]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
3fcfab16 | 24 | |
1da177e4 | 25 | #include <asm/uaccess.h> |
1da177e4 LT |
26 | |
27 | #include "delegation.h" | |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
def6ed7e | 30 | #include "nfs4_fs.h" |
074cc1de | 31 | #include "fscache.h" |
94ad1c80 | 32 | #include "pnfs.h" |
1da177e4 LT |
33 | |
34 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
35 | ||
36 | #define MIN_POOL_WRITE (32) | |
37 | #define MIN_POOL_COMMIT (4) | |
38 | ||
39 | /* | |
40 | * Local function declarations | |
41 | */ | |
c63c7b05 TM |
42 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
43 | struct inode *inode, int ioflags); | |
f8512ad0 | 44 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 TM |
45 | static const struct rpc_call_ops nfs_write_partial_ops; |
46 | static const struct rpc_call_ops nfs_write_full_ops; | |
47 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 | 48 | |
e18b890b | 49 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 50 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 51 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
52 | static mempool_t *nfs_commit_mempool; |
53 | ||
0b7c0153 | 54 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 55 | { |
0b7c0153 | 56 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 57 | |
1da177e4 LT |
58 | if (p) { |
59 | memset(p, 0, sizeof(*p)); | |
60 | INIT_LIST_HEAD(&p->pages); | |
61 | } | |
62 | return p; | |
63 | } | |
e0c2b380 | 64 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 65 | |
0b7c0153 | 66 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
67 | { |
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 | { |
383ba719 | 102 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
103 | nfs_writedata_free(wdata); |
104 | } | |
105 | ||
7b159fc1 TM |
106 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
107 | { | |
108 | ctx->error = error; | |
109 | smp_wmb(); | |
110 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
111 | } | |
112 | ||
277459d2 TM |
113 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
114 | { | |
115 | struct nfs_page *req = NULL; | |
116 | ||
117 | if (PagePrivate(page)) { | |
118 | req = (struct nfs_page *)page_private(page); | |
119 | if (req != NULL) | |
c03b4024 | 120 | kref_get(&req->wb_kref); |
277459d2 TM |
121 | } |
122 | return req; | |
123 | } | |
124 | ||
125 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
126 | { | |
587142f8 | 127 | struct inode *inode = page->mapping->host; |
277459d2 | 128 | struct nfs_page *req = NULL; |
277459d2 | 129 | |
587142f8 | 130 | spin_lock(&inode->i_lock); |
277459d2 | 131 | req = nfs_page_find_request_locked(page); |
587142f8 | 132 | spin_unlock(&inode->i_lock); |
277459d2 TM |
133 | return req; |
134 | } | |
135 | ||
1da177e4 LT |
136 | /* Adjust the file length if we're writing beyond the end */ |
137 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
138 | { | |
139 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
140 | loff_t end, i_size; |
141 | pgoff_t end_index; | |
1da177e4 | 142 | |
a3d01454 TM |
143 | spin_lock(&inode->i_lock); |
144 | i_size = i_size_read(inode); | |
145 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 146 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 147 | goto out; |
1da177e4 LT |
148 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
149 | if (i_size >= end) | |
a3d01454 | 150 | goto out; |
1da177e4 | 151 | i_size_write(inode, end); |
a3d01454 TM |
152 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
153 | out: | |
154 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
155 | } |
156 | ||
a301b777 TM |
157 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
158 | static void nfs_set_pageerror(struct page *page) | |
159 | { | |
160 | SetPageError(page); | |
161 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
162 | } | |
163 | ||
1da177e4 LT |
164 | /* We can set the PG_uptodate flag if we see that a write request |
165 | * covers the full page. | |
166 | */ | |
167 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
168 | { | |
1da177e4 LT |
169 | if (PageUptodate(page)) |
170 | return; | |
171 | if (base != 0) | |
172 | return; | |
49a70f27 | 173 | if (count != nfs_page_length(page)) |
1da177e4 | 174 | return; |
49a70f27 | 175 | SetPageUptodate(page); |
1da177e4 LT |
176 | } |
177 | ||
1da177e4 LT |
178 | static int wb_priority(struct writeback_control *wbc) |
179 | { | |
180 | if (wbc->for_reclaim) | |
c63c7b05 | 181 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 182 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
183 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
184 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
185 | } |
186 | ||
89a09141 PZ |
187 | /* |
188 | * NFS congestion control | |
189 | */ | |
190 | ||
191 | int nfs_congestion_kb; | |
192 | ||
193 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
194 | #define NFS_CONGESTION_OFF_THRESH \ | |
195 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
196 | ||
5a6d41b3 | 197 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 198 | { |
5a6d41b3 TM |
199 | int ret = test_set_page_writeback(page); |
200 | ||
201 | if (!ret) { | |
89a09141 PZ |
202 | struct inode *inode = page->mapping->host; |
203 | struct nfs_server *nfss = NFS_SERVER(inode); | |
204 | ||
a6305ddb | 205 | page_cache_get(page); |
277866a0 | 206 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
207 | NFS_CONGESTION_ON_THRESH) { |
208 | set_bdi_congested(&nfss->backing_dev_info, | |
209 | BLK_RW_ASYNC); | |
210 | } | |
89a09141 | 211 | } |
5a6d41b3 | 212 | return ret; |
89a09141 PZ |
213 | } |
214 | ||
215 | static void nfs_end_page_writeback(struct page *page) | |
216 | { | |
217 | struct inode *inode = page->mapping->host; | |
218 | struct nfs_server *nfss = NFS_SERVER(inode); | |
219 | ||
220 | end_page_writeback(page); | |
a6305ddb | 221 | page_cache_release(page); |
c4dc4bee | 222 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 223 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
224 | } |
225 | ||
cfb506e1 | 226 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 227 | { |
587142f8 | 228 | struct inode *inode = page->mapping->host; |
e261f51f | 229 | struct nfs_page *req; |
e261f51f TM |
230 | int ret; |
231 | ||
587142f8 | 232 | spin_lock(&inode->i_lock); |
074cc1de | 233 | for (;;) { |
e261f51f | 234 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
235 | if (req == NULL) |
236 | break; | |
9994b62b | 237 | if (nfs_lock_request_dontget(req)) |
e261f51f TM |
238 | break; |
239 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
9994b62b | 240 | * then the call to nfs_lock_request_dontget() will always |
e261f51f TM |
241 | * succeed provided that someone hasn't already marked the |
242 | * request as dirty (in which case we don't care). | |
243 | */ | |
587142f8 | 244 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
245 | if (!nonblock) |
246 | ret = nfs_wait_on_request(req); | |
247 | else | |
248 | ret = -EAGAIN; | |
e261f51f TM |
249 | nfs_release_request(req); |
250 | if (ret != 0) | |
074cc1de | 251 | return ERR_PTR(ret); |
587142f8 | 252 | spin_lock(&inode->i_lock); |
e261f51f | 253 | } |
587142f8 | 254 | spin_unlock(&inode->i_lock); |
074cc1de TM |
255 | return req; |
256 | } | |
257 | ||
258 | /* | |
259 | * Find an associated nfs write request, and prepare to flush it out | |
260 | * May return an error if the user signalled nfs_wait_on_request(). | |
261 | */ | |
262 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 263 | struct page *page, bool nonblock) |
074cc1de TM |
264 | { |
265 | struct nfs_page *req; | |
266 | int ret = 0; | |
267 | ||
cfb506e1 | 268 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
269 | if (!req) |
270 | goto out; | |
271 | ret = PTR_ERR(req); | |
272 | if (IS_ERR(req)) | |
273 | goto out; | |
274 | ||
275 | ret = nfs_set_page_writeback(page); | |
276 | BUG_ON(ret != 0); | |
277 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
278 | ||
f8512ad0 FI |
279 | if (!nfs_pageio_add_request(pgio, req)) { |
280 | nfs_redirty_request(req); | |
074cc1de | 281 | ret = pgio->pg_error; |
f8512ad0 | 282 | } |
074cc1de TM |
283 | out: |
284 | return ret; | |
e261f51f TM |
285 | } |
286 | ||
f758c885 | 287 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 288 | { |
1da177e4 | 289 | struct inode *inode = page->mapping->host; |
cfb506e1 | 290 | int ret; |
1da177e4 | 291 | |
91d5b470 CL |
292 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
293 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
294 | ||
7fe7f848 | 295 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 296 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
297 | if (ret == -EAGAIN) { |
298 | redirty_page_for_writepage(wbc, page); | |
299 | ret = 0; | |
300 | } | |
301 | return ret; | |
f758c885 | 302 | } |
7fe7f848 | 303 | |
f758c885 TM |
304 | /* |
305 | * Write an mmapped page to the server. | |
306 | */ | |
307 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
308 | { | |
309 | struct nfs_pageio_descriptor pgio; | |
310 | int err; | |
49a70f27 | 311 | |
f758c885 TM |
312 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
313 | err = nfs_do_writepage(page, wbc, &pgio); | |
314 | nfs_pageio_complete(&pgio); | |
315 | if (err < 0) | |
316 | return err; | |
317 | if (pgio.pg_error < 0) | |
318 | return pgio.pg_error; | |
319 | return 0; | |
4d770ccf TM |
320 | } |
321 | ||
322 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
323 | { | |
f758c885 | 324 | int ret; |
4d770ccf | 325 | |
f758c885 | 326 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 327 | unlock_page(page); |
f758c885 TM |
328 | return ret; |
329 | } | |
330 | ||
331 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
332 | { | |
333 | int ret; | |
334 | ||
335 | ret = nfs_do_writepage(page, wbc, data); | |
336 | unlock_page(page); | |
337 | return ret; | |
1da177e4 LT |
338 | } |
339 | ||
1da177e4 LT |
340 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
341 | { | |
1da177e4 | 342 | struct inode *inode = mapping->host; |
72cb77f4 | 343 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 344 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
345 | int err; |
346 | ||
72cb77f4 TM |
347 | /* Stop dirtying of new pages while we sync */ |
348 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
349 | nfs_wait_bit_killable, TASK_KILLABLE); | |
350 | if (err) | |
351 | goto out_err; | |
352 | ||
91d5b470 CL |
353 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
354 | ||
c63c7b05 | 355 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 356 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 357 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
358 | |
359 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
360 | smp_mb__after_clear_bit(); | |
361 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
362 | ||
f758c885 | 363 | if (err < 0) |
72cb77f4 TM |
364 | goto out_err; |
365 | err = pgio.pg_error; | |
366 | if (err < 0) | |
367 | goto out_err; | |
c63c7b05 | 368 | return 0; |
72cb77f4 TM |
369 | out_err: |
370 | return err; | |
1da177e4 LT |
371 | } |
372 | ||
373 | /* | |
374 | * Insert a write request into an inode | |
375 | */ | |
d6d6dc7c | 376 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
377 | { |
378 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
379 | |
380 | /* Lock the request! */ | |
381 | nfs_lock_request_dontget(req); | |
382 | ||
383 | spin_lock(&inode->i_lock); | |
4d65c520 | 384 | if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 385 | inode->i_version++; |
2df485a7 | 386 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 387 | SetPagePrivate(req->wb_page); |
277459d2 | 388 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 389 | nfsi->npages++; |
c03b4024 | 390 | kref_get(&req->wb_kref); |
e7d39069 | 391 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
392 | } |
393 | ||
394 | /* | |
89a09141 | 395 | * Remove a write request from an inode |
1da177e4 LT |
396 | */ |
397 | static void nfs_inode_remove_request(struct nfs_page *req) | |
398 | { | |
3d4ff43d | 399 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
400 | struct nfs_inode *nfsi = NFS_I(inode); |
401 | ||
402 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
403 | ||
587142f8 | 404 | spin_lock(&inode->i_lock); |
277459d2 | 405 | set_page_private(req->wb_page, 0); |
deb7d638 | 406 | ClearPagePrivate(req->wb_page); |
2df485a7 | 407 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 | 408 | nfsi->npages--; |
4d65c520 | 409 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
410 | nfs_release_request(req); |
411 | } | |
412 | ||
61822ab5 | 413 | static void |
6d884e8f | 414 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 415 | { |
61822ab5 TM |
416 | __set_page_dirty_nobuffers(req->wb_page); |
417 | } | |
418 | ||
1da177e4 | 419 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
8dd37758 TM |
420 | /** |
421 | * nfs_request_add_commit_list - add request to a commit list | |
422 | * @req: pointer to a struct nfs_page | |
423 | * @head: commit list head | |
424 | * | |
425 | * This sets the PG_CLEAN bit, updates the inode global count of | |
426 | * number of outstanding requests requiring a commit as well as | |
427 | * the MM page stats. | |
428 | * | |
429 | * The caller must _not_ hold the inode->i_lock, but must be | |
430 | * holding the nfs_page lock. | |
1da177e4 | 431 | */ |
8dd37758 TM |
432 | void |
433 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head) | |
1da177e4 | 434 | { |
3d4ff43d | 435 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 | 436 | |
e468bae9 | 437 | set_bit(PG_CLEAN, &(req)->wb_flags); |
8dd37758 TM |
438 | spin_lock(&inode->i_lock); |
439 | nfs_list_add_request(req, head); | |
440 | NFS_I(inode)->ncommit++; | |
587142f8 | 441 | spin_unlock(&inode->i_lock); |
fd39fc85 | 442 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 443 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 444 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 445 | } |
8dd37758 TM |
446 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
447 | ||
448 | /** | |
449 | * nfs_request_remove_commit_list - Remove request from a commit list | |
450 | * @req: pointer to a nfs_page | |
451 | * | |
452 | * This clears the PG_CLEAN bit, and updates the inode global count of | |
453 | * number of outstanding requests requiring a commit | |
454 | * It does not update the MM page stats. | |
455 | * | |
456 | * The caller _must_ hold the inode->i_lock and the nfs_page lock. | |
457 | */ | |
458 | void | |
459 | nfs_request_remove_commit_list(struct nfs_page *req) | |
460 | { | |
461 | struct inode *inode = req->wb_context->dentry->d_inode; | |
462 | ||
463 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) | |
464 | return; | |
465 | nfs_list_remove_request(req); | |
466 | NFS_I(inode)->ncommit--; | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
469 | ||
470 | ||
471 | /* | |
472 | * Add a request to the inode's commit list. | |
473 | */ | |
474 | static void | |
475 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) | |
476 | { | |
477 | struct inode *inode = req->wb_context->dentry->d_inode; | |
478 | ||
479 | if (pnfs_mark_request_commit(req, lseg)) | |
480 | return; | |
481 | nfs_request_add_commit_list(req, &NFS_I(inode)->commit_list); | |
482 | } | |
8e821cad | 483 | |
d6d6dc7c FI |
484 | static void |
485 | nfs_clear_page_commit(struct page *page) | |
486 | { | |
487 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
488 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
489 | } | |
490 | ||
8dd37758 | 491 | static void |
e468bae9 TM |
492 | nfs_clear_request_commit(struct nfs_page *req) |
493 | { | |
8dd37758 TM |
494 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
495 | struct inode *inode = req->wb_context->dentry->d_inode; | |
e468bae9 | 496 | |
8dd37758 TM |
497 | if (!pnfs_clear_request_commit(req)) { |
498 | spin_lock(&inode->i_lock); | |
499 | nfs_request_remove_commit_list(req); | |
500 | spin_unlock(&inode->i_lock); | |
501 | } | |
d6d6dc7c | 502 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 503 | } |
e468bae9 TM |
504 | } |
505 | ||
8e821cad TM |
506 | static inline |
507 | int nfs_write_need_commit(struct nfs_write_data *data) | |
508 | { | |
465d5243 FI |
509 | if (data->verf.committed == NFS_DATA_SYNC) |
510 | return data->lseg == NULL; | |
511 | else | |
512 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
513 | } |
514 | ||
515 | static inline | |
a861a1e1 FI |
516 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
517 | struct nfs_write_data *data) | |
8e821cad | 518 | { |
e468bae9 | 519 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
a861a1e1 | 520 | nfs_mark_request_commit(req, data->lseg); |
8e821cad TM |
521 | return 1; |
522 | } | |
523 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
6d884e8f | 524 | nfs_mark_request_dirty(req); |
8e821cad TM |
525 | return 1; |
526 | } | |
527 | return 0; | |
528 | } | |
529 | #else | |
8dd37758 | 530 | static void |
a861a1e1 | 531 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
8e821cad TM |
532 | { |
533 | } | |
534 | ||
8dd37758 | 535 | static void |
e468bae9 TM |
536 | nfs_clear_request_commit(struct nfs_page *req) |
537 | { | |
e468bae9 TM |
538 | } |
539 | ||
8e821cad TM |
540 | static inline |
541 | int nfs_write_need_commit(struct nfs_write_data *data) | |
542 | { | |
543 | return 0; | |
544 | } | |
545 | ||
546 | static inline | |
a861a1e1 FI |
547 | int nfs_reschedule_unstable_write(struct nfs_page *req, |
548 | struct nfs_write_data *data) | |
8e821cad TM |
549 | { |
550 | return 0; | |
551 | } | |
1da177e4 LT |
552 | #endif |
553 | ||
47c62564 | 554 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
555 | static int |
556 | nfs_need_commit(struct nfs_inode *nfsi) | |
557 | { | |
d6d6dc7c FI |
558 | return nfsi->ncommit > 0; |
559 | } | |
560 | ||
561 | /* i_lock held by caller */ | |
8dd37758 | 562 | static int |
3b3be88d TM |
563 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, int max, |
564 | spinlock_t *lock) | |
d6d6dc7c FI |
565 | { |
566 | struct nfs_page *req, *tmp; | |
567 | int ret = 0; | |
568 | ||
569 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
570 | if (!nfs_lock_request(req)) |
571 | continue; | |
3b3be88d TM |
572 | if (cond_resched_lock(lock)) |
573 | list_safe_reset_next(req, tmp, wb_list); | |
8dd37758 TM |
574 | nfs_request_remove_commit_list(req); |
575 | nfs_list_add_request(req, dst); | |
576 | ret++; | |
577 | if (ret == max) | |
578 | break; | |
d6d6dc7c FI |
579 | } |
580 | return ret; | |
fb8a1f11 TM |
581 | } |
582 | ||
1da177e4 LT |
583 | /* |
584 | * nfs_scan_commit - Scan an inode for commit requests | |
585 | * @inode: NFS inode to scan | |
586 | * @dst: destination list | |
1da177e4 LT |
587 | * |
588 | * Moves requests from the inode's 'commit' request list. | |
589 | * The requests are *not* checked to ensure that they form a contiguous set. | |
590 | */ | |
591 | static int | |
d6d6dc7c | 592 | nfs_scan_commit(struct inode *inode, struct list_head *dst) |
1da177e4 LT |
593 | { |
594 | struct nfs_inode *nfsi = NFS_I(inode); | |
d6d6dc7c | 595 | int ret = 0; |
fb8a1f11 | 596 | |
0d88f6e8 | 597 | spin_lock(&inode->i_lock); |
d6d6dc7c | 598 | if (nfsi->ncommit > 0) { |
8dd37758 | 599 | const int max = INT_MAX; |
d6d6dc7c | 600 | |
3b3be88d TM |
601 | ret = nfs_scan_commit_list(&nfsi->commit_list, dst, max, |
602 | &inode->i_lock); | |
603 | ret += pnfs_scan_commit_lists(inode, max - ret, | |
604 | &inode->i_lock); | |
d6d6dc7c | 605 | } |
0d88f6e8 | 606 | spin_unlock(&inode->i_lock); |
ff778d02 | 607 | return ret; |
1da177e4 | 608 | } |
d6d6dc7c | 609 | |
c42de9dd | 610 | #else |
fb8a1f11 TM |
611 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
612 | { | |
613 | return 0; | |
614 | } | |
615 | ||
d6d6dc7c | 616 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst) |
c42de9dd TM |
617 | { |
618 | return 0; | |
619 | } | |
1da177e4 LT |
620 | #endif |
621 | ||
1da177e4 | 622 | /* |
e7d39069 TM |
623 | * Search for an existing write request, and attempt to update |
624 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 625 | * |
e7d39069 TM |
626 | * If the attempt fails, then the existing request is flushed out |
627 | * to disk. | |
1da177e4 | 628 | */ |
e7d39069 TM |
629 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
630 | struct page *page, | |
631 | unsigned int offset, | |
632 | unsigned int bytes) | |
1da177e4 | 633 | { |
e7d39069 TM |
634 | struct nfs_page *req; |
635 | unsigned int rqend; | |
636 | unsigned int end; | |
637 | int error; | |
638 | ||
639 | if (!PagePrivate(page)) | |
640 | return NULL; | |
1da177e4 LT |
641 | |
642 | end = offset + bytes; | |
e7d39069 | 643 | spin_lock(&inode->i_lock); |
1da177e4 | 644 | |
1da177e4 | 645 | for (;;) { |
277459d2 | 646 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
647 | if (req == NULL) |
648 | goto out_unlock; | |
649 | ||
650 | rqend = req->wb_offset + req->wb_bytes; | |
651 | /* | |
652 | * Tell the caller to flush out the request if | |
653 | * the offsets are non-contiguous. | |
654 | * Note: nfs_flush_incompatible() will already | |
655 | * have flushed out requests having wrong owners. | |
656 | */ | |
e468bae9 | 657 | if (offset > rqend |
e7d39069 TM |
658 | || end < req->wb_offset) |
659 | goto out_flushme; | |
660 | ||
9994b62b | 661 | if (nfs_lock_request_dontget(req)) |
1da177e4 | 662 | break; |
1da177e4 | 663 | |
e7d39069 | 664 | /* The request is locked, so wait and then retry */ |
587142f8 | 665 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
666 | error = nfs_wait_on_request(req); |
667 | nfs_release_request(req); | |
668 | if (error != 0) | |
669 | goto out_err; | |
670 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
671 | } |
672 | ||
673 | /* Okay, the request matches. Update the region */ | |
674 | if (offset < req->wb_offset) { | |
675 | req->wb_offset = offset; | |
676 | req->wb_pgbase = offset; | |
1da177e4 | 677 | } |
1da177e4 LT |
678 | if (end > rqend) |
679 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
680 | else |
681 | req->wb_bytes = rqend - req->wb_offset; | |
682 | out_unlock: | |
683 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
684 | if (req) |
685 | nfs_clear_request_commit(req); | |
e7d39069 TM |
686 | return req; |
687 | out_flushme: | |
688 | spin_unlock(&inode->i_lock); | |
689 | nfs_release_request(req); | |
690 | error = nfs_wb_page(inode, page); | |
691 | out_err: | |
692 | return ERR_PTR(error); | |
693 | } | |
694 | ||
695 | /* | |
696 | * Try to update an existing write request, or create one if there is none. | |
697 | * | |
698 | * Note: Should always be called with the Page Lock held to prevent races | |
699 | * if we have to add a new request. Also assumes that the caller has | |
700 | * already called nfs_flush_incompatible() if necessary. | |
701 | */ | |
702 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
703 | struct page *page, unsigned int offset, unsigned int bytes) | |
704 | { | |
705 | struct inode *inode = page->mapping->host; | |
706 | struct nfs_page *req; | |
1da177e4 | 707 | |
e7d39069 TM |
708 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
709 | if (req != NULL) | |
710 | goto out; | |
711 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
712 | if (IS_ERR(req)) | |
713 | goto out; | |
d6d6dc7c | 714 | nfs_inode_add_request(inode, req); |
efc91ed0 | 715 | out: |
61e930a9 | 716 | return req; |
1da177e4 LT |
717 | } |
718 | ||
e7d39069 TM |
719 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
720 | unsigned int offset, unsigned int count) | |
721 | { | |
722 | struct nfs_page *req; | |
723 | ||
724 | req = nfs_setup_write_request(ctx, page, offset, count); | |
725 | if (IS_ERR(req)) | |
726 | return PTR_ERR(req); | |
727 | /* Update file length */ | |
728 | nfs_grow_file(page, offset, count); | |
729 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 730 | nfs_mark_request_dirty(req); |
9994b62b | 731 | nfs_unlock_request(req); |
e7d39069 TM |
732 | return 0; |
733 | } | |
734 | ||
1da177e4 LT |
735 | int nfs_flush_incompatible(struct file *file, struct page *page) |
736 | { | |
cd3758e3 | 737 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 738 | struct nfs_page *req; |
1a54533e | 739 | int do_flush, status; |
1da177e4 LT |
740 | /* |
741 | * Look for a request corresponding to this page. If there | |
742 | * is one, and it belongs to another file, we flush it out | |
743 | * before we try to copy anything into the page. Do this | |
744 | * due to the lack of an ACCESS-type call in NFSv2. | |
745 | * Also do the same if we find a request from an existing | |
746 | * dropped page. | |
747 | */ | |
1a54533e TM |
748 | do { |
749 | req = nfs_page_find_request(page); | |
750 | if (req == NULL) | |
751 | return 0; | |
f11ac8db TM |
752 | do_flush = req->wb_page != page || req->wb_context != ctx || |
753 | req->wb_lock_context->lockowner != current->files || | |
754 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 755 | nfs_release_request(req); |
1a54533e TM |
756 | if (!do_flush) |
757 | return 0; | |
758 | status = nfs_wb_page(page->mapping->host, page); | |
759 | } while (status == 0); | |
760 | return status; | |
1da177e4 LT |
761 | } |
762 | ||
5d47a356 TM |
763 | /* |
764 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
765 | * the PageUptodate() flag. In this case, we will need to turn off | |
766 | * write optimisations that depend on the page contents being correct. | |
767 | */ | |
768 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
769 | { | |
770 | return PageUptodate(page) && | |
771 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
772 | } | |
773 | ||
1da177e4 LT |
774 | /* |
775 | * Update and possibly write a cached page of an NFS file. | |
776 | * | |
777 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
778 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
779 | */ | |
780 | int nfs_updatepage(struct file *file, struct page *page, | |
781 | unsigned int offset, unsigned int count) | |
782 | { | |
cd3758e3 | 783 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 784 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
785 | int status = 0; |
786 | ||
91d5b470 CL |
787 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
788 | ||
48186c7d | 789 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
790 | file->f_path.dentry->d_parent->d_name.name, |
791 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 792 | (long long)(page_offset(page) + offset)); |
1da177e4 | 793 | |
1da177e4 | 794 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
795 | * is up to date, it may be more efficient to extend the write |
796 | * to cover the entire page in order to avoid fragmentation | |
797 | * inefficiencies. | |
1da177e4 | 798 | */ |
5d47a356 TM |
799 | if (nfs_write_pageuptodate(page, inode) && |
800 | inode->i_flock == NULL && | |
6b2f3d1f | 801 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 802 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 803 | offset = 0; |
1da177e4 LT |
804 | } |
805 | ||
e21195a7 | 806 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
807 | if (status < 0) |
808 | nfs_set_pageerror(page); | |
59b7c05f TM |
809 | else |
810 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 811 | |
48186c7d | 812 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 813 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
814 | return status; |
815 | } | |
816 | ||
a861a1e1 FI |
817 | static void nfs_writepage_release(struct nfs_page *req, |
818 | struct nfs_write_data *data) | |
1da177e4 | 819 | { |
a6305ddb | 820 | struct page *page = req->wb_page; |
1da177e4 | 821 | |
a861a1e1 | 822 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) |
8e821cad | 823 | nfs_inode_remove_request(req); |
9994b62b | 824 | nfs_unlock_request(req); |
a6305ddb | 825 | nfs_end_page_writeback(page); |
1da177e4 LT |
826 | } |
827 | ||
3ff7576d | 828 | static int flush_task_priority(int how) |
1da177e4 LT |
829 | { |
830 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
831 | case FLUSH_HIGHPRI: | |
832 | return RPC_PRIORITY_HIGH; | |
833 | case FLUSH_LOWPRI: | |
834 | return RPC_PRIORITY_LOW; | |
835 | } | |
836 | return RPC_PRIORITY_NORMAL; | |
837 | } | |
838 | ||
c5996c4e FI |
839 | int nfs_initiate_write(struct rpc_clnt *clnt, |
840 | struct nfs_write_data *data, | |
d138d5d1 AA |
841 | const struct rpc_call_ops *call_ops, |
842 | int how) | |
1da177e4 | 843 | { |
d138d5d1 | 844 | struct inode *inode = data->inode; |
3ff7576d | 845 | int priority = flush_task_priority(how); |
07737691 | 846 | struct rpc_task *task; |
bdc7f021 TM |
847 | struct rpc_message msg = { |
848 | .rpc_argp = &data->args, | |
849 | .rpc_resp = &data->res, | |
d138d5d1 | 850 | .rpc_cred = data->cred, |
bdc7f021 | 851 | }; |
84115e1c | 852 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 853 | .rpc_client = clnt, |
07737691 | 854 | .task = &data->task, |
bdc7f021 | 855 | .rpc_message = &msg, |
84115e1c TM |
856 | .callback_ops = call_ops, |
857 | .callback_data = data, | |
101070ca | 858 | .workqueue = nfsiod_workqueue, |
2c61be0a | 859 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 860 | .priority = priority, |
84115e1c | 861 | }; |
2c61be0a | 862 | int ret = 0; |
1da177e4 | 863 | |
d138d5d1 AA |
864 | /* Set up the initial task struct. */ |
865 | NFS_PROTO(inode)->write_setup(data, &msg); | |
866 | ||
867 | dprintk("NFS: %5u initiated write call " | |
868 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
869 | data->task.tk_pid, | |
870 | inode->i_sb->s_id, | |
871 | (long long)NFS_FILEID(inode), | |
872 | data->args.count, | |
873 | (unsigned long long)data->args.offset); | |
874 | ||
875 | task = rpc_run_task(&task_setup_data); | |
876 | if (IS_ERR(task)) { | |
877 | ret = PTR_ERR(task); | |
878 | goto out; | |
879 | } | |
880 | if (how & FLUSH_SYNC) { | |
881 | ret = rpc_wait_for_completion_task(task); | |
882 | if (ret == 0) | |
883 | ret = task->tk_status; | |
884 | } | |
885 | rpc_put_task(task); | |
886 | out: | |
887 | return ret; | |
888 | } | |
a69aef14 | 889 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
890 | |
891 | /* | |
892 | * Set up the argument/result storage required for the RPC call. | |
893 | */ | |
6e4efd56 | 894 | static void nfs_write_rpcsetup(struct nfs_page *req, |
d138d5d1 | 895 | struct nfs_write_data *data, |
d138d5d1 AA |
896 | unsigned int count, unsigned int offset, |
897 | int how) | |
898 | { | |
3d4ff43d | 899 | struct inode *inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 900 | |
1da177e4 LT |
901 | /* Set up the RPC argument and reply structs |
902 | * NB: take care not to mess about with data->commit et al. */ | |
903 | ||
904 | data->req = req; | |
3d4ff43d | 905 | data->inode = inode = req->wb_context->dentry->d_inode; |
d138d5d1 | 906 | data->cred = req->wb_context->cred; |
1da177e4 LT |
907 | |
908 | data->args.fh = NFS_FH(inode); | |
909 | data->args.offset = req_offset(req) + offset; | |
2bea038c BH |
910 | /* pnfs_set_layoutcommit needs this */ |
911 | data->mds_offset = data->args.offset; | |
1da177e4 LT |
912 | data->args.pgbase = req->wb_pgbase + offset; |
913 | data->args.pages = data->pagevec; | |
914 | data->args.count = count; | |
383ba719 | 915 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 916 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 917 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
918 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
919 | case 0: | |
920 | break; | |
921 | case FLUSH_COND_STABLE: | |
922 | if (nfs_need_commit(NFS_I(inode))) | |
923 | break; | |
924 | default: | |
925 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 926 | } |
1da177e4 LT |
927 | |
928 | data->res.fattr = &data->fattr; | |
929 | data->res.count = count; | |
930 | data->res.verf = &data->verf; | |
0e574af1 | 931 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 932 | } |
1da177e4 | 933 | |
6e4efd56 TM |
934 | static int nfs_do_write(struct nfs_write_data *data, |
935 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
936 | int how) |
937 | { | |
5f00bcb3 | 938 | struct inode *inode = data->args.context->dentry->d_inode; |
0382b744 | 939 | |
c5996c4e | 940 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how); |
1da177e4 LT |
941 | } |
942 | ||
275acaaf TM |
943 | static int nfs_do_multiple_writes(struct list_head *head, |
944 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
945 | int how) |
946 | { | |
947 | struct nfs_write_data *data; | |
948 | int ret = 0; | |
949 | ||
950 | while (!list_empty(head)) { | |
951 | int ret2; | |
952 | ||
953 | data = list_entry(head->next, struct nfs_write_data, list); | |
954 | list_del_init(&data->list); | |
955 | ||
dce81290 | 956 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
957 | if (ret == 0) |
958 | ret = ret2; | |
959 | } | |
960 | return ret; | |
961 | } | |
962 | ||
6d884e8f F |
963 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
964 | * call this on each, which will prepare them to be retried on next | |
965 | * writeback using standard nfs. | |
966 | */ | |
967 | static void nfs_redirty_request(struct nfs_page *req) | |
968 | { | |
a6305ddb TM |
969 | struct page *page = req->wb_page; |
970 | ||
6d884e8f | 971 | nfs_mark_request_dirty(req); |
9994b62b | 972 | nfs_unlock_request(req); |
a6305ddb | 973 | nfs_end_page_writeback(page); |
6d884e8f F |
974 | } |
975 | ||
1da177e4 LT |
976 | /* |
977 | * Generate multiple small requests to write out a single | |
978 | * contiguous dirty area on one page. | |
979 | */ | |
275acaaf | 980 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 | 981 | { |
c76069bd | 982 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
983 | struct page *page = req->wb_page; |
984 | struct nfs_write_data *data; | |
d097971d | 985 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 986 | unsigned int offset; |
1da177e4 | 987 | int requests = 0; |
dbae4c73 | 988 | int ret = 0; |
1da177e4 LT |
989 | |
990 | nfs_list_remove_request(req); | |
991 | ||
b31268ac TM |
992 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
993 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || | |
994 | desc->pg_count > wsize)) | |
995 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
996 | ||
997 | ||
275acaaf | 998 | offset = 0; |
c76069bd | 999 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1000 | do { |
1001 | size_t len = min(nbytes, wsize); | |
1002 | ||
8d5658c9 | 1003 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
1004 | if (!data) |
1005 | goto out_bad; | |
275acaaf | 1006 | data->pagevec[0] = page; |
f13c3620 | 1007 | nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags); |
275acaaf | 1008 | list_add(&data->list, res); |
1da177e4 | 1009 | requests++; |
e9f7bee1 | 1010 | nbytes -= len; |
275acaaf | 1011 | offset += len; |
e9f7bee1 | 1012 | } while (nbytes != 0); |
1da177e4 | 1013 | atomic_set(&req->wb_complete, requests); |
50828d7e | 1014 | desc->pg_rpc_callops = &nfs_write_partial_ops; |
dbae4c73 | 1015 | return ret; |
1da177e4 LT |
1016 | |
1017 | out_bad: | |
275acaaf TM |
1018 | while (!list_empty(res)) { |
1019 | data = list_entry(res->next, struct nfs_write_data, list); | |
6e4efd56 | 1020 | list_del(&data->list); |
8ccd271f | 1021 | nfs_writedata_release(data); |
1da177e4 | 1022 | } |
61822ab5 | 1023 | nfs_redirty_request(req); |
1da177e4 LT |
1024 | return -ENOMEM; |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Create an RPC task for the given write request and kick it. | |
1029 | * The page must have been locked by the caller. | |
1030 | * | |
1031 | * It may happen that the page we're passed is not marked dirty. | |
1032 | * This is the case if nfs_updatepage detects a conflicting request | |
1033 | * that has been written but not committed. | |
1034 | */ | |
275acaaf | 1035 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res) |
1da177e4 LT |
1036 | { |
1037 | struct nfs_page *req; | |
1038 | struct page **pages; | |
1039 | struct nfs_write_data *data; | |
c76069bd | 1040 | struct list_head *head = &desc->pg_list; |
3b609184 | 1041 | int ret = 0; |
1da177e4 | 1042 | |
c76069bd FI |
1043 | data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, |
1044 | desc->pg_count)); | |
44b83799 FI |
1045 | if (!data) { |
1046 | while (!list_empty(head)) { | |
1047 | req = nfs_list_entry(head->next); | |
1048 | nfs_list_remove_request(req); | |
1049 | nfs_redirty_request(req); | |
1050 | } | |
1051 | ret = -ENOMEM; | |
1052 | goto out; | |
1053 | } | |
1da177e4 | 1054 | pages = data->pagevec; |
1da177e4 LT |
1055 | while (!list_empty(head)) { |
1056 | req = nfs_list_entry(head->next); | |
1057 | nfs_list_remove_request(req); | |
1058 | nfs_list_add_request(req, &data->pages); | |
1da177e4 | 1059 | *pages++ = req->wb_page; |
1da177e4 LT |
1060 | } |
1061 | req = nfs_list_entry(data->pages.next); | |
1062 | ||
b31268ac TM |
1063 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1064 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) | |
1065 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1066 | ||
1da177e4 | 1067 | /* Set up the argument struct */ |
6e4efd56 | 1068 | nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags); |
275acaaf | 1069 | list_add(&data->list, res); |
50828d7e | 1070 | desc->pg_rpc_callops = &nfs_write_full_ops; |
44b83799 | 1071 | out: |
44b83799 | 1072 | return ret; |
1da177e4 LT |
1073 | } |
1074 | ||
dce81290 TM |
1075 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head) |
1076 | { | |
1077 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
1078 | return nfs_flush_multi(desc, head); | |
1079 | return nfs_flush_one(desc, head); | |
1080 | } | |
1081 | ||
1082 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1083 | { |
275acaaf TM |
1084 | LIST_HEAD(head); |
1085 | int ret; | |
1086 | ||
dce81290 | 1087 | ret = nfs_generic_flush(desc, &head); |
50828d7e TM |
1088 | if (ret == 0) |
1089 | ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops, | |
dce81290 | 1090 | desc->pg_ioflags); |
275acaaf | 1091 | return ret; |
1751c363 | 1092 | } |
1751c363 TM |
1093 | |
1094 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1095 | .pg_test = nfs_generic_pg_test, | |
1096 | .pg_doio = nfs_generic_pg_writepages, | |
1097 | }; | |
1098 | ||
e2fecb21 | 1099 | void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
c63c7b05 | 1100 | struct inode *inode, int ioflags) |
1da177e4 | 1101 | { |
1751c363 TM |
1102 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, |
1103 | NFS_SERVER(inode)->wsize, ioflags); | |
1104 | } | |
1da177e4 | 1105 | |
dce81290 TM |
1106 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1107 | { | |
1108 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1109 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1110 | } | |
1f945357 | 1111 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1112 | |
1751c363 TM |
1113 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1114 | struct inode *inode, int ioflags) | |
1115 | { | |
1116 | if (!pnfs_pageio_init_write(pgio, inode, ioflags)) | |
1117 | nfs_pageio_init_write_mds(pgio, inode, ioflags); | |
1da177e4 LT |
1118 | } |
1119 | ||
1120 | /* | |
1121 | * Handle a write reply that flushed part of a page. | |
1122 | */ | |
788e7a89 | 1123 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 1124 | { |
788e7a89 | 1125 | struct nfs_write_data *data = calldata; |
1da177e4 | 1126 | |
48186c7d CL |
1127 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1128 | task->tk_pid, | |
3d4ff43d | 1129 | data->req->wb_context->dentry->d_inode->i_sb->s_id, |
48186c7d | 1130 | (long long) |
3d4ff43d | 1131 | NFS_FILEID(data->req->wb_context->dentry->d_inode), |
48186c7d | 1132 | data->req->wb_bytes, (long long)req_offset(data->req)); |
1da177e4 | 1133 | |
c9d8f89d TM |
1134 | nfs_writeback_done(task, data); |
1135 | } | |
788e7a89 | 1136 | |
c9d8f89d TM |
1137 | static void nfs_writeback_release_partial(void *calldata) |
1138 | { | |
1139 | struct nfs_write_data *data = calldata; | |
1140 | struct nfs_page *req = data->req; | |
1141 | struct page *page = req->wb_page; | |
1142 | int status = data->task.tk_status; | |
1143 | ||
1144 | if (status < 0) { | |
a301b777 | 1145 | nfs_set_pageerror(page); |
c9d8f89d TM |
1146 | nfs_context_set_write_error(req->wb_context, status); |
1147 | dprintk(", error = %d\n", status); | |
8e821cad | 1148 | goto out; |
1da177e4 LT |
1149 | } |
1150 | ||
8e821cad | 1151 | if (nfs_write_need_commit(data)) { |
587142f8 | 1152 | struct inode *inode = page->mapping->host; |
8e821cad | 1153 | |
587142f8 | 1154 | spin_lock(&inode->i_lock); |
8e821cad TM |
1155 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1156 | /* Do nothing we need to resend the writes */ | |
1157 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1158 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1159 | dprintk(" defer commit\n"); | |
1160 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1161 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1162 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1163 | dprintk(" server reboot detected\n"); | |
1164 | } | |
587142f8 | 1165 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1166 | } else |
1167 | dprintk(" OK\n"); | |
1168 | ||
1169 | out: | |
1da177e4 | 1170 | if (atomic_dec_and_test(&req->wb_complete)) |
a861a1e1 | 1171 | nfs_writepage_release(req, data); |
c9d8f89d | 1172 | nfs_writedata_release(calldata); |
1da177e4 LT |
1173 | } |
1174 | ||
def6ed7e AA |
1175 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1176 | { | |
1177 | struct nfs_write_data *data = calldata; | |
c6cb80d0 | 1178 | NFS_PROTO(data->inode)->write_rpc_prepare(task, data); |
def6ed7e | 1179 | } |
def6ed7e | 1180 | |
0b7c0153 FI |
1181 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1182 | { | |
1183 | struct nfs_commit_data *data = calldata; | |
1184 | ||
1185 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1186 | } | |
1187 | ||
788e7a89 | 1188 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e | 1189 | .rpc_call_prepare = nfs_write_prepare, |
788e7a89 | 1190 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1191 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1192 | }; |
1193 | ||
1da177e4 LT |
1194 | /* |
1195 | * Handle a write reply that flushes a whole page. | |
1196 | * | |
1197 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1198 | * writebacks since the page->count is kept > 1 for as long | |
1199 | * as the page has a write request pending. | |
1200 | */ | |
788e7a89 | 1201 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1202 | { |
788e7a89 | 1203 | struct nfs_write_data *data = calldata; |
1da177e4 | 1204 | |
c9d8f89d TM |
1205 | nfs_writeback_done(task, data); |
1206 | } | |
1207 | ||
1208 | static void nfs_writeback_release_full(void *calldata) | |
1209 | { | |
1210 | struct nfs_write_data *data = calldata; | |
e2fecb21 | 1211 | int status = data->task.tk_status; |
788e7a89 | 1212 | |
1da177e4 LT |
1213 | /* Update attributes as result of writeback. */ |
1214 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1215 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1216 | struct page *page = req->wb_page; | |
1217 | ||
1da177e4 | 1218 | nfs_list_remove_request(req); |
1da177e4 | 1219 | |
48186c7d CL |
1220 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1221 | data->task.tk_pid, | |
3d4ff43d AV |
1222 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1223 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1224 | req->wb_bytes, |
1225 | (long long)req_offset(req)); | |
1226 | ||
c9d8f89d | 1227 | if (status < 0) { |
a301b777 | 1228 | nfs_set_pageerror(page); |
c9d8f89d TM |
1229 | nfs_context_set_write_error(req->wb_context, status); |
1230 | dprintk(", error = %d\n", status); | |
8e821cad | 1231 | goto remove_request; |
1da177e4 | 1232 | } |
1da177e4 | 1233 | |
8e821cad TM |
1234 | if (nfs_write_need_commit(data)) { |
1235 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
a861a1e1 | 1236 | nfs_mark_request_commit(req, data->lseg); |
8e821cad | 1237 | dprintk(" marked for commit\n"); |
1da177e4 LT |
1238 | goto next; |
1239 | } | |
8e821cad TM |
1240 | dprintk(" OK\n"); |
1241 | remove_request: | |
1da177e4 | 1242 | nfs_inode_remove_request(req); |
1da177e4 | 1243 | next: |
9994b62b | 1244 | nfs_unlock_request(req); |
a6305ddb | 1245 | nfs_end_page_writeback(page); |
1da177e4 | 1246 | } |
c9d8f89d | 1247 | nfs_writedata_release(calldata); |
1da177e4 LT |
1248 | } |
1249 | ||
788e7a89 | 1250 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e | 1251 | .rpc_call_prepare = nfs_write_prepare, |
788e7a89 | 1252 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1253 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1254 | }; |
1255 | ||
1256 | ||
1da177e4 LT |
1257 | /* |
1258 | * This function is called when the WRITE call is complete. | |
1259 | */ | |
13602896 | 1260 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1261 | { |
1da177e4 LT |
1262 | struct nfs_writeargs *argp = &data->args; |
1263 | struct nfs_writeres *resp = &data->res; | |
788e7a89 | 1264 | int status; |
1da177e4 | 1265 | |
a3f565b1 | 1266 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1267 | task->tk_pid, task->tk_status); |
1268 | ||
f551e44f CL |
1269 | /* |
1270 | * ->write_done will attempt to use post-op attributes to detect | |
1271 | * conflicting writes by other clients. A strict interpretation | |
1272 | * of close-to-open would allow us to continue caching even if | |
1273 | * another writer had changed the file, but some applications | |
1274 | * depend on tighter cache coherency when writing. | |
1275 | */ | |
788e7a89 TM |
1276 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1277 | if (status != 0) | |
13602896 | 1278 | return; |
91d5b470 CL |
1279 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1280 | ||
1da177e4 LT |
1281 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1282 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1283 | /* We tried a write call, but the server did not | |
1284 | * commit data to stable storage even though we | |
1285 | * requested it. | |
1286 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1287 | * the server reports NFS_DATA_SYNC, but performs | |
1288 | * NFS_FILE_SYNC. We therefore implement this checking | |
1289 | * as a dprintk() in order to avoid filling syslog. | |
1290 | */ | |
1291 | static unsigned long complain; | |
1292 | ||
a69aef14 | 1293 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1294 | if (time_before(complain, jiffies)) { |
48186c7d | 1295 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1296 | " (committed = %d) != (stable = %d)\n", |
2b72c9cc | 1297 | NFS_SERVER(data->inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1298 | resp->verf->committed, argp->stable); |
1299 | complain = jiffies + 300 * HZ; | |
1300 | } | |
1301 | } | |
1302 | #endif | |
1303 | /* Is this a short write? */ | |
1304 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1305 | static unsigned long complain; | |
1306 | ||
91d5b470 CL |
1307 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1308 | ||
1da177e4 LT |
1309 | /* Has the server at least made some progress? */ |
1310 | if (resp->count != 0) { | |
1311 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1312 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1313 | /* Resend from where the server left off */ | |
a69aef14 | 1314 | data->mds_offset += resp->count; |
1da177e4 LT |
1315 | argp->offset += resp->count; |
1316 | argp->pgbase += resp->count; | |
1317 | argp->count -= resp->count; | |
1318 | } else { | |
1319 | /* Resend as a stable write in order to avoid | |
1320 | * headaches in the case of a server crash. | |
1321 | */ | |
1322 | argp->stable = NFS_FILE_SYNC; | |
1323 | } | |
d00c5d43 | 1324 | rpc_restart_call_prepare(task); |
13602896 | 1325 | return; |
1da177e4 LT |
1326 | } |
1327 | if (time_before(complain, jiffies)) { | |
1328 | printk(KERN_WARNING | |
1329 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1330 | argp->count); | |
1331 | complain = jiffies + 300 * HZ; | |
1332 | } | |
1333 | /* Can't do anything about it except throw an error. */ | |
1334 | task->tk_status = -EIO; | |
1335 | } | |
13602896 | 1336 | return; |
1da177e4 LT |
1337 | } |
1338 | ||
1339 | ||
1340 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1341 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1342 | { | |
b8413f98 TM |
1343 | int ret; |
1344 | ||
71d0a611 TM |
1345 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1346 | return 1; | |
b8413f98 TM |
1347 | if (!may_wait) |
1348 | return 0; | |
1349 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1350 | NFS_INO_COMMIT, | |
1351 | nfs_wait_bit_killable, | |
1352 | TASK_KILLABLE); | |
1353 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1354 | } |
1355 | ||
e0c2b380 | 1356 | void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1357 | { |
1358 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1359 | smp_mb__after_clear_bit(); | |
1360 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1361 | } | |
e0c2b380 | 1362 | EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); |
71d0a611 | 1363 | |
0b7c0153 | 1364 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1365 | { |
0b7c0153 FI |
1366 | put_nfs_open_context(data->context); |
1367 | nfs_commit_free(data); | |
1da177e4 | 1368 | } |
e0c2b380 | 1369 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1370 | |
0b7c0153 | 1371 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd FI |
1372 | const struct rpc_call_ops *call_ops, |
1373 | int how) | |
1da177e4 | 1374 | { |
07737691 | 1375 | struct rpc_task *task; |
9ace33cd | 1376 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1377 | struct rpc_message msg = { |
1378 | .rpc_argp = &data->args, | |
1379 | .rpc_resp = &data->res, | |
9ace33cd | 1380 | .rpc_cred = data->cred, |
bdc7f021 | 1381 | }; |
84115e1c | 1382 | struct rpc_task_setup task_setup_data = { |
07737691 | 1383 | .task = &data->task, |
9ace33cd | 1384 | .rpc_client = clnt, |
bdc7f021 | 1385 | .rpc_message = &msg, |
9ace33cd | 1386 | .callback_ops = call_ops, |
84115e1c | 1387 | .callback_data = data, |
101070ca | 1388 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1389 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1390 | .priority = priority, |
84115e1c | 1391 | }; |
9ace33cd FI |
1392 | /* Set up the initial task struct. */ |
1393 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1394 | ||
1395 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1396 | ||
1397 | task = rpc_run_task(&task_setup_data); | |
1398 | if (IS_ERR(task)) | |
1399 | return PTR_ERR(task); | |
1400 | if (how & FLUSH_SYNC) | |
1401 | rpc_wait_for_completion_task(task); | |
1402 | rpc_put_task(task); | |
1403 | return 0; | |
1404 | } | |
e0c2b380 | 1405 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1406 | |
1407 | /* | |
1408 | * Set up the argument/result storage required for the RPC call. | |
1409 | */ | |
0b7c0153 | 1410 | void nfs_init_commit(struct nfs_commit_data *data, |
988b6dce FI |
1411 | struct list_head *head, |
1412 | struct pnfs_layout_segment *lseg) | |
9ace33cd FI |
1413 | { |
1414 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1415 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1416 | |
1417 | /* Set up the RPC argument and reply structs | |
1418 | * NB: take care not to mess about with data->commit et al. */ | |
1419 | ||
1420 | list_splice_init(head, &data->pages); | |
1da177e4 | 1421 | |
1da177e4 | 1422 | data->inode = inode; |
9ace33cd | 1423 | data->cred = first->wb_context->cred; |
988b6dce | 1424 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1425 | data->mds_ops = &nfs_commit_ops; |
1da177e4 LT |
1426 | |
1427 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1428 | /* Note: we always request a commit of the entire inode */ |
1429 | data->args.offset = 0; | |
1430 | data->args.count = 0; | |
0b7c0153 | 1431 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1432 | data->res.fattr = &data->fattr; |
1433 | data->res.verf = &data->verf; | |
0e574af1 | 1434 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1435 | } |
e0c2b380 | 1436 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1437 | |
e0c2b380 | 1438 | void nfs_retry_commit(struct list_head *page_list, |
a861a1e1 | 1439 | struct pnfs_layout_segment *lseg) |
64bfeb49 FI |
1440 | { |
1441 | struct nfs_page *req; | |
1442 | ||
1443 | while (!list_empty(page_list)) { | |
1444 | req = nfs_list_entry(page_list->next); | |
1445 | nfs_list_remove_request(req); | |
a861a1e1 | 1446 | nfs_mark_request_commit(req, lseg); |
64bfeb49 FI |
1447 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1448 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1449 | BDI_RECLAIMABLE); | |
9994b62b | 1450 | nfs_unlock_request(req); |
64bfeb49 FI |
1451 | } |
1452 | } | |
e0c2b380 | 1453 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1454 | |
1da177e4 LT |
1455 | /* |
1456 | * Commit dirty pages | |
1457 | */ | |
1458 | static int | |
40859d7e | 1459 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 | 1460 | { |
0b7c0153 | 1461 | struct nfs_commit_data *data; |
1da177e4 | 1462 | |
c9d8f89d | 1463 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1464 | |
1465 | if (!data) | |
1466 | goto out_bad; | |
1467 | ||
1468 | /* Set up the argument struct */ | |
988b6dce | 1469 | nfs_init_commit(data, head, NULL); |
0b7c0153 | 1470 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, how); |
1da177e4 | 1471 | out_bad: |
a861a1e1 | 1472 | nfs_retry_commit(head, NULL); |
71d0a611 | 1473 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1474 | return -ENOMEM; |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * COMMIT call returned | |
1479 | */ | |
788e7a89 | 1480 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1481 | { |
0b7c0153 | 1482 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1483 | |
a3f565b1 | 1484 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1485 | task->tk_pid, task->tk_status); |
1486 | ||
788e7a89 | 1487 | /* Call the NFS version-specific code */ |
c0d0e96b | 1488 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1489 | } |
1490 | ||
0b7c0153 | 1491 | void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1492 | { |
5917ce84 | 1493 | struct nfs_page *req; |
c9d8f89d | 1494 | int status = data->task.tk_status; |
788e7a89 | 1495 | |
1da177e4 LT |
1496 | while (!list_empty(&data->pages)) { |
1497 | req = nfs_list_entry(data->pages.next); | |
1498 | nfs_list_remove_request(req); | |
d6d6dc7c | 1499 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1500 | |
48186c7d | 1501 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1502 | req->wb_context->dentry->d_sb->s_id, |
1503 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1504 | req->wb_bytes, |
1505 | (long long)req_offset(req)); | |
c9d8f89d TM |
1506 | if (status < 0) { |
1507 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1508 | nfs_inode_remove_request(req); |
c9d8f89d | 1509 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1510 | goto next; |
1511 | } | |
1512 | ||
1513 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1514 | * returned by the server against all stored verfs. */ | |
1515 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1516 | /* We have a match */ | |
1517 | nfs_inode_remove_request(req); | |
1518 | dprintk(" OK\n"); | |
1519 | goto next; | |
1520 | } | |
1521 | /* We have a mismatch. Write the page again */ | |
1522 | dprintk(" mismatch\n"); | |
6d884e8f | 1523 | nfs_mark_request_dirty(req); |
1da177e4 | 1524 | next: |
9994b62b | 1525 | nfs_unlock_request(req); |
1da177e4 | 1526 | } |
5917ce84 | 1527 | } |
e0c2b380 | 1528 | EXPORT_SYMBOL_GPL(nfs_commit_release_pages); |
5917ce84 FI |
1529 | |
1530 | static void nfs_commit_release(void *calldata) | |
1531 | { | |
0b7c0153 | 1532 | struct nfs_commit_data *data = calldata; |
5917ce84 FI |
1533 | |
1534 | nfs_commit_release_pages(data); | |
71d0a611 | 1535 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1536 | nfs_commitdata_release(calldata); |
1da177e4 | 1537 | } |
788e7a89 TM |
1538 | |
1539 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1540 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1541 | .rpc_call_done = nfs_commit_done, |
1542 | .rpc_release = nfs_commit_release, | |
1543 | }; | |
1da177e4 | 1544 | |
b608b283 | 1545 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1546 | { |
1da177e4 | 1547 | LIST_HEAD(head); |
71d0a611 | 1548 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1549 | int res; |
1da177e4 | 1550 | |
b8413f98 TM |
1551 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1552 | if (res <= 0) | |
c5efa5fc | 1553 | goto out_mark_dirty; |
d6d6dc7c | 1554 | res = nfs_scan_commit(inode, &head); |
1da177e4 | 1555 | if (res) { |
a861a1e1 FI |
1556 | int error; |
1557 | ||
1558 | error = pnfs_commit_list(inode, &head, how); | |
1559 | if (error == PNFS_NOT_ATTEMPTED) | |
1560 | error = nfs_commit_list(inode, &head, how); | |
3da28eb1 TM |
1561 | if (error < 0) |
1562 | return error; | |
b8413f98 | 1563 | if (!may_wait) |
c5efa5fc | 1564 | goto out_mark_dirty; |
b8413f98 TM |
1565 | error = wait_on_bit(&NFS_I(inode)->flags, |
1566 | NFS_INO_COMMIT, | |
1567 | nfs_wait_bit_killable, | |
1568 | TASK_KILLABLE); | |
1569 | if (error < 0) | |
1570 | return error; | |
71d0a611 TM |
1571 | } else |
1572 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1573 | return res; |
1574 | /* Note: If we exit without ensuring that the commit is complete, | |
1575 | * we must mark the inode as dirty. Otherwise, future calls to | |
1576 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1577 | * that the data is on the disk. | |
1578 | */ | |
1579 | out_mark_dirty: | |
1580 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1581 | return res; |
1582 | } | |
8fc795f7 TM |
1583 | |
1584 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1585 | { | |
420e3646 TM |
1586 | struct nfs_inode *nfsi = NFS_I(inode); |
1587 | int flags = FLUSH_SYNC; | |
1588 | int ret = 0; | |
8fc795f7 | 1589 | |
3236c3e1 JL |
1590 | /* no commits means nothing needs to be done */ |
1591 | if (!nfsi->ncommit) | |
1592 | return ret; | |
1593 | ||
a00dd6c0 JL |
1594 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1595 | /* Don't commit yet if this is a non-blocking flush and there | |
1596 | * are a lot of outstanding writes for this mapping. | |
1597 | */ | |
1598 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1599 | goto out_mark_dirty; | |
420e3646 | 1600 | |
a00dd6c0 | 1601 | /* don't wait for the COMMIT response */ |
420e3646 | 1602 | flags = 0; |
a00dd6c0 JL |
1603 | } |
1604 | ||
420e3646 TM |
1605 | ret = nfs_commit_inode(inode, flags); |
1606 | if (ret >= 0) { | |
1607 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1608 | if (ret < wbc->nr_to_write) | |
1609 | wbc->nr_to_write -= ret; | |
1610 | else | |
1611 | wbc->nr_to_write = 0; | |
1612 | } | |
8fc795f7 | 1613 | return 0; |
420e3646 TM |
1614 | } |
1615 | out_mark_dirty: | |
8fc795f7 TM |
1616 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1617 | return ret; | |
1618 | } | |
c63c7b05 | 1619 | #else |
8fc795f7 TM |
1620 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1621 | { | |
1622 | return 0; | |
1623 | } | |
1da177e4 LT |
1624 | #endif |
1625 | ||
8fc795f7 TM |
1626 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1627 | { | |
863a3c6c AA |
1628 | int ret; |
1629 | ||
1630 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1631 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1632 | int status; |
1633 | bool sync = true; | |
863a3c6c | 1634 | |
846d5a09 | 1635 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1636 | sync = false; |
863a3c6c AA |
1637 | |
1638 | status = pnfs_layoutcommit_inode(inode, sync); | |
1639 | if (status < 0) | |
1640 | return status; | |
1641 | } | |
1642 | return ret; | |
8fc795f7 TM |
1643 | } |
1644 | ||
acdc53b2 TM |
1645 | /* |
1646 | * flush the inode to disk. | |
1647 | */ | |
1648 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1649 | { |
1650 | struct writeback_control wbc = { | |
72cb77f4 | 1651 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1652 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1653 | .range_start = 0, |
1654 | .range_end = LLONG_MAX, | |
34901f70 | 1655 | }; |
34901f70 | 1656 | |
acdc53b2 | 1657 | return sync_inode(inode, &wbc); |
1c75950b TM |
1658 | } |
1659 | ||
1b3b4a1a TM |
1660 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1661 | { | |
1662 | struct nfs_page *req; | |
1b3b4a1a TM |
1663 | int ret = 0; |
1664 | ||
1665 | BUG_ON(!PageLocked(page)); | |
1666 | for (;;) { | |
ba8b06e6 | 1667 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1668 | req = nfs_page_find_request(page); |
1669 | if (req == NULL) | |
1b3b4a1a | 1670 | break; |
1b3b4a1a | 1671 | if (nfs_lock_request_dontget(req)) { |
8dd37758 | 1672 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1673 | nfs_inode_remove_request(req); |
1674 | /* | |
1675 | * In case nfs_inode_remove_request has marked the | |
1676 | * page as being dirty | |
1677 | */ | |
1678 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1679 | nfs_unlock_request(req); | |
1680 | break; | |
1681 | } | |
1682 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1683 | nfs_release_request(req); |
1b3b4a1a | 1684 | if (ret < 0) |
c988950e | 1685 | break; |
1b3b4a1a | 1686 | } |
1b3b4a1a TM |
1687 | return ret; |
1688 | } | |
1689 | ||
7f2f12d9 TM |
1690 | /* |
1691 | * Write back all requests on one page - we do this before reading it. | |
1692 | */ | |
1693 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1694 | { |
1695 | loff_t range_start = page_offset(page); | |
1696 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1697 | struct writeback_control wbc = { |
4d770ccf | 1698 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1699 | .nr_to_write = 0, |
4d770ccf TM |
1700 | .range_start = range_start, |
1701 | .range_end = range_end, | |
1702 | }; | |
1703 | int ret; | |
1c75950b | 1704 | |
0522f6ad | 1705 | for (;;) { |
ba8b06e6 | 1706 | wait_on_page_writeback(page); |
73e3302f TM |
1707 | if (clear_page_dirty_for_io(page)) { |
1708 | ret = nfs_writepage_locked(page, &wbc); | |
1709 | if (ret < 0) | |
1710 | goto out_error; | |
0522f6ad | 1711 | continue; |
7f2f12d9 | 1712 | } |
0522f6ad TM |
1713 | if (!PagePrivate(page)) |
1714 | break; | |
1715 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1716 | if (ret < 0) |
73e3302f | 1717 | goto out_error; |
7f2f12d9 | 1718 | } |
73e3302f TM |
1719 | return 0; |
1720 | out_error: | |
4d770ccf | 1721 | return ret; |
1c75950b TM |
1722 | } |
1723 | ||
074cc1de TM |
1724 | #ifdef CONFIG_MIGRATION |
1725 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1726 | struct page *page, enum migrate_mode mode) |
074cc1de | 1727 | { |
2da95652 JL |
1728 | /* |
1729 | * If PagePrivate is set, then the page is currently associated with | |
1730 | * an in-progress read or write request. Don't try to migrate it. | |
1731 | * | |
1732 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1733 | * that we can safely release the inode reference while holding | |
1734 | * the page lock. | |
1735 | */ | |
1736 | if (PagePrivate(page)) | |
1737 | return -EBUSY; | |
074cc1de | 1738 | |
7549ad5f | 1739 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1740 | |
a6bc32b8 | 1741 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1742 | } |
1743 | #endif | |
1744 | ||
f7b422b1 | 1745 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1746 | { |
1747 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1748 | sizeof(struct nfs_write_data), | |
1749 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1750 | NULL); |
1da177e4 LT |
1751 | if (nfs_wdata_cachep == NULL) |
1752 | return -ENOMEM; | |
1753 | ||
93d2341c MD |
1754 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1755 | nfs_wdata_cachep); | |
1da177e4 LT |
1756 | if (nfs_wdata_mempool == NULL) |
1757 | return -ENOMEM; | |
1758 | ||
0b7c0153 FI |
1759 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1760 | sizeof(struct nfs_commit_data), | |
1761 | 0, SLAB_HWCACHE_ALIGN, | |
1762 | NULL); | |
1763 | if (nfs_cdata_cachep == NULL) | |
1764 | return -ENOMEM; | |
1765 | ||
93d2341c MD |
1766 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1767 | nfs_wdata_cachep); | |
1da177e4 LT |
1768 | if (nfs_commit_mempool == NULL) |
1769 | return -ENOMEM; | |
1770 | ||
89a09141 PZ |
1771 | /* |
1772 | * NFS congestion size, scale with available memory. | |
1773 | * | |
1774 | * 64MB: 8192k | |
1775 | * 128MB: 11585k | |
1776 | * 256MB: 16384k | |
1777 | * 512MB: 23170k | |
1778 | * 1GB: 32768k | |
1779 | * 2GB: 46340k | |
1780 | * 4GB: 65536k | |
1781 | * 8GB: 92681k | |
1782 | * 16GB: 131072k | |
1783 | * | |
1784 | * This allows larger machines to have larger/more transfers. | |
1785 | * Limit the default to 256M | |
1786 | */ | |
1787 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1788 | if (nfs_congestion_kb > 256*1024) | |
1789 | nfs_congestion_kb = 256*1024; | |
1790 | ||
1da177e4 LT |
1791 | return 0; |
1792 | } | |
1793 | ||
266bee88 | 1794 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1795 | { |
1796 | mempool_destroy(nfs_commit_mempool); | |
1797 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1798 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1799 | } |
1800 |