]>
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> |
1da177e4 LT |
16 | |
17 | #include <linux/sunrpc/clnt.h> | |
18 | #include <linux/nfs_fs.h> | |
19 | #include <linux/nfs_mount.h> | |
20 | #include <linux/nfs_page.h> | |
3fcfab16 AM |
21 | #include <linux/backing-dev.h> |
22 | ||
1da177e4 | 23 | #include <asm/uaccess.h> |
1da177e4 LT |
24 | |
25 | #include "delegation.h" | |
49a70f27 | 26 | #include "internal.h" |
91d5b470 | 27 | #include "iostat.h" |
def6ed7e | 28 | #include "nfs4_fs.h" |
1da177e4 LT |
29 | |
30 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
31 | ||
32 | #define MIN_POOL_WRITE (32) | |
33 | #define MIN_POOL_COMMIT (4) | |
34 | ||
35 | /* | |
36 | * Local function declarations | |
37 | */ | |
c63c7b05 TM |
38 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
39 | struct inode *inode, int ioflags); | |
f8512ad0 | 40 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 TM |
41 | static const struct rpc_call_ops nfs_write_partial_ops; |
42 | static const struct rpc_call_ops nfs_write_full_ops; | |
43 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 | 44 | |
e18b890b | 45 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 46 | static mempool_t *nfs_wdata_mempool; |
1da177e4 LT |
47 | static mempool_t *nfs_commit_mempool; |
48 | ||
c9d8f89d | 49 | struct nfs_write_data *nfs_commitdata_alloc(void) |
1da177e4 | 50 | { |
e6b4f8da | 51 | struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 52 | |
1da177e4 LT |
53 | if (p) { |
54 | memset(p, 0, sizeof(*p)); | |
55 | INIT_LIST_HEAD(&p->pages); | |
5f7dbd5c | 56 | p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE; |
1da177e4 LT |
57 | } |
58 | return p; | |
59 | } | |
60 | ||
5e4424af | 61 | void nfs_commit_free(struct nfs_write_data *p) |
1da177e4 | 62 | { |
40859d7e CL |
63 | if (p && (p->pagevec != &p->page_array[0])) |
64 | kfree(p->pagevec); | |
1da177e4 LT |
65 | mempool_free(p, nfs_commit_mempool); |
66 | } | |
67 | ||
8d5658c9 | 68 | struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) |
3feb2d49 | 69 | { |
e6b4f8da | 70 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
71 | |
72 | if (p) { | |
73 | memset(p, 0, sizeof(*p)); | |
74 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 75 | p->npages = pagecount; |
5f7dbd5c | 76 | p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE; |
0d0b5cb3 CL |
77 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
78 | p->pagevec = p->page_array; | |
3feb2d49 | 79 | else { |
0d0b5cb3 CL |
80 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
81 | if (!p->pagevec) { | |
3feb2d49 TM |
82 | mempool_free(p, nfs_wdata_mempool); |
83 | p = NULL; | |
84 | } | |
85 | } | |
86 | } | |
87 | return p; | |
88 | } | |
89 | ||
1ae88b2e | 90 | void nfs_writedata_free(struct nfs_write_data *p) |
3feb2d49 TM |
91 | { |
92 | if (p && (p->pagevec != &p->page_array[0])) | |
93 | kfree(p->pagevec); | |
94 | mempool_free(p, nfs_wdata_mempool); | |
95 | } | |
96 | ||
1ae88b2e | 97 | static void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 98 | { |
383ba719 | 99 | put_nfs_open_context(wdata->args.context); |
1da177e4 LT |
100 | nfs_writedata_free(wdata); |
101 | } | |
102 | ||
7b159fc1 TM |
103 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
104 | { | |
105 | ctx->error = error; | |
106 | smp_wmb(); | |
107 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
108 | } | |
109 | ||
277459d2 TM |
110 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
111 | { | |
112 | struct nfs_page *req = NULL; | |
113 | ||
114 | if (PagePrivate(page)) { | |
115 | req = (struct nfs_page *)page_private(page); | |
116 | if (req != NULL) | |
c03b4024 | 117 | kref_get(&req->wb_kref); |
277459d2 TM |
118 | } |
119 | return req; | |
120 | } | |
121 | ||
122 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
123 | { | |
587142f8 | 124 | struct inode *inode = page->mapping->host; |
277459d2 | 125 | struct nfs_page *req = NULL; |
277459d2 | 126 | |
587142f8 | 127 | spin_lock(&inode->i_lock); |
277459d2 | 128 | req = nfs_page_find_request_locked(page); |
587142f8 | 129 | spin_unlock(&inode->i_lock); |
277459d2 TM |
130 | return req; |
131 | } | |
132 | ||
1da177e4 LT |
133 | /* Adjust the file length if we're writing beyond the end */ |
134 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
135 | { | |
136 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
137 | loff_t end, i_size; |
138 | pgoff_t end_index; | |
1da177e4 | 139 | |
a3d01454 TM |
140 | spin_lock(&inode->i_lock); |
141 | i_size = i_size_read(inode); | |
142 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 143 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 144 | goto out; |
1da177e4 LT |
145 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
146 | if (i_size >= end) | |
a3d01454 | 147 | goto out; |
1da177e4 | 148 | i_size_write(inode, end); |
a3d01454 TM |
149 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
150 | out: | |
151 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
152 | } |
153 | ||
a301b777 TM |
154 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
155 | static void nfs_set_pageerror(struct page *page) | |
156 | { | |
157 | SetPageError(page); | |
158 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
159 | } | |
160 | ||
1da177e4 LT |
161 | /* We can set the PG_uptodate flag if we see that a write request |
162 | * covers the full page. | |
163 | */ | |
164 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
165 | { | |
1da177e4 LT |
166 | if (PageUptodate(page)) |
167 | return; | |
168 | if (base != 0) | |
169 | return; | |
49a70f27 | 170 | if (count != nfs_page_length(page)) |
1da177e4 | 171 | return; |
49a70f27 | 172 | SetPageUptodate(page); |
1da177e4 LT |
173 | } |
174 | ||
1da177e4 LT |
175 | static int wb_priority(struct writeback_control *wbc) |
176 | { | |
177 | if (wbc->for_reclaim) | |
c63c7b05 | 178 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
1da177e4 LT |
179 | if (wbc->for_kupdate) |
180 | return FLUSH_LOWPRI; | |
181 | return 0; | |
182 | } | |
183 | ||
89a09141 PZ |
184 | /* |
185 | * NFS congestion control | |
186 | */ | |
187 | ||
188 | int nfs_congestion_kb; | |
189 | ||
190 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
191 | #define NFS_CONGESTION_OFF_THRESH \ | |
192 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
193 | ||
5a6d41b3 | 194 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 195 | { |
5a6d41b3 TM |
196 | int ret = test_set_page_writeback(page); |
197 | ||
198 | if (!ret) { | |
89a09141 PZ |
199 | struct inode *inode = page->mapping->host; |
200 | struct nfs_server *nfss = NFS_SERVER(inode); | |
201 | ||
277866a0 | 202 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
203 | NFS_CONGESTION_ON_THRESH) { |
204 | set_bdi_congested(&nfss->backing_dev_info, | |
205 | BLK_RW_ASYNC); | |
206 | } | |
89a09141 | 207 | } |
5a6d41b3 | 208 | return ret; |
89a09141 PZ |
209 | } |
210 | ||
211 | static void nfs_end_page_writeback(struct page *page) | |
212 | { | |
213 | struct inode *inode = page->mapping->host; | |
214 | struct nfs_server *nfss = NFS_SERVER(inode); | |
215 | ||
216 | end_page_writeback(page); | |
c4dc4bee | 217 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 218 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
219 | } |
220 | ||
e261f51f TM |
221 | /* |
222 | * Find an associated nfs write request, and prepare to flush it out | |
9cccef95 | 223 | * May return an error if the user signalled nfs_wait_on_request(). |
e261f51f | 224 | */ |
c63c7b05 TM |
225 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, |
226 | struct page *page) | |
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); |
e261f51f TM |
233 | for(;;) { |
234 | req = nfs_page_find_request_locked(page); | |
235 | if (req == NULL) { | |
587142f8 | 236 | spin_unlock(&inode->i_lock); |
9cccef95 | 237 | return 0; |
e261f51f | 238 | } |
acee478a | 239 | if (nfs_set_page_tag_locked(req)) |
e261f51f TM |
240 | break; |
241 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
acee478a | 242 | * then the call to nfs_set_page_tag_locked() will always |
e261f51f TM |
243 | * succeed provided that someone hasn't already marked the |
244 | * request as dirty (in which case we don't care). | |
245 | */ | |
587142f8 | 246 | spin_unlock(&inode->i_lock); |
e261f51f TM |
247 | ret = nfs_wait_on_request(req); |
248 | nfs_release_request(req); | |
249 | if (ret != 0) | |
250 | return ret; | |
587142f8 | 251 | spin_lock(&inode->i_lock); |
e261f51f | 252 | } |
e468bae9 | 253 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
587142f8 | 254 | spin_unlock(&inode->i_lock); |
e468bae9 | 255 | BUG(); |
612c9384 | 256 | } |
c63c7b05 | 257 | if (nfs_set_page_writeback(page) != 0) { |
587142f8 | 258 | spin_unlock(&inode->i_lock); |
c63c7b05 TM |
259 | BUG(); |
260 | } | |
587142f8 | 261 | spin_unlock(&inode->i_lock); |
f8512ad0 FI |
262 | if (!nfs_pageio_add_request(pgio, req)) { |
263 | nfs_redirty_request(req); | |
f8512ad0 FI |
264 | return pgio->pg_error; |
265 | } | |
9cccef95 | 266 | return 0; |
e261f51f TM |
267 | } |
268 | ||
f758c885 | 269 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 270 | { |
1da177e4 | 271 | struct inode *inode = page->mapping->host; |
1da177e4 | 272 | |
91d5b470 CL |
273 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
274 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
275 | ||
7fe7f848 | 276 | nfs_pageio_cond_complete(pgio, page->index); |
f758c885 TM |
277 | return nfs_page_async_flush(pgio, page); |
278 | } | |
7fe7f848 | 279 | |
f758c885 TM |
280 | /* |
281 | * Write an mmapped page to the server. | |
282 | */ | |
283 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
284 | { | |
285 | struct nfs_pageio_descriptor pgio; | |
286 | int err; | |
49a70f27 | 287 | |
f758c885 TM |
288 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
289 | err = nfs_do_writepage(page, wbc, &pgio); | |
290 | nfs_pageio_complete(&pgio); | |
291 | if (err < 0) | |
292 | return err; | |
293 | if (pgio.pg_error < 0) | |
294 | return pgio.pg_error; | |
295 | return 0; | |
4d770ccf TM |
296 | } |
297 | ||
298 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
299 | { | |
f758c885 | 300 | int ret; |
4d770ccf | 301 | |
f758c885 | 302 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 303 | unlock_page(page); |
f758c885 TM |
304 | return ret; |
305 | } | |
306 | ||
307 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
308 | { | |
309 | int ret; | |
310 | ||
311 | ret = nfs_do_writepage(page, wbc, data); | |
312 | unlock_page(page); | |
313 | return ret; | |
1da177e4 LT |
314 | } |
315 | ||
1da177e4 LT |
316 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
317 | { | |
1da177e4 | 318 | struct inode *inode = mapping->host; |
72cb77f4 | 319 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 320 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
321 | int err; |
322 | ||
72cb77f4 TM |
323 | /* Stop dirtying of new pages while we sync */ |
324 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
325 | nfs_wait_bit_killable, TASK_KILLABLE); | |
326 | if (err) | |
327 | goto out_err; | |
328 | ||
91d5b470 CL |
329 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
330 | ||
c63c7b05 | 331 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 332 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 333 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
334 | |
335 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
336 | smp_mb__after_clear_bit(); | |
337 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
338 | ||
f758c885 | 339 | if (err < 0) |
72cb77f4 TM |
340 | goto out_err; |
341 | err = pgio.pg_error; | |
342 | if (err < 0) | |
343 | goto out_err; | |
c63c7b05 | 344 | return 0; |
72cb77f4 TM |
345 | out_err: |
346 | return err; | |
1da177e4 LT |
347 | } |
348 | ||
349 | /* | |
350 | * Insert a write request into an inode | |
351 | */ | |
e7d39069 | 352 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
353 | { |
354 | struct nfs_inode *nfsi = NFS_I(inode); | |
355 | int error; | |
356 | ||
e7d39069 TM |
357 | error = radix_tree_preload(GFP_NOFS); |
358 | if (error != 0) | |
359 | goto out; | |
360 | ||
361 | /* Lock the request! */ | |
362 | nfs_lock_request_dontget(req); | |
363 | ||
364 | spin_lock(&inode->i_lock); | |
1da177e4 | 365 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); |
27852596 | 366 | BUG_ON(error); |
1da177e4 LT |
367 | if (!nfsi->npages) { |
368 | igrab(inode); | |
1da177e4 LT |
369 | if (nfs_have_delegation(inode, FMODE_WRITE)) |
370 | nfsi->change_attr++; | |
371 | } | |
deb7d638 | 372 | SetPagePrivate(req->wb_page); |
277459d2 | 373 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 374 | nfsi->npages++; |
c03b4024 | 375 | kref_get(&req->wb_kref); |
27852596 NP |
376 | radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, |
377 | NFS_PAGE_TAG_LOCKED); | |
e7d39069 TM |
378 | spin_unlock(&inode->i_lock); |
379 | radix_tree_preload_end(); | |
380 | out: | |
381 | return error; | |
1da177e4 LT |
382 | } |
383 | ||
384 | /* | |
89a09141 | 385 | * Remove a write request from an inode |
1da177e4 LT |
386 | */ |
387 | static void nfs_inode_remove_request(struct nfs_page *req) | |
388 | { | |
88be9f99 | 389 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
390 | struct nfs_inode *nfsi = NFS_I(inode); |
391 | ||
392 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
393 | ||
587142f8 | 394 | spin_lock(&inode->i_lock); |
277459d2 | 395 | set_page_private(req->wb_page, 0); |
deb7d638 | 396 | ClearPagePrivate(req->wb_page); |
1da177e4 LT |
397 | radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); |
398 | nfsi->npages--; | |
399 | if (!nfsi->npages) { | |
587142f8 | 400 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
401 | iput(inode); |
402 | } else | |
587142f8 | 403 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
404 | nfs_clear_request(req); |
405 | nfs_release_request(req); | |
406 | } | |
407 | ||
61822ab5 | 408 | static void |
6d884e8f | 409 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 410 | { |
61822ab5 TM |
411 | __set_page_dirty_nobuffers(req->wb_page); |
412 | } | |
413 | ||
1da177e4 LT |
414 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
415 | /* | |
416 | * Add a request to the inode's commit list. | |
417 | */ | |
418 | static void | |
419 | nfs_mark_request_commit(struct nfs_page *req) | |
420 | { | |
88be9f99 | 421 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
1da177e4 LT |
422 | struct nfs_inode *nfsi = NFS_I(inode); |
423 | ||
587142f8 | 424 | spin_lock(&inode->i_lock); |
e468bae9 | 425 | set_bit(PG_CLEAN, &(req)->wb_flags); |
5c369683 TM |
426 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
427 | req->wb_index, | |
428 | NFS_PAGE_TAG_COMMIT); | |
587142f8 | 429 | spin_unlock(&inode->i_lock); |
fd39fc85 | 430 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 431 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 432 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 433 | } |
8e821cad | 434 | |
e468bae9 TM |
435 | static int |
436 | nfs_clear_request_commit(struct nfs_page *req) | |
437 | { | |
438 | struct page *page = req->wb_page; | |
439 | ||
440 | if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { | |
441 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
442 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
443 | return 1; | |
444 | } | |
445 | return 0; | |
446 | } | |
447 | ||
8e821cad TM |
448 | static inline |
449 | int nfs_write_need_commit(struct nfs_write_data *data) | |
450 | { | |
451 | return data->verf.committed != NFS_FILE_SYNC; | |
452 | } | |
453 | ||
454 | static inline | |
455 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
456 | { | |
e468bae9 | 457 | if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { |
8e821cad TM |
458 | nfs_mark_request_commit(req); |
459 | return 1; | |
460 | } | |
461 | if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { | |
6d884e8f | 462 | nfs_mark_request_dirty(req); |
8e821cad TM |
463 | return 1; |
464 | } | |
465 | return 0; | |
466 | } | |
467 | #else | |
468 | static inline void | |
469 | nfs_mark_request_commit(struct nfs_page *req) | |
470 | { | |
471 | } | |
472 | ||
e468bae9 TM |
473 | static inline int |
474 | nfs_clear_request_commit(struct nfs_page *req) | |
475 | { | |
476 | return 0; | |
477 | } | |
478 | ||
8e821cad TM |
479 | static inline |
480 | int nfs_write_need_commit(struct nfs_write_data *data) | |
481 | { | |
482 | return 0; | |
483 | } | |
484 | ||
485 | static inline | |
486 | int nfs_reschedule_unstable_write(struct nfs_page *req) | |
487 | { | |
488 | return 0; | |
489 | } | |
1da177e4 LT |
490 | #endif |
491 | ||
492 | /* | |
493 | * Wait for a request to complete. | |
494 | * | |
150030b7 | 495 | * Interruptible by fatal signals only. |
1da177e4 | 496 | */ |
ca52fec1 | 497 | static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
498 | { |
499 | struct nfs_inode *nfsi = NFS_I(inode); | |
500 | struct nfs_page *req; | |
ca52fec1 | 501 | pgoff_t idx_end, next; |
1da177e4 LT |
502 | unsigned int res = 0; |
503 | int error; | |
504 | ||
505 | if (npages == 0) | |
506 | idx_end = ~0; | |
507 | else | |
508 | idx_end = idx_start + npages - 1; | |
509 | ||
1da177e4 | 510 | next = idx_start; |
9fd367f0 | 511 | while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { |
1da177e4 LT |
512 | if (req->wb_index > idx_end) |
513 | break; | |
514 | ||
515 | next = req->wb_index + 1; | |
c6a556b8 | 516 | BUG_ON(!NFS_WBACK_BUSY(req)); |
1da177e4 | 517 | |
c03b4024 | 518 | kref_get(&req->wb_kref); |
587142f8 | 519 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
520 | error = nfs_wait_on_request(req); |
521 | nfs_release_request(req); | |
587142f8 | 522 | spin_lock(&inode->i_lock); |
1da177e4 LT |
523 | if (error < 0) |
524 | return error; | |
1da177e4 LT |
525 | res++; |
526 | } | |
1da177e4 LT |
527 | return res; |
528 | } | |
529 | ||
83715ad5 TM |
530 | static void nfs_cancel_commit_list(struct list_head *head) |
531 | { | |
532 | struct nfs_page *req; | |
533 | ||
534 | while(!list_empty(head)) { | |
535 | req = nfs_list_entry(head->next); | |
536 | nfs_list_remove_request(req); | |
e468bae9 | 537 | nfs_clear_request_commit(req); |
83715ad5 | 538 | nfs_inode_remove_request(req); |
b6dff26a | 539 | nfs_unlock_request(req); |
83715ad5 TM |
540 | } |
541 | } | |
542 | ||
47c62564 | 543 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
544 | static int |
545 | nfs_need_commit(struct nfs_inode *nfsi) | |
546 | { | |
547 | return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); | |
548 | } | |
549 | ||
1da177e4 LT |
550 | /* |
551 | * nfs_scan_commit - Scan an inode for commit requests | |
552 | * @inode: NFS inode to scan | |
553 | * @dst: destination list | |
554 | * @idx_start: lower bound of page->index to scan. | |
555 | * @npages: idx_start + npages sets the upper bound to scan. | |
556 | * | |
557 | * Moves requests from the inode's 'commit' request list. | |
558 | * The requests are *not* checked to ensure that they form a contiguous set. | |
559 | */ | |
560 | static int | |
ca52fec1 | 561 | nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
1da177e4 LT |
562 | { |
563 | struct nfs_inode *nfsi = NFS_I(inode); | |
3da28eb1 | 564 | |
fb8a1f11 TM |
565 | if (!nfs_need_commit(nfsi)) |
566 | return 0; | |
567 | ||
568 | return nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); | |
1da177e4 | 569 | } |
c42de9dd | 570 | #else |
fb8a1f11 TM |
571 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
572 | { | |
573 | return 0; | |
574 | } | |
575 | ||
ca52fec1 | 576 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) |
c42de9dd TM |
577 | { |
578 | return 0; | |
579 | } | |
1da177e4 LT |
580 | #endif |
581 | ||
1da177e4 | 582 | /* |
e7d39069 TM |
583 | * Search for an existing write request, and attempt to update |
584 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 585 | * |
e7d39069 TM |
586 | * If the attempt fails, then the existing request is flushed out |
587 | * to disk. | |
1da177e4 | 588 | */ |
e7d39069 TM |
589 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
590 | struct page *page, | |
591 | unsigned int offset, | |
592 | unsigned int bytes) | |
1da177e4 | 593 | { |
e7d39069 TM |
594 | struct nfs_page *req; |
595 | unsigned int rqend; | |
596 | unsigned int end; | |
597 | int error; | |
598 | ||
599 | if (!PagePrivate(page)) | |
600 | return NULL; | |
1da177e4 LT |
601 | |
602 | end = offset + bytes; | |
e7d39069 | 603 | spin_lock(&inode->i_lock); |
1da177e4 | 604 | |
1da177e4 | 605 | for (;;) { |
277459d2 | 606 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
607 | if (req == NULL) |
608 | goto out_unlock; | |
609 | ||
610 | rqend = req->wb_offset + req->wb_bytes; | |
611 | /* | |
612 | * Tell the caller to flush out the request if | |
613 | * the offsets are non-contiguous. | |
614 | * Note: nfs_flush_incompatible() will already | |
615 | * have flushed out requests having wrong owners. | |
616 | */ | |
e468bae9 | 617 | if (offset > rqend |
e7d39069 TM |
618 | || end < req->wb_offset) |
619 | goto out_flushme; | |
620 | ||
621 | if (nfs_set_page_tag_locked(req)) | |
1da177e4 | 622 | break; |
1da177e4 | 623 | |
e7d39069 | 624 | /* The request is locked, so wait and then retry */ |
587142f8 | 625 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
626 | error = nfs_wait_on_request(req); |
627 | nfs_release_request(req); | |
628 | if (error != 0) | |
629 | goto out_err; | |
630 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
631 | } |
632 | ||
e468bae9 TM |
633 | if (nfs_clear_request_commit(req)) |
634 | radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, | |
635 | req->wb_index, NFS_PAGE_TAG_COMMIT); | |
636 | ||
1da177e4 LT |
637 | /* Okay, the request matches. Update the region */ |
638 | if (offset < req->wb_offset) { | |
639 | req->wb_offset = offset; | |
640 | req->wb_pgbase = offset; | |
1da177e4 | 641 | } |
1da177e4 LT |
642 | if (end > rqend) |
643 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
644 | else |
645 | req->wb_bytes = rqend - req->wb_offset; | |
646 | out_unlock: | |
647 | spin_unlock(&inode->i_lock); | |
648 | return req; | |
649 | out_flushme: | |
650 | spin_unlock(&inode->i_lock); | |
651 | nfs_release_request(req); | |
652 | error = nfs_wb_page(inode, page); | |
653 | out_err: | |
654 | return ERR_PTR(error); | |
655 | } | |
656 | ||
657 | /* | |
658 | * Try to update an existing write request, or create one if there is none. | |
659 | * | |
660 | * Note: Should always be called with the Page Lock held to prevent races | |
661 | * if we have to add a new request. Also assumes that the caller has | |
662 | * already called nfs_flush_incompatible() if necessary. | |
663 | */ | |
664 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
665 | struct page *page, unsigned int offset, unsigned int bytes) | |
666 | { | |
667 | struct inode *inode = page->mapping->host; | |
668 | struct nfs_page *req; | |
669 | int error; | |
1da177e4 | 670 | |
e7d39069 TM |
671 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
672 | if (req != NULL) | |
673 | goto out; | |
674 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
675 | if (IS_ERR(req)) | |
676 | goto out; | |
677 | error = nfs_inode_add_request(inode, req); | |
678 | if (error != 0) { | |
679 | nfs_release_request(req); | |
680 | req = ERR_PTR(error); | |
681 | } | |
efc91ed0 | 682 | out: |
61e930a9 | 683 | return req; |
1da177e4 LT |
684 | } |
685 | ||
e7d39069 TM |
686 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
687 | unsigned int offset, unsigned int count) | |
688 | { | |
689 | struct nfs_page *req; | |
690 | ||
691 | req = nfs_setup_write_request(ctx, page, offset, count); | |
692 | if (IS_ERR(req)) | |
693 | return PTR_ERR(req); | |
694 | /* Update file length */ | |
695 | nfs_grow_file(page, offset, count); | |
696 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
697 | nfs_clear_page_tag_locked(req); | |
698 | return 0; | |
699 | } | |
700 | ||
1da177e4 LT |
701 | int nfs_flush_incompatible(struct file *file, struct page *page) |
702 | { | |
cd3758e3 | 703 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 704 | struct nfs_page *req; |
1a54533e | 705 | int do_flush, status; |
1da177e4 LT |
706 | /* |
707 | * Look for a request corresponding to this page. If there | |
708 | * is one, and it belongs to another file, we flush it out | |
709 | * before we try to copy anything into the page. Do this | |
710 | * due to the lack of an ACCESS-type call in NFSv2. | |
711 | * Also do the same if we find a request from an existing | |
712 | * dropped page. | |
713 | */ | |
1a54533e TM |
714 | do { |
715 | req = nfs_page_find_request(page); | |
716 | if (req == NULL) | |
717 | return 0; | |
e468bae9 | 718 | do_flush = req->wb_page != page || req->wb_context != ctx; |
1da177e4 | 719 | nfs_release_request(req); |
1a54533e TM |
720 | if (!do_flush) |
721 | return 0; | |
722 | status = nfs_wb_page(page->mapping->host, page); | |
723 | } while (status == 0); | |
724 | return status; | |
1da177e4 LT |
725 | } |
726 | ||
5d47a356 TM |
727 | /* |
728 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
729 | * the PageUptodate() flag. In this case, we will need to turn off | |
730 | * write optimisations that depend on the page contents being correct. | |
731 | */ | |
732 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
733 | { | |
734 | return PageUptodate(page) && | |
735 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
736 | } | |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Update and possibly write a cached page of an NFS file. | |
740 | * | |
741 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
742 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
743 | */ | |
744 | int nfs_updatepage(struct file *file, struct page *page, | |
745 | unsigned int offset, unsigned int count) | |
746 | { | |
cd3758e3 | 747 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 748 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
749 | int status = 0; |
750 | ||
91d5b470 CL |
751 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
752 | ||
48186c7d | 753 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
754 | file->f_path.dentry->d_parent->d_name.name, |
755 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 756 | (long long)(page_offset(page) + offset)); |
1da177e4 | 757 | |
1da177e4 | 758 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
759 | * is up to date, it may be more efficient to extend the write |
760 | * to cover the entire page in order to avoid fragmentation | |
761 | * inefficiencies. | |
1da177e4 | 762 | */ |
5d47a356 TM |
763 | if (nfs_write_pageuptodate(page, inode) && |
764 | inode->i_flock == NULL && | |
4b5621f6 | 765 | !(file->f_flags & O_SYNC)) { |
49a70f27 | 766 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 767 | offset = 0; |
1da177e4 LT |
768 | } |
769 | ||
e21195a7 | 770 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
771 | if (status < 0) |
772 | nfs_set_pageerror(page); | |
773 | else | |
774 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 775 | |
48186c7d | 776 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 777 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
778 | return status; |
779 | } | |
780 | ||
781 | static void nfs_writepage_release(struct nfs_page *req) | |
782 | { | |
1da177e4 | 783 | |
7e5f6146 | 784 | if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { |
8e821cad TM |
785 | nfs_end_page_writeback(req->wb_page); |
786 | nfs_inode_remove_request(req); | |
787 | } else | |
788 | nfs_end_page_writeback(req->wb_page); | |
9fd367f0 | 789 | nfs_clear_page_tag_locked(req); |
1da177e4 LT |
790 | } |
791 | ||
3ff7576d | 792 | static int flush_task_priority(int how) |
1da177e4 LT |
793 | { |
794 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
795 | case FLUSH_HIGHPRI: | |
796 | return RPC_PRIORITY_HIGH; | |
797 | case FLUSH_LOWPRI: | |
798 | return RPC_PRIORITY_LOW; | |
799 | } | |
800 | return RPC_PRIORITY_NORMAL; | |
801 | } | |
802 | ||
803 | /* | |
804 | * Set up the argument/result storage required for the RPC call. | |
805 | */ | |
dbae4c73 | 806 | static int nfs_write_rpcsetup(struct nfs_page *req, |
1da177e4 | 807 | struct nfs_write_data *data, |
788e7a89 | 808 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
809 | unsigned int count, unsigned int offset, |
810 | int how) | |
811 | { | |
84115e1c TM |
812 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
813 | int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
3ff7576d | 814 | int priority = flush_task_priority(how); |
07737691 | 815 | struct rpc_task *task; |
bdc7f021 TM |
816 | struct rpc_message msg = { |
817 | .rpc_argp = &data->args, | |
818 | .rpc_resp = &data->res, | |
819 | .rpc_cred = req->wb_context->cred, | |
820 | }; | |
84115e1c TM |
821 | struct rpc_task_setup task_setup_data = { |
822 | .rpc_client = NFS_CLIENT(inode), | |
07737691 | 823 | .task = &data->task, |
bdc7f021 | 824 | .rpc_message = &msg, |
84115e1c TM |
825 | .callback_ops = call_ops, |
826 | .callback_data = data, | |
101070ca | 827 | .workqueue = nfsiod_workqueue, |
84115e1c | 828 | .flags = flags, |
3ff7576d | 829 | .priority = priority, |
84115e1c | 830 | }; |
1da177e4 LT |
831 | |
832 | /* Set up the RPC argument and reply structs | |
833 | * NB: take care not to mess about with data->commit et al. */ | |
834 | ||
835 | data->req = req; | |
88be9f99 | 836 | data->inode = inode = req->wb_context->path.dentry->d_inode; |
bdc7f021 | 837 | data->cred = msg.rpc_cred; |
1da177e4 LT |
838 | |
839 | data->args.fh = NFS_FH(inode); | |
840 | data->args.offset = req_offset(req) + offset; | |
841 | data->args.pgbase = req->wb_pgbase + offset; | |
842 | data->args.pages = data->pagevec; | |
843 | data->args.count = count; | |
383ba719 | 844 | data->args.context = get_nfs_open_context(req->wb_context); |
bdc7f021 TM |
845 | data->args.stable = NFS_UNSTABLE; |
846 | if (how & FLUSH_STABLE) { | |
847 | data->args.stable = NFS_DATA_SYNC; | |
fb8a1f11 | 848 | if (!nfs_need_commit(NFS_I(inode))) |
bdc7f021 TM |
849 | data->args.stable = NFS_FILE_SYNC; |
850 | } | |
1da177e4 LT |
851 | |
852 | data->res.fattr = &data->fattr; | |
853 | data->res.count = count; | |
854 | data->res.verf = &data->verf; | |
0e574af1 | 855 | nfs_fattr_init(&data->fattr); |
1da177e4 | 856 | |
788e7a89 | 857 | /* Set up the initial task struct. */ |
bdc7f021 | 858 | NFS_PROTO(inode)->write_setup(data, &msg); |
1da177e4 | 859 | |
a3f565b1 | 860 | dprintk("NFS: %5u initiated write call " |
48186c7d | 861 | "(req %s/%lld, %u bytes @ offset %llu)\n", |
0bbacc40 | 862 | data->task.tk_pid, |
1da177e4 LT |
863 | inode->i_sb->s_id, |
864 | (long long)NFS_FILEID(inode), | |
865 | count, | |
866 | (unsigned long long)data->args.offset); | |
1da177e4 | 867 | |
07737691 | 868 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
869 | if (IS_ERR(task)) |
870 | return PTR_ERR(task); | |
871 | rpc_put_task(task); | |
872 | return 0; | |
1da177e4 LT |
873 | } |
874 | ||
6d884e8f F |
875 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
876 | * call this on each, which will prepare them to be retried on next | |
877 | * writeback using standard nfs. | |
878 | */ | |
879 | static void nfs_redirty_request(struct nfs_page *req) | |
880 | { | |
881 | nfs_mark_request_dirty(req); | |
882 | nfs_end_page_writeback(req->wb_page); | |
883 | nfs_clear_page_tag_locked(req); | |
884 | } | |
885 | ||
1da177e4 LT |
886 | /* |
887 | * Generate multiple small requests to write out a single | |
888 | * contiguous dirty area on one page. | |
889 | */ | |
8d5658c9 | 890 | static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) |
1da177e4 LT |
891 | { |
892 | struct nfs_page *req = nfs_list_entry(head->next); | |
893 | struct page *page = req->wb_page; | |
894 | struct nfs_write_data *data; | |
e9f7bee1 TM |
895 | size_t wsize = NFS_SERVER(inode)->wsize, nbytes; |
896 | unsigned int offset; | |
1da177e4 | 897 | int requests = 0; |
dbae4c73 | 898 | int ret = 0; |
1da177e4 LT |
899 | LIST_HEAD(list); |
900 | ||
901 | nfs_list_remove_request(req); | |
902 | ||
bcb71bba | 903 | nbytes = count; |
e9f7bee1 TM |
904 | do { |
905 | size_t len = min(nbytes, wsize); | |
906 | ||
8d5658c9 | 907 | data = nfs_writedata_alloc(1); |
1da177e4 LT |
908 | if (!data) |
909 | goto out_bad; | |
910 | list_add(&data->pages, &list); | |
911 | requests++; | |
e9f7bee1 TM |
912 | nbytes -= len; |
913 | } while (nbytes != 0); | |
1da177e4 LT |
914 | atomic_set(&req->wb_complete, requests); |
915 | ||
916 | ClearPageError(page); | |
1da177e4 | 917 | offset = 0; |
bcb71bba | 918 | nbytes = count; |
1da177e4 | 919 | do { |
dbae4c73 TM |
920 | int ret2; |
921 | ||
1da177e4 LT |
922 | data = list_entry(list.next, struct nfs_write_data, pages); |
923 | list_del_init(&data->pages); | |
924 | ||
925 | data->pagevec[0] = page; | |
1da177e4 | 926 | |
bcb71bba TM |
927 | if (nbytes < wsize) |
928 | wsize = nbytes; | |
dbae4c73 | 929 | ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, |
bcb71bba | 930 | wsize, offset, how); |
dbae4c73 TM |
931 | if (ret == 0) |
932 | ret = ret2; | |
bcb71bba TM |
933 | offset += wsize; |
934 | nbytes -= wsize; | |
1da177e4 LT |
935 | } while (nbytes != 0); |
936 | ||
dbae4c73 | 937 | return ret; |
1da177e4 LT |
938 | |
939 | out_bad: | |
940 | while (!list_empty(&list)) { | |
941 | data = list_entry(list.next, struct nfs_write_data, pages); | |
942 | list_del(&data->pages); | |
8aca67f0 | 943 | nfs_writedata_release(data); |
1da177e4 | 944 | } |
61822ab5 | 945 | nfs_redirty_request(req); |
1da177e4 LT |
946 | return -ENOMEM; |
947 | } | |
948 | ||
949 | /* | |
950 | * Create an RPC task for the given write request and kick it. | |
951 | * The page must have been locked by the caller. | |
952 | * | |
953 | * It may happen that the page we're passed is not marked dirty. | |
954 | * This is the case if nfs_updatepage detects a conflicting request | |
955 | * that has been written but not committed. | |
956 | */ | |
8d5658c9 | 957 | static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) |
1da177e4 LT |
958 | { |
959 | struct nfs_page *req; | |
960 | struct page **pages; | |
961 | struct nfs_write_data *data; | |
1da177e4 | 962 | |
8d5658c9 | 963 | data = nfs_writedata_alloc(npages); |
1da177e4 LT |
964 | if (!data) |
965 | goto out_bad; | |
966 | ||
967 | pages = data->pagevec; | |
1da177e4 LT |
968 | while (!list_empty(head)) { |
969 | req = nfs_list_entry(head->next); | |
970 | nfs_list_remove_request(req); | |
971 | nfs_list_add_request(req, &data->pages); | |
972 | ClearPageError(req->wb_page); | |
1da177e4 | 973 | *pages++ = req->wb_page; |
1da177e4 LT |
974 | } |
975 | req = nfs_list_entry(data->pages.next); | |
976 | ||
1da177e4 | 977 | /* Set up the argument struct */ |
dbae4c73 | 978 | return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); |
1da177e4 LT |
979 | out_bad: |
980 | while (!list_empty(head)) { | |
10afec90 | 981 | req = nfs_list_entry(head->next); |
1da177e4 | 982 | nfs_list_remove_request(req); |
61822ab5 | 983 | nfs_redirty_request(req); |
1da177e4 LT |
984 | } |
985 | return -ENOMEM; | |
986 | } | |
987 | ||
c63c7b05 TM |
988 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
989 | struct inode *inode, int ioflags) | |
1da177e4 | 990 | { |
bf4285e7 | 991 | size_t wsize = NFS_SERVER(inode)->wsize; |
1da177e4 | 992 | |
bcb71bba | 993 | if (wsize < PAGE_CACHE_SIZE) |
c63c7b05 | 994 | nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); |
bcb71bba | 995 | else |
c63c7b05 | 996 | nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); |
1da177e4 LT |
997 | } |
998 | ||
999 | /* | |
1000 | * Handle a write reply that flushed part of a page. | |
1001 | */ | |
788e7a89 | 1002 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 1003 | { |
788e7a89 | 1004 | struct nfs_write_data *data = calldata; |
1da177e4 | 1005 | |
48186c7d CL |
1006 | dprintk("NFS: %5u write(%s/%lld %d@%lld)", |
1007 | task->tk_pid, | |
1008 | data->req->wb_context->path.dentry->d_inode->i_sb->s_id, | |
1009 | (long long) | |
1010 | NFS_FILEID(data->req->wb_context->path.dentry->d_inode), | |
1011 | data->req->wb_bytes, (long long)req_offset(data->req)); | |
1da177e4 | 1012 | |
c9d8f89d TM |
1013 | nfs_writeback_done(task, data); |
1014 | } | |
788e7a89 | 1015 | |
c9d8f89d TM |
1016 | static void nfs_writeback_release_partial(void *calldata) |
1017 | { | |
1018 | struct nfs_write_data *data = calldata; | |
1019 | struct nfs_page *req = data->req; | |
1020 | struct page *page = req->wb_page; | |
1021 | int status = data->task.tk_status; | |
1022 | ||
1023 | if (status < 0) { | |
a301b777 | 1024 | nfs_set_pageerror(page); |
c9d8f89d TM |
1025 | nfs_context_set_write_error(req->wb_context, status); |
1026 | dprintk(", error = %d\n", status); | |
8e821cad | 1027 | goto out; |
1da177e4 LT |
1028 | } |
1029 | ||
8e821cad | 1030 | if (nfs_write_need_commit(data)) { |
587142f8 | 1031 | struct inode *inode = page->mapping->host; |
8e821cad | 1032 | |
587142f8 | 1033 | spin_lock(&inode->i_lock); |
8e821cad TM |
1034 | if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { |
1035 | /* Do nothing we need to resend the writes */ | |
1036 | } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { | |
1037 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1038 | dprintk(" defer commit\n"); | |
1039 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
1040 | set_bit(PG_NEED_RESCHED, &req->wb_flags); | |
1041 | clear_bit(PG_NEED_COMMIT, &req->wb_flags); | |
1042 | dprintk(" server reboot detected\n"); | |
1043 | } | |
587142f8 | 1044 | spin_unlock(&inode->i_lock); |
8e821cad TM |
1045 | } else |
1046 | dprintk(" OK\n"); | |
1047 | ||
1048 | out: | |
1da177e4 LT |
1049 | if (atomic_dec_and_test(&req->wb_complete)) |
1050 | nfs_writepage_release(req); | |
c9d8f89d | 1051 | nfs_writedata_release(calldata); |
1da177e4 LT |
1052 | } |
1053 | ||
def6ed7e AA |
1054 | #if defined(CONFIG_NFS_V4_1) |
1055 | void nfs_write_prepare(struct rpc_task *task, void *calldata) | |
1056 | { | |
1057 | struct nfs_write_data *data = calldata; | |
1058 | struct nfs_client *clp = (NFS_SERVER(data->inode))->nfs_client; | |
1059 | ||
1060 | if (nfs4_setup_sequence(clp, &data->args.seq_args, | |
1061 | &data->res.seq_res, 1, task)) | |
1062 | return; | |
1063 | rpc_call_start(task); | |
1064 | } | |
1065 | #endif /* CONFIG_NFS_V4_1 */ | |
1066 | ||
788e7a89 | 1067 | static const struct rpc_call_ops nfs_write_partial_ops = { |
def6ed7e AA |
1068 | #if defined(CONFIG_NFS_V4_1) |
1069 | .rpc_call_prepare = nfs_write_prepare, | |
1070 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1071 | .rpc_call_done = nfs_writeback_done_partial, |
c9d8f89d | 1072 | .rpc_release = nfs_writeback_release_partial, |
788e7a89 TM |
1073 | }; |
1074 | ||
1da177e4 LT |
1075 | /* |
1076 | * Handle a write reply that flushes a whole page. | |
1077 | * | |
1078 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1079 | * writebacks since the page->count is kept > 1 for as long | |
1080 | * as the page has a write request pending. | |
1081 | */ | |
788e7a89 | 1082 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1083 | { |
788e7a89 | 1084 | struct nfs_write_data *data = calldata; |
1da177e4 | 1085 | |
c9d8f89d TM |
1086 | nfs_writeback_done(task, data); |
1087 | } | |
1088 | ||
1089 | static void nfs_writeback_release_full(void *calldata) | |
1090 | { | |
1091 | struct nfs_write_data *data = calldata; | |
1092 | int status = data->task.tk_status; | |
788e7a89 | 1093 | |
1da177e4 LT |
1094 | /* Update attributes as result of writeback. */ |
1095 | while (!list_empty(&data->pages)) { | |
c9d8f89d TM |
1096 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
1097 | struct page *page = req->wb_page; | |
1098 | ||
1da177e4 | 1099 | nfs_list_remove_request(req); |
1da177e4 | 1100 | |
48186c7d CL |
1101 | dprintk("NFS: %5u write (%s/%lld %d@%lld)", |
1102 | data->task.tk_pid, | |
88be9f99 TM |
1103 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1104 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1105 | req->wb_bytes, |
1106 | (long long)req_offset(req)); | |
1107 | ||
c9d8f89d | 1108 | if (status < 0) { |
a301b777 | 1109 | nfs_set_pageerror(page); |
c9d8f89d TM |
1110 | nfs_context_set_write_error(req->wb_context, status); |
1111 | dprintk(", error = %d\n", status); | |
8e821cad | 1112 | goto remove_request; |
1da177e4 | 1113 | } |
1da177e4 | 1114 | |
8e821cad TM |
1115 | if (nfs_write_need_commit(data)) { |
1116 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1117 | nfs_mark_request_commit(req); | |
1118 | nfs_end_page_writeback(page); | |
1119 | dprintk(" marked for commit\n"); | |
1da177e4 LT |
1120 | goto next; |
1121 | } | |
8e821cad TM |
1122 | dprintk(" OK\n"); |
1123 | remove_request: | |
1124 | nfs_end_page_writeback(page); | |
1da177e4 | 1125 | nfs_inode_remove_request(req); |
1da177e4 | 1126 | next: |
9fd367f0 | 1127 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1128 | } |
c9d8f89d | 1129 | nfs_writedata_release(calldata); |
1da177e4 LT |
1130 | } |
1131 | ||
788e7a89 | 1132 | static const struct rpc_call_ops nfs_write_full_ops = { |
def6ed7e AA |
1133 | #if defined(CONFIG_NFS_V4_1) |
1134 | .rpc_call_prepare = nfs_write_prepare, | |
1135 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 | 1136 | .rpc_call_done = nfs_writeback_done_full, |
c9d8f89d | 1137 | .rpc_release = nfs_writeback_release_full, |
788e7a89 TM |
1138 | }; |
1139 | ||
1140 | ||
1da177e4 LT |
1141 | /* |
1142 | * This function is called when the WRITE call is complete. | |
1143 | */ | |
462d5b32 | 1144 | int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1145 | { |
1da177e4 LT |
1146 | struct nfs_writeargs *argp = &data->args; |
1147 | struct nfs_writeres *resp = &data->res; | |
eedc020e | 1148 | struct nfs_server *server = NFS_SERVER(data->inode); |
788e7a89 | 1149 | int status; |
1da177e4 | 1150 | |
a3f565b1 | 1151 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1152 | task->tk_pid, task->tk_status); |
1153 | ||
f551e44f CL |
1154 | /* |
1155 | * ->write_done will attempt to use post-op attributes to detect | |
1156 | * conflicting writes by other clients. A strict interpretation | |
1157 | * of close-to-open would allow us to continue caching even if | |
1158 | * another writer had changed the file, but some applications | |
1159 | * depend on tighter cache coherency when writing. | |
1160 | */ | |
788e7a89 TM |
1161 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1162 | if (status != 0) | |
1163 | return status; | |
91d5b470 CL |
1164 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1165 | ||
1da177e4 LT |
1166 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1167 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1168 | /* We tried a write call, but the server did not | |
1169 | * commit data to stable storage even though we | |
1170 | * requested it. | |
1171 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1172 | * the server reports NFS_DATA_SYNC, but performs | |
1173 | * NFS_FILE_SYNC. We therefore implement this checking | |
1174 | * as a dprintk() in order to avoid filling syslog. | |
1175 | */ | |
1176 | static unsigned long complain; | |
1177 | ||
1178 | if (time_before(complain, jiffies)) { | |
48186c7d | 1179 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1180 | " (committed = %d) != (stable = %d)\n", |
eedc020e | 1181 | server->nfs_client->cl_hostname, |
1da177e4 LT |
1182 | resp->verf->committed, argp->stable); |
1183 | complain = jiffies + 300 * HZ; | |
1184 | } | |
1185 | } | |
1186 | #endif | |
1187 | /* Is this a short write? */ | |
1188 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1189 | static unsigned long complain; | |
1190 | ||
91d5b470 CL |
1191 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1192 | ||
1da177e4 LT |
1193 | /* Has the server at least made some progress? */ |
1194 | if (resp->count != 0) { | |
1195 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1196 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1197 | /* Resend from where the server left off */ | |
1198 | argp->offset += resp->count; | |
1199 | argp->pgbase += resp->count; | |
1200 | argp->count -= resp->count; | |
1201 | } else { | |
1202 | /* Resend as a stable write in order to avoid | |
1203 | * headaches in the case of a server crash. | |
1204 | */ | |
1205 | argp->stable = NFS_FILE_SYNC; | |
1206 | } | |
eedc020e | 1207 | nfs4_restart_rpc(task, server->nfs_client); |
788e7a89 | 1208 | return -EAGAIN; |
1da177e4 LT |
1209 | } |
1210 | if (time_before(complain, jiffies)) { | |
1211 | printk(KERN_WARNING | |
1212 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1213 | argp->count); | |
1214 | complain = jiffies + 300 * HZ; | |
1215 | } | |
1216 | /* Can't do anything about it except throw an error. */ | |
1217 | task->tk_status = -EIO; | |
1218 | } | |
eedc020e | 1219 | nfs4_sequence_free_slot(server->nfs_client, &data->res.seq_res); |
788e7a89 | 1220 | return 0; |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | ||
1224 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
c9d8f89d | 1225 | void nfs_commitdata_release(void *data) |
1da177e4 | 1226 | { |
383ba719 TM |
1227 | struct nfs_write_data *wdata = data; |
1228 | ||
1229 | put_nfs_open_context(wdata->args.context); | |
1da177e4 LT |
1230 | nfs_commit_free(wdata); |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Set up the argument/result storage required for the RPC call. | |
1235 | */ | |
dbae4c73 | 1236 | static int nfs_commit_rpcsetup(struct list_head *head, |
788e7a89 TM |
1237 | struct nfs_write_data *data, |
1238 | int how) | |
1da177e4 | 1239 | { |
84115e1c TM |
1240 | struct nfs_page *first = nfs_list_entry(head->next); |
1241 | struct inode *inode = first->wb_context->path.dentry->d_inode; | |
1242 | int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
3ff7576d | 1243 | int priority = flush_task_priority(how); |
07737691 | 1244 | struct rpc_task *task; |
bdc7f021 TM |
1245 | struct rpc_message msg = { |
1246 | .rpc_argp = &data->args, | |
1247 | .rpc_resp = &data->res, | |
1248 | .rpc_cred = first->wb_context->cred, | |
1249 | }; | |
84115e1c | 1250 | struct rpc_task_setup task_setup_data = { |
07737691 | 1251 | .task = &data->task, |
84115e1c | 1252 | .rpc_client = NFS_CLIENT(inode), |
bdc7f021 | 1253 | .rpc_message = &msg, |
84115e1c TM |
1254 | .callback_ops = &nfs_commit_ops, |
1255 | .callback_data = data, | |
101070ca | 1256 | .workqueue = nfsiod_workqueue, |
84115e1c | 1257 | .flags = flags, |
3ff7576d | 1258 | .priority = priority, |
84115e1c | 1259 | }; |
1da177e4 LT |
1260 | |
1261 | /* Set up the RPC argument and reply structs | |
1262 | * NB: take care not to mess about with data->commit et al. */ | |
1263 | ||
1264 | list_splice_init(head, &data->pages); | |
1da177e4 | 1265 | |
1da177e4 | 1266 | data->inode = inode; |
bdc7f021 | 1267 | data->cred = msg.rpc_cred; |
1da177e4 LT |
1268 | |
1269 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1270 | /* Note: we always request a commit of the entire inode */ |
1271 | data->args.offset = 0; | |
1272 | data->args.count = 0; | |
383ba719 | 1273 | data->args.context = get_nfs_open_context(first->wb_context); |
3da28eb1 | 1274 | data->res.count = 0; |
1da177e4 LT |
1275 | data->res.fattr = &data->fattr; |
1276 | data->res.verf = &data->verf; | |
0e574af1 | 1277 | nfs_fattr_init(&data->fattr); |
788e7a89 TM |
1278 | |
1279 | /* Set up the initial task struct. */ | |
bdc7f021 | 1280 | NFS_PROTO(inode)->commit_setup(data, &msg); |
1da177e4 | 1281 | |
a3f565b1 | 1282 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); |
bdc7f021 | 1283 | |
07737691 | 1284 | task = rpc_run_task(&task_setup_data); |
dbae4c73 TM |
1285 | if (IS_ERR(task)) |
1286 | return PTR_ERR(task); | |
1287 | rpc_put_task(task); | |
1288 | return 0; | |
1da177e4 LT |
1289 | } |
1290 | ||
1291 | /* | |
1292 | * Commit dirty pages | |
1293 | */ | |
1294 | static int | |
40859d7e | 1295 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1296 | { |
1297 | struct nfs_write_data *data; | |
1298 | struct nfs_page *req; | |
1299 | ||
c9d8f89d | 1300 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1301 | |
1302 | if (!data) | |
1303 | goto out_bad; | |
1304 | ||
1305 | /* Set up the argument struct */ | |
dbae4c73 | 1306 | return nfs_commit_rpcsetup(head, data, how); |
1da177e4 LT |
1307 | out_bad: |
1308 | while (!list_empty(head)) { | |
1309 | req = nfs_list_entry(head->next); | |
1310 | nfs_list_remove_request(req); | |
1311 | nfs_mark_request_commit(req); | |
83715ad5 | 1312 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 PZ |
1313 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, |
1314 | BDI_RECLAIMABLE); | |
9fd367f0 | 1315 | nfs_clear_page_tag_locked(req); |
1da177e4 LT |
1316 | } |
1317 | return -ENOMEM; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * COMMIT call returned | |
1322 | */ | |
788e7a89 | 1323 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1324 | { |
963d8fe5 | 1325 | struct nfs_write_data *data = calldata; |
1da177e4 | 1326 | |
a3f565b1 | 1327 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1328 | task->tk_pid, task->tk_status); |
1329 | ||
788e7a89 TM |
1330 | /* Call the NFS version-specific code */ |
1331 | if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) | |
1332 | return; | |
c9d8f89d TM |
1333 | } |
1334 | ||
1335 | static void nfs_commit_release(void *calldata) | |
1336 | { | |
1337 | struct nfs_write_data *data = calldata; | |
1338 | struct nfs_page *req; | |
1339 | int status = data->task.tk_status; | |
788e7a89 | 1340 | |
1da177e4 LT |
1341 | while (!list_empty(&data->pages)) { |
1342 | req = nfs_list_entry(data->pages.next); | |
1343 | nfs_list_remove_request(req); | |
e468bae9 | 1344 | nfs_clear_request_commit(req); |
1da177e4 | 1345 | |
48186c7d | 1346 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
88be9f99 TM |
1347 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
1348 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
1349 | req->wb_bytes, |
1350 | (long long)req_offset(req)); | |
c9d8f89d TM |
1351 | if (status < 0) { |
1352 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1353 | nfs_inode_remove_request(req); |
c9d8f89d | 1354 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1355 | goto next; |
1356 | } | |
1357 | ||
1358 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1359 | * returned by the server against all stored verfs. */ | |
1360 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1361 | /* We have a match */ | |
1362 | nfs_inode_remove_request(req); | |
1363 | dprintk(" OK\n"); | |
1364 | goto next; | |
1365 | } | |
1366 | /* We have a mismatch. Write the page again */ | |
1367 | dprintk(" mismatch\n"); | |
6d884e8f | 1368 | nfs_mark_request_dirty(req); |
1da177e4 | 1369 | next: |
9fd367f0 | 1370 | nfs_clear_page_tag_locked(req); |
1da177e4 | 1371 | } |
c9d8f89d | 1372 | nfs_commitdata_release(calldata); |
1da177e4 | 1373 | } |
788e7a89 TM |
1374 | |
1375 | static const struct rpc_call_ops nfs_commit_ops = { | |
21d9a851 AA |
1376 | #if defined(CONFIG_NFS_V4_1) |
1377 | .rpc_call_prepare = nfs_write_prepare, | |
1378 | #endif /* CONFIG_NFS_V4_1 */ | |
788e7a89 TM |
1379 | .rpc_call_done = nfs_commit_done, |
1380 | .rpc_release = nfs_commit_release, | |
1381 | }; | |
1da177e4 | 1382 | |
3da28eb1 | 1383 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1384 | { |
1da177e4 | 1385 | LIST_HEAD(head); |
7d46a49f | 1386 | int res; |
1da177e4 | 1387 | |
587142f8 | 1388 | spin_lock(&inode->i_lock); |
3da28eb1 | 1389 | res = nfs_scan_commit(inode, &head, 0, 0); |
587142f8 | 1390 | spin_unlock(&inode->i_lock); |
1da177e4 | 1391 | if (res) { |
7d46a49f | 1392 | int error = nfs_commit_list(inode, &head, how); |
3da28eb1 TM |
1393 | if (error < 0) |
1394 | return error; | |
1395 | } | |
1da177e4 LT |
1396 | return res; |
1397 | } | |
c63c7b05 TM |
1398 | #else |
1399 | static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) | |
1400 | { | |
1401 | return 0; | |
1402 | } | |
1da177e4 LT |
1403 | #endif |
1404 | ||
1c75950b | 1405 | long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) |
1da177e4 | 1406 | { |
1c75950b | 1407 | struct inode *inode = mapping->host; |
ca52fec1 | 1408 | pgoff_t idx_start, idx_end; |
1c75950b | 1409 | unsigned int npages = 0; |
c42de9dd | 1410 | LIST_HEAD(head); |
70b9ecbd | 1411 | int nocommit = how & FLUSH_NOCOMMIT; |
3f442547 | 1412 | long pages, ret; |
1da177e4 | 1413 | |
1c75950b TM |
1414 | /* FIXME */ |
1415 | if (wbc->range_cyclic) | |
1416 | idx_start = 0; | |
1417 | else { | |
1418 | idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
1419 | idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
1420 | if (idx_end > idx_start) { | |
ca52fec1 | 1421 | pgoff_t l_npages = 1 + idx_end - idx_start; |
1c75950b TM |
1422 | npages = l_npages; |
1423 | if (sizeof(npages) != sizeof(l_npages) && | |
ca52fec1 | 1424 | (pgoff_t)npages != l_npages) |
1c75950b TM |
1425 | npages = 0; |
1426 | } | |
1427 | } | |
c42de9dd | 1428 | how &= ~FLUSH_NOCOMMIT; |
587142f8 | 1429 | spin_lock(&inode->i_lock); |
1da177e4 | 1430 | do { |
c42de9dd TM |
1431 | ret = nfs_wait_on_requests_locked(inode, idx_start, npages); |
1432 | if (ret != 0) | |
70b9ecbd | 1433 | continue; |
c42de9dd TM |
1434 | if (nocommit) |
1435 | break; | |
d2ccddf0 | 1436 | pages = nfs_scan_commit(inode, &head, idx_start, npages); |
724c439c | 1437 | if (pages == 0) |
c42de9dd | 1438 | break; |
d2ccddf0 | 1439 | if (how & FLUSH_INVALIDATE) { |
587142f8 | 1440 | spin_unlock(&inode->i_lock); |
83715ad5 | 1441 | nfs_cancel_commit_list(&head); |
e8e058e8 | 1442 | ret = pages; |
587142f8 | 1443 | spin_lock(&inode->i_lock); |
d2ccddf0 TM |
1444 | continue; |
1445 | } | |
1446 | pages += nfs_scan_commit(inode, &head, 0, 0); | |
587142f8 | 1447 | spin_unlock(&inode->i_lock); |
c42de9dd | 1448 | ret = nfs_commit_list(inode, &head, how); |
587142f8 TM |
1449 | spin_lock(&inode->i_lock); |
1450 | ||
c42de9dd | 1451 | } while (ret >= 0); |
587142f8 | 1452 | spin_unlock(&inode->i_lock); |
c42de9dd | 1453 | return ret; |
1da177e4 LT |
1454 | } |
1455 | ||
34901f70 | 1456 | static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) |
1c75950b | 1457 | { |
1c75950b TM |
1458 | int ret; |
1459 | ||
34901f70 | 1460 | ret = nfs_writepages(mapping, wbc); |
61822ab5 TM |
1461 | if (ret < 0) |
1462 | goto out; | |
34901f70 | 1463 | ret = nfs_sync_mapping_wait(mapping, wbc, how); |
ed90ef51 TM |
1464 | if (ret < 0) |
1465 | goto out; | |
1466 | return 0; | |
61822ab5 | 1467 | out: |
e507d9eb | 1468 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
1c75950b TM |
1469 | return ret; |
1470 | } | |
1471 | ||
34901f70 TM |
1472 | /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ |
1473 | static int nfs_write_mapping(struct address_space *mapping, int how) | |
1474 | { | |
1475 | struct writeback_control wbc = { | |
1476 | .bdi = mapping->backing_dev_info, | |
72cb77f4 | 1477 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1478 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1479 | .range_start = 0, |
1480 | .range_end = LLONG_MAX, | |
34901f70 | 1481 | .for_writepages = 1, |
34901f70 | 1482 | }; |
34901f70 | 1483 | |
34901f70 TM |
1484 | return __nfs_write_mapping(mapping, &wbc, how); |
1485 | } | |
1486 | ||
ed90ef51 TM |
1487 | /* |
1488 | * flush the inode to disk. | |
1489 | */ | |
1490 | int nfs_wb_all(struct inode *inode) | |
1c75950b | 1491 | { |
ed90ef51 TM |
1492 | return nfs_write_mapping(inode->i_mapping, 0); |
1493 | } | |
1c75950b | 1494 | |
ed90ef51 TM |
1495 | int nfs_wb_nocommit(struct inode *inode) |
1496 | { | |
1497 | return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); | |
1c75950b TM |
1498 | } |
1499 | ||
1b3b4a1a TM |
1500 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1501 | { | |
1502 | struct nfs_page *req; | |
1503 | loff_t range_start = page_offset(page); | |
1504 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
1505 | struct writeback_control wbc = { | |
1506 | .bdi = page->mapping->backing_dev_info, | |
1507 | .sync_mode = WB_SYNC_ALL, | |
1508 | .nr_to_write = LONG_MAX, | |
1509 | .range_start = range_start, | |
1510 | .range_end = range_end, | |
1511 | }; | |
1512 | int ret = 0; | |
1513 | ||
1514 | BUG_ON(!PageLocked(page)); | |
1515 | for (;;) { | |
1516 | req = nfs_page_find_request(page); | |
1517 | if (req == NULL) | |
1518 | goto out; | |
e468bae9 | 1519 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
1b3b4a1a TM |
1520 | nfs_release_request(req); |
1521 | break; | |
1522 | } | |
1523 | if (nfs_lock_request_dontget(req)) { | |
1524 | nfs_inode_remove_request(req); | |
1525 | /* | |
1526 | * In case nfs_inode_remove_request has marked the | |
1527 | * page as being dirty | |
1528 | */ | |
1529 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1530 | nfs_unlock_request(req); | |
1531 | break; | |
1532 | } | |
1533 | ret = nfs_wait_on_request(req); | |
1534 | if (ret < 0) | |
1535 | goto out; | |
1536 | } | |
1537 | if (!PagePrivate(page)) | |
1538 | return 0; | |
1539 | ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); | |
1540 | out: | |
1541 | return ret; | |
1542 | } | |
1543 | ||
5334eb13 AB |
1544 | static int nfs_wb_page_priority(struct inode *inode, struct page *page, |
1545 | int how) | |
1c75950b TM |
1546 | { |
1547 | loff_t range_start = page_offset(page); | |
1548 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf TM |
1549 | struct writeback_control wbc = { |
1550 | .bdi = page->mapping->backing_dev_info, | |
1551 | .sync_mode = WB_SYNC_ALL, | |
1552 | .nr_to_write = LONG_MAX, | |
1553 | .range_start = range_start, | |
1554 | .range_end = range_end, | |
1555 | }; | |
1556 | int ret; | |
1c75950b | 1557 | |
73e3302f TM |
1558 | do { |
1559 | if (clear_page_dirty_for_io(page)) { | |
1560 | ret = nfs_writepage_locked(page, &wbc); | |
1561 | if (ret < 0) | |
1562 | goto out_error; | |
1563 | } else if (!PagePrivate(page)) | |
1564 | break; | |
1565 | ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); | |
4d770ccf | 1566 | if (ret < 0) |
73e3302f TM |
1567 | goto out_error; |
1568 | } while (PagePrivate(page)); | |
1569 | return 0; | |
1570 | out_error: | |
e507d9eb | 1571 | __mark_inode_dirty(inode, I_DIRTY_PAGES); |
4d770ccf | 1572 | return ret; |
1c75950b TM |
1573 | } |
1574 | ||
1575 | /* | |
1576 | * Write back all requests on one page - we do this before reading it. | |
1577 | */ | |
1578 | int nfs_wb_page(struct inode *inode, struct page* page) | |
1579 | { | |
4d770ccf | 1580 | return nfs_wb_page_priority(inode, page, FLUSH_STABLE); |
1c75950b TM |
1581 | } |
1582 | ||
f7b422b1 | 1583 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1584 | { |
1585 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1586 | sizeof(struct nfs_write_data), | |
1587 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1588 | NULL); |
1da177e4 LT |
1589 | if (nfs_wdata_cachep == NULL) |
1590 | return -ENOMEM; | |
1591 | ||
93d2341c MD |
1592 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1593 | nfs_wdata_cachep); | |
1da177e4 LT |
1594 | if (nfs_wdata_mempool == NULL) |
1595 | return -ENOMEM; | |
1596 | ||
93d2341c MD |
1597 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1598 | nfs_wdata_cachep); | |
1da177e4 LT |
1599 | if (nfs_commit_mempool == NULL) |
1600 | return -ENOMEM; | |
1601 | ||
89a09141 PZ |
1602 | /* |
1603 | * NFS congestion size, scale with available memory. | |
1604 | * | |
1605 | * 64MB: 8192k | |
1606 | * 128MB: 11585k | |
1607 | * 256MB: 16384k | |
1608 | * 512MB: 23170k | |
1609 | * 1GB: 32768k | |
1610 | * 2GB: 46340k | |
1611 | * 4GB: 65536k | |
1612 | * 8GB: 92681k | |
1613 | * 16GB: 131072k | |
1614 | * | |
1615 | * This allows larger machines to have larger/more transfers. | |
1616 | * Limit the default to 256M | |
1617 | */ | |
1618 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1619 | if (nfs_congestion_kb > 256*1024) | |
1620 | nfs_congestion_kb = 256*1024; | |
1621 | ||
1da177e4 LT |
1622 | return 0; |
1623 | } | |
1624 | ||
266bee88 | 1625 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1626 | { |
1627 | mempool_destroy(nfs_commit_mempool); | |
1628 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1629 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1630 | } |
1631 |