]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
3fcfab16 | 24 | |
1da177e4 | 25 | #include <asm/uaccess.h> |
1da177e4 LT |
26 | |
27 | #include "delegation.h" | |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
def6ed7e | 30 | #include "nfs4_fs.h" |
074cc1de | 31 | #include "fscache.h" |
94ad1c80 | 32 | #include "pnfs.h" |
1da177e4 LT |
33 | |
34 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
35 | ||
36 | #define MIN_POOL_WRITE (32) | |
37 | #define MIN_POOL_COMMIT (4) | |
38 | ||
39 | /* | |
40 | * Local function declarations | |
41 | */ | |
f8512ad0 | 42 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 43 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 44 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 45 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 46 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
1da177e4 | 47 | |
e18b890b | 48 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 49 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 50 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
51 | static mempool_t *nfs_commit_mempool; |
52 | ||
0b7c0153 | 53 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 54 | { |
0b7c0153 | 55 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 56 | |
1da177e4 LT |
57 | if (p) { |
58 | memset(p, 0, sizeof(*p)); | |
59 | INIT_LIST_HEAD(&p->pages); | |
60 | } | |
61 | return p; | |
62 | } | |
e0c2b380 | 63 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 64 | |
0b7c0153 | 65 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
66 | { |
67 | mempool_free(p, nfs_commit_mempool); | |
68 | } | |
e0c2b380 | 69 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 70 | |
6c75dc0d | 71 | struct nfs_write_header *nfs_writehdr_alloc(void) |
3feb2d49 | 72 | { |
cd841605 | 73 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
74 | |
75 | if (p) { | |
cd841605 | 76 | struct nfs_pgio_header *hdr = &p->header; |
cd841605 | 77 | |
3feb2d49 | 78 | memset(p, 0, sizeof(*p)); |
cd841605 | 79 | INIT_LIST_HEAD(&hdr->pages); |
6c75dc0d FI |
80 | INIT_LIST_HEAD(&hdr->rpc_list); |
81 | spin_lock_init(&hdr->lock); | |
82 | atomic_set(&hdr->refcnt, 0); | |
3feb2d49 TM |
83 | } |
84 | return p; | |
85 | } | |
86 | ||
1763da12 FI |
87 | static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
88 | unsigned int pagecount) | |
6c75dc0d FI |
89 | { |
90 | struct nfs_write_data *data, *prealloc; | |
91 | ||
92 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
93 | if (prealloc->header == NULL) | |
94 | data = prealloc; | |
95 | else | |
96 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
97 | if (!data) | |
98 | goto out; | |
99 | ||
100 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
101 | data->header = hdr; | |
102 | atomic_inc(&hdr->refcnt); | |
103 | } else { | |
104 | if (data != prealloc) | |
105 | kfree(data); | |
106 | data = NULL; | |
107 | } | |
108 | out: | |
109 | return data; | |
110 | } | |
111 | ||
cd841605 | 112 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 113 | { |
cd841605 FI |
114 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
115 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 TM |
116 | } |
117 | ||
dce81290 | 118 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 119 | { |
6c75dc0d FI |
120 | struct nfs_pgio_header *hdr = wdata->header; |
121 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
122 | ||
383ba719 | 123 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
124 | if (wdata->pages.pagevec != wdata->pages.page_array) |
125 | kfree(wdata->pages.pagevec); | |
6c75dc0d FI |
126 | if (wdata != &write_header->rpc_data) |
127 | kfree(wdata); | |
128 | else | |
129 | wdata->header = NULL; | |
130 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 131 | hdr->completion_ops->completion(hdr); |
1da177e4 LT |
132 | } |
133 | ||
7b159fc1 TM |
134 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
135 | { | |
136 | ctx->error = error; | |
137 | smp_wmb(); | |
138 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
139 | } | |
140 | ||
277459d2 TM |
141 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
142 | { | |
143 | struct nfs_page *req = NULL; | |
144 | ||
145 | if (PagePrivate(page)) { | |
146 | req = (struct nfs_page *)page_private(page); | |
147 | if (req != NULL) | |
c03b4024 | 148 | kref_get(&req->wb_kref); |
277459d2 TM |
149 | } |
150 | return req; | |
151 | } | |
152 | ||
153 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
154 | { | |
587142f8 | 155 | struct inode *inode = page->mapping->host; |
277459d2 | 156 | struct nfs_page *req = NULL; |
277459d2 | 157 | |
587142f8 | 158 | spin_lock(&inode->i_lock); |
277459d2 | 159 | req = nfs_page_find_request_locked(page); |
587142f8 | 160 | spin_unlock(&inode->i_lock); |
277459d2 TM |
161 | return req; |
162 | } | |
163 | ||
1da177e4 LT |
164 | /* Adjust the file length if we're writing beyond the end */ |
165 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
166 | { | |
167 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
168 | loff_t end, i_size; |
169 | pgoff_t end_index; | |
1da177e4 | 170 | |
a3d01454 TM |
171 | spin_lock(&inode->i_lock); |
172 | i_size = i_size_read(inode); | |
173 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 174 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 175 | goto out; |
1da177e4 LT |
176 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
177 | if (i_size >= end) | |
a3d01454 | 178 | goto out; |
1da177e4 | 179 | i_size_write(inode, end); |
a3d01454 TM |
180 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
181 | out: | |
182 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
183 | } |
184 | ||
a301b777 TM |
185 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
186 | static void nfs_set_pageerror(struct page *page) | |
187 | { | |
188 | SetPageError(page); | |
189 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
190 | } | |
191 | ||
1da177e4 LT |
192 | /* We can set the PG_uptodate flag if we see that a write request |
193 | * covers the full page. | |
194 | */ | |
195 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
196 | { | |
1da177e4 LT |
197 | if (PageUptodate(page)) |
198 | return; | |
199 | if (base != 0) | |
200 | return; | |
49a70f27 | 201 | if (count != nfs_page_length(page)) |
1da177e4 | 202 | return; |
49a70f27 | 203 | SetPageUptodate(page); |
1da177e4 LT |
204 | } |
205 | ||
1da177e4 LT |
206 | static int wb_priority(struct writeback_control *wbc) |
207 | { | |
208 | if (wbc->for_reclaim) | |
c63c7b05 | 209 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 210 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
211 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
212 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
213 | } |
214 | ||
89a09141 PZ |
215 | /* |
216 | * NFS congestion control | |
217 | */ | |
218 | ||
219 | int nfs_congestion_kb; | |
220 | ||
221 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
222 | #define NFS_CONGESTION_OFF_THRESH \ | |
223 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
224 | ||
5a6d41b3 | 225 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 226 | { |
5a6d41b3 TM |
227 | int ret = test_set_page_writeback(page); |
228 | ||
229 | if (!ret) { | |
89a09141 PZ |
230 | struct inode *inode = page->mapping->host; |
231 | struct nfs_server *nfss = NFS_SERVER(inode); | |
232 | ||
277866a0 | 233 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
234 | NFS_CONGESTION_ON_THRESH) { |
235 | set_bdi_congested(&nfss->backing_dev_info, | |
236 | BLK_RW_ASYNC); | |
237 | } | |
89a09141 | 238 | } |
5a6d41b3 | 239 | return ret; |
89a09141 PZ |
240 | } |
241 | ||
242 | static void nfs_end_page_writeback(struct page *page) | |
243 | { | |
244 | struct inode *inode = page->mapping->host; | |
245 | struct nfs_server *nfss = NFS_SERVER(inode); | |
246 | ||
247 | end_page_writeback(page); | |
c4dc4bee | 248 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 249 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
250 | } |
251 | ||
cfb506e1 | 252 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 253 | { |
587142f8 | 254 | struct inode *inode = page->mapping->host; |
e261f51f | 255 | struct nfs_page *req; |
e261f51f TM |
256 | int ret; |
257 | ||
587142f8 | 258 | spin_lock(&inode->i_lock); |
074cc1de | 259 | for (;;) { |
e261f51f | 260 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
261 | if (req == NULL) |
262 | break; | |
7ad84aa9 | 263 | if (nfs_lock_request(req)) |
e261f51f TM |
264 | break; |
265 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 266 | * then the call to nfs_lock_request() will always |
e261f51f TM |
267 | * succeed provided that someone hasn't already marked the |
268 | * request as dirty (in which case we don't care). | |
269 | */ | |
587142f8 | 270 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
271 | if (!nonblock) |
272 | ret = nfs_wait_on_request(req); | |
273 | else | |
274 | ret = -EAGAIN; | |
e261f51f TM |
275 | nfs_release_request(req); |
276 | if (ret != 0) | |
074cc1de | 277 | return ERR_PTR(ret); |
587142f8 | 278 | spin_lock(&inode->i_lock); |
e261f51f | 279 | } |
587142f8 | 280 | spin_unlock(&inode->i_lock); |
074cc1de TM |
281 | return req; |
282 | } | |
283 | ||
284 | /* | |
285 | * Find an associated nfs write request, and prepare to flush it out | |
286 | * May return an error if the user signalled nfs_wait_on_request(). | |
287 | */ | |
288 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 289 | struct page *page, bool nonblock) |
074cc1de TM |
290 | { |
291 | struct nfs_page *req; | |
292 | int ret = 0; | |
293 | ||
cfb506e1 | 294 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
295 | if (!req) |
296 | goto out; | |
297 | ret = PTR_ERR(req); | |
298 | if (IS_ERR(req)) | |
299 | goto out; | |
300 | ||
301 | ret = nfs_set_page_writeback(page); | |
302 | BUG_ON(ret != 0); | |
303 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
304 | ||
f8512ad0 FI |
305 | if (!nfs_pageio_add_request(pgio, req)) { |
306 | nfs_redirty_request(req); | |
074cc1de | 307 | ret = pgio->pg_error; |
f8512ad0 | 308 | } |
074cc1de TM |
309 | out: |
310 | return ret; | |
e261f51f TM |
311 | } |
312 | ||
f758c885 | 313 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 314 | { |
1da177e4 | 315 | struct inode *inode = page->mapping->host; |
cfb506e1 | 316 | int ret; |
1da177e4 | 317 | |
91d5b470 CL |
318 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
319 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
320 | ||
7fe7f848 | 321 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 322 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
323 | if (ret == -EAGAIN) { |
324 | redirty_page_for_writepage(wbc, page); | |
325 | ret = 0; | |
326 | } | |
327 | return ret; | |
f758c885 | 328 | } |
7fe7f848 | 329 | |
f758c885 TM |
330 | /* |
331 | * Write an mmapped page to the server. | |
332 | */ | |
333 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
334 | { | |
335 | struct nfs_pageio_descriptor pgio; | |
336 | int err; | |
49a70f27 | 337 | |
061ae2ed FI |
338 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
339 | &nfs_async_write_completion_ops); | |
f758c885 TM |
340 | err = nfs_do_writepage(page, wbc, &pgio); |
341 | nfs_pageio_complete(&pgio); | |
342 | if (err < 0) | |
343 | return err; | |
344 | if (pgio.pg_error < 0) | |
345 | return pgio.pg_error; | |
346 | return 0; | |
4d770ccf TM |
347 | } |
348 | ||
349 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
350 | { | |
f758c885 | 351 | int ret; |
4d770ccf | 352 | |
f758c885 | 353 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 354 | unlock_page(page); |
f758c885 TM |
355 | return ret; |
356 | } | |
357 | ||
358 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
359 | { | |
360 | int ret; | |
361 | ||
362 | ret = nfs_do_writepage(page, wbc, data); | |
363 | unlock_page(page); | |
364 | return ret; | |
1da177e4 LT |
365 | } |
366 | ||
1da177e4 LT |
367 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
368 | { | |
1da177e4 | 369 | struct inode *inode = mapping->host; |
72cb77f4 | 370 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 371 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
372 | int err; |
373 | ||
72cb77f4 TM |
374 | /* Stop dirtying of new pages while we sync */ |
375 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
376 | nfs_wait_bit_killable, TASK_KILLABLE); | |
377 | if (err) | |
378 | goto out_err; | |
379 | ||
91d5b470 CL |
380 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
381 | ||
061ae2ed FI |
382 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), |
383 | &nfs_async_write_completion_ops); | |
f758c885 | 384 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 385 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
386 | |
387 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
388 | smp_mb__after_clear_bit(); | |
389 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
390 | ||
f758c885 | 391 | if (err < 0) |
72cb77f4 TM |
392 | goto out_err; |
393 | err = pgio.pg_error; | |
394 | if (err < 0) | |
395 | goto out_err; | |
c63c7b05 | 396 | return 0; |
72cb77f4 TM |
397 | out_err: |
398 | return err; | |
1da177e4 LT |
399 | } |
400 | ||
401 | /* | |
402 | * Insert a write request into an inode | |
403 | */ | |
d6d6dc7c | 404 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
405 | { |
406 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
407 | |
408 | /* Lock the request! */ | |
7ad84aa9 | 409 | nfs_lock_request(req); |
e7d39069 TM |
410 | |
411 | spin_lock(&inode->i_lock); | |
4d65c520 | 412 | if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 413 | inode->i_version++; |
2df485a7 | 414 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 415 | SetPagePrivate(req->wb_page); |
277459d2 | 416 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 417 | nfsi->npages++; |
c03b4024 | 418 | kref_get(&req->wb_kref); |
e7d39069 | 419 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
420 | } |
421 | ||
422 | /* | |
89a09141 | 423 | * Remove a write request from an inode |
1da177e4 LT |
424 | */ |
425 | static void nfs_inode_remove_request(struct nfs_page *req) | |
426 | { | |
3d4ff43d | 427 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
428 | struct nfs_inode *nfsi = NFS_I(inode); |
429 | ||
430 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
431 | ||
587142f8 | 432 | spin_lock(&inode->i_lock); |
277459d2 | 433 | set_page_private(req->wb_page, 0); |
deb7d638 | 434 | ClearPagePrivate(req->wb_page); |
2df485a7 | 435 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 | 436 | nfsi->npages--; |
4d65c520 | 437 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
438 | nfs_release_request(req); |
439 | } | |
440 | ||
61822ab5 | 441 | static void |
6d884e8f | 442 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 443 | { |
61822ab5 TM |
444 | __set_page_dirty_nobuffers(req->wb_page); |
445 | } | |
446 | ||
1da177e4 | 447 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
8dd37758 TM |
448 | /** |
449 | * nfs_request_add_commit_list - add request to a commit list | |
450 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
451 | * @dst: commit list head |
452 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 453 | * |
ea2cf228 | 454 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
455 | * number of outstanding requests requiring a commit as well as |
456 | * the MM page stats. | |
457 | * | |
ea2cf228 | 458 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 459 | * holding the nfs_page lock. |
1da177e4 | 460 | */ |
8dd37758 | 461 | void |
ea2cf228 FI |
462 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
463 | struct nfs_commit_info *cinfo) | |
1da177e4 | 464 | { |
e468bae9 | 465 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
466 | spin_lock(cinfo->lock); |
467 | nfs_list_add_request(req, dst); | |
468 | cinfo->mds->ncommit++; | |
469 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
470 | if (!cinfo->dreq) { |
471 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
472 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
473 | BDI_RECLAIMABLE); | |
474 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
475 | I_DIRTY_DATASYNC); | |
476 | } | |
1da177e4 | 477 | } |
8dd37758 TM |
478 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
479 | ||
480 | /** | |
481 | * nfs_request_remove_commit_list - Remove request from a commit list | |
482 | * @req: pointer to a nfs_page | |
ea2cf228 | 483 | * @cinfo: holds list lock and accounting info |
8dd37758 | 484 | * |
ea2cf228 | 485 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
486 | * number of outstanding requests requiring a commit |
487 | * It does not update the MM page stats. | |
488 | * | |
ea2cf228 | 489 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
490 | */ |
491 | void | |
ea2cf228 FI |
492 | nfs_request_remove_commit_list(struct nfs_page *req, |
493 | struct nfs_commit_info *cinfo) | |
8dd37758 | 494 | { |
8dd37758 TM |
495 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
496 | return; | |
497 | nfs_list_remove_request(req); | |
ea2cf228 | 498 | cinfo->mds->ncommit--; |
8dd37758 TM |
499 | } |
500 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
501 | ||
ea2cf228 FI |
502 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
503 | struct inode *inode) | |
504 | { | |
505 | cinfo->lock = &inode->i_lock; | |
506 | cinfo->mds = &NFS_I(inode)->commit_info; | |
507 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 508 | cinfo->dreq = NULL; |
f453a54a | 509 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
510 | } |
511 | ||
512 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
513 | struct inode *inode, | |
514 | struct nfs_direct_req *dreq) | |
515 | { | |
1763da12 FI |
516 | if (dreq) |
517 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
518 | else | |
519 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
520 | } |
521 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
522 | |
523 | /* | |
524 | * Add a request to the inode's commit list. | |
525 | */ | |
1763da12 | 526 | void |
ea2cf228 FI |
527 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
528 | struct nfs_commit_info *cinfo) | |
8dd37758 | 529 | { |
ea2cf228 | 530 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 531 | return; |
ea2cf228 | 532 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 533 | } |
8e821cad | 534 | |
d6d6dc7c FI |
535 | static void |
536 | nfs_clear_page_commit(struct page *page) | |
537 | { | |
538 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
539 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
540 | } | |
541 | ||
8dd37758 | 542 | static void |
e468bae9 TM |
543 | nfs_clear_request_commit(struct nfs_page *req) |
544 | { | |
8dd37758 TM |
545 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
546 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 547 | struct nfs_commit_info cinfo; |
e468bae9 | 548 | |
ea2cf228 FI |
549 | nfs_init_cinfo_from_inode(&cinfo, inode); |
550 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
551 | spin_lock(cinfo.lock); | |
552 | nfs_request_remove_commit_list(req, &cinfo); | |
553 | spin_unlock(cinfo.lock); | |
8dd37758 | 554 | } |
d6d6dc7c | 555 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 556 | } |
e468bae9 TM |
557 | } |
558 | ||
8e821cad TM |
559 | static inline |
560 | int nfs_write_need_commit(struct nfs_write_data *data) | |
561 | { | |
465d5243 | 562 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
563 | return data->header->lseg == NULL; |
564 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
565 | } |
566 | ||
8e821cad | 567 | #else |
68cd6fa4 BS |
568 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
569 | struct inode *inode) | |
570 | { | |
571 | } | |
572 | ||
573 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
574 | struct inode *inode, | |
575 | struct nfs_direct_req *dreq) | |
576 | { | |
577 | } | |
578 | ||
1763da12 | 579 | void |
ea2cf228 FI |
580 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
581 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
582 | { |
583 | } | |
584 | ||
8dd37758 | 585 | static void |
e468bae9 TM |
586 | nfs_clear_request_commit(struct nfs_page *req) |
587 | { | |
e468bae9 TM |
588 | } |
589 | ||
8e821cad TM |
590 | static inline |
591 | int nfs_write_need_commit(struct nfs_write_data *data) | |
592 | { | |
593 | return 0; | |
594 | } | |
595 | ||
6c75dc0d FI |
596 | #endif |
597 | ||
061ae2ed | 598 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 599 | { |
ea2cf228 | 600 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
601 | unsigned long bytes = 0; |
602 | ||
603 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
604 | goto out; | |
ea2cf228 | 605 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
606 | while (!list_empty(&hdr->pages)) { |
607 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
608 | |
609 | bytes += req->wb_bytes; | |
610 | nfs_list_remove_request(req); | |
611 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
612 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 613 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
614 | nfs_context_set_write_error(req->wb_context, hdr->error); |
615 | goto remove_req; | |
616 | } | |
617 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
618 | nfs_mark_request_dirty(req); | |
619 | goto next; | |
620 | } | |
621 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
ea2cf228 | 622 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
623 | goto next; |
624 | } | |
625 | remove_req: | |
626 | nfs_inode_remove_request(req); | |
627 | next: | |
3aff4ebb | 628 | nfs_unlock_request_dont_release(req); |
d1182b33 | 629 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 630 | nfs_release_request(req); |
6c75dc0d FI |
631 | } |
632 | out: | |
633 | hdr->release(hdr); | |
8e821cad | 634 | } |
1da177e4 | 635 | |
47c62564 | 636 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
ea2cf228 FI |
637 | static unsigned long |
638 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) | |
fb8a1f11 | 639 | { |
ea2cf228 | 640 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
641 | } |
642 | ||
ea2cf228 | 643 | /* cinfo->lock held by caller */ |
1763da12 | 644 | int |
ea2cf228 FI |
645 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
646 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
647 | { |
648 | struct nfs_page *req, *tmp; | |
649 | int ret = 0; | |
650 | ||
651 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
652 | if (!nfs_lock_request(req)) |
653 | continue; | |
7ad84aa9 | 654 | kref_get(&req->wb_kref); |
ea2cf228 | 655 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 656 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 657 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
658 | nfs_list_add_request(req, dst); |
659 | ret++; | |
1763da12 | 660 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 661 | break; |
d6d6dc7c FI |
662 | } |
663 | return ret; | |
fb8a1f11 TM |
664 | } |
665 | ||
1da177e4 LT |
666 | /* |
667 | * nfs_scan_commit - Scan an inode for commit requests | |
668 | * @inode: NFS inode to scan | |
ea2cf228 FI |
669 | * @dst: mds destination list |
670 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
671 | * |
672 | * Moves requests from the inode's 'commit' request list. | |
673 | * The requests are *not* checked to ensure that they form a contiguous set. | |
674 | */ | |
1763da12 | 675 | int |
ea2cf228 FI |
676 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
677 | struct nfs_commit_info *cinfo) | |
1da177e4 | 678 | { |
d6d6dc7c | 679 | int ret = 0; |
fb8a1f11 | 680 | |
ea2cf228 FI |
681 | spin_lock(cinfo->lock); |
682 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 683 | const int max = INT_MAX; |
d6d6dc7c | 684 | |
ea2cf228 FI |
685 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
686 | cinfo, max); | |
687 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 688 | } |
ea2cf228 | 689 | spin_unlock(cinfo->lock); |
ff778d02 | 690 | return ret; |
1da177e4 | 691 | } |
d6d6dc7c | 692 | |
c42de9dd | 693 | #else |
ea2cf228 | 694 | static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
695 | { |
696 | return 0; | |
697 | } | |
698 | ||
1763da12 FI |
699 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
700 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
701 | { |
702 | return 0; | |
703 | } | |
1da177e4 LT |
704 | #endif |
705 | ||
1da177e4 | 706 | /* |
e7d39069 TM |
707 | * Search for an existing write request, and attempt to update |
708 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 709 | * |
e7d39069 TM |
710 | * If the attempt fails, then the existing request is flushed out |
711 | * to disk. | |
1da177e4 | 712 | */ |
e7d39069 TM |
713 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
714 | struct page *page, | |
715 | unsigned int offset, | |
716 | unsigned int bytes) | |
1da177e4 | 717 | { |
e7d39069 TM |
718 | struct nfs_page *req; |
719 | unsigned int rqend; | |
720 | unsigned int end; | |
721 | int error; | |
722 | ||
723 | if (!PagePrivate(page)) | |
724 | return NULL; | |
1da177e4 LT |
725 | |
726 | end = offset + bytes; | |
e7d39069 | 727 | spin_lock(&inode->i_lock); |
1da177e4 | 728 | |
1da177e4 | 729 | for (;;) { |
277459d2 | 730 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
731 | if (req == NULL) |
732 | goto out_unlock; | |
733 | ||
734 | rqend = req->wb_offset + req->wb_bytes; | |
735 | /* | |
736 | * Tell the caller to flush out the request if | |
737 | * the offsets are non-contiguous. | |
738 | * Note: nfs_flush_incompatible() will already | |
739 | * have flushed out requests having wrong owners. | |
740 | */ | |
e468bae9 | 741 | if (offset > rqend |
e7d39069 TM |
742 | || end < req->wb_offset) |
743 | goto out_flushme; | |
744 | ||
7ad84aa9 | 745 | if (nfs_lock_request(req)) |
1da177e4 | 746 | break; |
1da177e4 | 747 | |
e7d39069 | 748 | /* The request is locked, so wait and then retry */ |
587142f8 | 749 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
750 | error = nfs_wait_on_request(req); |
751 | nfs_release_request(req); | |
752 | if (error != 0) | |
753 | goto out_err; | |
754 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
755 | } |
756 | ||
757 | /* Okay, the request matches. Update the region */ | |
758 | if (offset < req->wb_offset) { | |
759 | req->wb_offset = offset; | |
760 | req->wb_pgbase = offset; | |
1da177e4 | 761 | } |
1da177e4 LT |
762 | if (end > rqend) |
763 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
764 | else |
765 | req->wb_bytes = rqend - req->wb_offset; | |
766 | out_unlock: | |
767 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
768 | if (req) |
769 | nfs_clear_request_commit(req); | |
e7d39069 TM |
770 | return req; |
771 | out_flushme: | |
772 | spin_unlock(&inode->i_lock); | |
773 | nfs_release_request(req); | |
774 | error = nfs_wb_page(inode, page); | |
775 | out_err: | |
776 | return ERR_PTR(error); | |
777 | } | |
778 | ||
779 | /* | |
780 | * Try to update an existing write request, or create one if there is none. | |
781 | * | |
782 | * Note: Should always be called with the Page Lock held to prevent races | |
783 | * if we have to add a new request. Also assumes that the caller has | |
784 | * already called nfs_flush_incompatible() if necessary. | |
785 | */ | |
786 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
787 | struct page *page, unsigned int offset, unsigned int bytes) | |
788 | { | |
789 | struct inode *inode = page->mapping->host; | |
790 | struct nfs_page *req; | |
1da177e4 | 791 | |
e7d39069 TM |
792 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
793 | if (req != NULL) | |
794 | goto out; | |
795 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
796 | if (IS_ERR(req)) | |
797 | goto out; | |
d6d6dc7c | 798 | nfs_inode_add_request(inode, req); |
efc91ed0 | 799 | out: |
61e930a9 | 800 | return req; |
1da177e4 LT |
801 | } |
802 | ||
e7d39069 TM |
803 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
804 | unsigned int offset, unsigned int count) | |
805 | { | |
806 | struct nfs_page *req; | |
807 | ||
808 | req = nfs_setup_write_request(ctx, page, offset, count); | |
809 | if (IS_ERR(req)) | |
810 | return PTR_ERR(req); | |
811 | /* Update file length */ | |
812 | nfs_grow_file(page, offset, count); | |
813 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 814 | nfs_mark_request_dirty(req); |
9994b62b | 815 | nfs_unlock_request(req); |
e7d39069 TM |
816 | return 0; |
817 | } | |
818 | ||
1da177e4 LT |
819 | int nfs_flush_incompatible(struct file *file, struct page *page) |
820 | { | |
cd3758e3 | 821 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 822 | struct nfs_page *req; |
1a54533e | 823 | int do_flush, status; |
1da177e4 LT |
824 | /* |
825 | * Look for a request corresponding to this page. If there | |
826 | * is one, and it belongs to another file, we flush it out | |
827 | * before we try to copy anything into the page. Do this | |
828 | * due to the lack of an ACCESS-type call in NFSv2. | |
829 | * Also do the same if we find a request from an existing | |
830 | * dropped page. | |
831 | */ | |
1a54533e TM |
832 | do { |
833 | req = nfs_page_find_request(page); | |
834 | if (req == NULL) | |
835 | return 0; | |
f11ac8db TM |
836 | do_flush = req->wb_page != page || req->wb_context != ctx || |
837 | req->wb_lock_context->lockowner != current->files || | |
838 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 839 | nfs_release_request(req); |
1a54533e TM |
840 | if (!do_flush) |
841 | return 0; | |
842 | status = nfs_wb_page(page->mapping->host, page); | |
843 | } while (status == 0); | |
844 | return status; | |
1da177e4 LT |
845 | } |
846 | ||
5d47a356 TM |
847 | /* |
848 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
849 | * the PageUptodate() flag. In this case, we will need to turn off | |
850 | * write optimisations that depend on the page contents being correct. | |
851 | */ | |
8d197a56 | 852 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 853 | { |
8d197a56 TM |
854 | if (nfs_have_delegated_attributes(inode)) |
855 | goto out; | |
856 | if (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE) | |
857 | return false; | |
858 | out: | |
859 | return PageUptodate(page) != 0; | |
5d47a356 TM |
860 | } |
861 | ||
1da177e4 LT |
862 | /* |
863 | * Update and possibly write a cached page of an NFS file. | |
864 | * | |
865 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
866 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
867 | */ | |
868 | int nfs_updatepage(struct file *file, struct page *page, | |
869 | unsigned int offset, unsigned int count) | |
870 | { | |
cd3758e3 | 871 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 872 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
873 | int status = 0; |
874 | ||
91d5b470 CL |
875 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
876 | ||
48186c7d | 877 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
878 | file->f_path.dentry->d_parent->d_name.name, |
879 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 880 | (long long)(page_offset(page) + offset)); |
1da177e4 | 881 | |
1da177e4 | 882 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
883 | * is up to date, it may be more efficient to extend the write |
884 | * to cover the entire page in order to avoid fragmentation | |
885 | * inefficiencies. | |
1da177e4 | 886 | */ |
5d47a356 TM |
887 | if (nfs_write_pageuptodate(page, inode) && |
888 | inode->i_flock == NULL && | |
6b2f3d1f | 889 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 890 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 891 | offset = 0; |
1da177e4 LT |
892 | } |
893 | ||
e21195a7 | 894 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
895 | if (status < 0) |
896 | nfs_set_pageerror(page); | |
59b7c05f TM |
897 | else |
898 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 899 | |
48186c7d | 900 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 901 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
902 | return status; |
903 | } | |
904 | ||
3ff7576d | 905 | static int flush_task_priority(int how) |
1da177e4 LT |
906 | { |
907 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
908 | case FLUSH_HIGHPRI: | |
909 | return RPC_PRIORITY_HIGH; | |
910 | case FLUSH_LOWPRI: | |
911 | return RPC_PRIORITY_LOW; | |
912 | } | |
913 | return RPC_PRIORITY_NORMAL; | |
914 | } | |
915 | ||
c5996c4e FI |
916 | int nfs_initiate_write(struct rpc_clnt *clnt, |
917 | struct nfs_write_data *data, | |
d138d5d1 AA |
918 | const struct rpc_call_ops *call_ops, |
919 | int how) | |
1da177e4 | 920 | { |
cd841605 | 921 | struct inode *inode = data->header->inode; |
3ff7576d | 922 | int priority = flush_task_priority(how); |
07737691 | 923 | struct rpc_task *task; |
bdc7f021 TM |
924 | struct rpc_message msg = { |
925 | .rpc_argp = &data->args, | |
926 | .rpc_resp = &data->res, | |
cd841605 | 927 | .rpc_cred = data->header->cred, |
bdc7f021 | 928 | }; |
84115e1c | 929 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 930 | .rpc_client = clnt, |
07737691 | 931 | .task = &data->task, |
bdc7f021 | 932 | .rpc_message = &msg, |
84115e1c TM |
933 | .callback_ops = call_ops, |
934 | .callback_data = data, | |
101070ca | 935 | .workqueue = nfsiod_workqueue, |
2c61be0a | 936 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 937 | .priority = priority, |
84115e1c | 938 | }; |
2c61be0a | 939 | int ret = 0; |
1da177e4 | 940 | |
d138d5d1 AA |
941 | /* Set up the initial task struct. */ |
942 | NFS_PROTO(inode)->write_setup(data, &msg); | |
943 | ||
944 | dprintk("NFS: %5u initiated write call " | |
945 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
946 | data->task.tk_pid, | |
947 | inode->i_sb->s_id, | |
948 | (long long)NFS_FILEID(inode), | |
949 | data->args.count, | |
950 | (unsigned long long)data->args.offset); | |
951 | ||
952 | task = rpc_run_task(&task_setup_data); | |
953 | if (IS_ERR(task)) { | |
954 | ret = PTR_ERR(task); | |
955 | goto out; | |
956 | } | |
957 | if (how & FLUSH_SYNC) { | |
958 | ret = rpc_wait_for_completion_task(task); | |
959 | if (ret == 0) | |
960 | ret = task->tk_status; | |
961 | } | |
962 | rpc_put_task(task); | |
963 | out: | |
964 | return ret; | |
965 | } | |
a69aef14 | 966 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
967 | |
968 | /* | |
969 | * Set up the argument/result storage required for the RPC call. | |
970 | */ | |
6c75dc0d | 971 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 | 972 | unsigned int count, unsigned int offset, |
ea2cf228 | 973 | int how, struct nfs_commit_info *cinfo) |
d138d5d1 | 974 | { |
6c75dc0d | 975 | struct nfs_page *req = data->header->req; |
d138d5d1 | 976 | |
1da177e4 LT |
977 | /* Set up the RPC argument and reply structs |
978 | * NB: take care not to mess about with data->commit et al. */ | |
979 | ||
6c75dc0d | 980 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 981 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
982 | /* pnfs_set_layoutcommit needs this */ |
983 | data->mds_offset = data->args.offset; | |
1da177e4 | 984 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 985 | data->args.pages = data->pages.pagevec; |
1da177e4 | 986 | data->args.count = count; |
383ba719 | 987 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 988 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 989 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
990 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
991 | case 0: | |
992 | break; | |
993 | case FLUSH_COND_STABLE: | |
ea2cf228 | 994 | if (nfs_reqs_to_commit(cinfo)) |
87ed5eb4 TM |
995 | break; |
996 | default: | |
997 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 998 | } |
1da177e4 LT |
999 | |
1000 | data->res.fattr = &data->fattr; | |
1001 | data->res.count = count; | |
1002 | data->res.verf = &data->verf; | |
0e574af1 | 1003 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 1004 | } |
1da177e4 | 1005 | |
6e4efd56 TM |
1006 | static int nfs_do_write(struct nfs_write_data *data, |
1007 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
1008 | int how) |
1009 | { | |
cd841605 | 1010 | struct inode *inode = data->header->inode; |
0382b744 | 1011 | |
c5996c4e | 1012 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how); |
1da177e4 LT |
1013 | } |
1014 | ||
275acaaf TM |
1015 | static int nfs_do_multiple_writes(struct list_head *head, |
1016 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
1017 | int how) |
1018 | { | |
1019 | struct nfs_write_data *data; | |
1020 | int ret = 0; | |
1021 | ||
1022 | while (!list_empty(head)) { | |
1023 | int ret2; | |
1024 | ||
6c75dc0d | 1025 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
1026 | list_del_init(&data->list); |
1027 | ||
dce81290 | 1028 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
1029 | if (ret == 0) |
1030 | ret = ret2; | |
1031 | } | |
1032 | return ret; | |
1033 | } | |
1034 | ||
6d884e8f F |
1035 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1036 | * call this on each, which will prepare them to be retried on next | |
1037 | * writeback using standard nfs. | |
1038 | */ | |
1039 | static void nfs_redirty_request(struct nfs_page *req) | |
1040 | { | |
1041 | nfs_mark_request_dirty(req); | |
3aff4ebb | 1042 | nfs_unlock_request_dont_release(req); |
d1182b33 | 1043 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 1044 | nfs_release_request(req); |
6d884e8f F |
1045 | } |
1046 | ||
061ae2ed | 1047 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1048 | { |
1049 | struct nfs_page *req; | |
1050 | ||
1051 | while (!list_empty(head)) { | |
1052 | req = nfs_list_entry(head->next); | |
1053 | nfs_list_remove_request(req); | |
1054 | nfs_redirty_request(req); | |
1055 | } | |
1056 | } | |
1057 | ||
061ae2ed FI |
1058 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1059 | .error_cleanup = nfs_async_write_error, | |
1060 | .completion = nfs_write_completion, | |
1061 | }; | |
1062 | ||
25b11dcd TM |
1063 | static void nfs_flush_error(struct nfs_pageio_descriptor *desc, |
1064 | struct nfs_pgio_header *hdr) | |
1065 | { | |
1066 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1067 | while (!list_empty(&hdr->rpc_list)) { | |
1068 | struct nfs_write_data *data = list_first_entry(&hdr->rpc_list, | |
1069 | struct nfs_write_data, list); | |
1070 | list_del(&data->list); | |
1071 | nfs_writedata_release(data); | |
1072 | } | |
1073 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
1074 | } | |
1075 | ||
1da177e4 LT |
1076 | /* |
1077 | * Generate multiple small requests to write out a single | |
1078 | * contiguous dirty area on one page. | |
1079 | */ | |
6c75dc0d FI |
1080 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1081 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1082 | { |
6c75dc0d | 1083 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1084 | struct page *page = req->wb_page; |
1085 | struct nfs_write_data *data; | |
d097971d | 1086 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1087 | unsigned int offset; |
1da177e4 | 1088 | int requests = 0; |
ea2cf228 | 1089 | struct nfs_commit_info cinfo; |
1da177e4 | 1090 | |
ea2cf228 | 1091 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
1da177e4 | 1092 | |
b31268ac | 1093 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1094 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || |
b31268ac TM |
1095 | desc->pg_count > wsize)) |
1096 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1097 | ||
1098 | ||
275acaaf | 1099 | offset = 0; |
c76069bd | 1100 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1101 | do { |
1102 | size_t len = min(nbytes, wsize); | |
1103 | ||
6c75dc0d | 1104 | data = nfs_writedata_alloc(hdr, 1); |
25b11dcd TM |
1105 | if (!data) { |
1106 | nfs_flush_error(desc, hdr); | |
1107 | return -ENOMEM; | |
1108 | } | |
30dd374f | 1109 | data->pages.pagevec[0] = page; |
ea2cf228 | 1110 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); |
6c75dc0d | 1111 | list_add(&data->list, &hdr->rpc_list); |
1da177e4 | 1112 | requests++; |
e9f7bee1 | 1113 | nbytes -= len; |
275acaaf | 1114 | offset += len; |
e9f7bee1 | 1115 | } while (nbytes != 0); |
25b11dcd TM |
1116 | nfs_list_remove_request(req); |
1117 | nfs_list_add_request(req, &hdr->pages); | |
6c75dc0d | 1118 | desc->pg_rpc_callops = &nfs_write_common_ops; |
25b11dcd | 1119 | return 0; |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | /* | |
1123 | * Create an RPC task for the given write request and kick it. | |
1124 | * The page must have been locked by the caller. | |
1125 | * | |
1126 | * It may happen that the page we're passed is not marked dirty. | |
1127 | * This is the case if nfs_updatepage detects a conflicting request | |
1128 | * that has been written but not committed. | |
1129 | */ | |
6c75dc0d FI |
1130 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1131 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1132 | { |
1133 | struct nfs_page *req; | |
1134 | struct page **pages; | |
1135 | struct nfs_write_data *data; | |
c76069bd | 1136 | struct list_head *head = &desc->pg_list; |
ea2cf228 | 1137 | struct nfs_commit_info cinfo; |
1da177e4 | 1138 | |
6c75dc0d FI |
1139 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1140 | desc->pg_count)); | |
1141 | if (!data) { | |
25b11dcd TM |
1142 | nfs_flush_error(desc, hdr); |
1143 | return -ENOMEM; | |
44b83799 | 1144 | } |
6c75dc0d | 1145 | |
ea2cf228 | 1146 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
30dd374f | 1147 | pages = data->pages.pagevec; |
1da177e4 LT |
1148 | while (!list_empty(head)) { |
1149 | req = nfs_list_entry(head->next); | |
1150 | nfs_list_remove_request(req); | |
6c75dc0d | 1151 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1152 | *pages++ = req->wb_page; |
1da177e4 | 1153 | } |
1da177e4 | 1154 | |
b31268ac | 1155 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1156 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) |
b31268ac TM |
1157 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; |
1158 | ||
1da177e4 | 1159 | /* Set up the argument struct */ |
ea2cf228 | 1160 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); |
6c75dc0d FI |
1161 | list_add(&data->list, &hdr->rpc_list); |
1162 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
25b11dcd | 1163 | return 0; |
1da177e4 LT |
1164 | } |
1165 | ||
6c75dc0d FI |
1166 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1167 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1168 | { |
1169 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1170 | return nfs_flush_multi(desc, hdr); |
1171 | return nfs_flush_one(desc, hdr); | |
dce81290 TM |
1172 | } |
1173 | ||
1174 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1175 | { |
6c75dc0d FI |
1176 | struct nfs_write_header *whdr; |
1177 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1178 | int ret; |
1179 | ||
6c75dc0d FI |
1180 | whdr = nfs_writehdr_alloc(); |
1181 | if (!whdr) { | |
9b5415b5 | 1182 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
6c75dc0d FI |
1183 | return -ENOMEM; |
1184 | } | |
1185 | hdr = &whdr->header; | |
1186 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1187 | atomic_inc(&hdr->refcnt); | |
1188 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1189 | if (ret == 0) |
6c75dc0d FI |
1190 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1191 | desc->pg_rpc_callops, | |
1192 | desc->pg_ioflags); | |
6c75dc0d | 1193 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 1194 | hdr->completion_ops->completion(hdr); |
275acaaf | 1195 | return ret; |
1751c363 | 1196 | } |
1751c363 TM |
1197 | |
1198 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1199 | .pg_test = nfs_generic_pg_test, | |
1200 | .pg_doio = nfs_generic_pg_writepages, | |
1201 | }; | |
1202 | ||
e2fecb21 | 1203 | void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
061ae2ed FI |
1204 | struct inode *inode, int ioflags, |
1205 | const struct nfs_pgio_completion_ops *compl_ops) | |
1da177e4 | 1206 | { |
061ae2ed | 1207 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops, |
1751c363 TM |
1208 | NFS_SERVER(inode)->wsize, ioflags); |
1209 | } | |
1da177e4 | 1210 | |
dce81290 TM |
1211 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1212 | { | |
1213 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1214 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1215 | } | |
1f945357 | 1216 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1217 | |
1763da12 FI |
1218 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1219 | struct inode *inode, int ioflags, | |
1220 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 | 1221 | { |
061ae2ed FI |
1222 | if (!pnfs_pageio_init_write(pgio, inode, ioflags, compl_ops)) |
1223 | nfs_pageio_init_write_mds(pgio, inode, ioflags, compl_ops); | |
1da177e4 LT |
1224 | } |
1225 | ||
def6ed7e AA |
1226 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1227 | { | |
1228 | struct nfs_write_data *data = calldata; | |
cd841605 | 1229 | NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); |
def6ed7e | 1230 | } |
def6ed7e | 1231 | |
0b7c0153 FI |
1232 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1233 | { | |
1234 | struct nfs_commit_data *data = calldata; | |
1235 | ||
1236 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1237 | } | |
1238 | ||
1da177e4 LT |
1239 | /* |
1240 | * Handle a write reply that flushes a whole page. | |
1241 | * | |
1242 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1243 | * writebacks since the page->count is kept > 1 for as long | |
1244 | * as the page has a write request pending. | |
1245 | */ | |
6c75dc0d | 1246 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1247 | { |
788e7a89 | 1248 | struct nfs_write_data *data = calldata; |
1da177e4 | 1249 | |
c9d8f89d TM |
1250 | nfs_writeback_done(task, data); |
1251 | } | |
1252 | ||
6c75dc0d | 1253 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1254 | { |
1255 | struct nfs_write_data *data = calldata; | |
cd841605 | 1256 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1257 | int status = data->task.tk_status; |
6c75dc0d | 1258 | struct nfs_page *req = hdr->req; |
788e7a89 | 1259 | |
6c75dc0d FI |
1260 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1261 | spin_lock(&hdr->lock); | |
1262 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1263 | ; /* Do nothing */ | |
1264 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
8e821cad | 1265 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); |
6c75dc0d FI |
1266 | else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) |
1267 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); | |
1268 | spin_unlock(&hdr->lock); | |
1da177e4 | 1269 | } |
cd841605 | 1270 | nfs_writedata_release(data); |
1da177e4 LT |
1271 | } |
1272 | ||
6c75dc0d | 1273 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1274 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1275 | .rpc_call_done = nfs_writeback_done_common, |
1276 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1277 | }; |
1278 | ||
1279 | ||
1da177e4 LT |
1280 | /* |
1281 | * This function is called when the WRITE call is complete. | |
1282 | */ | |
13602896 | 1283 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1284 | { |
1da177e4 LT |
1285 | struct nfs_writeargs *argp = &data->args; |
1286 | struct nfs_writeres *resp = &data->res; | |
cd841605 | 1287 | struct inode *inode = data->header->inode; |
788e7a89 | 1288 | int status; |
1da177e4 | 1289 | |
a3f565b1 | 1290 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1291 | task->tk_pid, task->tk_status); |
1292 | ||
f551e44f CL |
1293 | /* |
1294 | * ->write_done will attempt to use post-op attributes to detect | |
1295 | * conflicting writes by other clients. A strict interpretation | |
1296 | * of close-to-open would allow us to continue caching even if | |
1297 | * another writer had changed the file, but some applications | |
1298 | * depend on tighter cache coherency when writing. | |
1299 | */ | |
cd841605 | 1300 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1301 | if (status != 0) |
13602896 | 1302 | return; |
cd841605 | 1303 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1304 | |
1da177e4 LT |
1305 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1306 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1307 | /* We tried a write call, but the server did not | |
1308 | * commit data to stable storage even though we | |
1309 | * requested it. | |
1310 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1311 | * the server reports NFS_DATA_SYNC, but performs | |
1312 | * NFS_FILE_SYNC. We therefore implement this checking | |
1313 | * as a dprintk() in order to avoid filling syslog. | |
1314 | */ | |
1315 | static unsigned long complain; | |
1316 | ||
a69aef14 | 1317 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1318 | if (time_before(complain, jiffies)) { |
48186c7d | 1319 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1320 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1321 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1322 | resp->verf->committed, argp->stable); |
1323 | complain = jiffies + 300 * HZ; | |
1324 | } | |
1325 | } | |
1326 | #endif | |
6c75dc0d FI |
1327 | if (task->tk_status < 0) |
1328 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); | |
1329 | else if (resp->count < argp->count) { | |
1da177e4 LT |
1330 | static unsigned long complain; |
1331 | ||
6c75dc0d | 1332 | /* This a short write! */ |
cd841605 | 1333 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1334 | |
1da177e4 | 1335 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1336 | if (resp->count == 0) { |
1337 | if (time_before(complain, jiffies)) { | |
1338 | printk(KERN_WARNING | |
1339 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1340 | argp->count); | |
1341 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1342 | } |
6c75dc0d FI |
1343 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1344 | task->tk_status = -EIO; | |
13602896 | 1345 | return; |
1da177e4 | 1346 | } |
6c75dc0d FI |
1347 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1348 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1349 | /* Resend from where the server left off */ | |
1350 | data->mds_offset += resp->count; | |
1351 | argp->offset += resp->count; | |
1352 | argp->pgbase += resp->count; | |
1353 | argp->count -= resp->count; | |
1354 | } else { | |
1355 | /* Resend as a stable write in order to avoid | |
1356 | * headaches in the case of a server crash. | |
1357 | */ | |
1358 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1359 | } |
6c75dc0d | 1360 | rpc_restart_call_prepare(task); |
1da177e4 | 1361 | } |
1da177e4 LT |
1362 | } |
1363 | ||
1364 | ||
1365 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1366 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1367 | { | |
b8413f98 TM |
1368 | int ret; |
1369 | ||
71d0a611 TM |
1370 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1371 | return 1; | |
b8413f98 TM |
1372 | if (!may_wait) |
1373 | return 0; | |
1374 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1375 | NFS_INO_COMMIT, | |
1376 | nfs_wait_bit_killable, | |
1377 | TASK_KILLABLE); | |
1378 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1379 | } |
1380 | ||
f453a54a | 1381 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1382 | { |
1383 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1384 | smp_mb__after_clear_bit(); | |
1385 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1386 | } | |
1387 | ||
0b7c0153 | 1388 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1389 | { |
0b7c0153 FI |
1390 | put_nfs_open_context(data->context); |
1391 | nfs_commit_free(data); | |
1da177e4 | 1392 | } |
e0c2b380 | 1393 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1394 | |
0b7c0153 | 1395 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd FI |
1396 | const struct rpc_call_ops *call_ops, |
1397 | int how) | |
1da177e4 | 1398 | { |
07737691 | 1399 | struct rpc_task *task; |
9ace33cd | 1400 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1401 | struct rpc_message msg = { |
1402 | .rpc_argp = &data->args, | |
1403 | .rpc_resp = &data->res, | |
9ace33cd | 1404 | .rpc_cred = data->cred, |
bdc7f021 | 1405 | }; |
84115e1c | 1406 | struct rpc_task_setup task_setup_data = { |
07737691 | 1407 | .task = &data->task, |
9ace33cd | 1408 | .rpc_client = clnt, |
bdc7f021 | 1409 | .rpc_message = &msg, |
9ace33cd | 1410 | .callback_ops = call_ops, |
84115e1c | 1411 | .callback_data = data, |
101070ca | 1412 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1413 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1414 | .priority = priority, |
84115e1c | 1415 | }; |
9ace33cd FI |
1416 | /* Set up the initial task struct. */ |
1417 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1418 | ||
1419 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1420 | ||
1421 | task = rpc_run_task(&task_setup_data); | |
1422 | if (IS_ERR(task)) | |
1423 | return PTR_ERR(task); | |
1424 | if (how & FLUSH_SYNC) | |
1425 | rpc_wait_for_completion_task(task); | |
1426 | rpc_put_task(task); | |
1427 | return 0; | |
1428 | } | |
e0c2b380 | 1429 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1430 | |
1431 | /* | |
1432 | * Set up the argument/result storage required for the RPC call. | |
1433 | */ | |
0b7c0153 | 1434 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1435 | struct list_head *head, |
1436 | struct pnfs_layout_segment *lseg, | |
1437 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1438 | { |
1439 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1440 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1441 | |
1442 | /* Set up the RPC argument and reply structs | |
1443 | * NB: take care not to mess about with data->commit et al. */ | |
1444 | ||
1445 | list_splice_init(head, &data->pages); | |
1da177e4 | 1446 | |
1da177e4 | 1447 | data->inode = inode; |
9ace33cd | 1448 | data->cred = first->wb_context->cred; |
988b6dce | 1449 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1450 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1451 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1452 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1453 | |
1454 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1455 | /* Note: we always request a commit of the entire inode */ |
1456 | data->args.offset = 0; | |
1457 | data->args.count = 0; | |
0b7c0153 | 1458 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1459 | data->res.fattr = &data->fattr; |
1460 | data->res.verf = &data->verf; | |
0e574af1 | 1461 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1462 | } |
e0c2b380 | 1463 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1464 | |
e0c2b380 | 1465 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1466 | struct pnfs_layout_segment *lseg, |
1467 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1468 | { |
1469 | struct nfs_page *req; | |
1470 | ||
1471 | while (!list_empty(page_list)) { | |
1472 | req = nfs_list_entry(page_list->next); | |
1473 | nfs_list_remove_request(req); | |
ea2cf228 | 1474 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1475 | if (!cinfo->dreq) { |
1476 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
1477 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1478 | BDI_RECLAIMABLE); | |
1479 | } | |
9994b62b | 1480 | nfs_unlock_request(req); |
64bfeb49 FI |
1481 | } |
1482 | } | |
e0c2b380 | 1483 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1484 | |
1da177e4 LT |
1485 | /* |
1486 | * Commit dirty pages | |
1487 | */ | |
1488 | static int | |
ea2cf228 FI |
1489 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1490 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1491 | { |
0b7c0153 | 1492 | struct nfs_commit_data *data; |
1da177e4 | 1493 | |
c9d8f89d | 1494 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1495 | |
1496 | if (!data) | |
1497 | goto out_bad; | |
1498 | ||
1499 | /* Set up the argument struct */ | |
f453a54a FI |
1500 | nfs_init_commit(data, head, NULL, cinfo); |
1501 | atomic_inc(&cinfo->mds->rpcs_out); | |
0b7c0153 | 1502 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, how); |
1da177e4 | 1503 | out_bad: |
ea2cf228 | 1504 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1505 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1506 | return -ENOMEM; |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * COMMIT call returned | |
1511 | */ | |
788e7a89 | 1512 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1513 | { |
0b7c0153 | 1514 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1515 | |
a3f565b1 | 1516 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1517 | task->tk_pid, task->tk_status); |
1518 | ||
788e7a89 | 1519 | /* Call the NFS version-specific code */ |
c0d0e96b | 1520 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1521 | } |
1522 | ||
f453a54a | 1523 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1524 | { |
5917ce84 | 1525 | struct nfs_page *req; |
c9d8f89d | 1526 | int status = data->task.tk_status; |
f453a54a | 1527 | struct nfs_commit_info cinfo; |
788e7a89 | 1528 | |
1da177e4 LT |
1529 | while (!list_empty(&data->pages)) { |
1530 | req = nfs_list_entry(data->pages.next); | |
1531 | nfs_list_remove_request(req); | |
d6d6dc7c | 1532 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1533 | |
48186c7d | 1534 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1535 | req->wb_context->dentry->d_sb->s_id, |
1536 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1537 | req->wb_bytes, |
1538 | (long long)req_offset(req)); | |
c9d8f89d TM |
1539 | if (status < 0) { |
1540 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1541 | nfs_inode_remove_request(req); |
c9d8f89d | 1542 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1543 | goto next; |
1544 | } | |
1545 | ||
1546 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1547 | * returned by the server against all stored verfs. */ | |
1548 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1549 | /* We have a match */ | |
1550 | nfs_inode_remove_request(req); | |
1551 | dprintk(" OK\n"); | |
1552 | goto next; | |
1553 | } | |
1554 | /* We have a mismatch. Write the page again */ | |
1555 | dprintk(" mismatch\n"); | |
6d884e8f | 1556 | nfs_mark_request_dirty(req); |
1da177e4 | 1557 | next: |
9994b62b | 1558 | nfs_unlock_request(req); |
1da177e4 | 1559 | } |
f453a54a FI |
1560 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1561 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1562 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1563 | } |
1564 | ||
1565 | static void nfs_commit_release(void *calldata) | |
1566 | { | |
0b7c0153 | 1567 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1568 | |
f453a54a | 1569 | data->completion_ops->completion(data); |
c9d8f89d | 1570 | nfs_commitdata_release(calldata); |
1da177e4 | 1571 | } |
788e7a89 TM |
1572 | |
1573 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1574 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1575 | .rpc_call_done = nfs_commit_done, |
1576 | .rpc_release = nfs_commit_release, | |
1577 | }; | |
1da177e4 | 1578 | |
f453a54a FI |
1579 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1580 | .completion = nfs_commit_release_pages, | |
1581 | .error_cleanup = nfs_commit_clear_lock, | |
1582 | }; | |
1583 | ||
1763da12 FI |
1584 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1585 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1586 | { |
1587 | int status; | |
1588 | ||
ea2cf228 | 1589 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1590 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1591 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1592 | return status; |
1593 | } | |
1594 | ||
b608b283 | 1595 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1596 | { |
1da177e4 | 1597 | LIST_HEAD(head); |
ea2cf228 | 1598 | struct nfs_commit_info cinfo; |
71d0a611 | 1599 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1600 | int res; |
1da177e4 | 1601 | |
b8413f98 TM |
1602 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1603 | if (res <= 0) | |
c5efa5fc | 1604 | goto out_mark_dirty; |
ea2cf228 FI |
1605 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1606 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1607 | if (res) { |
a861a1e1 FI |
1608 | int error; |
1609 | ||
ea2cf228 | 1610 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1611 | if (error < 0) |
1612 | return error; | |
b8413f98 | 1613 | if (!may_wait) |
c5efa5fc | 1614 | goto out_mark_dirty; |
b8413f98 TM |
1615 | error = wait_on_bit(&NFS_I(inode)->flags, |
1616 | NFS_INO_COMMIT, | |
1617 | nfs_wait_bit_killable, | |
1618 | TASK_KILLABLE); | |
1619 | if (error < 0) | |
1620 | return error; | |
71d0a611 TM |
1621 | } else |
1622 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1623 | return res; |
1624 | /* Note: If we exit without ensuring that the commit is complete, | |
1625 | * we must mark the inode as dirty. Otherwise, future calls to | |
1626 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1627 | * that the data is on the disk. | |
1628 | */ | |
1629 | out_mark_dirty: | |
1630 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1631 | return res; |
1632 | } | |
8fc795f7 TM |
1633 | |
1634 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1635 | { | |
420e3646 TM |
1636 | struct nfs_inode *nfsi = NFS_I(inode); |
1637 | int flags = FLUSH_SYNC; | |
1638 | int ret = 0; | |
8fc795f7 | 1639 | |
3236c3e1 | 1640 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1641 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1642 | return ret; |
1643 | ||
a00dd6c0 JL |
1644 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1645 | /* Don't commit yet if this is a non-blocking flush and there | |
1646 | * are a lot of outstanding writes for this mapping. | |
1647 | */ | |
ea2cf228 | 1648 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1649 | goto out_mark_dirty; |
420e3646 | 1650 | |
a00dd6c0 | 1651 | /* don't wait for the COMMIT response */ |
420e3646 | 1652 | flags = 0; |
a00dd6c0 JL |
1653 | } |
1654 | ||
420e3646 TM |
1655 | ret = nfs_commit_inode(inode, flags); |
1656 | if (ret >= 0) { | |
1657 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1658 | if (ret < wbc->nr_to_write) | |
1659 | wbc->nr_to_write -= ret; | |
1660 | else | |
1661 | wbc->nr_to_write = 0; | |
1662 | } | |
8fc795f7 | 1663 | return 0; |
420e3646 TM |
1664 | } |
1665 | out_mark_dirty: | |
8fc795f7 TM |
1666 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1667 | return ret; | |
1668 | } | |
c63c7b05 | 1669 | #else |
8fc795f7 TM |
1670 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1671 | { | |
1672 | return 0; | |
1673 | } | |
1da177e4 LT |
1674 | #endif |
1675 | ||
8fc795f7 TM |
1676 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1677 | { | |
863a3c6c AA |
1678 | int ret; |
1679 | ||
1680 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1681 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1682 | int status; |
1683 | bool sync = true; | |
863a3c6c | 1684 | |
846d5a09 | 1685 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1686 | sync = false; |
863a3c6c AA |
1687 | |
1688 | status = pnfs_layoutcommit_inode(inode, sync); | |
1689 | if (status < 0) | |
1690 | return status; | |
1691 | } | |
1692 | return ret; | |
8fc795f7 TM |
1693 | } |
1694 | ||
acdc53b2 TM |
1695 | /* |
1696 | * flush the inode to disk. | |
1697 | */ | |
1698 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1699 | { |
1700 | struct writeback_control wbc = { | |
72cb77f4 | 1701 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1702 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1703 | .range_start = 0, |
1704 | .range_end = LLONG_MAX, | |
34901f70 | 1705 | }; |
34901f70 | 1706 | |
acdc53b2 | 1707 | return sync_inode(inode, &wbc); |
1c75950b TM |
1708 | } |
1709 | ||
1b3b4a1a TM |
1710 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1711 | { | |
1712 | struct nfs_page *req; | |
1b3b4a1a TM |
1713 | int ret = 0; |
1714 | ||
1715 | BUG_ON(!PageLocked(page)); | |
1716 | for (;;) { | |
ba8b06e6 | 1717 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1718 | req = nfs_page_find_request(page); |
1719 | if (req == NULL) | |
1b3b4a1a | 1720 | break; |
7ad84aa9 | 1721 | if (nfs_lock_request(req)) { |
8dd37758 | 1722 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1723 | nfs_inode_remove_request(req); |
1724 | /* | |
1725 | * In case nfs_inode_remove_request has marked the | |
1726 | * page as being dirty | |
1727 | */ | |
1728 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1729 | nfs_unlock_request(req); | |
1730 | break; | |
1731 | } | |
1732 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1733 | nfs_release_request(req); |
1b3b4a1a | 1734 | if (ret < 0) |
c988950e | 1735 | break; |
1b3b4a1a | 1736 | } |
1b3b4a1a TM |
1737 | return ret; |
1738 | } | |
1739 | ||
7f2f12d9 TM |
1740 | /* |
1741 | * Write back all requests on one page - we do this before reading it. | |
1742 | */ | |
1743 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1744 | { |
1745 | loff_t range_start = page_offset(page); | |
1746 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1747 | struct writeback_control wbc = { |
4d770ccf | 1748 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1749 | .nr_to_write = 0, |
4d770ccf TM |
1750 | .range_start = range_start, |
1751 | .range_end = range_end, | |
1752 | }; | |
1753 | int ret; | |
1c75950b | 1754 | |
0522f6ad | 1755 | for (;;) { |
ba8b06e6 | 1756 | wait_on_page_writeback(page); |
73e3302f TM |
1757 | if (clear_page_dirty_for_io(page)) { |
1758 | ret = nfs_writepage_locked(page, &wbc); | |
1759 | if (ret < 0) | |
1760 | goto out_error; | |
0522f6ad | 1761 | continue; |
7f2f12d9 | 1762 | } |
0522f6ad TM |
1763 | if (!PagePrivate(page)) |
1764 | break; | |
1765 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1766 | if (ret < 0) |
73e3302f | 1767 | goto out_error; |
7f2f12d9 | 1768 | } |
73e3302f TM |
1769 | return 0; |
1770 | out_error: | |
4d770ccf | 1771 | return ret; |
1c75950b TM |
1772 | } |
1773 | ||
074cc1de TM |
1774 | #ifdef CONFIG_MIGRATION |
1775 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1776 | struct page *page, enum migrate_mode mode) |
074cc1de | 1777 | { |
2da95652 JL |
1778 | /* |
1779 | * If PagePrivate is set, then the page is currently associated with | |
1780 | * an in-progress read or write request. Don't try to migrate it. | |
1781 | * | |
1782 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1783 | * that we can safely release the inode reference while holding | |
1784 | * the page lock. | |
1785 | */ | |
1786 | if (PagePrivate(page)) | |
1787 | return -EBUSY; | |
074cc1de | 1788 | |
7549ad5f | 1789 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1790 | |
a6bc32b8 | 1791 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1792 | } |
1793 | #endif | |
1794 | ||
f7b422b1 | 1795 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1796 | { |
1797 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1798 | sizeof(struct nfs_write_header), |
1da177e4 | 1799 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1800 | NULL); |
1da177e4 LT |
1801 | if (nfs_wdata_cachep == NULL) |
1802 | return -ENOMEM; | |
1803 | ||
93d2341c MD |
1804 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1805 | nfs_wdata_cachep); | |
1da177e4 LT |
1806 | if (nfs_wdata_mempool == NULL) |
1807 | return -ENOMEM; | |
1808 | ||
0b7c0153 FI |
1809 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1810 | sizeof(struct nfs_commit_data), | |
1811 | 0, SLAB_HWCACHE_ALIGN, | |
1812 | NULL); | |
1813 | if (nfs_cdata_cachep == NULL) | |
1814 | return -ENOMEM; | |
1815 | ||
93d2341c MD |
1816 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1817 | nfs_wdata_cachep); | |
1da177e4 LT |
1818 | if (nfs_commit_mempool == NULL) |
1819 | return -ENOMEM; | |
1820 | ||
89a09141 PZ |
1821 | /* |
1822 | * NFS congestion size, scale with available memory. | |
1823 | * | |
1824 | * 64MB: 8192k | |
1825 | * 128MB: 11585k | |
1826 | * 256MB: 16384k | |
1827 | * 512MB: 23170k | |
1828 | * 1GB: 32768k | |
1829 | * 2GB: 46340k | |
1830 | * 4GB: 65536k | |
1831 | * 8GB: 92681k | |
1832 | * 16GB: 131072k | |
1833 | * | |
1834 | * This allows larger machines to have larger/more transfers. | |
1835 | * Limit the default to 256M | |
1836 | */ | |
1837 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1838 | if (nfs_congestion_kb > 256*1024) | |
1839 | nfs_congestion_kb = 256*1024; | |
1840 | ||
1da177e4 LT |
1841 | return 0; |
1842 | } | |
1843 | ||
266bee88 | 1844 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1845 | { |
1846 | mempool_destroy(nfs_commit_mempool); | |
1847 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1848 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1849 | } |
1850 |