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 TM |
430 | __set_page_dirty_nobuffers(req->wb_page); |
431 | } | |
432 | ||
1da177e4 LT |
433 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
434 | /* | |
435 | * Add a request to the inode's commit list. | |
436 | */ | |
437 | static void | |
a861a1e1 | 438 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
1da177e4 | 439 | { |
3d4ff43d | 440 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
441 | struct nfs_inode *nfsi = NFS_I(inode); |
442 | ||
587142f8 | 443 | spin_lock(&inode->i_lock); |
e468bae9 | 444 | set_bit(PG_CLEAN, &(req)->wb_flags); |
5c369683 TM |
445 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
446 | req->wb_index, | |
447 | NFS_PAGE_TAG_COMMIT); | |
ff778d02 | 448 | nfsi->ncommit++; |
587142f8 | 449 | spin_unlock(&inode->i_lock); |
a861a1e1 | 450 | pnfs_mark_request_commit(req, lseg); |
fd39fc85 | 451 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 452 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 453 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 454 | } |
8e821cad | 455 | |
e468bae9 TM |
456 | static int |
457 | nfs_clear_request_commit(struct nfs_page *req) | |
458 | { | |
459 | struct page *page = req->wb_page; | |
460 | ||
461 | if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { | |
462 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
463 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
464 | return 1; | |
465 | } | |
466 | return 0; | |
467 | } | |
468 | ||
8e821cad TM |
469 | static inline |
470 | int nfs_write_need_commit(struct nfs_write_data *data) | |
471 | { | |
465d5243 FI |
472 | if (data->verf.committed == NFS_DATA_SYNC) |
473 | return data->lseg == NULL; | |
474 | else | |
475 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
476 | } |
477 | ||
478 | static inline | |
a861a1e1 FI |
479 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
480 | struct nfs_write_data *data) | |
8e821cad | 481 | { |
e468bae9 | 482 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
a861a1e1 | 483 | nfs_mark_request_commit(req, data->lseg); |
8e821cad TM |
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 | |
a861a1e1 | 494 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
8e821cad TM |
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 | |
a861a1e1 FI |
511 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
512 | struct nfs_write_data *data) | |
8e821cad TM |
513 | { |
514 | return 0; | |
515 | } | |
1da177e4 LT |
516 | #endif |
517 | ||
47c62564 | 518 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
519 | static int |
520 | nfs_need_commit(struct nfs_inode *nfsi) | |
521 | { | |
522 | return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); | |
523 | } | |
524 | ||
1da177e4 LT |
525 | /* |
526 | * nfs_scan_commit - Scan an inode for commit requests | |
527 | * @inode: NFS inode to scan | |
528 | * @dst: destination list | |
529 | * @idx_start: lower bound of page->index to scan. | |
530 | * @npages: idx_start + npages sets the upper bound to scan. | |
531 | * | |
532 | * Moves requests from the inode's 'commit' request list. | |
533 | * The requests are *not* checked to ensure that they form a contiguous set. | |
534 | */ | |
535 | static int | |
ca52fec1 | 536 | nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
537 | { |
538 | struct nfs_inode *nfsi = NFS_I(inode); | |
ff778d02 | 539 | int ret; |
3da28eb1 | 540 | |
fb8a1f11 TM |
541 | if (!nfs_need_commit(nfsi)) |
542 | return 0; | |
543 | ||
0d88f6e8 | 544 | spin_lock(&inode->i_lock); |
ff778d02 TM |
545 | ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); |
546 | if (ret > 0) | |
547 | nfsi->ncommit -= ret; | |
0d88f6e8 DC |
548 | spin_unlock(&inode->i_lock); |
549 | ||
2928db1f TM |
550 | if (nfs_need_commit(NFS_I(inode))) |
551 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
0d88f6e8 | 552 | |
ff778d02 | 553 | return ret; |
1da177e4 | 554 | } |
c42de9dd | 555 | #else |
fb8a1f11 TM |
556 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
557 | { | |
558 | return 0; | |
559 | } | |
560 | ||
ca52fec1 | 561 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
c42de9dd TM |
562 | { |
563 | return 0; | |
564 | } | |
1da177e4 LT |
565 | #endif |
566 | ||
1da177e4 | 567 | /* |
e7d39069 TM |
568 | * Search for an existing write request, and attempt to update |
569 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 570 | * |
e7d39069 TM |
571 | * If the attempt fails, then the existing request is flushed out |
572 | * to disk. | |
1da177e4 | 573 | */ |
e7d39069 TM |
574 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
575 | struct page *page, | |
576 | unsigned int offset, | |
577 | unsigned int bytes) | |
1da177e4 | 578 | { |
e7d39069 TM |
579 | struct nfs_page *req; |
580 | unsigned int rqend; | |
581 | unsigned int end; | |
582 | int error; | |
583 | ||
584 | if (!PagePrivate(page)) | |
585 | return NULL; | |
1da177e4 LT |
586 | |
587 | end = offset + bytes; | |
e7d39069 | 588 | spin_lock(&inode->i_lock); |
1da177e4 | 589 | |
1da177e4 | 590 | for (;;) { |
277459d2 | 591 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
592 | if (req == NULL) |
593 | goto out_unlock; | |
594 | ||
595 | rqend = req->wb_offset + req->wb_bytes; | |
596 | /* | |
597 | * Tell the caller to flush out the request if | |
598 | * the offsets are non-contiguous. | |
599 | * Note: nfs_flush_incompatible() will already | |
600 | * have flushed out requests having wrong owners. | |
601 | */ | |
e468bae9 | 602 | if (offset > rqend |
e7d39069 TM |
603 | || end < req->wb_offset) |
604 | goto out_flushme; | |
605 | ||
606 | if (nfs_set_page_tag_locked(req)) | |
1da177e4 | 607 | break; |
1da177e4 | 608 | |
e7d39069 | 609 | /* The request is locked, so wait and then retry */ |
587142f8 | 610 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
611 | error = nfs_wait_on_request(req); |
612 | nfs_release_request(req); | |
613 | if (error != 0) | |
614 | goto out_err; | |
615 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
616 | } |
617 | ||
ff778d02 | 618 | if (nfs_clear_request_commit(req) && |
a861a1e1 FI |
619 | radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, |
620 | req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) { | |
ff778d02 | 621 | NFS_I(inode)->ncommit--; |
a861a1e1 FI |
622 | pnfs_clear_request_commit(req); |
623 | } | |
e468bae9 | 624 | |
1da177e4 LT |
625 | /* Okay, the request matches. Update the region */ |
626 | if (offset < req->wb_offset) { | |
627 | req->wb_offset = offset; | |
628 | req->wb_pgbase = offset; | |
1da177e4 | 629 | } |
1da177e4 LT |
630 | if (end > rqend) |
631 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
632 | else |
633 | req->wb_bytes = rqend - req->wb_offset; | |
634 | out_unlock: | |
635 | spin_unlock(&inode->i_lock); | |
636 | return req; | |
637 | out_flushme: | |
638 | spin_unlock(&inode->i_lock); | |
639 | nfs_release_request(req); | |
640 | error = nfs_wb_page(inode, page); | |
641 | out_err: | |
642 | return ERR_PTR(error); | |
643 | } | |
644 | ||
645 | /* | |
646 | * Try to update an existing write request, or create one if there is none. | |
647 | * | |
648 | * Note: Should always be called with the Page Lock held to prevent races | |
649 | * if we have to add a new request. Also assumes that the caller has | |
650 | * already called nfs_flush_incompatible() if necessary. | |
651 | */ | |
652 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
653 | struct page *page, unsigned int offset, unsigned int bytes) | |
654 | { | |
655 | struct inode *inode = page->mapping->host; | |
656 | struct nfs_page *req; | |
657 | int error; | |
1da177e4 | 658 | |
e7d39069 TM |
659 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
660 | if (req != NULL) | |
661 | goto out; | |
662 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
663 | if (IS_ERR(req)) | |
664 | goto out; | |
665 | error = nfs_inode_add_request(inode, req); | |
666 | if (error != 0) { | |
667 | nfs_release_request(req); | |
668 | req = ERR_PTR(error); | |
669 | } | |
efc91ed0 | 670 | out: |
61e930a9 | 671 | return req; |
1da177e4 LT |
672 | } |
673 | ||
e7d39069 TM |
674 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
675 | unsigned int offset, unsigned int count) | |
676 | { | |
677 | struct nfs_page *req; | |
678 | ||
679 | req = nfs_setup_write_request(ctx, page, offset, count); | |
680 | if (IS_ERR(req)) | |
681 | return PTR_ERR(req); | |
682 | /* Update file length */ | |
683 | nfs_grow_file(page, offset, count); | |
684 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 685 | nfs_mark_request_dirty(req); |
e7d39069 TM |
686 | nfs_clear_page_tag_locked(req); |
687 | return 0; | |
688 | } | |
689 | ||
1da177e4 LT |
690 | int nfs_flush_incompatible(struct file *file, struct page *page) |
691 | { | |
cd3758e3 | 692 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 693 | struct nfs_page *req; |
1a54533e | 694 | int do_flush, status; |
1da177e4 LT |
695 | /* |
696 | * Look for a request corresponding to this page. If there | |
697 | * is one, and it belongs to another file, we flush it out | |
698 | * before we try to copy anything into the page. Do this | |
699 | * due to the lack of an ACCESS-type call in NFSv2. | |
700 | * Also do the same if we find a request from an existing | |
701 | * dropped page. | |
702 | */ | |
1a54533e TM |
703 | do { |
704 | req = nfs_page_find_request(page); | |
705 | if (req == NULL) | |
706 | return 0; | |
f11ac8db TM |
707 | do_flush = req->wb_page != page || req->wb_context != ctx || |
708 | req->wb_lock_context->lockowner != current->files || | |
709 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 710 | nfs_release_request(req); |
1a54533e TM |
711 | if (!do_flush) |
712 | return 0; | |
713 | status = nfs_wb_page(page->mapping->host, page); | |
714 | } while (status == 0); | |
715 | return status; | |
1da177e4 LT |
716 | } |
717 | ||
5d47a356 TM |
718 | /* |
719 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
720 | * the PageUptodate() flag. In this case, we will need to turn off | |
721 | * write optimisations that depend on the page contents being correct. | |
722 | */ | |
723 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
724 | { | |
725 | return PageUptodate(page) && | |
726 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
727 | } | |
728 | ||
1da177e4 LT |
729 | /* |
730 | * Update and possibly write a cached page of an NFS file. | |
731 | * | |
732 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
733 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
734 | */ | |
735 | int nfs_updatepage(struct file *file, struct page *page, | |
736 | unsigned int offset, unsigned int count) | |
737 | { | |
cd3758e3 | 738 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 739 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
740 | int status = 0; |
741 | ||
91d5b470 CL |
742 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
743 | ||
48186c7d | 744 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
745 | file->f_path.dentry->d_parent->d_name.name, |
746 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 747 | (long long)(page_offset(page) + offset)); |
1da177e4 | 748 | |
1da177e4 | 749 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
750 | * is up to date, it may be more efficient to extend the write |
751 | * to cover the entire page in order to avoid fragmentation | |
752 | * inefficiencies. | |
1da177e4 | 753 | */ |
5d47a356 TM |
754 | if (nfs_write_pageuptodate(page, inode) && |
755 | inode->i_flock == NULL && | |
6b2f3d1f | 756 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 757 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 758 | offset = 0; |
1da177e4 LT |
759 | } |
760 | ||
e21195a7 | 761 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
762 | if (status < 0) |
763 | nfs_set_pageerror(page); | |
59b7c05f TM |
764 | else |
765 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 766 | |
48186c7d | 767 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 768 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
769 | return status; |
770 | } | |
771 | ||
a861a1e1 FI |
772 | static void nfs_writepage_release(struct nfs_page *req, |
773 | struct nfs_write_data *data) | |
1da177e4 | 774 | { |
a6305ddb | 775 | struct page *page = req->wb_page; |
1da177e4 | 776 | |
a861a1e1 | 777 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) |
8e821cad | 778 | nfs_inode_remove_request(req); |
9fd367f0 | 779 | nfs_clear_page_tag_locked(req); |
a6305ddb | 780 | nfs_end_page_writeback(page); |
1da177e4 LT |
781 | } |
782 | ||
3ff7576d | 783 | static int flush_task_priority(int how) |
1da177e4 LT |
784 | { |
785 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
786 | case FLUSH_HIGHPRI: | |
787 | return RPC_PRIORITY_HIGH; | |
788 | case FLUSH_LOWPRI: | |
789 | return RPC_PRIORITY_LOW; | |
790 | } | |
791 | return RPC_PRIORITY_NORMAL; | |
792 | } | |
793 | ||
a69aef14 | 794 | int nfs_initiate_write(struct nfs_write_data *data, |
d138d5d1 AA |
795 | struct rpc_clnt *clnt, |
796 | const struct rpc_call_ops *call_ops, | |
797 | int how) | |
1da177e4 | 798 | { |
d138d5d1 | 799 | struct inode *inode = data->inode; |
3ff7576d | 800 | int priority = flush_task_priority(how); |
07737691 | 801 | struct rpc_task *task; |
bdc7f021 TM |
802 | struct rpc_message msg = { |
803 | .rpc_argp = &data->args, | |
804 | .rpc_resp = &data->res, | |
d138d5d1 | 805 | .rpc_cred = data->cred, |
bdc7f021 | 806 | }; |
84115e1c | 807 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 808 | .rpc_client = clnt, |
07737691 | 809 | .task = &data->task, |
bdc7f021 | 810 | .rpc_message = &msg, |
84115e1c TM |
811 | .callback_ops = call_ops, |
812 | .callback_data = data, | |
101070ca | 813 | .workqueue = nfsiod_workqueue, |
2c61be0a | 814 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 815 | .priority = priority, |
84115e1c | 816 | }; |
2c61be0a | 817 | int ret = 0; |
1da177e4 | 818 | |
d138d5d1 AA |
819 | /* Set up the initial task struct. */ |
820 | NFS_PROTO(inode)->write_setup(data, &msg); | |
821 | ||
822 | dprintk("NFS: %5u initiated write call " | |
823 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
824 | data->task.tk_pid, | |
825 | inode->i_sb->s_id, | |
826 | (long long)NFS_FILEID(inode), | |
827 | data->args.count, | |
828 | (unsigned long long)data->args.offset); | |
829 | ||
830 | task = rpc_run_task(&task_setup_data); | |
831 | if (IS_ERR(task)) { | |
832 | ret = PTR_ERR(task); | |
833 | goto out; | |
834 | } | |
835 | if (how & FLUSH_SYNC) { | |
836 | ret = rpc_wait_for_completion_task(task); | |
837 | if (ret == 0) | |
838 | ret = task->tk_status; | |
839 | } | |
840 | rpc_put_task(task); | |
841 | out: | |
842 | return ret; | |
843 | } | |
a69aef14 | 844 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
845 | |
846 | /* | |
847 | * Set up the argument/result storage required for the RPC call. | |
848 | */ | |
6e4efd56 | 849 | static void nfs_write_rpcsetup(struct nfs_page *req, |
d138d5d1 | 850 | struct nfs_write_data *data, |
d138d5d1 AA |
851 | unsigned int count, unsigned int offset, |
852 | int how) | |
853 | { | |
3d4ff43d | 854 | struct inode *inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 855 | |
1da177e4 LT |
856 | /* Set up the RPC argument and reply structs |
857 | * NB: take care not to mess about with data->commit et al. */ | |
858 | ||
859 | data->req = req; | |
3d4ff43d | 860 | data->inode = inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 861 | data->cred = req->wb_context->cred; |
1da177e4 LT |
862 | |
863 | data->args.fh = NFS_FH(inode); | |
864 | data->args.offset = req_offset(req) + offset; | |
2bea038c BH |
865 | /* pnfs_set_layoutcommit needs this */ |
866 | data->mds_offset = data->args.offset; | |
1da177e4 LT |
867 | data->args.pgbase = req->wb_pgbase + offset; |
868 | data->args.pages = data->pagevec; | |
869 | data->args.count = count; | |
383ba719 | 870 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 871 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 872 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
873 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
874 | case 0: | |
875 | break; | |
876 | case FLUSH_COND_STABLE: | |
877 | if (nfs_need_commit(NFS_I(inode))) | |
878 | break; | |
879 | default: | |
880 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 881 | } |
1da177e4 LT |
882 | |
883 | data->res.fattr = &data->fattr; | |
884 | data->res.count = count; | |
885 | data->res.verf = &data->verf; | |
0e574af1 | 886 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 887 | } |
1da177e4 | 888 | |
6e4efd56 TM |
889 | static int nfs_do_write(struct nfs_write_data *data, |
890 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
891 | int how) |
892 | { | |
5f00bcb3 | 893 | struct inode *inode = data->args.context->dentry->d_inode; |
0382b744 | 894 | |
d138d5d1 | 895 | return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); |
1da177e4 LT |
896 | } |
897 | ||
275acaaf TM |
898 | static int nfs_do_multiple_writes(struct list_head *head, |
899 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
900 | int how) |
901 | { | |
902 | struct nfs_write_data *data; | |
903 | int ret = 0; | |
904 | ||
905 | while (!list_empty(head)) { | |
906 | int ret2; | |
907 | ||
908 | data = list_entry(head->next, struct nfs_write_data, list); | |
909 | list_del_init(&data->list); | |
910 | ||
dce81290 | 911 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
912 | if (ret == 0) |
913 | ret = ret2; | |
914 | } | |
915 | return ret; | |
916 | } | |
917 | ||
6d884e8f F |
918 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
919 | * call this on each, which will prepare them to be retried on next | |
920 | * writeback using standard nfs. | |
921 | */ | |
922 | static void nfs_redirty_request(struct nfs_page *req) | |
923 | { | |
a6305ddb TM |
924 | struct page *page = req->wb_page; |
925 | ||
6d884e8f | 926 | nfs_mark_request_dirty(req); |
6d884e8f | 927 | nfs_clear_page_tag_locked(req); |
a6305ddb | 928 | nfs_end_page_writeback(page); |
6d884e8f F |
929 | } |
930 | ||
1da177e4 LT |
931 | /* |
932 | * Generate multiple small requests to write out a single | |
933 | * contiguous dirty area on one page. | |
934 | */ | |
275acaaf | 935 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 | 936 | { |
c76069bd | 937 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
938 | struct page *page = req->wb_page; |
939 | struct nfs_write_data *data; | |
d097971d | 940 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 941 | unsigned int offset; |
1da177e4 | 942 | int requests = 0; |
dbae4c73 | 943 | int ret = 0; |
1da177e4 LT |
944 | |
945 | nfs_list_remove_request(req); | |
946 | ||
b31268ac TM |
947 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
948 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || | |
949 | desc->pg_count > wsize)) | |
950 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
951 | ||
952 | ||
275acaaf | 953 | offset = 0; |
c76069bd | 954 | nbytes = desc->pg_count; |
e9f7bee1 TM |
955 | do { |
956 | size_t len = min(nbytes, wsize); | |
957 | ||
8d5658c9 | 958 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
959 | if (!data) |
960 | goto out_bad; | |
275acaaf | 961 | data->pagevec[0] = page; |
f13c3620 | 962 | nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags); |
275acaaf | 963 | list_add(&data->list, res); |
1da177e4 | 964 | requests++; |
e9f7bee1 | 965 | nbytes -= len; |
275acaaf | 966 | offset += len; |
e9f7bee1 | 967 | } while (nbytes != 0); |
1da177e4 | 968 | atomic_set(&req->wb_complete, requests); |
50828d7e | 969 | desc->pg_rpc_callops = &nfs_write_partial_ops; |
dbae4c73 | 970 | return ret; |
1da177e4 LT |
971 | |
972 | out_bad: | |
275acaaf TM |
973 | while (!list_empty(res)) { |
974 | data = list_entry(res->next, struct nfs_write_data, list); | |
6e4efd56 | 975 | list_del(&data->list); |
0da2a4ac | 976 | nfs_writedata_free(data); |
1da177e4 | 977 | } |
61822ab5 | 978 | nfs_redirty_request(req); |
1da177e4 LT |
979 | return -ENOMEM; |
980 | } | |
981 | ||
982 | /* | |
983 | * Create an RPC task for the given write request and kick it. | |
984 | * The page must have been locked by the caller. | |
985 | * | |
986 | * It may happen that the page we're passed is not marked dirty. | |
987 | * This is the case if nfs_updatepage detects a conflicting request | |
988 | * that has been written but not committed. | |
989 | */ | |
275acaaf | 990 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 LT |
991 | { |
992 | struct nfs_page *req; | |
993 | struct page **pages; | |
994 | struct nfs_write_data *data; | |
c76069bd | 995 | struct list_head *head = &desc->pg_list; |
3b609184 | 996 | int ret = 0; |
1da177e4 | 997 | |
c76069bd FI |
998 | data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, |
999 | desc->pg_count)); | |
44b83799 FI |
1000 | if (!data) { |
1001 | while (!list_empty(head)) { | |
1002 | req = nfs_list_entry(head->next); | |
1003 | nfs_list_remove_request(req); | |
1004 | nfs_redirty_request(req); | |
1005 | } | |
1006 | ret = -ENOMEM; | |
1007 | goto out; | |
1008 | } | |
1da177e4 | 1009 | pages = data->pagevec; |
1da177e4 LT |
1010 | while (!list_empty(head)) { |
1011 | req = nfs_list_entry(head->next); | |
1012 | nfs_list_remove_request(req); | |
1013 | nfs_list_add_request(req, &data->pages); | |
1014 | ClearPageError(req->wb_page); | |
1da177e4 | 1015 | *pages++ = req->wb_page; |
1da177e4 LT |
1016 | } |
1017 | req = nfs_list_entry(data->pages.next); | |
1018 | ||
b31268ac TM |
1019 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1020 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) | |
1021 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1022 | ||
1da177e4 | 1023 | /* Set up the argument struct */ |
6e4efd56 | 1024 | nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags); |
275acaaf | 1025 | list_add(&data->list, res); |
50828d7e | 1026 | desc->pg_rpc_callops = &nfs_write_full_ops; |
44b83799 | 1027 | out: |
44b83799 | 1028 | return ret; |
1da177e4 LT |
1029 | } |
1030 | ||
dce81290 TM |
1031 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head) |
1032 | { | |
1033 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
1034 | return nfs_flush_multi(desc, head); | |
1035 | return nfs_flush_one(desc, head); | |
1036 | } | |
1037 | ||
1038 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1039 | { |
275acaaf TM |
1040 | LIST_HEAD(head); |
1041 | int ret; | |
1042 | ||
dce81290 | 1043 | ret = nfs_generic_flush(desc, &head); |
50828d7e TM |
1044 | if (ret == 0) |
1045 | ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops, | |
dce81290 | 1046 | desc->pg_ioflags); |
275acaaf | 1047 | return ret; |
1751c363 | 1048 | } |
1751c363 TM |
1049 | |
1050 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1051 | .pg_test = nfs_generic_pg_test, | |
1052 | .pg_doio = nfs_generic_pg_writepages, | |
1053 | }; | |
1054 | ||
1f945357 | 1055 | static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
c63c7b05 | 1056 | struct inode *inode, int ioflags) |
1da177e4 | 1057 | { |
1751c363 TM |
1058 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, |
1059 | NFS_SERVER(inode)->wsize, ioflags); | |
1060 | } | |
1da177e4 | 1061 | |
dce81290 TM |
1062 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1063 | { | |
1064 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1065 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1066 | } | |
1f945357 | 1067 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1068 | |
1751c363 TM |
1069 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1070 | struct inode *inode, int ioflags) | |
1071 | { | |
1072 | if (!pnfs_pageio_init_write(pgio, inode, ioflags)) | |
1073 | nfs_pageio_init_write_mds(pgio, inode, ioflags); | |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | /* | |
1077 | * Handle a write reply that flushed part of a page. | |
1078 | */ | |
788e7a89 | 1079 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 1080 | { |
788e7a89 | 1081 | struct nfs_write_data *data = calldata; |
1da177e4 | 1082 | |
48186c7d CL |
1083 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1084 | task->tk_pid, | |
3d4ff43d | 1085 | data->req->wb_context->dentry->d_inode->i_sb->s_id, |
48186c7d | 1086 | (long long) |
3d4ff43d | 1087 | NFS_FILEID(data->req->wb_context->dentry->d_inode), |
48186c7d | 1088 | data->req->wb_bytes, (long long)req_offset(data->req)); |
1da177e4 | 1089 | |
c9d8f89d TM |
1090 | nfs_writeback_done(task, data); |
1091 | } | |
788e7a89 | 1092 | |
c9d8f89d TM |
1093 | static void nfs_writeback_release_partial(void *calldata) |
1094 | { | |
1095 | struct nfs_write_data *data = calldata; | |
1096 | struct nfs_page *req = data->req; | |
1097 | struct page *page = req->wb_page; | |
1098 | int status = data->task.tk_status; | |
1099 | ||
1100 | if (status < 0) { | |
a301b777 | 1101 | nfs_set_pageerror(page); |
c9d8f89d TM |
1102 | nfs_context_set_write_error(req->wb_context, status); |
1103 | dprintk(", error = %d\n", status); | |
8e821cad | 1104 | goto out; |
1da177e4 LT |
1105 | } |
1106 | ||
8e821cad | 1107 | if (nfs_write_need_commit(data)) { |
587142f8 | 1108 | struct inode *inode = page->mapping->host; |
8e821cad | 1109 | |
587142f8 | 1110 | spin_lock(&inode->i_lock); |
8e821cad TM |
1111 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1112 | /* Do nothing we need to resend the writes */ | |
1113 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1114 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1115 | dprintk(" defer commit\n"); | |
1116 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1117 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1118 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1119 | dprintk(" server reboot detected\n"); | |
1120 | } | |
587142f8 | 1121 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1122 | } else |
1123 | dprintk(" OK\n"); | |
1124 | ||
1125 | out: | |
1da177e4 | 1126 | if (atomic_dec_and_test(&req->wb_complete)) |
a861a1e1 | 1127 | nfs_writepage_release(req, data); |
c9d8f89d | 1128 | nfs_writedata_release(calldata); |
1da177e4 LT |
1129 | } |
1130 | ||
def6ed7e AA |
1131 | #if defined(CONFIG_NFS_V4_1) |
1132 | void nfs_write_prepare(struct rpc_task *task, void *calldata) | |
1133 | { | |
1134 | struct nfs_write_data *data = calldata; | |
def6ed7e | 1135 | |
035168ab TM |
1136 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
1137 | &data->args.seq_args, | |
def6ed7e AA |
1138 | &data->res.seq_res, 1, task)) |
1139 | return; | |
1140 | rpc_call_start(task); | |
1141 | } | |
1142 | #endif /* CONFIG_NFS_V4_1 */ | |
1143 | ||
788e7a89 | 1144 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e AA |
1145 | #if defined(CONFIG_NFS_V4_1) |
1146 | .rpc_call_prepare = nfs_write_prepare, | |
1147 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1148 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1149 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1150 | }; |
1151 | ||
1da177e4 LT |
1152 | /* |
1153 | * Handle a write reply that flushes a whole page. | |
1154 | * | |
1155 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1156 | * writebacks since the page->count is kept > 1 for as long | |
1157 | * as the page has a write request pending. | |
1158 | */ | |
788e7a89 | 1159 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1160 | { |
788e7a89 | 1161 | struct nfs_write_data *data = calldata; |
1da177e4 | 1162 | |
c9d8f89d TM |
1163 | nfs_writeback_done(task, data); |
1164 | } | |
1165 | ||
1166 | static void nfs_writeback_release_full(void *calldata) | |
1167 | { | |
1168 | struct nfs_write_data *data = calldata; | |
1169 | int status = data->task.tk_status; | |
788e7a89 | 1170 | |
1da177e4 LT |
1171 | /* Update attributes as result of writeback. */ |
1172 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1173 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1174 | struct page *page = req->wb_page; | |
1175 | ||
1da177e4 | 1176 | nfs_list_remove_request(req); |
1da177e4 | 1177 | |
48186c7d CL |
1178 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1179 | data->task.tk_pid, | |
3d4ff43d AV |
1180 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1181 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1182 | req->wb_bytes, |
1183 | (long long)req_offset(req)); | |
1184 | ||
c9d8f89d | 1185 | if (status < 0) { |
a301b777 | 1186 | nfs_set_pageerror(page); |
c9d8f89d TM |
1187 | nfs_context_set_write_error(req->wb_context, status); |
1188 | dprintk(", error = %d\n", status); | |
8e821cad | 1189 | goto remove_request; |
1da177e4 | 1190 | } |
1da177e4 | 1191 | |
8e821cad TM |
1192 | if (nfs_write_need_commit(data)) { |
1193 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
a861a1e1 | 1194 | nfs_mark_request_commit(req, data->lseg); |
8e821cad | 1195 | dprintk(" marked for commit\n"); |
1da177e4 LT |
1196 | goto next; |
1197 | } | |
8e821cad TM |
1198 | dprintk(" OK\n"); |
1199 | remove_request: | |
1da177e4 | 1200 | nfs_inode_remove_request(req); |
1da177e4 | 1201 | next: |
9fd367f0 | 1202 | nfs_clear_page_tag_locked(req); |
a6305ddb | 1203 | nfs_end_page_writeback(page); |
1da177e4 | 1204 | } |
c9d8f89d | 1205 | nfs_writedata_release(calldata); |
1da177e4 LT |
1206 | } |
1207 | ||
788e7a89 | 1208 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e AA |
1209 | #if defined(CONFIG_NFS_V4_1) |
1210 | .rpc_call_prepare = nfs_write_prepare, | |
1211 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1212 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1213 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1214 | }; |
1215 | ||
1216 | ||
1da177e4 LT |
1217 | /* |
1218 | * This function is called when the WRITE call is complete. | |
1219 | */ | |
13602896 | 1220 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1221 | { |
1da177e4 LT |
1222 | struct nfs_writeargs *argp = &data->args; |
1223 | struct nfs_writeres *resp = &data->res; | |
eedc020e | 1224 | struct nfs_server *server = NFS_SERVER(data->inode); |
788e7a89 | 1225 | int status; |
1da177e4 | 1226 | |
a3f565b1 | 1227 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1228 | task->tk_pid, task->tk_status); |
1229 | ||
f551e44f CL |
1230 | /* |
1231 | * ->write_done will attempt to use post-op attributes to detect | |
1232 | * conflicting writes by other clients. A strict interpretation | |
1233 | * of close-to-open would allow us to continue caching even if | |
1234 | * another writer had changed the file, but some applications | |
1235 | * depend on tighter cache coherency when writing. | |
1236 | */ | |
788e7a89 TM |
1237 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1238 | if (status != 0) | |
13602896 | 1239 | return; |
91d5b470 CL |
1240 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1241 | ||
1da177e4 LT |
1242 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1243 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1244 | /* We tried a write call, but the server did not | |
1245 | * commit data to stable storage even though we | |
1246 | * requested it. | |
1247 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1248 | * the server reports NFS_DATA_SYNC, but performs | |
1249 | * NFS_FILE_SYNC. We therefore implement this checking | |
1250 | * as a dprintk() in order to avoid filling syslog. | |
1251 | */ | |
1252 | static unsigned long complain; | |
1253 | ||
a69aef14 | 1254 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1255 | if (time_before(complain, jiffies)) { |
48186c7d | 1256 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1257 | " (committed = %d) != (stable = %d)\n", |
eedc020e | 1258 | server->nfs_client->cl_hostname, |
1da177e4 LT |
1259 | resp->verf->committed, argp->stable); |
1260 | complain = jiffies + 300 * HZ; | |
1261 | } | |
1262 | } | |
1263 | #endif | |
1264 | /* Is this a short write? */ | |
1265 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1266 | static unsigned long complain; | |
1267 | ||
91d5b470 CL |
1268 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1269 | ||
1da177e4 LT |
1270 | /* Has the server at least made some progress? */ |
1271 | if (resp->count != 0) { | |
1272 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1273 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1274 | /* Resend from where the server left off */ | |
a69aef14 | 1275 | data->mds_offset += resp->count; |
1da177e4 LT |
1276 | argp->offset += resp->count; |
1277 | argp->pgbase += resp->count; | |
1278 | argp->count -= resp->count; | |
1279 | } else { | |
1280 | /* Resend as a stable write in order to avoid | |
1281 | * headaches in the case of a server crash. | |
1282 | */ | |
1283 | argp->stable = NFS_FILE_SYNC; | |
1284 | } | |
0110ee15 | 1285 | nfs_restart_rpc(task, server->nfs_client); |
13602896 | 1286 | return; |
1da177e4 LT |
1287 | } |
1288 | if (time_before(complain, jiffies)) { | |
1289 | printk(KERN_WARNING | |
1290 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1291 | argp->count); | |
1292 | complain = jiffies + 300 * HZ; | |
1293 | } | |
1294 | /* Can't do anything about it except throw an error. */ | |
1295 | task->tk_status = -EIO; | |
1296 | } | |
13602896 | 1297 | return; |
1da177e4 LT |
1298 | } |
1299 | ||
1300 | ||
1301 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1302 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1303 | { | |
b8413f98 TM |
1304 | int ret; |
1305 | ||
71d0a611 TM |
1306 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1307 | return 1; | |
b8413f98 TM |
1308 | if (!may_wait) |
1309 | return 0; | |
1310 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1311 | NFS_INO_COMMIT, | |
1312 | nfs_wait_bit_killable, | |
1313 | TASK_KILLABLE); | |
1314 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1315 | } |
1316 | ||
e0c2b380 | 1317 | void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1318 | { |
1319 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1320 | smp_mb__after_clear_bit(); | |
1321 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1322 | } | |
e0c2b380 | 1323 | EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); |
71d0a611 | 1324 | |
e0c2b380 | 1325 | void nfs_commitdata_release(void *data) |
1da177e4 | 1326 | { |
383ba719 TM |
1327 | struct nfs_write_data *wdata = data; |
1328 | ||
988b6dce | 1329 | put_lseg(wdata->lseg); |
383ba719 | 1330 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
1331 | nfs_commit_free(wdata); |
1332 | } | |
e0c2b380 | 1333 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1334 | |
e0c2b380 | 1335 | int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, |
9ace33cd FI |
1336 | const struct rpc_call_ops *call_ops, |
1337 | int how) | |
1da177e4 | 1338 | { |
07737691 | 1339 | struct rpc_task *task; |
9ace33cd | 1340 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1341 | struct rpc_message msg = { |
1342 | .rpc_argp = &data->args, | |
1343 | .rpc_resp = &data->res, | |
9ace33cd | 1344 | .rpc_cred = data->cred, |
bdc7f021 | 1345 | }; |
84115e1c | 1346 | struct rpc_task_setup task_setup_data = { |
07737691 | 1347 | .task = &data->task, |
9ace33cd | 1348 | .rpc_client = clnt, |
bdc7f021 | 1349 | .rpc_message = &msg, |
9ace33cd | 1350 | .callback_ops = call_ops, |
84115e1c | 1351 | .callback_data = data, |
101070ca | 1352 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1353 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1354 | .priority = priority, |
84115e1c | 1355 | }; |
9ace33cd FI |
1356 | /* Set up the initial task struct. */ |
1357 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1358 | ||
1359 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1360 | ||
1361 | task = rpc_run_task(&task_setup_data); | |
1362 | if (IS_ERR(task)) | |
1363 | return PTR_ERR(task); | |
1364 | if (how & FLUSH_SYNC) | |
1365 | rpc_wait_for_completion_task(task); | |
1366 | rpc_put_task(task); | |
1367 | return 0; | |
1368 | } | |
e0c2b380 | 1369 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1370 | |
1371 | /* | |
1372 | * Set up the argument/result storage required for the RPC call. | |
1373 | */ | |
e0c2b380 | 1374 | void nfs_init_commit(struct nfs_write_data *data, |
988b6dce FI |
1375 | struct list_head *head, |
1376 | struct pnfs_layout_segment *lseg) | |
9ace33cd FI |
1377 | { |
1378 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1379 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1380 | |
1381 | /* Set up the RPC argument and reply structs | |
1382 | * NB: take care not to mess about with data->commit et al. */ | |
1383 | ||
1384 | list_splice_init(head, &data->pages); | |
1da177e4 | 1385 | |
1da177e4 | 1386 | data->inode = inode; |
9ace33cd | 1387 | data->cred = first->wb_context->cred; |
988b6dce | 1388 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1389 | data->mds_ops = &nfs_commit_ops; |
1da177e4 LT |
1390 | |
1391 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1392 | /* Note: we always request a commit of the entire inode */ |
1393 | data->args.offset = 0; | |
1394 | data->args.count = 0; | |
383ba719 | 1395 | data->args.context = get_nfs_open_context(first->wb_context); |
3da28eb1 | 1396 | data->res.count = 0; |
1da177e4 LT |
1397 | data->res.fattr = &data->fattr; |
1398 | data->res.verf = &data->verf; | |
0e574af1 | 1399 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1400 | } |
e0c2b380 | 1401 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1402 | |
e0c2b380 | 1403 | void nfs_retry_commit(struct list_head *page_list, |
a861a1e1 | 1404 | struct pnfs_layout_segment *lseg) |
64bfeb49 FI |
1405 | { |
1406 | struct nfs_page *req; | |
1407 | ||
1408 | while (!list_empty(page_list)) { | |
1409 | req = nfs_list_entry(page_list->next); | |
1410 | nfs_list_remove_request(req); | |
a861a1e1 | 1411 | nfs_mark_request_commit(req, lseg); |
64bfeb49 FI |
1412 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1413 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1414 | BDI_RECLAIMABLE); | |
1415 | nfs_clear_page_tag_locked(req); | |
1416 | } | |
1417 | } | |
e0c2b380 | 1418 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1419 | |
1da177e4 LT |
1420 | /* |
1421 | * Commit dirty pages | |
1422 | */ | |
1423 | static int | |
40859d7e | 1424 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1425 | { |
1426 | struct nfs_write_data *data; | |
1da177e4 | 1427 | |
c9d8f89d | 1428 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1429 | |
1430 | if (!data) | |
1431 | goto out_bad; | |
1432 | ||
1433 | /* Set up the argument struct */ | |
988b6dce | 1434 | nfs_init_commit(data, head, NULL); |
9ace33cd | 1435 | return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); |
1da177e4 | 1436 | out_bad: |
a861a1e1 | 1437 | nfs_retry_commit(head, NULL); |
71d0a611 | 1438 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1439 | return -ENOMEM; |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * COMMIT call returned | |
1444 | */ | |
788e7a89 | 1445 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1446 | { |
963d8fe5 | 1447 | struct nfs_write_data *data = calldata; |
1da177e4 | 1448 | |
a3f565b1 | 1449 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1450 | task->tk_pid, task->tk_status); |
1451 | ||
788e7a89 | 1452 | /* Call the NFS version-specific code */ |
c0d0e96b | 1453 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1454 | } |
1455 | ||
e0c2b380 | 1456 | void nfs_commit_release_pages(struct nfs_write_data *data) |
c9d8f89d | 1457 | { |
5917ce84 | 1458 | struct nfs_page *req; |
c9d8f89d | 1459 | int status = data->task.tk_status; |
788e7a89 | 1460 | |
1da177e4 LT |
1461 | while (!list_empty(&data->pages)) { |
1462 | req = nfs_list_entry(data->pages.next); | |
1463 | nfs_list_remove_request(req); | |
e468bae9 | 1464 | nfs_clear_request_commit(req); |
1da177e4 | 1465 | |
48186c7d | 1466 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1467 | req->wb_context->dentry->d_sb->s_id, |
1468 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1469 | req->wb_bytes, |
1470 | (long long)req_offset(req)); | |
c9d8f89d TM |
1471 | if (status < 0) { |
1472 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1473 | nfs_inode_remove_request(req); |
c9d8f89d | 1474 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1475 | goto next; |
1476 | } | |
1477 | ||
1478 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1479 | * returned by the server against all stored verfs. */ | |
1480 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1481 | /* We have a match */ | |
1482 | nfs_inode_remove_request(req); | |
1483 | dprintk(" OK\n"); | |
1484 | goto next; | |
1485 | } | |
1486 | /* We have a mismatch. Write the page again */ | |
1487 | dprintk(" mismatch\n"); | |
6d884e8f | 1488 | nfs_mark_request_dirty(req); |
1da177e4 | 1489 | next: |
9fd367f0 | 1490 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1491 | } |
5917ce84 | 1492 | } |
e0c2b380 | 1493 | EXPORT_SYMBOL_GPL(nfs_commit_release_pages); |
5917ce84 FI |
1494 | |
1495 | static void nfs_commit_release(void *calldata) | |
1496 | { | |
1497 | struct nfs_write_data *data = calldata; | |
1498 | ||
1499 | nfs_commit_release_pages(data); | |
71d0a611 | 1500 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1501 | nfs_commitdata_release(calldata); |
1da177e4 | 1502 | } |
788e7a89 TM |
1503 | |
1504 | static const struct rpc_call_ops nfs_commit_ops = { | |
21d9a851 AA |
1505 | #if defined(CONFIG_NFS_V4_1) |
1506 | .rpc_call_prepare = nfs_write_prepare, | |
1507 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 TM |
1508 | .rpc_call_done = nfs_commit_done, |
1509 | .rpc_release = nfs_commit_release, | |
1510 | }; | |
1da177e4 | 1511 | |
b608b283 | 1512 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1513 | { |
1da177e4 | 1514 | LIST_HEAD(head); |
71d0a611 | 1515 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1516 | int res; |
1da177e4 | 1517 | |
b8413f98 TM |
1518 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1519 | if (res <= 0) | |
c5efa5fc | 1520 | goto out_mark_dirty; |
3da28eb1 | 1521 | res = nfs_scan_commit(inode, &head, 0, 0); |
1da177e4 | 1522 | if (res) { |
a861a1e1 FI |
1523 | int error; |
1524 | ||
1525 | error = pnfs_commit_list(inode, &head, how); | |
1526 | if (error == PNFS_NOT_ATTEMPTED) | |
1527 | error = nfs_commit_list(inode, &head, how); | |
3da28eb1 TM |
1528 | if (error < 0) |
1529 | return error; | |
b8413f98 | 1530 | if (!may_wait) |
c5efa5fc | 1531 | goto out_mark_dirty; |
b8413f98 TM |
1532 | error = wait_on_bit(&NFS_I(inode)->flags, |
1533 | NFS_INO_COMMIT, | |
1534 | nfs_wait_bit_killable, | |
1535 | TASK_KILLABLE); | |
1536 | if (error < 0) | |
1537 | return error; | |
71d0a611 TM |
1538 | } else |
1539 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1540 | return res; |
1541 | /* Note: If we exit without ensuring that the commit is complete, | |
1542 | * we must mark the inode as dirty. Otherwise, future calls to | |
1543 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1544 | * that the data is on the disk. | |
1545 | */ | |
1546 | out_mark_dirty: | |
1547 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1548 | return res; |
1549 | } | |
8fc795f7 TM |
1550 | |
1551 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1552 | { | |
420e3646 TM |
1553 | struct nfs_inode *nfsi = NFS_I(inode); |
1554 | int flags = FLUSH_SYNC; | |
1555 | int ret = 0; | |
8fc795f7 | 1556 | |
3236c3e1 JL |
1557 | /* no commits means nothing needs to be done */ |
1558 | if (!nfsi->ncommit) | |
1559 | return ret; | |
1560 | ||
a00dd6c0 JL |
1561 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1562 | /* Don't commit yet if this is a non-blocking flush and there | |
1563 | * are a lot of outstanding writes for this mapping. | |
1564 | */ | |
1565 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1566 | goto out_mark_dirty; | |
420e3646 | 1567 | |
a00dd6c0 | 1568 | /* don't wait for the COMMIT response */ |
420e3646 | 1569 | flags = 0; |
a00dd6c0 JL |
1570 | } |
1571 | ||
420e3646 TM |
1572 | ret = nfs_commit_inode(inode, flags); |
1573 | if (ret >= 0) { | |
1574 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1575 | if (ret < wbc->nr_to_write) | |
1576 | wbc->nr_to_write -= ret; | |
1577 | else | |
1578 | wbc->nr_to_write = 0; | |
1579 | } | |
8fc795f7 | 1580 | return 0; |
420e3646 TM |
1581 | } |
1582 | out_mark_dirty: | |
8fc795f7 TM |
1583 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1584 | return ret; | |
1585 | } | |
c63c7b05 | 1586 | #else |
8fc795f7 TM |
1587 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1588 | { | |
1589 | return 0; | |
1590 | } | |
1da177e4 LT |
1591 | #endif |
1592 | ||
8fc795f7 TM |
1593 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1594 | { | |
863a3c6c AA |
1595 | int ret; |
1596 | ||
1597 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1598 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1599 | int status; |
1600 | bool sync = true; | |
863a3c6c | 1601 | |
846d5a09 | 1602 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1603 | sync = false; |
863a3c6c AA |
1604 | |
1605 | status = pnfs_layoutcommit_inode(inode, sync); | |
1606 | if (status < 0) | |
1607 | return status; | |
1608 | } | |
1609 | return ret; | |
8fc795f7 TM |
1610 | } |
1611 | ||
acdc53b2 TM |
1612 | /* |
1613 | * flush the inode to disk. | |
1614 | */ | |
1615 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1616 | { |
1617 | struct writeback_control wbc = { | |
72cb77f4 | 1618 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1619 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1620 | .range_start = 0, |
1621 | .range_end = LLONG_MAX, | |
34901f70 | 1622 | }; |
34901f70 | 1623 | |
acdc53b2 | 1624 | return sync_inode(inode, &wbc); |
1c75950b TM |
1625 | } |
1626 | ||
1b3b4a1a TM |
1627 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1628 | { | |
1629 | struct nfs_page *req; | |
1b3b4a1a TM |
1630 | int ret = 0; |
1631 | ||
1632 | BUG_ON(!PageLocked(page)); | |
1633 | for (;;) { | |
ba8b06e6 | 1634 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1635 | req = nfs_page_find_request(page); |
1636 | if (req == NULL) | |
1b3b4a1a | 1637 | break; |
1b3b4a1a TM |
1638 | if (nfs_lock_request_dontget(req)) { |
1639 | nfs_inode_remove_request(req); | |
1640 | /* | |
1641 | * In case nfs_inode_remove_request has marked the | |
1642 | * page as being dirty | |
1643 | */ | |
1644 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1645 | nfs_unlock_request(req); | |
1646 | break; | |
1647 | } | |
1648 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1649 | nfs_release_request(req); |
1b3b4a1a | 1650 | if (ret < 0) |
c988950e | 1651 | break; |
1b3b4a1a | 1652 | } |
1b3b4a1a TM |
1653 | return ret; |
1654 | } | |
1655 | ||
7f2f12d9 TM |
1656 | /* |
1657 | * Write back all requests on one page - we do this before reading it. | |
1658 | */ | |
1659 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1660 | { |
1661 | loff_t range_start = page_offset(page); | |
1662 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1663 | struct writeback_control wbc = { |
4d770ccf | 1664 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1665 | .nr_to_write = 0, |
4d770ccf TM |
1666 | .range_start = range_start, |
1667 | .range_end = range_end, | |
1668 | }; | |
1669 | int ret; | |
1c75950b | 1670 | |
0522f6ad | 1671 | for (;;) { |
ba8b06e6 | 1672 | wait_on_page_writeback(page); |
73e3302f TM |
1673 | if (clear_page_dirty_for_io(page)) { |
1674 | ret = nfs_writepage_locked(page, &wbc); | |
1675 | if (ret < 0) | |
1676 | goto out_error; | |
0522f6ad | 1677 | continue; |
7f2f12d9 | 1678 | } |
0522f6ad TM |
1679 | if (!PagePrivate(page)) |
1680 | break; | |
1681 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1682 | if (ret < 0) |
73e3302f | 1683 | goto out_error; |
7f2f12d9 | 1684 | } |
73e3302f TM |
1685 | return 0; |
1686 | out_error: | |
4d770ccf | 1687 | return ret; |
1c75950b TM |
1688 | } |
1689 | ||
074cc1de TM |
1690 | #ifdef CONFIG_MIGRATION |
1691 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
1692 | struct page *page) | |
1693 | { | |
1694 | struct nfs_page *req; | |
1695 | int ret; | |
1696 | ||
7549ad5f | 1697 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1698 | |
cfb506e1 | 1699 | req = nfs_find_and_lock_request(page, false); |
074cc1de TM |
1700 | ret = PTR_ERR(req); |
1701 | if (IS_ERR(req)) | |
1702 | goto out; | |
1703 | ||
1704 | ret = migrate_page(mapping, newpage, page); | |
1705 | if (!req) | |
1706 | goto out; | |
1707 | if (ret) | |
1708 | goto out_unlock; | |
1709 | page_cache_get(newpage); | |
190f38e5 | 1710 | spin_lock(&mapping->host->i_lock); |
074cc1de TM |
1711 | req->wb_page = newpage; |
1712 | SetPagePrivate(newpage); | |
190f38e5 | 1713 | set_page_private(newpage, (unsigned long)req); |
074cc1de TM |
1714 | ClearPagePrivate(page); |
1715 | set_page_private(page, 0); | |
190f38e5 | 1716 | spin_unlock(&mapping->host->i_lock); |
074cc1de TM |
1717 | page_cache_release(page); |
1718 | out_unlock: | |
1719 | nfs_clear_page_tag_locked(req); | |
074cc1de TM |
1720 | out: |
1721 | return ret; | |
1722 | } | |
1723 | #endif | |
1724 | ||
f7b422b1 | 1725 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1726 | { |
1727 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1728 | sizeof(struct nfs_write_data), | |
1729 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1730 | NULL); |
1da177e4 LT |
1731 | if (nfs_wdata_cachep == NULL) |
1732 | return -ENOMEM; | |
1733 | ||
93d2341c MD |
1734 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1735 | nfs_wdata_cachep); | |
1da177e4 LT |
1736 | if (nfs_wdata_mempool == NULL) |
1737 | return -ENOMEM; | |
1738 | ||
93d2341c MD |
1739 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1740 | nfs_wdata_cachep); | |
1da177e4 LT |
1741 | if (nfs_commit_mempool == NULL) |
1742 | return -ENOMEM; | |
1743 | ||
89a09141 PZ |
1744 | /* |
1745 | * NFS congestion size, scale with available memory. | |
1746 | * | |
1747 | * 64MB: 8192k | |
1748 | * 128MB: 11585k | |
1749 | * 256MB: 16384k | |
1750 | * 512MB: 23170k | |
1751 | * 1GB: 32768k | |
1752 | * 2GB: 46340k | |
1753 | * 4GB: 65536k | |
1754 | * 8GB: 92681k | |
1755 | * 16GB: 131072k | |
1756 | * | |
1757 | * This allows larger machines to have larger/more transfers. | |
1758 | * Limit the default to 256M | |
1759 | */ | |
1760 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1761 | if (nfs_congestion_kb > 256*1024) | |
1762 | nfs_congestion_kb = 256*1024; | |
1763 | ||
1da177e4 LT |
1764 | return 0; |
1765 | } | |
1766 | ||
266bee88 | 1767 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1768 | { |
1769 | mempool_destroy(nfs_commit_mempool); | |
1770 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1771 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1772 | } |
1773 |