | |
1da177e4 | 608 | break; |
1da177e4 | 609 | |
e7d39069 | 610 | /* The request is locked, so wait and then retry */ |
587142f8 | 611 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
612 | error = nfs_wait_on_request(req); |
613 | nfs_release_request(req); | |
614 | if (error != 0) | |
615 | goto out_err; | |
616 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
617 | } |
618 | ||
ff778d02 | 619 | if (nfs_clear_request_commit(req) && |
a861a1e1 FI |
620 | radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, |
621 | req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) { | |
ff778d02 | 622 | NFS_I(inode)->ncommit--; |
a861a1e1 FI |
623 | pnfs_clear_request_commit(req); |
624 | } | |
e468bae9 | 625 | |
1da177e4 LT |
626 | /* Okay, the request matches. Update the region */ |
627 | if (offset < req->wb_offset) { | |
628 | req->wb_offset = offset; | |
629 | req->wb_pgbase = offset; | |
1da177e4 | 630 | } |
1da177e4 LT |
631 | if (end > rqend) |
632 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
633 | else |
634 | req->wb_bytes = rqend - req->wb_offset; | |
635 | out_unlock: | |
636 | spin_unlock(&inode->i_lock); | |
637 | return req; | |
638 | out_flushme: | |
639 | spin_unlock(&inode->i_lock); | |
640 | nfs_release_request(req); | |
641 | error = nfs_wb_page(inode, page); | |
642 | out_err: | |
643 | return ERR_PTR(error); | |
644 | } | |
645 | ||
646 | /* | |
647 | * Try to update an existing write request, or create one if there is none. | |
648 | * | |
649 | * Note: Should always be called with the Page Lock held to prevent races | |
650 | * if we have to add a new request. Also assumes that the caller has | |
651 | * already called nfs_flush_incompatible() if necessary. | |
652 | */ | |
653 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
654 | struct page *page, unsigned int offset, unsigned int bytes) | |
655 | { | |
656 | struct inode *inode = page->mapping->host; | |
657 | struct nfs_page *req; | |
658 | int error; | |
1da177e4 | 659 | |
e7d39069 TM |
660 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
661 | if (req != NULL) | |
662 | goto out; | |
663 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
664 | if (IS_ERR(req)) | |
665 | goto out; | |
666 | error = nfs_inode_add_request(inode, req); | |
667 | if (error != 0) { | |
668 | nfs_release_request(req); | |
669 | req = ERR_PTR(error); | |
670 | } | |
efc91ed0 | 671 | out: |
61e930a9 | 672 | return req; |
1da177e4 LT |
673 | } |
674 | ||
e7d39069 TM |
675 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
676 | unsigned int offset, unsigned int count) | |
677 | { | |
678 | struct nfs_page *req; | |
679 | ||
680 | req = nfs_setup_write_request(ctx, page, offset, count); | |
681 | if (IS_ERR(req)) | |
682 | return PTR_ERR(req); | |
683 | /* Update file length */ | |
684 | nfs_grow_file(page, offset, count); | |
685 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 686 | nfs_mark_request_dirty(req); |
e7d39069 TM |
687 | nfs_clear_page_tag_locked(req); |
688 | return 0; | |
689 | } | |
690 | ||
1da177e4 LT |
691 | int nfs_flush_incompatible(struct file *file, struct page *page) |
692 | { | |
cd3758e3 | 693 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 694 | struct nfs_page *req; |
1a54533e | 695 | int do_flush, status; |
1da177e4 LT |
696 | /* |
697 | * Look for a request corresponding to this page. If there | |
698 | * is one, and it belongs to another file, we flush it out | |
699 | * before we try to copy anything into the page. Do this | |
700 | * due to the lack of an ACCESS-type call in NFSv2. | |
701 | * Also do the same if we find a request from an existing | |
702 | * dropped page. | |
703 | */ | |
1a54533e TM |
704 | do { |
705 | req = nfs_page_find_request(page); | |
706 | if (req == NULL) | |
707 | return 0; | |
f11ac8db TM |
708 | do_flush = req->wb_page != page || req->wb_context != ctx || |
709 | req->wb_lock_context->lockowner != current->files || | |
710 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 711 | nfs_release_request(req); |
1a54533e TM |
712 | if (!do_flush) |
713 | return 0; | |
714 | status = nfs_wb_page(page->mapping->host, page); | |
715 | } while (status == 0); | |
716 | return status; | |
1da177e4 LT |
717 | } |
718 | ||
5d47a356 TM |
719 | /* |
720 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
721 | * the PageUptodate() flag. In this case, we will need to turn off | |
722 | * write optimisations that depend on the page contents being correct. | |
723 | */ | |
724 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
725 | { | |
726 | return PageUptodate(page) && | |
727 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
728 | } | |
729 | ||
1da177e4 LT |
730 | /* |
731 | * Update and possibly write a cached page of an NFS file. | |
732 | * | |
733 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
734 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
735 | */ | |
736 | int nfs_updatepage(struct file *file, struct page *page, | |
737 | unsigned int offset, unsigned int count) | |
738 | { | |
cd3758e3 | 739 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 740 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
741 | int status = 0; |
742 | ||
91d5b470 CL |
743 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
744 | ||
48186c7d | 745 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
746 | file->f_path.dentry->d_parent->d_name.name, |
747 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 748 | (long long)(page_offset(page) + offset)); |
1da177e4 | 749 | |
1da177e4 | 750 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
751 | * is up to date, it may be more efficient to extend the write |
752 | * to cover the entire page in order to avoid fragmentation | |
753 | * inefficiencies. | |
1da177e4 | 754 | */ |
5d47a356 TM |
755 | if (nfs_write_pageuptodate(page, inode) && |
756 | inode->i_flock == NULL && | |
6b2f3d1f | 757 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 758 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 759 | offset = 0; |
1da177e4 LT |
760 | } |
761 | ||
e21195a7 | 762 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
763 | if (status < 0) |
764 | nfs_set_pageerror(page); | |
1da177e4 | 765 | |
48186c7d | 766 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 767 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
768 | return status; |
769 | } | |
770 | ||
a861a1e1 FI |
771 | static void nfs_writepage_release(struct nfs_page *req, |
772 | struct nfs_write_data *data) | |
1da177e4 | 773 | { |
a6305ddb | 774 | struct page *page = req->wb_page; |
1da177e4 | 775 | |
a861a1e1 | 776 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) |
8e821cad | 777 | nfs_inode_remove_request(req); |
9fd367f0 | 778 | nfs_clear_page_tag_locked(req); |
a6305ddb | 779 | nfs_end_page_writeback(page); |
1da177e4 LT |
780 | } |
781 | ||
3ff7576d | 782 | static int flush_task_priority(int how) |
1da177e4 LT |
783 | { |
784 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
785 | case FLUSH_HIGHPRI: | |
786 | return RPC_PRIORITY_HIGH; | |
787 | case FLUSH_LOWPRI: | |
788 | return RPC_PRIORITY_LOW; | |
789 | } | |
790 | return RPC_PRIORITY_NORMAL; | |
791 | } | |
792 | ||
a69aef14 | 793 | int nfs_initiate_write(struct nfs_write_data *data, |
d138d5d1 AA |
794 | struct rpc_clnt *clnt, |
795 | const struct rpc_call_ops *call_ops, | |
796 | int how) | |
1da177e4 | 797 | { |
d138d5d1 | 798 | struct inode *inode = data->inode; |
3ff7576d | 799 | int priority = flush_task_priority(how); |
07737691 | 800 | struct rpc_task *task; |
bdc7f021 TM |
801 | struct rpc_message msg = { |
802 | .rpc_argp = &data->args, | |
803 | .rpc_resp = &data->res, | |
d138d5d1 | 804 | .rpc_cred = data->cred, |
bdc7f021 | 805 | }; |
84115e1c | 806 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 807 | .rpc_client = clnt, |
07737691 | 808 | .task = &data->task, |
bdc7f021 | 809 | .rpc_message = &msg, |
84115e1c TM |
810 | .callback_ops = call_ops, |
811 | .callback_data = data, | |
101070ca | 812 | .workqueue = nfsiod_workqueue, |
2c61be0a | 813 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 814 | .priority = priority, |
84115e1c | 815 | }; |
2c61be0a | 816 | int ret = 0; |
1da177e4 | 817 | |
d138d5d1 AA |
818 | /* Set up the initial task struct. */ |
819 | NFS_PROTO(inode)->write_setup(data, &msg); | |
820 | ||
821 | dprintk("NFS: %5u initiated write call " | |
822 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
823 | data->task.tk_pid, | |
824 | inode->i_sb->s_id, | |
825 | (long long)NFS_FILEID(inode), | |
826 | data->args.count, | |
827 | (unsigned long long)data->args.offset); | |
828 | ||
829 | task = rpc_run_task(&task_setup_data); | |
830 | if (IS_ERR(task)) { | |
831 | ret = PTR_ERR(task); | |
832 | goto out; | |
833 | } | |
834 | if (how & FLUSH_SYNC) { | |
835 | ret = rpc_wait_for_completion_task(task); | |
836 | if (ret == 0) | |
837 | ret = task->tk_status; | |
838 | } | |
839 | rpc_put_task(task); | |
840 | out: | |
841 | return ret; | |
842 | } | |
a69aef14 | 843 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
844 | |
845 | /* | |
846 | * Set up the argument/result storage required for the RPC call. | |
847 | */ | |
6e4efd56 | 848 | static void nfs_write_rpcsetup(struct nfs_page *req, |
d138d5d1 | 849 | struct nfs_write_data *data, |
d138d5d1 AA |
850 | unsigned int count, unsigned int offset, |
851 | int how) | |
852 | { | |
3d4ff43d | 853 | struct inode *inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 854 | |
1da177e4 LT |
855 | /* Set up the RPC argument and reply structs |
856 | * NB: take care not to mess about with data->commit et al. */ | |
857 | ||
858 | data->req = req; | |
3d4ff43d | 859 | data->inode = inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 860 | data->cred = req->wb_context->cred; |
1da177e4 LT |
861 | |
862 | data->args.fh = NFS_FH(inode); | |
863 | data->args.offset = req_offset(req) + offset; | |
2bea038c BH |
864 | /* pnfs_set_layoutcommit needs this */ |
865 | data->mds_offset = data->args.offset; | |
1da177e4 LT |
866 | data->args.pgbase = req->wb_pgbase + offset; |
867 | data->args.pages = data->pagevec; | |
868 | data->args.count = count; | |
383ba719 | 869 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 870 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 871 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
872 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
873 | case 0: | |
874 | break; | |
875 | case FLUSH_COND_STABLE: | |
876 | if (nfs_need_commit(NFS_I(inode))) | |
877 | break; | |
878 | default: | |
879 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 880 | } |
1da177e4 LT |
881 | |
882 | data->res.fattr = &data->fattr; | |
883 | data->res.count = count; | |
884 | data->res.verf = &data->verf; | |
0e574af1 | 885 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 886 | } |
1da177e4 | 887 | |
6e4efd56 TM |
888 | static int nfs_do_write(struct nfs_write_data *data, |
889 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
890 | int how) |
891 | { | |
5f00bcb3 | 892 | struct inode *inode = data->args.context->dentry->d_inode; |
0382b744 | 893 | |
d138d5d1 | 894 | return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); |
1da177e4 LT |
895 | } |
896 | ||
275acaaf TM |
897 | static int nfs_do_multiple_writes(struct list_head *head, |
898 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
899 | int how) |
900 | { | |
901 | struct nfs_write_data *data; | |
902 | int ret = 0; | |
903 | ||
904 | while (!list_empty(head)) { | |
905 | int ret2; | |
906 | ||
907 | data = list_entry(head->next, struct nfs_write_data, list); | |
908 | list_del_init(&data->list); | |
909 | ||
dce81290 | 910 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
911 | if (ret == 0) |
912 | ret = ret2; | |
913 | } | |
914 | return ret; | |
915 | } | |
916 | ||
6d884e8f F |
917 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
918 | * call this on each, which will prepare them to be retried on next | |
919 | * writeback using standard nfs. | |
920 | */ | |
921 | static void nfs_redirty_request(struct nfs_page *req) | |
922 | { | |
a6305ddb TM |
923 | struct page *page = req->wb_page; |
924 | ||
6d884e8f | 925 | nfs_mark_request_dirty(req); |
6d884e8f | 926 | nfs_clear_page_tag_locked(req); |
a6305ddb | 927 | nfs_end_page_writeback(page); |
6d884e8f F |
928 | } |
929 | ||
1da177e4 LT |
930 | /* |
931 | * Generate multiple small requests to write out a single | |
932 | * contiguous dirty area on one page. | |
933 | */ | |
275acaaf | 934 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 | 935 | { |
c76069bd | 936 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
937 | struct page *page = req->wb_page; |
938 | struct nfs_write_data *data; | |
d097971d | 939 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 940 | unsigned int offset; |
1da177e4 | 941 | int requests = 0; |
dbae4c73 | 942 | int ret = 0; |
1da177e4 LT |
943 | |
944 | nfs_list_remove_request(req); | |
945 | ||
b31268ac TM |
946 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
947 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || | |
948 | desc->pg_count > wsize)) | |
949 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
950 | ||
951 | ||
275acaaf | 952 | offset = 0; |
c76069bd | 953 | nbytes = desc->pg_count; |
e9f7bee1 TM |
954 | do { |
955 | size_t len = min(nbytes, wsize); | |
956 | ||
8d5658c9 | 957 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
958 | if (!data) |
959 | goto out_bad; | |
275acaaf | 960 | data->pagevec[0] = page; |
f13c3620 | 961 | nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags); |
275acaaf | 962 | list_add(&data->list, res); |
1da177e4 | 963 | requests++; |
e9f7bee1 | 964 | nbytes -= len; |
275acaaf | 965 | offset += len; |
e9f7bee1 | 966 | } while (nbytes != 0); |
1da177e4 | 967 | atomic_set(&req->wb_complete, requests); |
50828d7e | 968 | desc->pg_rpc_callops = &nfs_write_partial_ops; |
dbae4c73 | 969 | return ret; |
1da177e4 LT |
970 | |
971 | out_bad: | |
275acaaf TM |
972 | while (!list_empty(res)) { |
973 | data = list_entry(res->next, struct nfs_write_data, list); | |
6e4efd56 | 974 | list_del(&data->list); |
0da2a4ac | 975 | nfs_writedata_free(data); |
1da177e4 | 976 | } |
61822ab5 | 977 | nfs_redirty_request(req); |
1da177e4 LT |
978 | return -ENOMEM; |
979 | } | |
980 | ||
981 | /* | |
982 | * Create an RPC task for the given write request and kick it. | |
983 | * The page must have been locked by the caller. | |
984 | * | |
985 | * It may happen that the page we're passed is not marked dirty. | |
986 | * This is the case if nfs_updatepage detects a conflicting request | |
987 | * that has been written but not committed. | |
988 | */ | |
275acaaf | 989 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 LT |
990 | { |
991 | struct nfs_page *req; | |
992 | struct page **pages; | |
993 | struct nfs_write_data *data; | |
c76069bd | 994 | struct list_head *head = &desc->pg_list; |
3b609184 | 995 | int ret = 0; |
1da177e4 | 996 | |
c76069bd FI |
997 | data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, |
998 | desc->pg_count)); | |
44b83799 FI |
999 | if (!data) { |
1000 | while (!list_empty(head)) { | |
1001 | req = nfs_list_entry(head->next); | |
1002 | nfs_list_remove_request(req); | |
1003 | nfs_redirty_request(req); | |
1004 | } | |
1005 | ret = -ENOMEM; | |
1006 | goto out; | |
1007 | } | |
1da177e4 | 1008 | pages = data->pagevec; |
1da177e4 LT |
1009 | while (!list_empty(head)) { |
1010 | req = nfs_list_entry(head->next); | |
1011 | nfs_list_remove_request(req); | |
1012 | nfs_list_add_request(req, &data->pages); | |
1013 | ClearPageError(req->wb_page); | |
1da177e4 | 1014 | *pages++ = req->wb_page; |
1da177e4 LT |
1015 | } |
1016 | req = nfs_list_entry(data->pages.next); | |
1017 | ||
b31268ac TM |
1018 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1019 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) | |
1020 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1021 | ||
1da177e4 | 1022 | /* Set up the argument struct */ |
6e4efd56 | 1023 | nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags); |
275acaaf | 1024 | list_add(&data->list, res); |
50828d7e | 1025 | desc->pg_rpc_callops = &nfs_write_full_ops; |
44b83799 | 1026 | out: |
44b83799 | 1027 | return ret; |
1da177e4 LT |
1028 | } |
1029 | ||
dce81290 TM |
1030 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head) |
1031 | { | |
1032 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
1033 | return nfs_flush_multi(desc, head); | |
1034 | return nfs_flush_one(desc, head); | |
1035 | } | |
1036 | ||
1037 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1038 | { |
275acaaf TM |
1039 | LIST_HEAD(head); |
1040 | int ret; | |
1041 | ||
dce81290 | 1042 | ret = nfs_generic_flush(desc, &head); |
50828d7e TM |
1043 | if (ret == 0) |
1044 | ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops, | |
dce81290 | 1045 | desc->pg_ioflags); |
275acaaf | 1046 | return ret; |
1751c363 | 1047 | } |
1751c363 TM |
1048 | |
1049 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1050 | .pg_test = nfs_generic_pg_test, | |
1051 | .pg_doio = nfs_generic_pg_writepages, | |
1052 | }; | |
1053 | ||
1f945357 | 1054 | static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
c63c7b05 | 1055 | struct inode *inode, int ioflags) |
1da177e4 | 1056 | { |
1751c363 TM |
1057 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, |
1058 | NFS_SERVER(inode)->wsize, ioflags); | |
1059 | } | |
1da177e4 | 1060 | |
dce81290 TM |
1061 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1062 | { | |
1063 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1064 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1065 | } | |
1f945357 | 1066 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1067 | |
1751c363 TM |
1068 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1069 | struct inode *inode, int ioflags) | |
1070 | { | |
1071 | if (!pnfs_pageio_init_write(pgio, inode, ioflags)) | |
1072 | nfs_pageio_init_write_mds(pgio, inode, ioflags); | |
1da177e4 LT |
1073 | } |
1074 | ||
1075 | /* | |
1076 | * Handle a write reply that flushed part of a page. | |
1077 | */ | |
788e7a89 | 1078 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 1079 | { |
788e7a89 | 1080 | struct nfs_write_data *data = calldata; |
1da177e4 | 1081 | |
48186c7d CL |
1082 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1083 | task->tk_pid, | |
3d4ff43d | 1084 | data->req->wb_context->dentry->d_inode->i_sb->s_id, |
48186c7d | 1085 | (long long) |
3d4ff43d | 1086 | NFS_FILEID(data->req->wb_context->dentry->d_inode), |
48186c7d | 1087 | data->req->wb_bytes, (long long)req_offset(data->req)); |
1da177e4 | 1088 | |
c9d8f89d TM |
1089 | nfs_writeback_done(task, data); |
1090 | } | |
788e7a89 | 1091 | |
c9d8f89d TM |
1092 | static void nfs_writeback_release_partial(void *calldata) |
1093 | { | |
1094 | struct nfs_write_data *data = calldata; | |
1095 | struct nfs_page *req = data->req; | |
1096 | struct page *page = req->wb_page; | |
1097 | int status = data->task.tk_status; | |
1098 | ||
1099 | if (status < 0) { | |
a301b777 | 1100 | nfs_set_pageerror(page); |
c9d8f89d TM |
1101 | nfs_context_set_write_error(req->wb_context, status); |
1102 | dprintk(", error = %d\n", status); | |
8e821cad | 1103 | goto out; |
1da177e4 LT |
1104 | } |
1105 | ||
8e821cad | 1106 | if (nfs_write_need_commit(data)) { |
587142f8 | 1107 | struct inode *inode = page->mapping->host; |
8e821cad | 1108 | |
587142f8 | 1109 | spin_lock(&inode->i_lock); |
8e821cad TM |
1110 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1111 | /* Do nothing we need to resend the writes */ | |
1112 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1113 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1114 | dprintk(" defer commit\n"); | |
1115 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1116 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1117 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1118 | dprintk(" server reboot detected\n"); | |
1119 | } | |
587142f8 | 1120 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1121 | } else |
1122 | dprintk(" OK\n"); | |
1123 | ||
1124 | out: | |
1da177e4 | 1125 | if (atomic_dec_and_test(&req->wb_complete)) |
a861a1e1 | 1126 | nfs_writepage_release(req, data); |
c9d8f89d | 1127 | nfs_writedata_release(calldata); |
1da177e4 LT |
1128 | } |
1129 | ||
def6ed7e AA |
1130 | #if defined(CONFIG_NFS_V4_1) |
1131 | void nfs_write_prepare(struct rpc_task *task, void *calldata) | |
1132 | { | |
1133 | struct nfs_write_data *data = calldata; | |
def6ed7e | 1134 | |
035168ab TM |
1135 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
1136 | &data->args.seq_args, | |
def6ed7e AA |
1137 | &data->res.seq_res, 1, task)) |
1138 | return; | |
1139 | rpc_call_start(task); | |
1140 | } | |
1141 | #endif /* CONFIG_NFS_V4_1 */ | |
1142 | ||
788e7a89 | 1143 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e AA |
1144 | #if defined(CONFIG_NFS_V4_1) |
1145 | .rpc_call_prepare = nfs_write_prepare, | |
1146 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1147 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1148 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1149 | }; |
1150 | ||
1da177e4 LT |
1151 | /* |
1152 | * Handle a write reply that flushes a whole page. | |
1153 | * | |
1154 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1155 | * writebacks since the page->count is kept > 1 for as long | |
1156 | * as the page has a write request pending. | |
1157 | */ | |
788e7a89 | 1158 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1159 | { |
788e7a89 | 1160 | struct nfs_write_data *data = calldata; |
1da177e4 | 1161 | |
c9d8f89d TM |
1162 | nfs_writeback_done(task, data); |
1163 | } | |
1164 | ||
1165 | static void nfs_writeback_release_full(void *calldata) | |
1166 | { | |
1167 | struct nfs_write_data *data = calldata; | |
1168 | int status = data->task.tk_status; | |
788e7a89 | 1169 | |
1da177e4 LT |
1170 | /* Update attributes as result of writeback. */ |
1171 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1172 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1173 | struct page *page = req->wb_page; | |
1174 | ||
1da177e4 | 1175 | nfs_list_remove_request(req); |
1da177e4 | 1176 | |
48186c7d CL |
1177 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1178 | data->task.tk_pid, | |
3d4ff43d AV |
1179 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1180 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1181 | req->wb_bytes, |
1182 | (long long)req_offset(req)); | |
1183 | ||
c9d8f89d | 1184 | if (status < 0) { |
a301b777 | 1185 | nfs_set_pageerror(page); |
c9d8f89d TM |
1186 | nfs_context_set_write_error(req->wb_context, status); |
1187 | dprintk(", error = %d\n", status); | |
8e821cad | 1188 | goto remove_request; |
1da177e4 | 1189 | } |
1da177e4 | 1190 | |
8e821cad TM |
1191 | if (nfs_write_need_commit(data)) { |
1192 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
a861a1e1 | 1193 | nfs_mark_request_commit(req, data->lseg); |
8e821cad | 1194 | dprintk(" marked for commit\n"); |
1da177e4 LT |
1195 | goto next; |
1196 | } | |
8e821cad TM |
1197 | dprintk(" OK\n"); |
1198 | remove_request: | |
1da177e4 | 1199 | nfs_inode_remove_request(req); |
1da177e4 | 1200 | next: |
9fd367f0 | 1201 | nfs_clear_page_tag_locked(req); |
a6305ddb | 1202 | nfs_end_page_writeback(page); |
1da177e4 | 1203 | } |
c9d8f89d | 1204 | nfs_writedata_release(calldata); |
1da177e4 LT |
1205 | } |
1206 | ||
788e7a89 | 1207 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e AA |
1208 | #if defined(CONFIG_NFS_V4_1) |
1209 | .rpc_call_prepare = nfs_write_prepare, | |
1210 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1211 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1212 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1213 | }; |
1214 | ||
1215 | ||
1da177e4 LT |
1216 | /* |
1217 | * This function is called when the WRITE call is complete. | |
1218 | */ | |
13602896 | 1219 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1220 | { |
1da177e4 LT |
1221 | struct nfs_writeargs *argp = &data->args; |
1222 | struct nfs_writeres *resp = &data->res; | |
eedc020e | 1223 | struct nfs_server *server = NFS_SERVER(data->inode); |
788e7a89 | 1224 | int status; |
1da177e4 | 1225 | |
a3f565b1 | 1226 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1227 | task->tk_pid, task->tk_status); |
1228 | ||
f551e44f CL |
1229 | /* |
1230 | * ->write_done will attempt to use post-op attributes to detect | |
1231 | * conflicting writes by other clients. A strict interpretation | |
1232 | * of close-to-open would allow us to continue caching even if | |
1233 | * another writer had changed the file, but some applications | |
1234 | * depend on tighter cache coherency when writing. | |
1235 | */ | |
788e7a89 TM |
1236 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1237 | if (status != 0) | |
13602896 | 1238 | return; |
91d5b470 CL |
1239 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1240 | ||
1da177e4 LT |
1241 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1242 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1243 | /* We tried a write call, but the server did not | |
1244 | * commit data to stable storage even though we | |
1245 | * requested it. | |
1246 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1247 | * the server reports NFS_DATA_SYNC, but performs | |
1248 | * NFS_FILE_SYNC. We therefore implement this checking | |
1249 | * as a dprintk() in order to avoid filling syslog. | |
1250 | */ | |
1251 | static unsigned long complain; | |
1252 | ||
a69aef14 | 1253 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1254 | if (time_before(complain, jiffies)) { |
48186c7d | 1255 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1256 | " (committed = %d) != (stable = %d)\n", |
eedc020e | 1257 | server->nfs_client->cl_hostname, |
1da177e4 LT |
1258 | resp->verf->committed, argp->stable); |
1259 | complain = jiffies + 300 * HZ; | |
1260 | } | |
1261 | } | |
1262 | #endif | |
1263 | /* Is this a short write? */ | |
1264 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1265 | static unsigned long complain; | |
1266 | ||
91d5b470 CL |
1267 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1268 | ||
1da177e4 LT |
1269 | /* Has the server at least made some progress? */ |
1270 | if (resp->count != 0) { | |
1271 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1272 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1273 | /* Resend from where the server left off */ | |
a69aef14 | 1274 | data->mds_offset += resp->count; |
1da177e4 LT |
1275 | argp->offset += resp->count; |
1276 | argp->pgbase += resp->count; | |
1277 | argp->count -= resp->count; | |
1278 | } else { | |
1279 | /* Resend as a stable write in order to avoid | |
1280 | * headaches in the case of a server crash. | |
1281 | */ | |
1282 | argp->stable = NFS_FILE_SYNC; | |
1283 | } | |
0110ee15 | 1284 | nfs_restart_rpc(task, server->nfs_client); |
13602896 | 1285 | return; |
1da177e4 LT |
1286 | } |
1287 | if (time_before(complain, jiffies)) { | |
1288 | printk(KERN_WARNING | |
1289 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1290 | argp->count); | |
1291 | complain = jiffies + 300 * HZ; | |
1292 | } | |
1293 | /* Can't do anything about it except throw an error. */ | |
1294 | task->tk_status = -EIO; | |
1295 | } | |
13602896 | 1296 | return; |
1da177e4 LT |
1297 | } |
1298 | ||
1299 | ||
1300 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1301 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1302 | { | |
b8413f98 TM |
1303 | int ret; |
1304 | ||
71d0a611 TM |
1305 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1306 | return 1; | |
b8413f98 TM |
1307 | if (!may_wait) |
1308 | return 0; | |
1309 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1310 | NFS_INO_COMMIT, | |
1311 | nfs_wait_bit_killable, | |
1312 | TASK_KILLABLE); | |
1313 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1314 | } |
1315 | ||
e0c2b380 | 1316 | void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1317 | { |
1318 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1319 | smp_mb__after_clear_bit(); | |
1320 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1321 | } | |
e0c2b380 | 1322 | EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); |
71d0a611 | 1323 | |
e0c2b380 | 1324 | void nfs_commitdata_release(void *data) |
1da177e4 | 1325 | { |
383ba719 TM |
1326 | struct nfs_write_data *wdata = data; |
1327 | ||
988b6dce | 1328 | put_lseg(wdata->lseg); |
383ba719 | 1329 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
1330 | nfs_commit_free(wdata); |
1331 | } | |
e0c2b380 | 1332 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1333 | |
e0c2b380 | 1334 | int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, |
9ace33cd FI |
1335 | const struct rpc_call_ops *call_ops, |
1336 | int how) | |
1da177e4 | 1337 | { |
07737691 | 1338 | struct rpc_task *task; |
9ace33cd | 1339 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1340 | struct rpc_message msg = { |
1341 | .rpc_argp = &data->args, | |
1342 | .rpc_resp = &data->res, | |
9ace33cd | 1343 | .rpc_cred = data->cred, |
bdc7f021 | 1344 | }; |
84115e1c | 1345 | struct rpc_task_setup task_setup_data = { |
07737691 | 1346 | .task = &data->task, |
9ace33cd | 1347 | .rpc_client = clnt, |
bdc7f021 | 1348 | .rpc_message = &msg, |
9ace33cd | 1349 | .callback_ops = call_ops, |
84115e1c | 1350 | .callback_data = data, |
101070ca | 1351 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1352 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1353 | .priority = priority, |
84115e1c | 1354 | }; |
9ace33cd FI |
1355 | /* Set up the initial task struct. */ |
1356 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1357 | ||
1358 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1359 | ||
1360 | task = rpc_run_task(&task_setup_data); | |
1361 | if (IS_ERR(task)) | |
1362 | return PTR_ERR(task); | |
1363 | if (how & FLUSH_SYNC) | |
1364 | rpc_wait_for_completion_task(task); | |
1365 | rpc_put_task(task); | |
1366 | return 0; | |
1367 | } | |
e0c2b380 | 1368 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1369 | |
1370 | /* | |
1371 | * Set up the argument/result storage required for the RPC call. | |
1372 | */ | |
e0c2b380 | 1373 | void nfs_init_commit(struct nfs_write_data *data, |
988b6dce FI |
1374 | struct list_head *head, |
1375 | struct pnfs_layout_segment *lseg) | |
9ace33cd FI |
1376 | { |
1377 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1378 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1379 | |
1380 | /* Set up the RPC argument and reply structs | |
1381 | * NB: take care not to mess about with data->commit et al. */ | |
1382 | ||
1383 | list_splice_init(head, &data->pages); | |
1da177e4 | 1384 | |
1da177e4 | 1385 | data->inode = inode; |
9ace33cd | 1386 | data->cred = first->wb_context->cred; |
988b6dce | 1387 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1388 | data->mds_ops = &nfs_commit_ops; |
1da177e4 LT |
1389 | |
1390 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1391 | /* Note: we always request a commit of the entire inode */ |
1392 | data->args.offset = 0; | |
1393 | data->args.count = 0; | |
383ba719 | 1394 | data->args.context = get_nfs_open_context(first->wb_context); |
3da28eb1 | 1395 | data->res.count = 0; |
1da177e4 LT |
1396 | data->res.fattr = &data->fattr; |
1397 | data->res.verf = &data->verf; | |
0e574af1 | 1398 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1399 | } |
e0c2b380 | 1400 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1401 | |
e0c2b380 | 1402 | void nfs_retry_commit(struct list_head *page_list, |
a861a1e1 | 1403 | struct pnfs_layout_segment *lseg) |
64bfeb49 FI |
1404 | { |
1405 | struct nfs_page *req; | |
1406 | ||
1407 | while (!list_empty(page_list)) { | |
1408 | req = nfs_list_entry(page_list->next); | |
1409 | nfs_list_remove_request(req); | |
a861a1e1 | 1410 | nfs_mark_request_commit(req, lseg); |
64bfeb49 FI |
1411 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1412 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1413 | BDI_RECLAIMABLE); | |
1414 | nfs_clear_page_tag_locked(req); | |
1415 | } | |
1416 | } | |
e0c2b380 | 1417 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1418 | |
1da177e4 LT |
1419 | /* |
1420 | * Commit dirty pages | |
1421 | */ | |
1422 | static int | |
40859d7e | 1423 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1424 | { |
1425 | struct nfs_write_data *data; | |
1da177e4 | 1426 | |
c9d8f89d | 1427 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1428 | |
1429 | if (!data) | |
1430 | goto out_bad; | |
1431 | ||
1432 | /* Set up the argument struct */ | |
988b6dce | 1433 | nfs_init_commit(data, head, NULL); |
9ace33cd | 1434 | return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); |
1da177e4 | 1435 | out_bad: |
a861a1e1 | 1436 | nfs_retry_commit(head, NULL); |
71d0a611 | 1437 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1438 | return -ENOMEM; |
1439 | } | |
1440 | ||
1441 | /* | |
1442 | * COMMIT call returned | |
1443 | */ | |
788e7a89 | 1444 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1445 | { |
963d8fe5 | 1446 | struct nfs_write_data *data = calldata; |
1da177e4 | 1447 | |
a3f565b1 | 1448 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1449 | task->tk_pid, task->tk_status); |
1450 | ||
788e7a89 | 1451 | /* Call the NFS version-specific code */ |
c0d0e96b | 1452 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1453 | } |
1454 | ||
e0c2b380 | 1455 | void nfs_commit_release_pages(struct nfs_write_data *data) |
c9d8f89d | 1456 | { |
5917ce84 | 1457 | struct nfs_page *req; |
c9d8f89d | 1458 | int status = data->task.tk_status; |
788e7a89 | 1459 | |
1da177e4 LT |
1460 | while (!list_empty(&data->pages)) { |
1461 | req = nfs_list_entry(data->pages.next); | |
1462 | nfs_list_remove_request(req); | |
e468bae9 | 1463 | nfs_clear_request_commit(req); |
1da177e4 | 1464 | |
48186c7d | 1465 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1466 | req->wb_context->dentry->d_sb->s_id, |
1467 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1468 | req->wb_bytes, |
1469 | (long long)req_offset(req)); | |
c9d8f89d TM |
1470 | if (status < 0) { |
1471 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1472 | nfs_inode_remove_request(req); |
c9d8f89d | 1473 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1474 | goto next; |
1475 | } | |
1476 | ||
1477 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1478 | * returned by the server against all stored verfs. */ | |
1479 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1480 | /* We have a match */ | |
1481 | nfs_inode_remove_request(req); | |
1482 | dprintk(" OK\n"); | |
1483 | goto next; | |
1484 | } | |
1485 | /* We have a mismatch. Write the page again */ | |
1486 | dprintk(" mismatch\n"); | |
6d884e8f | 1487 | nfs_mark_request_dirty(req); |
1da177e4 | 1488 | next: |
9fd367f0 | 1489 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1490 | } |
5917ce84 | 1491 | } |
e0c2b380 | 1492 | EXPORT_SYMBOL_GPL(nfs_commit_release_pages); |
5917ce84 FI |
1493 | |
1494 | static void nfs_commit_release(void *calldata) | |
1495 | { | |
1496 | struct nfs_write_data *data = calldata; | |
1497 | ||
1498 | nfs_commit_release_pages(data); | |
71d0a611 | 1499 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1500 | nfs_commitdata_release(calldata); |
1da177e4 | 1501 | } |
788e7a89 TM |
1502 | |
1503 | static const struct rpc_call_ops nfs_commit_ops = { | |
21d9a851 AA |
1504 | #if defined(CONFIG_NFS_V4_1) |
1505 | .rpc_call_prepare = nfs_write_prepare, | |
1506 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 TM |
1507 | .rpc_call_done = nfs_commit_done, |
1508 | .rpc_release = nfs_commit_release, | |
1509 | }; | |
1da177e4 | 1510 | |
b608b283 | 1511 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1512 | { |
1da177e4 | 1513 | LIST_HEAD(head); |
71d0a611 | 1514 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1515 | int res; |
1da177e4 | 1516 | |
b8413f98 TM |
1517 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1518 | if (res <= 0) | |
c5efa5fc | 1519 | goto out_mark_dirty; |
3da28eb1 | 1520 | res = nfs_scan_commit(inode, &head, 0, 0); |
1da177e4 | 1521 | if (res) { |
a861a1e1 FI |
1522 | int error; |
1523 | ||
1524 | error = pnfs_commit_list(inode, &head, how); | |
1525 | if (error == PNFS_NOT_ATTEMPTED) | |
1526 | error = nfs_commit_list(inode, &head, how); | |
3da28eb1 TM |
1527 | if (error < 0) |
1528 | return error; | |
b8413f98 | 1529 | if (!may_wait) |
c5efa5fc | 1530 | goto out_mark_dirty; |
b8413f98 TM |
1531 | error = wait_on_bit(&NFS_I(inode)->flags, |
1532 | NFS_INO_COMMIT, | |
1533 | nfs_wait_bit_killable, | |
1534 | TASK_KILLABLE); | |
1535 | if (error < 0) | |
1536 | return error; | |
71d0a611 TM |
1537 | } else |
1538 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1539 | return res; |
1540 | /* Note: If we exit without ensuring that the commit is complete, | |
1541 | * we must mark the inode as dirty. Otherwise, future calls to | |
1542 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1543 | * that the data is on the disk. | |
1544 | */ | |
1545 | out_mark_dirty: | |
1546 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1547 | return res; |
1548 | } | |
8fc795f7 TM |
1549 | |
1550 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1551 | { | |
420e3646 TM |
1552 | struct nfs_inode *nfsi = NFS_I(inode); |
1553 | int flags = FLUSH_SYNC; | |
1554 | int ret = 0; | |
8fc795f7 | 1555 | |
a00dd6c0 JL |
1556 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1557 | /* Don't commit yet if this is a non-blocking flush and there | |
1558 | * are a lot of outstanding writes for this mapping. | |
1559 | */ | |
1560 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1561 | goto out_mark_dirty; | |
420e3646 | 1562 | |
a00dd6c0 | 1563 | /* don't wait for the COMMIT response */ |
420e3646 | 1564 | flags = 0; |
a00dd6c0 JL |
1565 | } |
1566 | ||
420e3646 TM |
1567 | ret = nfs_commit_inode(inode, flags); |
1568 | if (ret >= 0) { | |
1569 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1570 | if (ret < wbc->nr_to_write) | |
1571 | wbc->nr_to_write -= ret; | |
1572 | else | |
1573 | wbc->nr_to_write = 0; | |
1574 | } | |
8fc795f7 | 1575 | return 0; |
420e3646 TM |
1576 | } |
1577 | out_mark_dirty: | |
8fc795f7 TM |
1578 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1579 | return ret; | |
1580 | } | |
c63c7b05 | 1581 | #else |
8fc795f7 TM |
1582 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1583 | { | |
1584 | return 0; | |
1585 | } | |
1da177e4 LT |
1586 | #endif |
1587 | ||
8fc795f7 TM |
1588 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1589 | { | |
863a3c6c AA |
1590 | int ret; |
1591 | ||
1592 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1593 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1594 | int status; |
1595 | bool sync = true; | |
863a3c6c | 1596 | |
846d5a09 | 1597 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1598 | sync = false; |
863a3c6c AA |
1599 | |
1600 | status = pnfs_layoutcommit_inode(inode, sync); | |
1601 | if (status < 0) | |
1602 | return status; | |
1603 | } | |
1604 | return ret; | |
8fc795f7 TM |
1605 | } |
1606 | ||
acdc53b2 TM |
1607 | /* |
1608 | * flush the inode to disk. | |
1609 | */ | |
1610 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1611 | { |
1612 | struct writeback_control wbc = { | |
72cb77f4 | 1613 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1614 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1615 | .range_start = 0, |
1616 | .range_end = LLONG_MAX, | |
34901f70 | 1617 | }; |
34901f70 | 1618 | |
acdc53b2 | 1619 | return sync_inode(inode, &wbc); |
1c75950b TM |
1620 | } |
1621 | ||
1b3b4a1a TM |
1622 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1623 | { | |
1624 | struct nfs_page *req; | |
1b3b4a1a TM |
1625 | int ret = 0; |
1626 | ||
1627 | BUG_ON(!PageLocked(page)); | |
1628 | for (;;) { | |
ba8b06e6 | 1629 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1630 | req = nfs_page_find_request(page); |
1631 | if (req == NULL) | |
1b3b4a1a | 1632 | break; |
1b3b4a1a TM |
1633 | if (nfs_lock_request_dontget(req)) { |
1634 | nfs_inode_remove_request(req); | |
1635 | /* | |
1636 | * In case nfs_inode_remove_request has marked the | |
1637 | * page as being dirty | |
1638 | */ | |
1639 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1640 | nfs_unlock_request(req); | |
1641 | break; | |
1642 | } | |
1643 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1644 | nfs_release_request(req); |
1b3b4a1a | 1645 | if (ret < 0) |
c988950e | 1646 | break; |
1b3b4a1a | 1647 | } |
1b3b4a1a TM |
1648 | return ret; |
1649 | } | |
1650 | ||
7f2f12d9 TM |
1651 | /* |
1652 | * Write back all requests on one page - we do this before reading it. | |
1653 | */ | |
1654 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1655 | { |
1656 | loff_t range_start = page_offset(page); | |
1657 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1658 | struct writeback_control wbc = { |
4d770ccf | 1659 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1660 | .nr_to_write = 0, |
4d770ccf TM |
1661 | .range_start = range_start, |
1662 | .range_end = range_end, | |
1663 | }; | |
1664 | int ret; | |
1c75950b | 1665 | |
0522f6ad | 1666 | for (;;) { |
ba8b06e6 | 1667 | wait_on_page_writeback(page); |
73e3302f TM |
1668 | if (clear_page_dirty_for_io(page)) { |
1669 | ret = nfs_writepage_locked(page, &wbc); | |
1670 | if (ret < 0) | |
1671 | goto out_error; | |
0522f6ad | 1672 | continue; |
7f2f12d9 | 1673 | } |
0522f6ad TM |
1674 | if (!PagePrivate(page)) |
1675 | break; | |
1676 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1677 | if (ret < 0) |
73e3302f | 1678 | goto out_error; |
7f2f12d9 | 1679 | } |
73e3302f TM |
1680 | return 0; |
1681 | out_error: | |
4d770ccf | 1682 | return ret; |
1c75950b TM |
1683 | } |
1684 | ||
074cc1de TM |
1685 | #ifdef CONFIG_MIGRATION |
1686 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
1687 | struct page *page) | |
1688 | { | |
1689 | struct nfs_page *req; | |
1690 | int ret; | |
1691 | ||
7549ad5f | 1692 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1693 | |
cfb506e1 | 1694 | req = nfs_find_and_lock_request(page, false); |
074cc1de TM |
1695 | ret = PTR_ERR(req); |
1696 | if (IS_ERR(req)) | |
1697 | goto out; | |
1698 | ||
1699 | ret = migrate_page(mapping, newpage, page); | |
1700 | if (!req) | |
1701 | goto out; | |
1702 | if (ret) | |
1703 | goto out_unlock; | |
1704 | page_cache_get(newpage); | |
190f38e5 | 1705 | spin_lock(&mapping->host->i_lock); |
074cc1de TM |
1706 | req->wb_page = newpage; |
1707 | SetPagePrivate(newpage); | |
190f38e5 | 1708 | set_page_private(newpage, (unsigned long)req); |
074cc1de TM |
1709 | ClearPagePrivate(page); |
1710 | set_page_private(page, 0); | |
190f38e5 | 1711 | spin_unlock(&mapping->host->i_lock); |
074cc1de TM |
1712 | page_cache_release(page); |
1713 | out_unlock: | |
1714 | nfs_clear_page_tag_locked(req); | |
074cc1de TM |
1715 | out: |
1716 | return ret; | |
1717 | } | |
1718 | #endif | |
1719 | ||
f7b422b1 | 1720 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1721 | { |
1722 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1723 | sizeof(struct nfs_write_data), | |
1724 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1725 | NULL); |
1da177e4 LT |
1726 | if (nfs_wdata_cachep == NULL) |
1727 | return -ENOMEM; | |
1728 | ||
93d2341c MD |
1729 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1730 | nfs_wdata_cachep); | |
1da177e4 LT |
1731 | if (nfs_wdata_mempool == NULL) |
1732 | return -ENOMEM; | |
1733 | ||
93d2341c MD |
1734 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1735 | nfs_wdata_cachep); | |
1da177e4 LT |
1736 | if (nfs_commit_mempool == NULL) |
1737 | return -ENOMEM; | |
1738 | ||
89a09141 PZ |
1739 | /* |
1740 | * NFS congestion size, scale with available memory. | |
1741 | * | |
1742 | * 64MB: 8192k | |
1743 | * 128MB: 11585k | |
1744 | * 256MB: 16384k | |
1745 | * 512MB: 23170k | |
1746 | * 1GB: 32768k | |
1747 | * 2GB: 46340k | |
1748 | * 4GB: 65536k | |
1749 | * 8GB: 92681k | |
1750 | * 16GB: 131072k | |
1751 | * | |
1752 | * This allows larger machines to have larger/more transfers. | |
1753 | * Limit the default to 256M | |
1754 | */ | |
1755 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1756 | if (nfs_congestion_kb > 256*1024) | |
1757 | nfs_congestion_kb = 256*1024; | |
1758 | ||
1da177e4 LT |
1759 | return 0; |
1760 | } | |
1761 | ||
266bee88 | 1762 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1763 | { |
1764 | mempool_destroy(nfs_commit_mempool); | |
1765 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1766 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1767 | } |
1768 |