Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 AM |
22 | #include <linux/backing-dev.h> |
23 | ||
1da177e4 | 24 | #include <asm/uaccess.h> |
1da177e4 LT |
25 | |
26 | #include "delegation.h" | |
49a70f27 | 27 | #include "internal.h" |
91d5b470 | 28 | #include "iostat.h" |
def6ed7e | 29 | #include "nfs4_fs.h" |
074cc1de | 30 | #include "fscache.h" |
94ad1c80 | 31 | #include "pnfs.h" |
1da177e4 LT |
32 | |
33 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
34 | ||
35 | #define MIN_POOL_WRITE (32) | |
36 | #define MIN_POOL_COMMIT (4) | |
37 | ||
38 | /* | |
39 | * Local function declarations | |
40 | */ | |
c63c7b05 TM |
41 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
42 | struct inode *inode, int ioflags); | |
f8512ad0 | 43 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 TM |
44 | static const struct rpc_call_ops nfs_write_partial_ops; |
45 | static const struct rpc_call_ops nfs_write_full_ops; | |
46 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 | 47 | |
e18b890b | 48 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 49 | static mempool_t *nfs_wdata_mempool; |
1da177e4 LT |
50 | static mempool_t *nfs_commit_mempool; |
51 | ||
c9d8f89d | 52 | struct nfs_write_data *nfs_commitdata_alloc(void) |
1da177e4 | 53 | { |
e6b4f8da | 54 | struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 55 | |
1da177e4 LT |
56 | if (p) { |
57 | memset(p, 0, sizeof(*p)); | |
58 | INIT_LIST_HEAD(&p->pages); | |
59 | } | |
60 | return p; | |
61 | } | |
62 | ||
5e4424af | 63 | void nfs_commit_free(struct nfs_write_data *p) |
1da177e4 | 64 | { |
40859d7e CL |
65 | if (p && (p->pagevec != &p->page_array[0])) |
66 | kfree(p->pagevec); | |
1da177e4 LT |
67 | mempool_free(p, nfs_commit_mempool); |
68 | } | |
69 | ||
8d5658c9 | 70 | struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) |
3feb2d49 | 71 | { |
e6b4f8da | 72 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
73 | |
74 | if (p) { | |
75 | memset(p, 0, sizeof(*p)); | |
76 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 77 | p->npages = pagecount; |
0d0b5cb3 CL |
78 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
79 | p->pagevec = p->page_array; | |
3feb2d49 | 80 | else { |
0d0b5cb3 CL |
81 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
82 | if (!p->pagevec) { | |
3feb2d49 TM |
83 | mempool_free(p, nfs_wdata_mempool); |
84 | p = NULL; | |
85 | } | |
86 | } | |
87 | } | |
88 | return p; | |
89 | } | |
90 | ||
1ae88b2e | 91 | void nfs_writedata_free(struct nfs_write_data *p) |
3feb2d49 TM |
92 | { |
93 | if (p && (p->pagevec != &p->page_array[0])) | |
94 | kfree(p->pagevec); | |
95 | mempool_free(p, nfs_wdata_mempool); | |
96 | } | |
97 | ||
1ae88b2e | 98 | static void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 99 | { |
5053aa56 | 100 | put_lseg(wdata->lseg); |
383ba719 | 101 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
102 | nfs_writedata_free(wdata); |
103 | } | |
104 | ||
7b159fc1 TM |
105 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
106 | { | |
107 | ctx->error = error; | |
108 | smp_wmb(); | |
109 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
110 | } | |
111 | ||
277459d2 TM |
112 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
113 | { | |
114 | struct nfs_page *req = NULL; | |
115 | ||
116 | if (PagePrivate(page)) { | |
117 | req = (struct nfs_page *)page_private(page); | |
118 | if (req != NULL) | |
c03b4024 | 119 | kref_get(&req->wb_kref); |
277459d2 TM |
120 | } |
121 | return req; | |
122 | } | |
123 | ||
124 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
125 | { | |
587142f8 | 126 | struct inode *inode = page->mapping->host; |
277459d2 | 127 | struct nfs_page *req = NULL; |
277459d2 | 128 | |
587142f8 | 129 | spin_lock(&inode->i_lock); |
277459d2 | 130 | req = nfs_page_find_request_locked(page); |
587142f8 | 131 | spin_unlock(&inode->i_lock); |
277459d2 TM |
132 | return req; |
133 | } | |
134 | ||
1da177e4 LT |
135 | /* Adjust the file length if we're writing beyond the end */ |
136 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
137 | { | |
138 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
139 | loff_t end, i_size; |
140 | pgoff_t end_index; | |
1da177e4 | 141 | |
a3d01454 TM |
142 | spin_lock(&inode->i_lock); |
143 | i_size = i_size_read(inode); | |
144 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 145 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 146 | goto out; |
1da177e4 LT |
147 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
148 | if (i_size >= end) | |
a3d01454 | 149 | goto out; |
1da177e4 | 150 | i_size_write(inode, end); |
a3d01454 TM |
151 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
152 | out: | |
153 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
154 | } |
155 | ||
a301b777 TM |
156 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
157 | static void nfs_set_pageerror(struct page *page) | |
158 | { | |
159 | SetPageError(page); | |
160 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
161 | } | |
162 | ||
1da177e4 LT |
163 | /* We can set the PG_uptodate flag if we see that a write request |
164 | * covers the full page. | |
165 | */ | |
166 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
167 | { | |
1da177e4 LT |
168 | if (PageUptodate(page)) |
169 | return; | |
170 | if (base != 0) | |
171 | return; | |
49a70f27 | 172 | if (count != nfs_page_length(page)) |
1da177e4 | 173 | return; |
49a70f27 | 174 | SetPageUptodate(page); |
1da177e4 LT |
175 | } |
176 | ||
1da177e4 LT |
177 | static int wb_priority(struct writeback_control *wbc) |
178 | { | |
179 | if (wbc->for_reclaim) | |
c63c7b05 | 180 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 181 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
182 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
183 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
184 | } |
185 | ||
89a09141 PZ |
186 | /* |
187 | * NFS congestion control | |
188 | */ | |
189 | ||
190 | int nfs_congestion_kb; | |
191 | ||
192 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
193 | #define NFS_CONGESTION_OFF_THRESH \ | |
194 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
195 | ||
5a6d41b3 | 196 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 197 | { |
5a6d41b3 TM |
198 | int ret = test_set_page_writeback(page); |
199 | ||
200 | if (!ret) { | |
89a09141 PZ |
201 | struct inode *inode = page->mapping->host; |
202 | struct nfs_server *nfss = NFS_SERVER(inode); | |
203 | ||
a6305ddb | 204 | page_cache_get(page); |
277866a0 | 205 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
206 | NFS_CONGESTION_ON_THRESH) { |
207 | set_bdi_congested(&nfss->backing_dev_info, | |
208 | BLK_RW_ASYNC); | |
209 | } | |
89a09141 | 210 | } |
5a6d41b3 | 211 | return ret; |
89a09141 PZ |
212 | } |
213 | ||
214 | static void nfs_end_page_writeback(struct page *page) | |
215 | { | |
216 | struct inode *inode = page->mapping->host; | |
217 | struct nfs_server *nfss = NFS_SERVER(inode); | |
218 | ||
219 | end_page_writeback(page); | |
a6305ddb | 220 | page_cache_release(page); |
c4dc4bee | 221 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 222 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
223 | } |
224 | ||
cfb506e1 | 225 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 226 | { |
587142f8 | 227 | struct inode *inode = page->mapping->host; |
e261f51f | 228 | struct nfs_page *req; |
e261f51f TM |
229 | int ret; |
230 | ||
587142f8 | 231 | spin_lock(&inode->i_lock); |
074cc1de | 232 | for (;;) { |
e261f51f | 233 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
234 | if (req == NULL) |
235 | break; | |
acee478a | 236 | if (nfs_set_page_tag_locked(req)) |
e261f51f TM |
237 | break; |
238 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
acee478a | 239 | * then the call to nfs_set_page_tag_locked() will always |
e261f51f TM |
240 | * succeed provided that someone hasn't already marked the |
241 | * request as dirty (in which case we don't care). | |
242 | */ | |
587142f8 | 243 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
244 | if (!nonblock) |
245 | ret = nfs_wait_on_request(req); | |
246 | else | |
247 | ret = -EAGAIN; | |
e261f51f TM |
248 | nfs_release_request(req); |
249 | if (ret != 0) | |
074cc1de | 250 | return ERR_PTR(ret); |
587142f8 | 251 | spin_lock(&inode->i_lock); |
e261f51f | 252 | } |
587142f8 | 253 | spin_unlock(&inode->i_lock); |
074cc1de TM |
254 | return req; |
255 | } | |
256 | ||
257 | /* | |
258 | * Find an associated nfs write request, and prepare to flush it out | |
259 | * May return an error if the user signalled nfs_wait_on_request(). | |
260 | */ | |
261 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 262 | struct page *page, bool nonblock) |
074cc1de TM |
263 | { |
264 | struct nfs_page *req; | |
265 | int ret = 0; | |
266 | ||
cfb506e1 | 267 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
268 | if (!req) |
269 | goto out; | |
270 | ret = PTR_ERR(req); | |
271 | if (IS_ERR(req)) | |
272 | goto out; | |
273 | ||
274 | ret = nfs_set_page_writeback(page); | |
275 | BUG_ON(ret != 0); | |
276 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
277 | ||
f8512ad0 FI |
278 | if (!nfs_pageio_add_request(pgio, req)) { |
279 | nfs_redirty_request(req); | |
074cc1de | 280 | ret = pgio->pg_error; |
f8512ad0 | 281 | } |
074cc1de TM |
282 | out: |
283 | return ret; | |
e261f51f TM |
284 | } |
285 | ||
f758c885 | 286 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 287 | { |
1da177e4 | 288 | struct inode *inode = page->mapping->host; |
cfb506e1 | 289 | int ret; |
1da177e4 | 290 | |
91d5b470 CL |
291 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
292 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
293 | ||
7fe7f848 | 294 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 295 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
296 | if (ret == -EAGAIN) { |
297 | redirty_page_for_writepage(wbc, page); | |
298 | ret = 0; | |
299 | } | |
300 | return ret; | |
f758c885 | 301 | } |
7fe7f848 | 302 | |
f758c885 TM |
303 | /* |
304 | * Write an mmapped page to the server. | |
305 | */ | |
306 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
307 | { | |
308 | struct nfs_pageio_descriptor pgio; | |
309 | int err; | |
49a70f27 | 310 | |
f758c885 TM |
311 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
312 | err = nfs_do_writepage(page, wbc, &pgio); | |
313 | nfs_pageio_complete(&pgio); | |
314 | if (err < 0) | |
315 | return err; | |
316 | if (pgio.pg_error < 0) | |
317 | return pgio.pg_error; | |
318 | return 0; | |
4d770ccf TM |
319 | } |
320 | ||
321 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
322 | { | |
f758c885 | 323 | int ret; |
4d770ccf | 324 | |
f758c885 | 325 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 326 | unlock_page(page); |
f758c885 TM |
327 | return ret; |
328 | } | |
329 | ||
330 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
331 | { | |
332 | int ret; | |
333 | ||
334 | ret = nfs_do_writepage(page, wbc, data); | |
335 | unlock_page(page); | |
336 | return ret; | |
1da177e4 LT |
337 | } |
338 | ||
1da177e4 LT |
339 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
340 | { | |
1da177e4 | 341 | struct inode *inode = mapping->host; |
72cb77f4 | 342 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 343 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
344 | int err; |
345 | ||
72cb77f4 TM |
346 | /* Stop dirtying of new pages while we sync */ |
347 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
348 | nfs_wait_bit_killable, TASK_KILLABLE); | |
349 | if (err) | |
350 | goto out_err; | |
351 | ||
91d5b470 CL |
352 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
353 | ||
c63c7b05 | 354 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 355 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 356 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
357 | |
358 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
359 | smp_mb__after_clear_bit(); | |
360 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
361 | ||
f758c885 | 362 | if (err < 0) |
72cb77f4 TM |
363 | goto out_err; |
364 | err = pgio.pg_error; | |
365 | if (err < 0) | |
366 | goto out_err; | |
c63c7b05 | 367 | return 0; |
72cb77f4 TM |
368 | out_err: |
369 | return err; | |
1da177e4 LT |
370 | } |
371 | ||
372 | /* | |
373 | * Insert a write request into an inode | |
374 | */ | |
e7d39069 | 375 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
376 | { |
377 | struct nfs_inode *nfsi = NFS_I(inode); | |
378 | int error; | |
379 | ||
e7d39069 TM |
380 | error = radix_tree_preload(GFP_NOFS); |
381 | if (error != 0) | |
382 | goto out; | |
383 | ||
384 | /* Lock the request! */ | |
385 | nfs_lock_request_dontget(req); | |
386 | ||
387 | spin_lock(&inode->i_lock); | |
1da177e4 | 388 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); |
27852596 | 389 | BUG_ON(error); |
1da177e4 LT |
390 | if (!nfsi->npages) { |
391 | igrab(inode); | |
1da177e4 LT |
392 | if (nfs_have_delegation(inode, FMODE_WRITE)) |
393 | nfsi->change_attr++; | |
394 | } | |
2df485a7 | 395 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 396 | SetPagePrivate(req->wb_page); |
277459d2 | 397 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 398 | nfsi->npages++; |
c03b4024 | 399 | kref_get(&req->wb_kref); |
27852596 NP |
400 | radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, |
401 | NFS_PAGE_TAG_LOCKED); | |
e7d39069 TM |
402 | spin_unlock(&inode->i_lock); |
403 | radix_tree_preload_end(); | |
404 | out: | |
405 | return error; | |
1da177e4 LT |
406 | } |
407 | ||
408 | /* | |
89a09141 | 409 | * Remove a write request from an inode |
1da177e4 LT |
410 | */ |
411 | static void nfs_inode_remove_request(struct nfs_page *req) | |
412 | { | |
88be9f99 | 413 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
414 | struct nfs_inode *nfsi = NFS_I(inode); |
415 | ||
416 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
417 | ||
587142f8 | 418 | spin_lock(&inode->i_lock); |
277459d2 | 419 | set_page_private(req->wb_page, 0); |
deb7d638 | 420 | ClearPagePrivate(req->wb_page); |
2df485a7 | 421 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 LT |
422 | radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); |
423 | nfsi->npages--; | |
424 | if (!nfsi->npages) { | |
587142f8 | 425 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
426 | iput(inode); |
427 | } else | |
587142f8 | 428 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
429 | nfs_release_request(req); |
430 | } | |
431 | ||
61822ab5 | 432 | static void |
6d884e8f | 433 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 434 | { |
61822ab5 | 435 | __set_page_dirty_nobuffers(req->wb_page); |
b80c3cb6 | 436 | __mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC); |
61822ab5 TM |
437 | } |
438 | ||
1da177e4 LT |
439 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
440 | /* | |
441 | * Add a request to the inode's commit list. | |
442 | */ | |
443 | static void | |
444 | nfs_mark_request_commit(struct nfs_page *req) | |
445 | { | |
88be9f99 | 446 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
447 | struct nfs_inode *nfsi = NFS_I(inode); |
448 | ||
587142f8 | 449 | spin_lock(&inode->i_lock); |
e468bae9 | 450 | set_bit(PG_CLEAN, &(req)->wb_flags); |
5c369683 TM |
451 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
452 | req->wb_index, | |
453 | NFS_PAGE_TAG_COMMIT); | |
ff778d02 | 454 | nfsi->ncommit++; |
587142f8 | 455 | spin_unlock(&inode->i_lock); |
fd39fc85 | 456 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 457 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 458 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 459 | } |
8e821cad | 460 | |
e468bae9 TM |
461 | static int |
462 | nfs_clear_request_commit(struct nfs_page *req) | |
463 | { | |
464 | struct page *page = req->wb_page; | |
465 | ||
466 | if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { | |
467 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
468 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
469 | return 1; | |
470 | } | |
471 | return 0; | |
472 | } | |
473 | ||
8e821cad TM |
474 | static inline |
475 | int nfs_write_need_commit(struct nfs_write_data *data) | |
476 | { | |
465d5243 FI |
477 | if (data->verf.committed == NFS_DATA_SYNC) |
478 | return data->lseg == NULL; | |
479 | else | |
480 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
481 | } |
482 | ||
483 | static inline | |
484 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
485 | { | |
e468bae9 | 486 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
8e821cad TM |
487 | nfs_mark_request_commit(req); |
488 | return 1; | |
489 | } | |
490 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
6d884e8f | 491 | nfs_mark_request_dirty(req); |
8e821cad TM |
492 | return 1; |
493 | } | |
494 | return 0; | |
495 | } | |
496 | #else | |
497 | static inline void | |
498 | nfs_mark_request_commit(struct nfs_page *req) | |
499 | { | |
500 | } | |
501 | ||
e468bae9 TM |
502 | static inline int |
503 | nfs_clear_request_commit(struct nfs_page *req) | |
504 | { | |
505 | return 0; | |
506 | } | |
507 | ||
8e821cad TM |
508 | static inline |
509 | int nfs_write_need_commit(struct nfs_write_data *data) | |
510 | { | |
511 | return 0; | |
512 | } | |
513 | ||
514 | static inline | |
515 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
516 | { | |
517 | return 0; | |
518 | } | |
1da177e4 LT |
519 | #endif |
520 | ||
47c62564 | 521 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
522 | static int |
523 | nfs_need_commit(struct nfs_inode *nfsi) | |
524 | { | |
525 | return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); | |
526 | } | |
527 | ||
1da177e4 LT |
528 | /* |
529 | * nfs_scan_commit - Scan an inode for commit requests | |
530 | * @inode: NFS inode to scan | |
531 | * @dst: destination list | |
532 | * @idx_start: lower bound of page->index to scan. | |
533 | * @npages: idx_start + npages sets the upper bound to scan. | |
534 | * | |
535 | * Moves requests from the inode's 'commit' request list. | |
536 | * The requests are *not* checked to ensure that they form a contiguous set. | |
537 | */ | |
538 | static int | |
ca52fec1 | 539 | nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
540 | { |
541 | struct nfs_inode *nfsi = NFS_I(inode); | |
ff778d02 | 542 | int ret; |
3da28eb1 | 543 | |
fb8a1f11 TM |
544 | if (!nfs_need_commit(nfsi)) |
545 | return 0; | |
546 | ||
ff778d02 TM |
547 | ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); |
548 | if (ret > 0) | |
549 | nfsi->ncommit -= ret; | |
2928db1f TM |
550 | if (nfs_need_commit(NFS_I(inode))) |
551 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
ff778d02 | 552 | return ret; |
1da177e4 | 553 | } |
c42de9dd | 554 | #else |
fb8a1f11 TM |
555 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
556 | { | |
557 | return 0; | |
558 | } | |
559 | ||
ca52fec1 | 560 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
c42de9dd TM |
561 | { |
562 | return 0; | |
563 | } | |
1da177e4 LT |
564 | #endif |
565 | ||
1da177e4 | 566 | /* |
e7d39069 TM |
567 | * Search for an existing write request, and attempt to update |
568 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 569 | * |
e7d39069 TM |
570 | * If the attempt fails, then the existing request is flushed out |
571 | * to disk. | |
1da177e4 | 572 | */ |
e7d39069 TM |
573 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
574 | struct page *page, | |
575 | unsigned int offset, | |
576 | unsigned int bytes) | |
1da177e4 | 577 | { |
e7d39069 TM |
578 | struct nfs_page *req; |
579 | unsigned int rqend; | |
580 | unsigned int end; | |
581 | int error; | |
582 | ||
583 | if (!PagePrivate(page)) | |
584 | return NULL; | |
1da177e4 LT |
585 | |
586 | end = offset + bytes; | |
e7d39069 | 587 | spin_lock(&inode->i_lock); |
1da177e4 | 588 | |
1da177e4 | 589 | for (;;) { |
277459d2 | 590 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
591 | if (req == NULL) |
592 | goto out_unlock; | |
593 | ||
594 | rqend = req->wb_offset + req->wb_bytes; | |
595 | /* | |
596 | * Tell the caller to flush out the request if | |
597 | * the offsets are non-contiguous. | |
598 | * Note: nfs_flush_incompatible() will already | |
599 | * have flushed out requests having wrong owners. | |
600 | */ | |
e468bae9 | 601 | if (offset > rqend |
e7d39069 TM |
602 | || end < req->wb_offset) |
603 | goto out_flushme; | |
604 | ||
605 | if (nfs_set_page_tag_locked(req)) | |
1da177e4 | 606 | break; |
1da177e4 | 607 | |
e7d39069 | 608 | /* The request is locked, so wait and then retry */ |
587142f8 | 609 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
610 | error = nfs_wait_on_request(req); |
611 | nfs_release_request(req); | |
612 | if (error != 0) | |
613 | goto out_err; | |
614 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
615 | } |
616 | ||
ff778d02 TM |
617 | if (nfs_clear_request_commit(req) && |
618 | radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, | |
619 | req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) | |
620 | NFS_I(inode)->ncommit--; | |
e468bae9 | 621 | |
1da177e4 LT |
622 | /* Okay, the request matches. Update the region */ |
623 | if (offset < req->wb_offset) { | |
624 | req->wb_offset = offset; | |
625 | req->wb_pgbase = offset; | |
1da177e4 | 626 | } |
1da177e4 LT |
627 | if (end > rqend) |
628 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
629 | else |
630 | req->wb_bytes = rqend - req->wb_offset; | |
631 | out_unlock: | |
632 | spin_unlock(&inode->i_lock); | |
633 | return req; | |
634 | out_flushme: | |
635 | spin_unlock(&inode->i_lock); | |
636 | nfs_release_request(req); | |
637 | error = nfs_wb_page(inode, page); | |
638 | out_err: | |
639 | return ERR_PTR(error); | |
640 | } | |
641 | ||
642 | /* | |
643 | * Try to update an existing write request, or create one if there is none. | |
644 | * | |
645 | * Note: Should always be called with the Page Lock held to prevent races | |
646 | * if we have to add a new request. Also assumes that the caller has | |
647 | * already called nfs_flush_incompatible() if necessary. | |
648 | */ | |
649 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
650 | struct page *page, unsigned int offset, unsigned int bytes) | |
651 | { | |
652 | struct inode *inode = page->mapping->host; | |
653 | struct nfs_page *req; | |
654 | int error; | |
1da177e4 | 655 | |
e7d39069 TM |
656 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
657 | if (req != NULL) | |
658 | goto out; | |
659 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
660 | if (IS_ERR(req)) | |
661 | goto out; | |
662 | error = nfs_inode_add_request(inode, req); | |
663 | if (error != 0) { | |
664 | nfs_release_request(req); | |
665 | req = ERR_PTR(error); | |
666 | } | |
efc91ed0 | 667 | out: |
61e930a9 | 668 | return req; |
1da177e4 LT |
669 | } |
670 | ||
e7d39069 TM |
671 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
672 | unsigned int offset, unsigned int count) | |
673 | { | |
674 | struct nfs_page *req; | |
675 | ||
676 | req = nfs_setup_write_request(ctx, page, offset, count); | |
677 | if (IS_ERR(req)) | |
678 | return PTR_ERR(req); | |
b80c3cb6 | 679 | nfs_mark_request_dirty(req); |
e7d39069 TM |
680 | /* Update file length */ |
681 | nfs_grow_file(page, offset, count); | |
682 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 683 | nfs_mark_request_dirty(req); |
e7d39069 TM |
684 | nfs_clear_page_tag_locked(req); |
685 | return 0; | |
686 | } | |
687 | ||
1da177e4 LT |
688 | int nfs_flush_incompatible(struct file *file, struct page *page) |
689 | { | |
cd3758e3 | 690 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 691 | struct nfs_page *req; |
1a54533e | 692 | int do_flush, status; |
1da177e4 LT |
693 | /* |
694 | * Look for a request corresponding to this page. If there | |
695 | * is one, and it belongs to another file, we flush it out | |
696 | * before we try to copy anything into the page. Do this | |
697 | * due to the lack of an ACCESS-type call in NFSv2. | |
698 | * Also do the same if we find a request from an existing | |
699 | * dropped page. | |
700 | */ | |
1a54533e TM |
701 | do { |
702 | req = nfs_page_find_request(page); | |
703 | if (req == NULL) | |
704 | return 0; | |
f11ac8db TM |
705 | do_flush = req->wb_page != page || req->wb_context != ctx || |
706 | req->wb_lock_context->lockowner != current->files || | |
707 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 708 | nfs_release_request(req); |
1a54533e TM |
709 | if (!do_flush) |
710 | return 0; | |
711 | status = nfs_wb_page(page->mapping->host, page); | |
712 | } while (status == 0); | |
713 | return status; | |
1da177e4 LT |
714 | } |
715 | ||
5d47a356 TM |
716 | /* |
717 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
718 | * the PageUptodate() flag. In this case, we will need to turn off | |
719 | * write optimisations that depend on the page contents being correct. | |
720 | */ | |
721 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
722 | { | |
723 | return PageUptodate(page) && | |
724 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
725 | } | |
726 | ||
1da177e4 LT |
727 | /* |
728 | * Update and possibly write a cached page of an NFS file. | |
729 | * | |
730 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
731 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
732 | */ | |
733 | int nfs_updatepage(struct file *file, struct page *page, | |
734 | unsigned int offset, unsigned int count) | |
735 | { | |
cd3758e3 | 736 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 737 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
738 | int status = 0; |
739 | ||
91d5b470 CL |
740 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
741 | ||
48186c7d | 742 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
743 | file->f_path.dentry->d_parent->d_name.name, |
744 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 745 | (long long)(page_offset(page) + offset)); |
1da177e4 | 746 | |
1da177e4 | 747 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
748 | * is up to date, it may be more efficient to extend the write |
749 | * to cover the entire page in order to avoid fragmentation | |
750 | * inefficiencies. | |
1da177e4 | 751 | */ |
5d47a356 TM |
752 | if (nfs_write_pageuptodate(page, inode) && |
753 | inode->i_flock == NULL && | |
6b2f3d1f | 754 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 755 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 756 | offset = 0; |
1da177e4 LT |
757 | } |
758 | ||
e21195a7 | 759 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
760 | if (status < 0) |
761 | nfs_set_pageerror(page); | |
1da177e4 | 762 | |
48186c7d | 763 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 764 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
765 | return status; |
766 | } | |
767 | ||
768 | static void nfs_writepage_release(struct nfs_page *req) | |
769 | { | |
a6305ddb | 770 | struct page *page = req->wb_page; |
1da177e4 | 771 | |
a6305ddb | 772 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) |
8e821cad | 773 | nfs_inode_remove_request(req); |
9fd367f0 | 774 | nfs_clear_page_tag_locked(req); |
a6305ddb | 775 | nfs_end_page_writeback(page); |
1da177e4 LT |
776 | } |
777 | ||
3ff7576d | 778 | static int flush_task_priority(int how) |
1da177e4 LT |
779 | { |
780 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
781 | case FLUSH_HIGHPRI: | |
782 | return RPC_PRIORITY_HIGH; | |
783 | case FLUSH_LOWPRI: | |
784 | return RPC_PRIORITY_LOW; | |
785 | } | |
786 | return RPC_PRIORITY_NORMAL; | |
787 | } | |
788 | ||
a69aef14 | 789 | int nfs_initiate_write(struct nfs_write_data *data, |
d138d5d1 AA |
790 | struct rpc_clnt *clnt, |
791 | const struct rpc_call_ops *call_ops, | |
792 | int how) | |
1da177e4 | 793 | { |
d138d5d1 | 794 | struct inode *inode = data->inode; |
3ff7576d | 795 | int priority = flush_task_priority(how); |
07737691 | 796 | struct rpc_task *task; |
bdc7f021 TM |
797 | struct rpc_message msg = { |
798 | .rpc_argp = &data->args, | |
799 | .rpc_resp = &data->res, | |
d138d5d1 | 800 | .rpc_cred = data->cred, |
bdc7f021 | 801 | }; |
84115e1c | 802 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 803 | .rpc_client = clnt, |
07737691 | 804 | .task = &data->task, |
bdc7f021 | 805 | .rpc_message = &msg, |
84115e1c TM |
806 | .callback_ops = call_ops, |
807 | .callback_data = data, | |
101070ca | 808 | .workqueue = nfsiod_workqueue, |
2c61be0a | 809 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 810 | .priority = priority, |
84115e1c | 811 | }; |
2c61be0a | 812 | int ret = 0; |
1da177e4 | 813 | |
d138d5d1 AA |
814 | /* Set up the initial task struct. */ |
815 | NFS_PROTO(inode)->write_setup(data, &msg); | |
816 | ||
817 | dprintk("NFS: %5u initiated write call " | |
818 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
819 | data->task.tk_pid, | |
820 | inode->i_sb->s_id, | |
821 | (long long)NFS_FILEID(inode), | |
822 | data->args.count, | |
823 | (unsigned long long)data->args.offset); | |
824 | ||
825 | task = rpc_run_task(&task_setup_data); | |
826 | if (IS_ERR(task)) { | |
827 | ret = PTR_ERR(task); | |
828 | goto out; | |
829 | } | |
830 | if (how & FLUSH_SYNC) { | |
831 | ret = rpc_wait_for_completion_task(task); | |
832 | if (ret == 0) | |
833 | ret = task->tk_status; | |
834 | } | |
835 | rpc_put_task(task); | |
836 | out: | |
837 | return ret; | |
838 | } | |
a69aef14 | 839 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
840 | |
841 | /* | |
842 | * Set up the argument/result storage required for the RPC call. | |
843 | */ | |
844 | static int nfs_write_rpcsetup(struct nfs_page *req, | |
845 | struct nfs_write_data *data, | |
846 | const struct rpc_call_ops *call_ops, | |
847 | unsigned int count, unsigned int offset, | |
5053aa56 | 848 | struct pnfs_layout_segment *lseg, |
d138d5d1 AA |
849 | int how) |
850 | { | |
851 | struct inode *inode = req->wb_context->path.dentry->d_inode; | |
852 | ||
1da177e4 LT |
853 | /* Set up the RPC argument and reply structs |
854 | * NB: take care not to mess about with data->commit et al. */ | |
855 | ||
856 | data->req = req; | |
88be9f99 | 857 | data->inode = inode = req->wb_context->path.dentry->d_inode; |
d138d5d1 | 858 | data->cred = req->wb_context->cred; |
5053aa56 | 859 | data->lseg = get_lseg(lseg); |
1da177e4 LT |
860 | |
861 | data->args.fh = NFS_FH(inode); | |
862 | data->args.offset = req_offset(req) + offset; | |
863 | data->args.pgbase = req->wb_pgbase + offset; | |
864 | data->args.pages = data->pagevec; | |
865 | data->args.count = count; | |
383ba719 | 866 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 867 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 868 | data->args.stable = NFS_UNSTABLE; |
b31268ac | 869 | if (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
bdc7f021 | 870 | data->args.stable = NFS_DATA_SYNC; |
fb8a1f11 | 871 | if (!nfs_need_commit(NFS_I(inode))) |
bdc7f021 TM |
872 | data->args.stable = NFS_FILE_SYNC; |
873 | } | |
1da177e4 LT |
874 | |
875 | data->res.fattr = &data->fattr; | |
876 | data->res.count = count; | |
877 | data->res.verf = &data->verf; | |
0e574af1 | 878 | nfs_fattr_init(&data->fattr); |
1da177e4 | 879 | |
0382b744 AA |
880 | if (data->lseg && |
881 | (pnfs_try_to_write_data(data, call_ops, how) == PNFS_ATTEMPTED)) | |
882 | return 0; | |
883 | ||
d138d5d1 | 884 | return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); |
1da177e4 LT |
885 | } |
886 | ||
6d884e8f F |
887 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
888 | * call this on each, which will prepare them to be retried on next | |
889 | * writeback using standard nfs. | |
890 | */ | |
891 | static void nfs_redirty_request(struct nfs_page *req) | |
892 | { | |
a6305ddb TM |
893 | struct page *page = req->wb_page; |
894 | ||
6d884e8f | 895 | nfs_mark_request_dirty(req); |
6d884e8f | 896 | nfs_clear_page_tag_locked(req); |
a6305ddb | 897 | nfs_end_page_writeback(page); |
6d884e8f F |
898 | } |
899 | ||
1da177e4 LT |
900 | /* |
901 | * Generate multiple small requests to write out a single | |
902 | * contiguous dirty area on one page. | |
903 | */ | |
c76069bd | 904 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc) |
1da177e4 | 905 | { |
c76069bd | 906 | struct nfs_page *req = nfs_list_entry(desc->pg_list.next); |
1da177e4 LT |
907 | struct page *page = req->wb_page; |
908 | struct nfs_write_data *data; | |
c76069bd | 909 | size_t wsize = NFS_SERVER(desc->pg_inode)->wsize, nbytes; |
e9f7bee1 | 910 | unsigned int offset; |
1da177e4 | 911 | int requests = 0; |
dbae4c73 | 912 | int ret = 0; |
c76069bd | 913 | struct pnfs_layout_segment *lseg; |
1da177e4 LT |
914 | LIST_HEAD(list); |
915 | ||
916 | nfs_list_remove_request(req); | |
917 | ||
b31268ac TM |
918 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
919 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || | |
920 | desc->pg_count > wsize)) | |
921 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
922 | ||
923 | ||
c76069bd | 924 | nbytes = desc->pg_count; |
e9f7bee1 TM |
925 | do { |
926 | size_t len = min(nbytes, wsize); | |
927 | ||
8d5658c9 | 928 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
929 | if (!data) |
930 | goto out_bad; | |
931 | list_add(&data->pages, &list); | |
932 | requests++; | |
e9f7bee1 TM |
933 | nbytes -= len; |
934 | } while (nbytes != 0); | |
1da177e4 LT |
935 | atomic_set(&req->wb_complete, requests); |
936 | ||
c76069bd FI |
937 | BUG_ON(desc->pg_lseg); |
938 | lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW); | |
1da177e4 | 939 | ClearPageError(page); |
1da177e4 | 940 | offset = 0; |
c76069bd | 941 | nbytes = desc->pg_count; |
1da177e4 | 942 | do { |
dbae4c73 TM |
943 | int ret2; |
944 | ||
1da177e4 LT |
945 | data = list_entry(list.next, struct nfs_write_data, pages); |
946 | list_del_init(&data->pages); | |
947 | ||
948 | data->pagevec[0] = page; | |
1da177e4 | 949 | |
bcb71bba TM |
950 | if (nbytes < wsize) |
951 | wsize = nbytes; | |
dbae4c73 | 952 | ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, |
c76069bd | 953 | wsize, offset, lseg, desc->pg_ioflags); |
dbae4c73 TM |
954 | if (ret == 0) |
955 | ret = ret2; | |
bcb71bba TM |
956 | offset += wsize; |
957 | nbytes -= wsize; | |
1da177e4 LT |
958 | } while (nbytes != 0); |
959 | ||
44b83799 | 960 | put_lseg(lseg); |
36fe432d | 961 | desc->pg_lseg = NULL; |
dbae4c73 | 962 | return ret; |
1da177e4 LT |
963 | |
964 | out_bad: | |
965 | while (!list_empty(&list)) { | |
966 | data = list_entry(list.next, struct nfs_write_data, pages); | |
967 | list_del(&data->pages); | |
0da2a4ac | 968 | nfs_writedata_free(data); |
1da177e4 | 969 | } |
61822ab5 | 970 | nfs_redirty_request(req); |
1da177e4 LT |
971 | return -ENOMEM; |
972 | } | |
973 | ||
974 | /* | |
975 | * Create an RPC task for the given write request and kick it. | |
976 | * The page must have been locked by the caller. | |
977 | * | |
978 | * It may happen that the page we're passed is not marked dirty. | |
979 | * This is the case if nfs_updatepage detects a conflicting request | |
980 | * that has been written but not committed. | |
981 | */ | |
c76069bd | 982 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc) |
1da177e4 LT |
983 | { |
984 | struct nfs_page *req; | |
985 | struct page **pages; | |
986 | struct nfs_write_data *data; | |
c76069bd FI |
987 | struct list_head *head = &desc->pg_list; |
988 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
44b83799 | 989 | int ret; |
1da177e4 | 990 | |
c76069bd FI |
991 | data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, |
992 | desc->pg_count)); | |
44b83799 FI |
993 | if (!data) { |
994 | while (!list_empty(head)) { | |
995 | req = nfs_list_entry(head->next); | |
996 | nfs_list_remove_request(req); | |
997 | nfs_redirty_request(req); | |
998 | } | |
999 | ret = -ENOMEM; | |
1000 | goto out; | |
1001 | } | |
1da177e4 | 1002 | pages = data->pagevec; |
1da177e4 LT |
1003 | while (!list_empty(head)) { |
1004 | req = nfs_list_entry(head->next); | |
1005 | nfs_list_remove_request(req); | |
1006 | nfs_list_add_request(req, &data->pages); | |
1007 | ClearPageError(req->wb_page); | |
1da177e4 | 1008 | *pages++ = req->wb_page; |
1da177e4 LT |
1009 | } |
1010 | req = nfs_list_entry(data->pages.next); | |
44b83799 | 1011 | if ((!lseg) && list_is_singular(&data->pages)) |
c76069bd | 1012 | lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW); |
1da177e4 | 1013 | |
b31268ac TM |
1014 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1015 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) | |
1016 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1017 | ||
1da177e4 | 1018 | /* Set up the argument struct */ |
c76069bd | 1019 | ret = nfs_write_rpcsetup(req, data, &nfs_write_full_ops, desc->pg_count, 0, lseg, desc->pg_ioflags); |
44b83799 FI |
1020 | out: |
1021 | put_lseg(lseg); /* Cleans any gotten in ->pg_test */ | |
36fe432d | 1022 | desc->pg_lseg = NULL; |
44b83799 | 1023 | return ret; |
1da177e4 LT |
1024 | } |
1025 | ||
c63c7b05 TM |
1026 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1027 | struct inode *inode, int ioflags) | |
1da177e4 | 1028 | { |
bf4285e7 | 1029 | size_t wsize = NFS_SERVER(inode)->wsize; |
1da177e4 | 1030 | |
44b83799 | 1031 | pnfs_pageio_init_write(pgio, inode); |
94ad1c80 | 1032 | |
bcb71bba | 1033 | if (wsize < PAGE_CACHE_SIZE) |
c63c7b05 | 1034 | nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); |
bcb71bba | 1035 | else |
c63c7b05 | 1036 | nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | /* | |
1040 | * Handle a write reply that flushed part of a page. | |
1041 | */ | |
788e7a89 | 1042 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 1043 | { |
788e7a89 | 1044 | struct nfs_write_data *data = calldata; |
1da177e4 | 1045 | |
48186c7d CL |
1046 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1047 | task->tk_pid, | |
1048 | data->req->wb_context->path.dentry->d_inode->i_sb->s_id, | |
1049 | (long long) | |
1050 | NFS_FILEID(data->req->wb_context->path.dentry->d_inode), | |
1051 | data->req->wb_bytes, (long long)req_offset(data->req)); | |
1da177e4 | 1052 | |
c9d8f89d TM |
1053 | nfs_writeback_done(task, data); |
1054 | } | |
788e7a89 | 1055 | |
c9d8f89d TM |
1056 | static void nfs_writeback_release_partial(void *calldata) |
1057 | { | |
1058 | struct nfs_write_data *data = calldata; | |
1059 | struct nfs_page *req = data->req; | |
1060 | struct page *page = req->wb_page; | |
1061 | int status = data->task.tk_status; | |
1062 | ||
1063 | if (status < 0) { | |
a301b777 | 1064 | nfs_set_pageerror(page); |
c9d8f89d TM |
1065 | nfs_context_set_write_error(req->wb_context, status); |
1066 | dprintk(", error = %d\n", status); | |
8e821cad | 1067 | goto out; |
1da177e4 LT |
1068 | } |
1069 | ||
8e821cad | 1070 | if (nfs_write_need_commit(data)) { |
587142f8 | 1071 | struct inode *inode = page->mapping->host; |
8e821cad | 1072 | |
587142f8 | 1073 | spin_lock(&inode->i_lock); |
8e821cad TM |
1074 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1075 | /* Do nothing we need to resend the writes */ | |
1076 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1077 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1078 | dprintk(" defer commit\n"); | |
1079 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1080 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1081 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1082 | dprintk(" server reboot detected\n"); | |
1083 | } | |
587142f8 | 1084 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1085 | } else |
1086 | dprintk(" OK\n"); | |
1087 | ||
1088 | out: | |
1da177e4 LT |
1089 | if (atomic_dec_and_test(&req->wb_complete)) |
1090 | nfs_writepage_release(req); | |
c9d8f89d | 1091 | nfs_writedata_release(calldata); |
1da177e4 LT |
1092 | } |
1093 | ||
def6ed7e AA |
1094 | #if defined(CONFIG_NFS_V4_1) |
1095 | void nfs_write_prepare(struct rpc_task *task, void *calldata) | |
1096 | { | |
1097 | struct nfs_write_data *data = calldata; | |
def6ed7e | 1098 | |
035168ab TM |
1099 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
1100 | &data->args.seq_args, | |
def6ed7e AA |
1101 | &data->res.seq_res, 1, task)) |
1102 | return; | |
1103 | rpc_call_start(task); | |
1104 | } | |
1105 | #endif /* CONFIG_NFS_V4_1 */ | |
1106 | ||
788e7a89 | 1107 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e AA |
1108 | #if defined(CONFIG_NFS_V4_1) |
1109 | .rpc_call_prepare = nfs_write_prepare, | |
1110 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1111 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1112 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1113 | }; |
1114 | ||
1da177e4 LT |
1115 | /* |
1116 | * Handle a write reply that flushes a whole page. | |
1117 | * | |
1118 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1119 | * writebacks since the page->count is kept > 1 for as long | |
1120 | * as the page has a write request pending. | |
1121 | */ | |
788e7a89 | 1122 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1123 | { |
788e7a89 | 1124 | struct nfs_write_data *data = calldata; |
1da177e4 | 1125 | |
c9d8f89d TM |
1126 | nfs_writeback_done(task, data); |
1127 | } | |
1128 | ||
1129 | static void nfs_writeback_release_full(void *calldata) | |
1130 | { | |
1131 | struct nfs_write_data *data = calldata; | |
1132 | int status = data->task.tk_status; | |
788e7a89 | 1133 | |
1da177e4 LT |
1134 | /* Update attributes as result of writeback. */ |
1135 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1136 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1137 | struct page *page = req->wb_page; | |
1138 | ||
1da177e4 | 1139 | nfs_list_remove_request(req); |
1da177e4 | 1140 | |
48186c7d CL |
1141 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1142 | data->task.tk_pid, | |
88be9f99 TM |
1143 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1144 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1145 | req->wb_bytes, |
1146 | (long long)req_offset(req)); | |
1147 | ||
c9d8f89d | 1148 | if (status < 0) { |
a301b777 | 1149 | nfs_set_pageerror(page); |
c9d8f89d TM |
1150 | nfs_context_set_write_error(req->wb_context, status); |
1151 | dprintk(", error = %d\n", status); | |
8e821cad | 1152 | goto remove_request; |
1da177e4 | 1153 | } |
1da177e4 | 1154 | |
8e821cad TM |
1155 | if (nfs_write_need_commit(data)) { |
1156 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1157 | nfs_mark_request_commit(req); | |
8e821cad | 1158 | dprintk(" marked for commit\n"); |
1da177e4 LT |
1159 | goto next; |
1160 | } | |
8e821cad TM |
1161 | dprintk(" OK\n"); |
1162 | remove_request: | |
1da177e4 | 1163 | nfs_inode_remove_request(req); |
1da177e4 | 1164 | next: |
9fd367f0 | 1165 | nfs_clear_page_tag_locked(req); |
a6305ddb | 1166 | nfs_end_page_writeback(page); |
1da177e4 | 1167 | } |
c9d8f89d | 1168 | nfs_writedata_release(calldata); |
1da177e4 LT |
1169 | } |
1170 | ||
788e7a89 | 1171 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e AA |
1172 | #if defined(CONFIG_NFS_V4_1) |
1173 | .rpc_call_prepare = nfs_write_prepare, | |
1174 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1175 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1176 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1177 | }; |
1178 | ||
1179 | ||
1da177e4 LT |
1180 | /* |
1181 | * This function is called when the WRITE call is complete. | |
1182 | */ | |
13602896 | 1183 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1184 | { |
1da177e4 LT |
1185 | struct nfs_writeargs *argp = &data->args; |
1186 | struct nfs_writeres *resp = &data->res; | |
eedc020e | 1187 | struct nfs_server *server = NFS_SERVER(data->inode); |
788e7a89 | 1188 | int status; |
1da177e4 | 1189 | |
a3f565b1 | 1190 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1191 | task->tk_pid, task->tk_status); |
1192 | ||
f551e44f CL |
1193 | /* |
1194 | * ->write_done will attempt to use post-op attributes to detect | |
1195 | * conflicting writes by other clients. A strict interpretation | |
1196 | * of close-to-open would allow us to continue caching even if | |
1197 | * another writer had changed the file, but some applications | |
1198 | * depend on tighter cache coherency when writing. | |
1199 | */ | |
788e7a89 TM |
1200 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1201 | if (status != 0) | |
13602896 | 1202 | return; |
91d5b470 CL |
1203 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1204 | ||
1da177e4 LT |
1205 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1206 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1207 | /* We tried a write call, but the server did not | |
1208 | * commit data to stable storage even though we | |
1209 | * requested it. | |
1210 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1211 | * the server reports NFS_DATA_SYNC, but performs | |
1212 | * NFS_FILE_SYNC. We therefore implement this checking | |
1213 | * as a dprintk() in order to avoid filling syslog. | |
1214 | */ | |
1215 | static unsigned long complain; | |
1216 | ||
a69aef14 | 1217 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1218 | if (time_before(complain, jiffies)) { |
48186c7d | 1219 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1220 | " (committed = %d) != (stable = %d)\n", |
eedc020e | 1221 | server->nfs_client->cl_hostname, |
1da177e4 LT |
1222 | resp->verf->committed, argp->stable); |
1223 | complain = jiffies + 300 * HZ; | |
1224 | } | |
1225 | } | |
1226 | #endif | |
1227 | /* Is this a short write? */ | |
1228 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1229 | static unsigned long complain; | |
1230 | ||
91d5b470 CL |
1231 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1232 | ||
1da177e4 LT |
1233 | /* Has the server at least made some progress? */ |
1234 | if (resp->count != 0) { | |
1235 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1236 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1237 | /* Resend from where the server left off */ | |
a69aef14 | 1238 | data->mds_offset += resp->count; |
1da177e4 LT |
1239 | argp->offset += resp->count; |
1240 | argp->pgbase += resp->count; | |
1241 | argp->count -= resp->count; | |
1242 | } else { | |
1243 | /* Resend as a stable write in order to avoid | |
1244 | * headaches in the case of a server crash. | |
1245 | */ | |
1246 | argp->stable = NFS_FILE_SYNC; | |
1247 | } | |
0110ee15 | 1248 | nfs_restart_rpc(task, server->nfs_client); |
13602896 | 1249 | return; |
1da177e4 LT |
1250 | } |
1251 | if (time_before(complain, jiffies)) { | |
1252 | printk(KERN_WARNING | |
1253 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1254 | argp->count); | |
1255 | complain = jiffies + 300 * HZ; | |
1256 | } | |
1257 | /* Can't do anything about it except throw an error. */ | |
1258 | task->tk_status = -EIO; | |
1259 | } | |
13602896 | 1260 | return; |
1da177e4 LT |
1261 | } |
1262 | ||
1263 | ||
1264 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1265 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1266 | { | |
b8413f98 TM |
1267 | int ret; |
1268 | ||
71d0a611 TM |
1269 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1270 | return 1; | |
b8413f98 TM |
1271 | if (!may_wait) |
1272 | return 0; | |
1273 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1274 | NFS_INO_COMMIT, | |
1275 | nfs_wait_bit_killable, | |
1276 | TASK_KILLABLE); | |
1277 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1278 | } |
1279 | ||
1280 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) | |
1281 | { | |
1282 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1283 | smp_mb__after_clear_bit(); | |
1284 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1285 | } | |
1286 | ||
1287 | ||
0aa05887 | 1288 | static void nfs_commitdata_release(void *data) |
1da177e4 | 1289 | { |
383ba719 TM |
1290 | struct nfs_write_data *wdata = data; |
1291 | ||
1292 | put_nfs_open_context(wdata->args.context); | |
1da177e4 LT |
1293 | nfs_commit_free(wdata); |
1294 | } | |
1295 | ||
9ace33cd FI |
1296 | static int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, |
1297 | const struct rpc_call_ops *call_ops, | |
1298 | int how) | |
1da177e4 | 1299 | { |
07737691 | 1300 | struct rpc_task *task; |
9ace33cd | 1301 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1302 | struct rpc_message msg = { |
1303 | .rpc_argp = &data->args, | |
1304 | .rpc_resp = &data->res, | |
9ace33cd | 1305 | .rpc_cred = data->cred, |
bdc7f021 | 1306 | }; |
84115e1c | 1307 | struct rpc_task_setup task_setup_data = { |
07737691 | 1308 | .task = &data->task, |
9ace33cd | 1309 | .rpc_client = clnt, |
bdc7f021 | 1310 | .rpc_message = &msg, |
9ace33cd | 1311 | .callback_ops = call_ops, |
84115e1c | 1312 | .callback_data = data, |
101070ca | 1313 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1314 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1315 | .priority = priority, |
84115e1c | 1316 | }; |
9ace33cd FI |
1317 | /* Set up the initial task struct. */ |
1318 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1319 | ||
1320 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1321 | ||
1322 | task = rpc_run_task(&task_setup_data); | |
1323 | if (IS_ERR(task)) | |
1324 | return PTR_ERR(task); | |
1325 | if (how & FLUSH_SYNC) | |
1326 | rpc_wait_for_completion_task(task); | |
1327 | rpc_put_task(task); | |
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * Set up the argument/result storage required for the RPC call. | |
1333 | */ | |
1334 | static void nfs_init_commit(struct nfs_write_data *data, | |
1335 | struct list_head *head) | |
1336 | { | |
1337 | struct nfs_page *first = nfs_list_entry(head->next); | |
1338 | struct inode *inode = first->wb_context->path.dentry->d_inode; | |
1da177e4 LT |
1339 | |
1340 | /* Set up the RPC argument and reply structs | |
1341 | * NB: take care not to mess about with data->commit et al. */ | |
1342 | ||
1343 | list_splice_init(head, &data->pages); | |
1da177e4 | 1344 | |
1da177e4 | 1345 | data->inode = inode; |
9ace33cd FI |
1346 | data->cred = first->wb_context->cred; |
1347 | data->mds_ops = &nfs_commit_ops; | |
1da177e4 LT |
1348 | |
1349 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1350 | /* Note: we always request a commit of the entire inode */ |
1351 | data->args.offset = 0; | |
1352 | data->args.count = 0; | |
383ba719 | 1353 | data->args.context = get_nfs_open_context(first->wb_context); |
3da28eb1 | 1354 | data->res.count = 0; |
1da177e4 LT |
1355 | data->res.fattr = &data->fattr; |
1356 | data->res.verf = &data->verf; | |
0e574af1 | 1357 | nfs_fattr_init(&data->fattr); |
1da177e4 LT |
1358 | } |
1359 | ||
1360 | /* | |
1361 | * Commit dirty pages | |
1362 | */ | |
1363 | static int | |
40859d7e | 1364 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1365 | { |
1366 | struct nfs_write_data *data; | |
1367 | struct nfs_page *req; | |
1368 | ||
c9d8f89d | 1369 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1370 | |
1371 | if (!data) | |
1372 | goto out_bad; | |
1373 | ||
1374 | /* Set up the argument struct */ | |
9ace33cd FI |
1375 | nfs_init_commit(data, head); |
1376 | return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); | |
1da177e4 LT |
1377 | out_bad: |
1378 | while (!list_empty(head)) { | |
1379 | req = nfs_list_entry(head->next); | |
1380 | nfs_list_remove_request(req); | |
1381 | nfs_mark_request_commit(req); | |
83715ad5 | 1382 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 PZ |
1383 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, |
1384 | BDI_RECLAIMABLE); | |
9fd367f0 | 1385 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1386 | } |
71d0a611 | 1387 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1388 | return -ENOMEM; |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * COMMIT call returned | |
1393 | */ | |
788e7a89 | 1394 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1395 | { |
963d8fe5 | 1396 | struct nfs_write_data *data = calldata; |
1da177e4 | 1397 | |
a3f565b1 | 1398 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1399 | task->tk_pid, task->tk_status); |
1400 | ||
788e7a89 TM |
1401 | /* Call the NFS version-specific code */ |
1402 | if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) | |
1403 | return; | |
c9d8f89d TM |
1404 | } |
1405 | ||
1406 | static void nfs_commit_release(void *calldata) | |
1407 | { | |
1408 | struct nfs_write_data *data = calldata; | |
1409 | struct nfs_page *req; | |
1410 | int status = data->task.tk_status; | |
788e7a89 | 1411 | |
1da177e4 LT |
1412 | while (!list_empty(&data->pages)) { |
1413 | req = nfs_list_entry(data->pages.next); | |
1414 | nfs_list_remove_request(req); | |
e468bae9 | 1415 | nfs_clear_request_commit(req); |
1da177e4 | 1416 | |
48186c7d | 1417 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
88be9f99 TM |
1418 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1419 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1420 | req->wb_bytes, |
1421 | (long long)req_offset(req)); | |
c9d8f89d TM |
1422 | if (status < 0) { |
1423 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1424 | nfs_inode_remove_request(req); |
c9d8f89d | 1425 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1426 | goto next; |
1427 | } | |
1428 | ||
1429 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1430 | * returned by the server against all stored verfs. */ | |
1431 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1432 | /* We have a match */ | |
1433 | nfs_inode_remove_request(req); | |
1434 | dprintk(" OK\n"); | |
1435 | goto next; | |
1436 | } | |
1437 | /* We have a mismatch. Write the page again */ | |
1438 | dprintk(" mismatch\n"); | |
6d884e8f | 1439 | nfs_mark_request_dirty(req); |
1da177e4 | 1440 | next: |
9fd367f0 | 1441 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1442 | } |
71d0a611 | 1443 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1444 | nfs_commitdata_release(calldata); |
1da177e4 | 1445 | } |
788e7a89 TM |
1446 | |
1447 | static const struct rpc_call_ops nfs_commit_ops = { | |
21d9a851 AA |
1448 | #if defined(CONFIG_NFS_V4_1) |
1449 | .rpc_call_prepare = nfs_write_prepare, | |
1450 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 TM |
1451 | .rpc_call_done = nfs_commit_done, |
1452 | .rpc_release = nfs_commit_release, | |
1453 | }; | |
1da177e4 | 1454 | |
b608b283 | 1455 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1456 | { |
1da177e4 | 1457 | LIST_HEAD(head); |
71d0a611 | 1458 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1459 | int res; |
1da177e4 | 1460 | |
b8413f98 TM |
1461 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1462 | if (res <= 0) | |
c5efa5fc | 1463 | goto out_mark_dirty; |
587142f8 | 1464 | spin_lock(&inode->i_lock); |
3da28eb1 | 1465 | res = nfs_scan_commit(inode, &head, 0, 0); |
587142f8 | 1466 | spin_unlock(&inode->i_lock); |
1da177e4 | 1467 | if (res) { |
7d46a49f | 1468 | int error = nfs_commit_list(inode, &head, how); |
3da28eb1 TM |
1469 | if (error < 0) |
1470 | return error; | |
b8413f98 | 1471 | if (!may_wait) |
c5efa5fc | 1472 | goto out_mark_dirty; |
b8413f98 TM |
1473 | error = wait_on_bit(&NFS_I(inode)->flags, |
1474 | NFS_INO_COMMIT, | |
1475 | nfs_wait_bit_killable, | |
1476 | TASK_KILLABLE); | |
1477 | if (error < 0) | |
1478 | return error; | |
71d0a611 TM |
1479 | } else |
1480 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1481 | return res; |
1482 | /* Note: If we exit without ensuring that the commit is complete, | |
1483 | * we must mark the inode as dirty. Otherwise, future calls to | |
1484 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1485 | * that the data is on the disk. | |
1486 | */ | |
1487 | out_mark_dirty: | |
1488 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1489 | return res; |
1490 | } | |
8fc795f7 TM |
1491 | |
1492 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1493 | { | |
420e3646 TM |
1494 | struct nfs_inode *nfsi = NFS_I(inode); |
1495 | int flags = FLUSH_SYNC; | |
1496 | int ret = 0; | |
8fc795f7 | 1497 | |
a00dd6c0 JL |
1498 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1499 | /* Don't commit yet if this is a non-blocking flush and there | |
1500 | * are a lot of outstanding writes for this mapping. | |
1501 | */ | |
1502 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1503 | goto out_mark_dirty; | |
420e3646 | 1504 | |
a00dd6c0 | 1505 | /* don't wait for the COMMIT response */ |
420e3646 | 1506 | flags = 0; |
a00dd6c0 JL |
1507 | } |
1508 | ||
420e3646 TM |
1509 | ret = nfs_commit_inode(inode, flags); |
1510 | if (ret >= 0) { | |
1511 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1512 | if (ret < wbc->nr_to_write) | |
1513 | wbc->nr_to_write -= ret; | |
1514 | else | |
1515 | wbc->nr_to_write = 0; | |
1516 | } | |
8fc795f7 | 1517 | return 0; |
420e3646 TM |
1518 | } |
1519 | out_mark_dirty: | |
8fc795f7 TM |
1520 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1521 | return ret; | |
1522 | } | |
c63c7b05 | 1523 | #else |
8fc795f7 TM |
1524 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1525 | { | |
1526 | return 0; | |
1527 | } | |
1da177e4 LT |
1528 | #endif |
1529 | ||
8fc795f7 TM |
1530 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1531 | { | |
1532 | return nfs_commit_unstable_pages(inode, wbc); | |
1533 | } | |
1534 | ||
acdc53b2 TM |
1535 | /* |
1536 | * flush the inode to disk. | |
1537 | */ | |
1538 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1539 | { |
1540 | struct writeback_control wbc = { | |
72cb77f4 | 1541 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1542 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1543 | .range_start = 0, |
1544 | .range_end = LLONG_MAX, | |
34901f70 | 1545 | }; |
34901f70 | 1546 | |
acdc53b2 | 1547 | return sync_inode(inode, &wbc); |
1c75950b TM |
1548 | } |
1549 | ||
1b3b4a1a TM |
1550 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1551 | { | |
1552 | struct nfs_page *req; | |
1b3b4a1a TM |
1553 | int ret = 0; |
1554 | ||
1555 | BUG_ON(!PageLocked(page)); | |
1556 | for (;;) { | |
ba8b06e6 | 1557 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1558 | req = nfs_page_find_request(page); |
1559 | if (req == NULL) | |
1b3b4a1a | 1560 | break; |
1b3b4a1a TM |
1561 | if (nfs_lock_request_dontget(req)) { |
1562 | nfs_inode_remove_request(req); | |
1563 | /* | |
1564 | * In case nfs_inode_remove_request has marked the | |
1565 | * page as being dirty | |
1566 | */ | |
1567 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1568 | nfs_unlock_request(req); | |
1569 | break; | |
1570 | } | |
1571 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1572 | nfs_release_request(req); |
1b3b4a1a | 1573 | if (ret < 0) |
c988950e | 1574 | break; |
1b3b4a1a | 1575 | } |
1b3b4a1a TM |
1576 | return ret; |
1577 | } | |
1578 | ||
7f2f12d9 TM |
1579 | /* |
1580 | * Write back all requests on one page - we do this before reading it. | |
1581 | */ | |
1582 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1583 | { |
1584 | loff_t range_start = page_offset(page); | |
1585 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1586 | struct writeback_control wbc = { |
4d770ccf | 1587 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1588 | .nr_to_write = 0, |
4d770ccf TM |
1589 | .range_start = range_start, |
1590 | .range_end = range_end, | |
1591 | }; | |
1592 | int ret; | |
1c75950b | 1593 | |
0522f6ad | 1594 | for (;;) { |
ba8b06e6 | 1595 | wait_on_page_writeback(page); |
73e3302f TM |
1596 | if (clear_page_dirty_for_io(page)) { |
1597 | ret = nfs_writepage_locked(page, &wbc); | |
1598 | if (ret < 0) | |
1599 | goto out_error; | |
0522f6ad | 1600 | continue; |
7f2f12d9 | 1601 | } |
0522f6ad TM |
1602 | if (!PagePrivate(page)) |
1603 | break; | |
1604 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1605 | if (ret < 0) |
73e3302f | 1606 | goto out_error; |
7f2f12d9 | 1607 | } |
73e3302f TM |
1608 | return 0; |
1609 | out_error: | |
4d770ccf | 1610 | return ret; |
1c75950b TM |
1611 | } |
1612 | ||
074cc1de TM |
1613 | #ifdef CONFIG_MIGRATION |
1614 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
1615 | struct page *page) | |
1616 | { | |
1617 | struct nfs_page *req; | |
1618 | int ret; | |
1619 | ||
7549ad5f | 1620 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1621 | |
cfb506e1 | 1622 | req = nfs_find_and_lock_request(page, false); |
074cc1de TM |
1623 | ret = PTR_ERR(req); |
1624 | if (IS_ERR(req)) | |
1625 | goto out; | |
1626 | ||
1627 | ret = migrate_page(mapping, newpage, page); | |
1628 | if (!req) | |
1629 | goto out; | |
1630 | if (ret) | |
1631 | goto out_unlock; | |
1632 | page_cache_get(newpage); | |
190f38e5 | 1633 | spin_lock(&mapping->host->i_lock); |
074cc1de TM |
1634 | req->wb_page = newpage; |
1635 | SetPagePrivate(newpage); | |
190f38e5 | 1636 | set_page_private(newpage, (unsigned long)req); |
074cc1de TM |
1637 | ClearPagePrivate(page); |
1638 | set_page_private(page, 0); | |
190f38e5 | 1639 | spin_unlock(&mapping->host->i_lock); |
074cc1de TM |
1640 | page_cache_release(page); |
1641 | out_unlock: | |
1642 | nfs_clear_page_tag_locked(req); | |
074cc1de TM |
1643 | out: |
1644 | return ret; | |
1645 | } | |
1646 | #endif | |
1647 | ||
f7b422b1 | 1648 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1649 | { |
1650 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1651 | sizeof(struct nfs_write_data), | |
1652 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1653 | NULL); |
1da177e4 LT |
1654 | if (nfs_wdata_cachep == NULL) |
1655 | return -ENOMEM; | |
1656 | ||
93d2341c MD |
1657 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1658 | nfs_wdata_cachep); | |
1da177e4 LT |
1659 | if (nfs_wdata_mempool == NULL) |
1660 | return -ENOMEM; | |
1661 | ||
93d2341c MD |
1662 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1663 | nfs_wdata_cachep); | |
1da177e4 LT |
1664 | if (nfs_commit_mempool == NULL) |
1665 | return -ENOMEM; | |
1666 | ||
89a09141 PZ |
1667 | /* |
1668 | * NFS congestion size, scale with available memory. | |
1669 | * | |
1670 | * 64MB: 8192k | |
1671 | * 128MB: 11585k | |
1672 | * 256MB: 16384k | |
1673 | * 512MB: 23170k | |
1674 | * 1GB: 32768k | |
1675 | * 2GB: 46340k | |
1676 | * 4GB: 65536k | |
1677 | * 8GB: 92681k | |
1678 | * 16GB: 131072k | |
1679 | * | |
1680 | * This allows larger machines to have larger/more transfers. | |
1681 | * Limit the default to 256M | |
1682 | */ | |
1683 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1684 | if (nfs_congestion_kb > 256*1024) | |
1685 | nfs_congestion_kb = 256*1024; | |
1686 | ||
1da177e4 LT |
1687 | return 0; |
1688 | } | |
1689 | ||
266bee88 | 1690 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1691 | { |
1692 | mempool_destroy(nfs_commit_mempool); | |
1693 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1694 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1695 | } |
1696 |