Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
3fcfab16 | 24 | |
1da177e4 | 25 | #include <asm/uaccess.h> |
1da177e4 LT |
26 | |
27 | #include "delegation.h" | |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
def6ed7e | 30 | #include "nfs4_fs.h" |
074cc1de | 31 | #include "fscache.h" |
94ad1c80 | 32 | #include "pnfs.h" |
1da177e4 LT |
33 | |
34 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
35 | ||
36 | #define MIN_POOL_WRITE (32) | |
37 | #define MIN_POOL_COMMIT (4) | |
38 | ||
39 | /* | |
40 | * Local function declarations | |
41 | */ | |
c63c7b05 | 42 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
061ae2ed FI |
43 | struct inode *inode, int ioflags, |
44 | const struct nfs_pgio_completion_ops *compl_ops); | |
f8512ad0 | 45 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 46 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 47 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 48 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
1da177e4 | 49 | |
e18b890b | 50 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 51 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 52 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
53 | static mempool_t *nfs_commit_mempool; |
54 | ||
0b7c0153 | 55 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 56 | { |
0b7c0153 | 57 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 58 | |
1da177e4 LT |
59 | if (p) { |
60 | memset(p, 0, sizeof(*p)); | |
61 | INIT_LIST_HEAD(&p->pages); | |
62 | } | |
63 | return p; | |
64 | } | |
e0c2b380 | 65 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 66 | |
0b7c0153 | 67 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
68 | { |
69 | mempool_free(p, nfs_commit_mempool); | |
70 | } | |
e0c2b380 | 71 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 72 | |
6c75dc0d | 73 | struct nfs_write_header *nfs_writehdr_alloc(void) |
3feb2d49 | 74 | { |
cd841605 | 75 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
76 | |
77 | if (p) { | |
cd841605 | 78 | struct nfs_pgio_header *hdr = &p->header; |
cd841605 | 79 | |
3feb2d49 | 80 | memset(p, 0, sizeof(*p)); |
cd841605 | 81 | INIT_LIST_HEAD(&hdr->pages); |
6c75dc0d FI |
82 | INIT_LIST_HEAD(&hdr->rpc_list); |
83 | spin_lock_init(&hdr->lock); | |
84 | atomic_set(&hdr->refcnt, 0); | |
3feb2d49 TM |
85 | } |
86 | return p; | |
87 | } | |
88 | ||
6c75dc0d FI |
89 | struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
90 | unsigned int pagecount) | |
91 | { | |
92 | struct nfs_write_data *data, *prealloc; | |
93 | ||
94 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
95 | if (prealloc->header == NULL) | |
96 | data = prealloc; | |
97 | else | |
98 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
99 | if (!data) | |
100 | goto out; | |
101 | ||
102 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
103 | data->header = hdr; | |
104 | atomic_inc(&hdr->refcnt); | |
105 | } else { | |
106 | if (data != prealloc) | |
107 | kfree(data); | |
108 | data = NULL; | |
109 | } | |
110 | out: | |
111 | return data; | |
112 | } | |
113 | ||
cd841605 | 114 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 115 | { |
cd841605 FI |
116 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
117 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 TM |
118 | } |
119 | ||
dce81290 | 120 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 121 | { |
6c75dc0d FI |
122 | struct nfs_pgio_header *hdr = wdata->header; |
123 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
124 | ||
383ba719 | 125 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
126 | if (wdata->pages.pagevec != wdata->pages.page_array) |
127 | kfree(wdata->pages.pagevec); | |
6c75dc0d FI |
128 | if (wdata != &write_header->rpc_data) |
129 | kfree(wdata); | |
130 | else | |
131 | wdata->header = NULL; | |
132 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 133 | hdr->completion_ops->completion(hdr); |
1da177e4 LT |
134 | } |
135 | ||
7b159fc1 TM |
136 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
137 | { | |
138 | ctx->error = error; | |
139 | smp_wmb(); | |
140 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
141 | } | |
142 | ||
277459d2 TM |
143 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
144 | { | |
145 | struct nfs_page *req = NULL; | |
146 | ||
147 | if (PagePrivate(page)) { | |
148 | req = (struct nfs_page *)page_private(page); | |
149 | if (req != NULL) | |
c03b4024 | 150 | kref_get(&req->wb_kref); |
277459d2 TM |
151 | } |
152 | return req; | |
153 | } | |
154 | ||
155 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
156 | { | |
587142f8 | 157 | struct inode *inode = page->mapping->host; |
277459d2 | 158 | struct nfs_page *req = NULL; |
277459d2 | 159 | |
587142f8 | 160 | spin_lock(&inode->i_lock); |
277459d2 | 161 | req = nfs_page_find_request_locked(page); |
587142f8 | 162 | spin_unlock(&inode->i_lock); |
277459d2 TM |
163 | return req; |
164 | } | |
165 | ||
1da177e4 LT |
166 | /* Adjust the file length if we're writing beyond the end */ |
167 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
168 | { | |
169 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
170 | loff_t end, i_size; |
171 | pgoff_t end_index; | |
1da177e4 | 172 | |
a3d01454 TM |
173 | spin_lock(&inode->i_lock); |
174 | i_size = i_size_read(inode); | |
175 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 176 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 177 | goto out; |
1da177e4 LT |
178 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
179 | if (i_size >= end) | |
a3d01454 | 180 | goto out; |
1da177e4 | 181 | i_size_write(inode, end); |
a3d01454 TM |
182 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
183 | out: | |
184 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
185 | } |
186 | ||
a301b777 TM |
187 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
188 | static void nfs_set_pageerror(struct page *page) | |
189 | { | |
190 | SetPageError(page); | |
191 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
192 | } | |
193 | ||
1da177e4 LT |
194 | /* We can set the PG_uptodate flag if we see that a write request |
195 | * covers the full page. | |
196 | */ | |
197 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
198 | { | |
1da177e4 LT |
199 | if (PageUptodate(page)) |
200 | return; | |
201 | if (base != 0) | |
202 | return; | |
49a70f27 | 203 | if (count != nfs_page_length(page)) |
1da177e4 | 204 | return; |
49a70f27 | 205 | SetPageUptodate(page); |
1da177e4 LT |
206 | } |
207 | ||
1da177e4 LT |
208 | static int wb_priority(struct writeback_control *wbc) |
209 | { | |
210 | if (wbc->for_reclaim) | |
c63c7b05 | 211 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 212 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
213 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
214 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
215 | } |
216 | ||
89a09141 PZ |
217 | /* |
218 | * NFS congestion control | |
219 | */ | |
220 | ||
221 | int nfs_congestion_kb; | |
222 | ||
223 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
224 | #define NFS_CONGESTION_OFF_THRESH \ | |
225 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
226 | ||
5a6d41b3 | 227 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 228 | { |
5a6d41b3 TM |
229 | int ret = test_set_page_writeback(page); |
230 | ||
231 | if (!ret) { | |
89a09141 PZ |
232 | struct inode *inode = page->mapping->host; |
233 | struct nfs_server *nfss = NFS_SERVER(inode); | |
234 | ||
a6305ddb | 235 | page_cache_get(page); |
277866a0 | 236 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
237 | NFS_CONGESTION_ON_THRESH) { |
238 | set_bdi_congested(&nfss->backing_dev_info, | |
239 | BLK_RW_ASYNC); | |
240 | } | |
89a09141 | 241 | } |
5a6d41b3 | 242 | return ret; |
89a09141 PZ |
243 | } |
244 | ||
245 | static void nfs_end_page_writeback(struct page *page) | |
246 | { | |
247 | struct inode *inode = page->mapping->host; | |
248 | struct nfs_server *nfss = NFS_SERVER(inode); | |
249 | ||
250 | end_page_writeback(page); | |
a6305ddb | 251 | page_cache_release(page); |
c4dc4bee | 252 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 253 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
254 | } |
255 | ||
cfb506e1 | 256 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 257 | { |
587142f8 | 258 | struct inode *inode = page->mapping->host; |
e261f51f | 259 | struct nfs_page *req; |
e261f51f TM |
260 | int ret; |
261 | ||
587142f8 | 262 | spin_lock(&inode->i_lock); |
074cc1de | 263 | for (;;) { |
e261f51f | 264 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
265 | if (req == NULL) |
266 | break; | |
9994b62b | 267 | if (nfs_lock_request_dontget(req)) |
e261f51f TM |
268 | break; |
269 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
9994b62b | 270 | * then the call to nfs_lock_request_dontget() will always |
e261f51f TM |
271 | * succeed provided that someone hasn't already marked the |
272 | * request as dirty (in which case we don't care). | |
273 | */ | |
587142f8 | 274 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
275 | if (!nonblock) |
276 | ret = nfs_wait_on_request(req); | |
277 | else | |
278 | ret = -EAGAIN; | |
e261f51f TM |
279 | nfs_release_request(req); |
280 | if (ret != 0) | |
074cc1de | 281 | return ERR_PTR(ret); |
587142f8 | 282 | spin_lock(&inode->i_lock); |
e261f51f | 283 | } |
587142f8 | 284 | spin_unlock(&inode->i_lock); |
074cc1de TM |
285 | return req; |
286 | } | |
287 | ||
288 | /* | |
289 | * Find an associated nfs write request, and prepare to flush it out | |
290 | * May return an error if the user signalled nfs_wait_on_request(). | |
291 | */ | |
292 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 293 | struct page *page, bool nonblock) |
074cc1de TM |
294 | { |
295 | struct nfs_page *req; | |
296 | int ret = 0; | |
297 | ||
cfb506e1 | 298 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
299 | if (!req) |
300 | goto out; | |
301 | ret = PTR_ERR(req); | |
302 | if (IS_ERR(req)) | |
303 | goto out; | |
304 | ||
305 | ret = nfs_set_page_writeback(page); | |
306 | BUG_ON(ret != 0); | |
307 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
308 | ||
f8512ad0 FI |
309 | if (!nfs_pageio_add_request(pgio, req)) { |
310 | nfs_redirty_request(req); | |
074cc1de | 311 | ret = pgio->pg_error; |
f8512ad0 | 312 | } |
074cc1de TM |
313 | out: |
314 | return ret; | |
e261f51f TM |
315 | } |
316 | ||
f758c885 | 317 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 318 | { |
1da177e4 | 319 | struct inode *inode = page->mapping->host; |
cfb506e1 | 320 | int ret; |
1da177e4 | 321 | |
91d5b470 CL |
322 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
323 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
324 | ||
7fe7f848 | 325 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 326 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
327 | if (ret == -EAGAIN) { |
328 | redirty_page_for_writepage(wbc, page); | |
329 | ret = 0; | |
330 | } | |
331 | return ret; | |
f758c885 | 332 | } |
7fe7f848 | 333 | |
f758c885 TM |
334 | /* |
335 | * Write an mmapped page to the server. | |
336 | */ | |
337 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
338 | { | |
339 | struct nfs_pageio_descriptor pgio; | |
340 | int err; | |
49a70f27 | 341 | |
061ae2ed FI |
342 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
343 | &nfs_async_write_completion_ops); | |
f758c885 TM |
344 | err = nfs_do_writepage(page, wbc, &pgio); |
345 | nfs_pageio_complete(&pgio); | |
346 | if (err < 0) | |
347 | return err; | |
348 | if (pgio.pg_error < 0) | |
349 | return pgio.pg_error; | |
350 | return 0; | |
4d770ccf TM |
351 | } |
352 | ||
353 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
354 | { | |
f758c885 | 355 | int ret; |
4d770ccf | 356 | |
f758c885 | 357 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 358 | unlock_page(page); |
f758c885 TM |
359 | return ret; |
360 | } | |
361 | ||
362 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
363 | { | |
364 | int ret; | |
365 | ||
366 | ret = nfs_do_writepage(page, wbc, data); | |
367 | unlock_page(page); | |
368 | return ret; | |
1da177e4 LT |
369 | } |
370 | ||
1da177e4 LT |
371 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
372 | { | |
1da177e4 | 373 | struct inode *inode = mapping->host; |
72cb77f4 | 374 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 375 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
376 | int err; |
377 | ||
72cb77f4 TM |
378 | /* Stop dirtying of new pages while we sync */ |
379 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
380 | nfs_wait_bit_killable, TASK_KILLABLE); | |
381 | if (err) | |
382 | goto out_err; | |
383 | ||
91d5b470 CL |
384 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
385 | ||
061ae2ed FI |
386 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), |
387 | &nfs_async_write_completion_ops); | |
f758c885 | 388 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 389 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
390 | |
391 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
392 | smp_mb__after_clear_bit(); | |
393 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
394 | ||
f758c885 | 395 | if (err < 0) |
72cb77f4 TM |
396 | goto out_err; |
397 | err = pgio.pg_error; | |
398 | if (err < 0) | |
399 | goto out_err; | |
c63c7b05 | 400 | return 0; |
72cb77f4 TM |
401 | out_err: |
402 | return err; | |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * Insert a write request into an inode | |
407 | */ | |
d6d6dc7c | 408 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
409 | { |
410 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
411 | |
412 | /* Lock the request! */ | |
413 | nfs_lock_request_dontget(req); | |
414 | ||
415 | spin_lock(&inode->i_lock); | |
4d65c520 | 416 | if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 417 | inode->i_version++; |
2df485a7 | 418 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 419 | SetPagePrivate(req->wb_page); |
277459d2 | 420 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 421 | nfsi->npages++; |
c03b4024 | 422 | kref_get(&req->wb_kref); |
e7d39069 | 423 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
424 | } |
425 | ||
426 | /* | |
89a09141 | 427 | * Remove a write request from an inode |
1da177e4 LT |
428 | */ |
429 | static void nfs_inode_remove_request(struct nfs_page *req) | |
430 | { | |
3d4ff43d | 431 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
432 | struct nfs_inode *nfsi = NFS_I(inode); |
433 | ||
434 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
435 | ||
587142f8 | 436 | spin_lock(&inode->i_lock); |
277459d2 | 437 | set_page_private(req->wb_page, 0); |
deb7d638 | 438 | ClearPagePrivate(req->wb_page); |
2df485a7 | 439 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 | 440 | nfsi->npages--; |
4d65c520 | 441 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
442 | nfs_release_request(req); |
443 | } | |
444 | ||
61822ab5 | 445 | static void |
6d884e8f | 446 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 447 | { |
61822ab5 TM |
448 | __set_page_dirty_nobuffers(req->wb_page); |
449 | } | |
450 | ||
1da177e4 | 451 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
8dd37758 TM |
452 | /** |
453 | * nfs_request_add_commit_list - add request to a commit list | |
454 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
455 | * @dst: commit list head |
456 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 457 | * |
ea2cf228 | 458 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
459 | * number of outstanding requests requiring a commit as well as |
460 | * the MM page stats. | |
461 | * | |
ea2cf228 | 462 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 463 | * holding the nfs_page lock. |
1da177e4 | 464 | */ |
8dd37758 | 465 | void |
ea2cf228 FI |
466 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
467 | struct nfs_commit_info *cinfo) | |
1da177e4 | 468 | { |
e468bae9 | 469 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
470 | spin_lock(cinfo->lock); |
471 | nfs_list_add_request(req, dst); | |
472 | cinfo->mds->ncommit++; | |
473 | spin_unlock(cinfo->lock); | |
fd39fc85 | 474 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 475 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
ea2cf228 | 476 | __mark_inode_dirty(req->wb_context->dentry->d_inode, I_DIRTY_DATASYNC); |
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); | |
508 | } | |
509 | ||
510 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
511 | struct inode *inode, | |
512 | struct nfs_direct_req *dreq) | |
513 | { | |
514 | nfs_init_cinfo_from_inode(cinfo, inode); | |
515 | } | |
516 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
517 | |
518 | /* | |
519 | * Add a request to the inode's commit list. | |
520 | */ | |
521 | static void | |
ea2cf228 FI |
522 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
523 | struct nfs_commit_info *cinfo) | |
8dd37758 | 524 | { |
ea2cf228 | 525 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 526 | return; |
ea2cf228 | 527 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 528 | } |
8e821cad | 529 | |
d6d6dc7c FI |
530 | static void |
531 | nfs_clear_page_commit(struct page *page) | |
532 | { | |
533 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
534 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
535 | } | |
536 | ||
8dd37758 | 537 | static void |
e468bae9 TM |
538 | nfs_clear_request_commit(struct nfs_page *req) |
539 | { | |
8dd37758 TM |
540 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
541 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 542 | struct nfs_commit_info cinfo; |
e468bae9 | 543 | |
ea2cf228 FI |
544 | nfs_init_cinfo_from_inode(&cinfo, inode); |
545 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
546 | spin_lock(cinfo.lock); | |
547 | nfs_request_remove_commit_list(req, &cinfo); | |
548 | spin_unlock(cinfo.lock); | |
8dd37758 | 549 | } |
d6d6dc7c | 550 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 551 | } |
e468bae9 TM |
552 | } |
553 | ||
8e821cad TM |
554 | static inline |
555 | int nfs_write_need_commit(struct nfs_write_data *data) | |
556 | { | |
465d5243 | 557 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
558 | return data->header->lseg == NULL; |
559 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
560 | } |
561 | ||
8e821cad | 562 | #else |
8dd37758 | 563 | static void |
ea2cf228 FI |
564 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
565 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
566 | { |
567 | } | |
568 | ||
8dd37758 | 569 | static void |
e468bae9 TM |
570 | nfs_clear_request_commit(struct nfs_page *req) |
571 | { | |
e468bae9 TM |
572 | } |
573 | ||
8e821cad TM |
574 | static inline |
575 | int nfs_write_need_commit(struct nfs_write_data *data) | |
576 | { | |
577 | return 0; | |
578 | } | |
579 | ||
6c75dc0d FI |
580 | #endif |
581 | ||
061ae2ed | 582 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 583 | { |
ea2cf228 | 584 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
585 | unsigned long bytes = 0; |
586 | ||
587 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
588 | goto out; | |
ea2cf228 | 589 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
590 | while (!list_empty(&hdr->pages)) { |
591 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
592 | struct page *page = req->wb_page; | |
593 | ||
594 | bytes += req->wb_bytes; | |
595 | nfs_list_remove_request(req); | |
596 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
597 | (hdr->good_bytes < bytes)) { | |
598 | nfs_set_pageerror(page); | |
599 | nfs_context_set_write_error(req->wb_context, hdr->error); | |
600 | goto remove_req; | |
601 | } | |
602 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
603 | nfs_mark_request_dirty(req); | |
604 | goto next; | |
605 | } | |
606 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
ea2cf228 | 607 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
608 | goto next; |
609 | } | |
610 | remove_req: | |
611 | nfs_inode_remove_request(req); | |
612 | next: | |
613 | nfs_unlock_request(req); | |
614 | nfs_end_page_writeback(page); | |
615 | } | |
616 | out: | |
617 | hdr->release(hdr); | |
8e821cad | 618 | } |
1da177e4 | 619 | |
47c62564 | 620 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
ea2cf228 FI |
621 | static unsigned long |
622 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) | |
fb8a1f11 | 623 | { |
ea2cf228 | 624 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
625 | } |
626 | ||
ea2cf228 | 627 | /* cinfo->lock held by caller */ |
8dd37758 | 628 | static int |
ea2cf228 FI |
629 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
630 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
631 | { |
632 | struct nfs_page *req, *tmp; | |
633 | int ret = 0; | |
634 | ||
635 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
636 | if (!nfs_lock_request(req)) |
637 | continue; | |
ea2cf228 | 638 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 639 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 640 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
641 | nfs_list_add_request(req, dst); |
642 | ret++; | |
643 | if (ret == max) | |
644 | break; | |
d6d6dc7c FI |
645 | } |
646 | return ret; | |
fb8a1f11 TM |
647 | } |
648 | ||
1da177e4 LT |
649 | /* |
650 | * nfs_scan_commit - Scan an inode for commit requests | |
651 | * @inode: NFS inode to scan | |
ea2cf228 FI |
652 | * @dst: mds destination list |
653 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
654 | * |
655 | * Moves requests from the inode's 'commit' request list. | |
656 | * The requests are *not* checked to ensure that they form a contiguous set. | |
657 | */ | |
658 | static int | |
ea2cf228 FI |
659 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
660 | struct nfs_commit_info *cinfo) | |
1da177e4 | 661 | { |
d6d6dc7c | 662 | int ret = 0; |
fb8a1f11 | 663 | |
ea2cf228 FI |
664 | spin_lock(cinfo->lock); |
665 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 666 | const int max = INT_MAX; |
d6d6dc7c | 667 | |
ea2cf228 FI |
668 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
669 | cinfo, max); | |
670 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 671 | } |
ea2cf228 | 672 | spin_unlock(cinfo->lock); |
ff778d02 | 673 | return ret; |
1da177e4 | 674 | } |
d6d6dc7c | 675 | |
c42de9dd | 676 | #else |
ea2cf228 | 677 | static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
678 | { |
679 | return 0; | |
680 | } | |
681 | ||
ea2cf228 FI |
682 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
683 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
684 | { |
685 | return 0; | |
686 | } | |
1da177e4 LT |
687 | #endif |
688 | ||
1da177e4 | 689 | /* |
e7d39069 TM |
690 | * Search for an existing write request, and attempt to update |
691 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 692 | * |
e7d39069 TM |
693 | * If the attempt fails, then the existing request is flushed out |
694 | * to disk. | |
1da177e4 | 695 | */ |
e7d39069 TM |
696 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
697 | struct page *page, | |
698 | unsigned int offset, | |
699 | unsigned int bytes) | |
1da177e4 | 700 | { |
e7d39069 TM |
701 | struct nfs_page *req; |
702 | unsigned int rqend; | |
703 | unsigned int end; | |
704 | int error; | |
705 | ||
706 | if (!PagePrivate(page)) | |
707 | return NULL; | |
1da177e4 LT |
708 | |
709 | end = offset + bytes; | |
e7d39069 | 710 | spin_lock(&inode->i_lock); |
1da177e4 | 711 | |
1da177e4 | 712 | for (;;) { |
277459d2 | 713 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
714 | if (req == NULL) |
715 | goto out_unlock; | |
716 | ||
717 | rqend = req->wb_offset + req->wb_bytes; | |
718 | /* | |
719 | * Tell the caller to flush out the request if | |
720 | * the offsets are non-contiguous. | |
721 | * Note: nfs_flush_incompatible() will already | |
722 | * have flushed out requests having wrong owners. | |
723 | */ | |
e468bae9 | 724 | if (offset > rqend |
e7d39069 TM |
725 | || end < req->wb_offset) |
726 | goto out_flushme; | |
727 | ||
9994b62b | 728 | if (nfs_lock_request_dontget(req)) |
1da177e4 | 729 | break; |
1da177e4 | 730 | |
e7d39069 | 731 | /* The request is locked, so wait and then retry */ |
587142f8 | 732 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
733 | error = nfs_wait_on_request(req); |
734 | nfs_release_request(req); | |
735 | if (error != 0) | |
736 | goto out_err; | |
737 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
738 | } |
739 | ||
740 | /* Okay, the request matches. Update the region */ | |
741 | if (offset < req->wb_offset) { | |
742 | req->wb_offset = offset; | |
743 | req->wb_pgbase = offset; | |
1da177e4 | 744 | } |
1da177e4 LT |
745 | if (end > rqend) |
746 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
747 | else |
748 | req->wb_bytes = rqend - req->wb_offset; | |
749 | out_unlock: | |
750 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
751 | if (req) |
752 | nfs_clear_request_commit(req); | |
e7d39069 TM |
753 | return req; |
754 | out_flushme: | |
755 | spin_unlock(&inode->i_lock); | |
756 | nfs_release_request(req); | |
757 | error = nfs_wb_page(inode, page); | |
758 | out_err: | |
759 | return ERR_PTR(error); | |
760 | } | |
761 | ||
762 | /* | |
763 | * Try to update an existing write request, or create one if there is none. | |
764 | * | |
765 | * Note: Should always be called with the Page Lock held to prevent races | |
766 | * if we have to add a new request. Also assumes that the caller has | |
767 | * already called nfs_flush_incompatible() if necessary. | |
768 | */ | |
769 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
770 | struct page *page, unsigned int offset, unsigned int bytes) | |
771 | { | |
772 | struct inode *inode = page->mapping->host; | |
773 | struct nfs_page *req; | |
1da177e4 | 774 | |
e7d39069 TM |
775 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
776 | if (req != NULL) | |
777 | goto out; | |
778 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
779 | if (IS_ERR(req)) | |
780 | goto out; | |
d6d6dc7c | 781 | nfs_inode_add_request(inode, req); |
efc91ed0 | 782 | out: |
61e930a9 | 783 | return req; |
1da177e4 LT |
784 | } |
785 | ||
e7d39069 TM |
786 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
787 | unsigned int offset, unsigned int count) | |
788 | { | |
789 | struct nfs_page *req; | |
790 | ||
791 | req = nfs_setup_write_request(ctx, page, offset, count); | |
792 | if (IS_ERR(req)) | |
793 | return PTR_ERR(req); | |
794 | /* Update file length */ | |
795 | nfs_grow_file(page, offset, count); | |
796 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 797 | nfs_mark_request_dirty(req); |
9994b62b | 798 | nfs_unlock_request(req); |
e7d39069 TM |
799 | return 0; |
800 | } | |
801 | ||
1da177e4 LT |
802 | int nfs_flush_incompatible(struct file *file, struct page *page) |
803 | { | |
cd3758e3 | 804 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 805 | struct nfs_page *req; |
1a54533e | 806 | int do_flush, status; |
1da177e4 LT |
807 | /* |
808 | * Look for a request corresponding to this page. If there | |
809 | * is one, and it belongs to another file, we flush it out | |
810 | * before we try to copy anything into the page. Do this | |
811 | * due to the lack of an ACCESS-type call in NFSv2. | |
812 | * Also do the same if we find a request from an existing | |
813 | * dropped page. | |
814 | */ | |
1a54533e TM |
815 | do { |
816 | req = nfs_page_find_request(page); | |
817 | if (req == NULL) | |
818 | return 0; | |
f11ac8db TM |
819 | do_flush = req->wb_page != page || req->wb_context != ctx || |
820 | req->wb_lock_context->lockowner != current->files || | |
821 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 822 | nfs_release_request(req); |
1a54533e TM |
823 | if (!do_flush) |
824 | return 0; | |
825 | status = nfs_wb_page(page->mapping->host, page); | |
826 | } while (status == 0); | |
827 | return status; | |
1da177e4 LT |
828 | } |
829 | ||
5d47a356 TM |
830 | /* |
831 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
832 | * the PageUptodate() flag. In this case, we will need to turn off | |
833 | * write optimisations that depend on the page contents being correct. | |
834 | */ | |
835 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
836 | { | |
837 | return PageUptodate(page) && | |
838 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
839 | } | |
840 | ||
1da177e4 LT |
841 | /* |
842 | * Update and possibly write a cached page of an NFS file. | |
843 | * | |
844 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
845 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
846 | */ | |
847 | int nfs_updatepage(struct file *file, struct page *page, | |
848 | unsigned int offset, unsigned int count) | |
849 | { | |
cd3758e3 | 850 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 851 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
852 | int status = 0; |
853 | ||
91d5b470 CL |
854 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
855 | ||
48186c7d | 856 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
857 | file->f_path.dentry->d_parent->d_name.name, |
858 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 859 | (long long)(page_offset(page) + offset)); |
1da177e4 | 860 | |
1da177e4 | 861 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
862 | * is up to date, it may be more efficient to extend the write |
863 | * to cover the entire page in order to avoid fragmentation | |
864 | * inefficiencies. | |
1da177e4 | 865 | */ |
5d47a356 TM |
866 | if (nfs_write_pageuptodate(page, inode) && |
867 | inode->i_flock == NULL && | |
6b2f3d1f | 868 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 869 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 870 | offset = 0; |
1da177e4 LT |
871 | } |
872 | ||
e21195a7 | 873 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
874 | if (status < 0) |
875 | nfs_set_pageerror(page); | |
59b7c05f TM |
876 | else |
877 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 878 | |
48186c7d | 879 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 880 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
881 | return status; |
882 | } | |
883 | ||
3ff7576d | 884 | static int flush_task_priority(int how) |
1da177e4 LT |
885 | { |
886 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
887 | case FLUSH_HIGHPRI: | |
888 | return RPC_PRIORITY_HIGH; | |
889 | case FLUSH_LOWPRI: | |
890 | return RPC_PRIORITY_LOW; | |
891 | } | |
892 | return RPC_PRIORITY_NORMAL; | |
893 | } | |
894 | ||
c5996c4e FI |
895 | int nfs_initiate_write(struct rpc_clnt *clnt, |
896 | struct nfs_write_data *data, | |
d138d5d1 AA |
897 | const struct rpc_call_ops *call_ops, |
898 | int how) | |
1da177e4 | 899 | { |
cd841605 | 900 | struct inode *inode = data->header->inode; |
3ff7576d | 901 | int priority = flush_task_priority(how); |
07737691 | 902 | struct rpc_task *task; |
bdc7f021 TM |
903 | struct rpc_message msg = { |
904 | .rpc_argp = &data->args, | |
905 | .rpc_resp = &data->res, | |
cd841605 | 906 | .rpc_cred = data->header->cred, |
bdc7f021 | 907 | }; |
84115e1c | 908 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 909 | .rpc_client = clnt, |
07737691 | 910 | .task = &data->task, |
bdc7f021 | 911 | .rpc_message = &msg, |
84115e1c TM |
912 | .callback_ops = call_ops, |
913 | .callback_data = data, | |
101070ca | 914 | .workqueue = nfsiod_workqueue, |
2c61be0a | 915 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 916 | .priority = priority, |
84115e1c | 917 | }; |
2c61be0a | 918 | int ret = 0; |
1da177e4 | 919 | |
d138d5d1 AA |
920 | /* Set up the initial task struct. */ |
921 | NFS_PROTO(inode)->write_setup(data, &msg); | |
922 | ||
923 | dprintk("NFS: %5u initiated write call " | |
924 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
925 | data->task.tk_pid, | |
926 | inode->i_sb->s_id, | |
927 | (long long)NFS_FILEID(inode), | |
928 | data->args.count, | |
929 | (unsigned long long)data->args.offset); | |
930 | ||
931 | task = rpc_run_task(&task_setup_data); | |
932 | if (IS_ERR(task)) { | |
933 | ret = PTR_ERR(task); | |
934 | goto out; | |
935 | } | |
936 | if (how & FLUSH_SYNC) { | |
937 | ret = rpc_wait_for_completion_task(task); | |
938 | if (ret == 0) | |
939 | ret = task->tk_status; | |
940 | } | |
941 | rpc_put_task(task); | |
942 | out: | |
943 | return ret; | |
944 | } | |
a69aef14 | 945 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
946 | |
947 | /* | |
948 | * Set up the argument/result storage required for the RPC call. | |
949 | */ | |
6c75dc0d | 950 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 | 951 | unsigned int count, unsigned int offset, |
ea2cf228 | 952 | int how, struct nfs_commit_info *cinfo) |
d138d5d1 | 953 | { |
6c75dc0d | 954 | struct nfs_page *req = data->header->req; |
d138d5d1 | 955 | |
1da177e4 LT |
956 | /* Set up the RPC argument and reply structs |
957 | * NB: take care not to mess about with data->commit et al. */ | |
958 | ||
6c75dc0d | 959 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 960 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
961 | /* pnfs_set_layoutcommit needs this */ |
962 | data->mds_offset = data->args.offset; | |
1da177e4 | 963 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 964 | data->args.pages = data->pages.pagevec; |
1da177e4 | 965 | data->args.count = count; |
383ba719 | 966 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 967 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 968 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
969 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
970 | case 0: | |
971 | break; | |
972 | case FLUSH_COND_STABLE: | |
ea2cf228 | 973 | if (nfs_reqs_to_commit(cinfo)) |
87ed5eb4 TM |
974 | break; |
975 | default: | |
976 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 977 | } |
1da177e4 LT |
978 | |
979 | data->res.fattr = &data->fattr; | |
980 | data->res.count = count; | |
981 | data->res.verf = &data->verf; | |
0e574af1 | 982 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 983 | } |
1da177e4 | 984 | |
6e4efd56 TM |
985 | static int nfs_do_write(struct nfs_write_data *data, |
986 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
987 | int how) |
988 | { | |
cd841605 | 989 | struct inode *inode = data->header->inode; |
0382b744 | 990 | |
c5996c4e | 991 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how); |
1da177e4 LT |
992 | } |
993 | ||
275acaaf TM |
994 | static int nfs_do_multiple_writes(struct list_head *head, |
995 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
996 | int how) |
997 | { | |
998 | struct nfs_write_data *data; | |
999 | int ret = 0; | |
1000 | ||
1001 | while (!list_empty(head)) { | |
1002 | int ret2; | |
1003 | ||
6c75dc0d | 1004 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
1005 | list_del_init(&data->list); |
1006 | ||
dce81290 | 1007 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
1008 | if (ret == 0) |
1009 | ret = ret2; | |
1010 | } | |
1011 | return ret; | |
1012 | } | |
1013 | ||
6d884e8f F |
1014 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1015 | * call this on each, which will prepare them to be retried on next | |
1016 | * writeback using standard nfs. | |
1017 | */ | |
1018 | static void nfs_redirty_request(struct nfs_page *req) | |
1019 | { | |
a6305ddb TM |
1020 | struct page *page = req->wb_page; |
1021 | ||
6d884e8f | 1022 | nfs_mark_request_dirty(req); |
9994b62b | 1023 | nfs_unlock_request(req); |
a6305ddb | 1024 | nfs_end_page_writeback(page); |
6d884e8f F |
1025 | } |
1026 | ||
061ae2ed | 1027 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1028 | { |
1029 | struct nfs_page *req; | |
1030 | ||
1031 | while (!list_empty(head)) { | |
1032 | req = nfs_list_entry(head->next); | |
1033 | nfs_list_remove_request(req); | |
1034 | nfs_redirty_request(req); | |
1035 | } | |
1036 | } | |
1037 | ||
061ae2ed FI |
1038 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1039 | .error_cleanup = nfs_async_write_error, | |
1040 | .completion = nfs_write_completion, | |
1041 | }; | |
1042 | ||
1da177e4 LT |
1043 | /* |
1044 | * Generate multiple small requests to write out a single | |
1045 | * contiguous dirty area on one page. | |
1046 | */ | |
6c75dc0d FI |
1047 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1048 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1049 | { |
6c75dc0d | 1050 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1051 | struct page *page = req->wb_page; |
1052 | struct nfs_write_data *data; | |
d097971d | 1053 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1054 | unsigned int offset; |
1da177e4 | 1055 | int requests = 0; |
dbae4c73 | 1056 | int ret = 0; |
ea2cf228 | 1057 | struct nfs_commit_info cinfo; |
1da177e4 | 1058 | |
ea2cf228 | 1059 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
1da177e4 | 1060 | nfs_list_remove_request(req); |
6c75dc0d | 1061 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1062 | |
b31268ac | 1063 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1064 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || |
b31268ac TM |
1065 | desc->pg_count > wsize)) |
1066 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1067 | ||
1068 | ||
275acaaf | 1069 | offset = 0; |
c76069bd | 1070 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1071 | do { |
1072 | size_t len = min(nbytes, wsize); | |
1073 | ||
6c75dc0d FI |
1074 | data = nfs_writedata_alloc(hdr, 1); |
1075 | if (!data) | |
1da177e4 | 1076 | goto out_bad; |
30dd374f | 1077 | data->pages.pagevec[0] = page; |
ea2cf228 | 1078 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); |
6c75dc0d | 1079 | list_add(&data->list, &hdr->rpc_list); |
1da177e4 | 1080 | requests++; |
e9f7bee1 | 1081 | nbytes -= len; |
275acaaf | 1082 | offset += len; |
e9f7bee1 | 1083 | } while (nbytes != 0); |
6c75dc0d | 1084 | desc->pg_rpc_callops = &nfs_write_common_ops; |
dbae4c73 | 1085 | return ret; |
1da177e4 LT |
1086 | |
1087 | out_bad: | |
6c75dc0d FI |
1088 | while (!list_empty(&hdr->rpc_list)) { |
1089 | data = list_first_entry(&hdr->rpc_list, struct nfs_write_data, list); | |
6e4efd56 | 1090 | list_del(&data->list); |
8ccd271f | 1091 | nfs_writedata_release(data); |
1da177e4 | 1092 | } |
061ae2ed | 1093 | desc->pg_completion_ops->error_cleanup(&hdr->pages); |
1da177e4 LT |
1094 | return -ENOMEM; |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * Create an RPC task for the given write request and kick it. | |
1099 | * The page must have been locked by the caller. | |
1100 | * | |
1101 | * It may happen that the page we're passed is not marked dirty. | |
1102 | * This is the case if nfs_updatepage detects a conflicting request | |
1103 | * that has been written but not committed. | |
1104 | */ | |
6c75dc0d FI |
1105 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1106 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1107 | { |
1108 | struct nfs_page *req; | |
1109 | struct page **pages; | |
1110 | struct nfs_write_data *data; | |
c76069bd | 1111 | struct list_head *head = &desc->pg_list; |
3b609184 | 1112 | int ret = 0; |
ea2cf228 | 1113 | struct nfs_commit_info cinfo; |
1da177e4 | 1114 | |
6c75dc0d FI |
1115 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1116 | desc->pg_count)); | |
1117 | if (!data) { | |
061ae2ed | 1118 | desc->pg_completion_ops->error_cleanup(head); |
44b83799 FI |
1119 | ret = -ENOMEM; |
1120 | goto out; | |
1121 | } | |
6c75dc0d | 1122 | |
ea2cf228 | 1123 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
30dd374f | 1124 | pages = data->pages.pagevec; |
1da177e4 LT |
1125 | while (!list_empty(head)) { |
1126 | req = nfs_list_entry(head->next); | |
1127 | nfs_list_remove_request(req); | |
6c75dc0d | 1128 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1129 | *pages++ = req->wb_page; |
1da177e4 | 1130 | } |
1da177e4 | 1131 | |
b31268ac | 1132 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1133 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) |
b31268ac TM |
1134 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; |
1135 | ||
1da177e4 | 1136 | /* Set up the argument struct */ |
ea2cf228 | 1137 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); |
6c75dc0d FI |
1138 | list_add(&data->list, &hdr->rpc_list); |
1139 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
44b83799 | 1140 | out: |
44b83799 | 1141 | return ret; |
1da177e4 LT |
1142 | } |
1143 | ||
6c75dc0d FI |
1144 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1145 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1146 | { |
1147 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1148 | return nfs_flush_multi(desc, hdr); |
1149 | return nfs_flush_one(desc, hdr); | |
dce81290 TM |
1150 | } |
1151 | ||
1152 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1153 | { |
6c75dc0d FI |
1154 | struct nfs_write_header *whdr; |
1155 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1156 | int ret; |
1157 | ||
6c75dc0d FI |
1158 | whdr = nfs_writehdr_alloc(); |
1159 | if (!whdr) { | |
061ae2ed | 1160 | desc->pg_completion_ops->error_cleanup(&hdr->pages); |
6c75dc0d FI |
1161 | return -ENOMEM; |
1162 | } | |
1163 | hdr = &whdr->header; | |
1164 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1165 | atomic_inc(&hdr->refcnt); | |
1166 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1167 | if (ret == 0) |
6c75dc0d FI |
1168 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1169 | desc->pg_rpc_callops, | |
1170 | desc->pg_ioflags); | |
1171 | else | |
1172 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1173 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1174 | hdr->completion_ops->completion(hdr); |
275acaaf | 1175 | return ret; |
1751c363 | 1176 | } |
1751c363 TM |
1177 | |
1178 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1179 | .pg_test = nfs_generic_pg_test, | |
1180 | .pg_doio = nfs_generic_pg_writepages, | |
1181 | }; | |
1182 | ||
e2fecb21 | 1183 | void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
061ae2ed FI |
1184 | struct inode *inode, int ioflags, |
1185 | const struct nfs_pgio_completion_ops *compl_ops) | |
1da177e4 | 1186 | { |
061ae2ed | 1187 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops, |
1751c363 TM |
1188 | NFS_SERVER(inode)->wsize, ioflags); |
1189 | } | |
1da177e4 | 1190 | |
dce81290 TM |
1191 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1192 | { | |
1193 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1194 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1195 | } | |
1f945357 | 1196 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1197 | |
1751c363 | 1198 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
061ae2ed FI |
1199 | struct inode *inode, int ioflags, |
1200 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 | 1201 | { |
061ae2ed FI |
1202 | if (!pnfs_pageio_init_write(pgio, inode, ioflags, compl_ops)) |
1203 | nfs_pageio_init_write_mds(pgio, inode, ioflags, compl_ops); | |
1da177e4 LT |
1204 | } |
1205 | ||
def6ed7e AA |
1206 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1207 | { | |
1208 | struct nfs_write_data *data = calldata; | |
cd841605 | 1209 | NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); |
def6ed7e | 1210 | } |
def6ed7e | 1211 | |
0b7c0153 FI |
1212 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1213 | { | |
1214 | struct nfs_commit_data *data = calldata; | |
1215 | ||
1216 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1217 | } | |
1218 | ||
1da177e4 LT |
1219 | /* |
1220 | * Handle a write reply that flushes a whole page. | |
1221 | * | |
1222 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1223 | * writebacks since the page->count is kept > 1 for as long | |
1224 | * as the page has a write request pending. | |
1225 | */ | |
6c75dc0d | 1226 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1227 | { |
788e7a89 | 1228 | struct nfs_write_data *data = calldata; |
1da177e4 | 1229 | |
c9d8f89d TM |
1230 | nfs_writeback_done(task, data); |
1231 | } | |
1232 | ||
6c75dc0d | 1233 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1234 | { |
1235 | struct nfs_write_data *data = calldata; | |
cd841605 | 1236 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1237 | int status = data->task.tk_status; |
6c75dc0d | 1238 | struct nfs_page *req = hdr->req; |
788e7a89 | 1239 | |
6c75dc0d FI |
1240 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1241 | spin_lock(&hdr->lock); | |
1242 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1243 | ; /* Do nothing */ | |
1244 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
8e821cad | 1245 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); |
6c75dc0d FI |
1246 | else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) |
1247 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); | |
1248 | spin_unlock(&hdr->lock); | |
1da177e4 | 1249 | } |
cd841605 | 1250 | nfs_writedata_release(data); |
1da177e4 LT |
1251 | } |
1252 | ||
6c75dc0d | 1253 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1254 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1255 | .rpc_call_done = nfs_writeback_done_common, |
1256 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1257 | }; |
1258 | ||
1259 | ||
1da177e4 LT |
1260 | /* |
1261 | * This function is called when the WRITE call is complete. | |
1262 | */ | |
13602896 | 1263 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1264 | { |
1da177e4 LT |
1265 | struct nfs_writeargs *argp = &data->args; |
1266 | struct nfs_writeres *resp = &data->res; | |
cd841605 | 1267 | struct inode *inode = data->header->inode; |
788e7a89 | 1268 | int status; |
1da177e4 | 1269 | |
a3f565b1 | 1270 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1271 | task->tk_pid, task->tk_status); |
1272 | ||
f551e44f CL |
1273 | /* |
1274 | * ->write_done will attempt to use post-op attributes to detect | |
1275 | * conflicting writes by other clients. A strict interpretation | |
1276 | * of close-to-open would allow us to continue caching even if | |
1277 | * another writer had changed the file, but some applications | |
1278 | * depend on tighter cache coherency when writing. | |
1279 | */ | |
cd841605 | 1280 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1281 | if (status != 0) |
13602896 | 1282 | return; |
cd841605 | 1283 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1284 | |
1da177e4 LT |
1285 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1286 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1287 | /* We tried a write call, but the server did not | |
1288 | * commit data to stable storage even though we | |
1289 | * requested it. | |
1290 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1291 | * the server reports NFS_DATA_SYNC, but performs | |
1292 | * NFS_FILE_SYNC. We therefore implement this checking | |
1293 | * as a dprintk() in order to avoid filling syslog. | |
1294 | */ | |
1295 | static unsigned long complain; | |
1296 | ||
a69aef14 | 1297 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1298 | if (time_before(complain, jiffies)) { |
48186c7d | 1299 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1300 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1301 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1302 | resp->verf->committed, argp->stable); |
1303 | complain = jiffies + 300 * HZ; | |
1304 | } | |
1305 | } | |
1306 | #endif | |
6c75dc0d FI |
1307 | if (task->tk_status < 0) |
1308 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); | |
1309 | else if (resp->count < argp->count) { | |
1da177e4 LT |
1310 | static unsigned long complain; |
1311 | ||
6c75dc0d | 1312 | /* This a short write! */ |
cd841605 | 1313 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1314 | |
1da177e4 | 1315 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1316 | if (resp->count == 0) { |
1317 | if (time_before(complain, jiffies)) { | |
1318 | printk(KERN_WARNING | |
1319 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1320 | argp->count); | |
1321 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1322 | } |
6c75dc0d FI |
1323 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1324 | task->tk_status = -EIO; | |
13602896 | 1325 | return; |
1da177e4 | 1326 | } |
6c75dc0d FI |
1327 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1328 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1329 | /* Resend from where the server left off */ | |
1330 | data->mds_offset += resp->count; | |
1331 | argp->offset += resp->count; | |
1332 | argp->pgbase += resp->count; | |
1333 | argp->count -= resp->count; | |
1334 | } else { | |
1335 | /* Resend as a stable write in order to avoid | |
1336 | * headaches in the case of a server crash. | |
1337 | */ | |
1338 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1339 | } |
6c75dc0d | 1340 | rpc_restart_call_prepare(task); |
1da177e4 | 1341 | } |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | ||
1345 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1346 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1347 | { | |
b8413f98 TM |
1348 | int ret; |
1349 | ||
71d0a611 TM |
1350 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1351 | return 1; | |
b8413f98 TM |
1352 | if (!may_wait) |
1353 | return 0; | |
1354 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1355 | NFS_INO_COMMIT, | |
1356 | nfs_wait_bit_killable, | |
1357 | TASK_KILLABLE); | |
1358 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1359 | } |
1360 | ||
e0c2b380 | 1361 | void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1362 | { |
1363 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1364 | smp_mb__after_clear_bit(); | |
1365 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1366 | } | |
e0c2b380 | 1367 | EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); |
71d0a611 | 1368 | |
0b7c0153 | 1369 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1370 | { |
0b7c0153 FI |
1371 | put_nfs_open_context(data->context); |
1372 | nfs_commit_free(data); | |
1da177e4 | 1373 | } |
e0c2b380 | 1374 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1375 | |
0b7c0153 | 1376 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd FI |
1377 | const struct rpc_call_ops *call_ops, |
1378 | int how) | |
1da177e4 | 1379 | { |
07737691 | 1380 | struct rpc_task *task; |
9ace33cd | 1381 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1382 | struct rpc_message msg = { |
1383 | .rpc_argp = &data->args, | |
1384 | .rpc_resp = &data->res, | |
9ace33cd | 1385 | .rpc_cred = data->cred, |
bdc7f021 | 1386 | }; |
84115e1c | 1387 | struct rpc_task_setup task_setup_data = { |
07737691 | 1388 | .task = &data->task, |
9ace33cd | 1389 | .rpc_client = clnt, |
bdc7f021 | 1390 | .rpc_message = &msg, |
9ace33cd | 1391 | .callback_ops = call_ops, |
84115e1c | 1392 | .callback_data = data, |
101070ca | 1393 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1394 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1395 | .priority = priority, |
84115e1c | 1396 | }; |
9ace33cd FI |
1397 | /* Set up the initial task struct. */ |
1398 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1399 | ||
1400 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1401 | ||
1402 | task = rpc_run_task(&task_setup_data); | |
1403 | if (IS_ERR(task)) | |
1404 | return PTR_ERR(task); | |
1405 | if (how & FLUSH_SYNC) | |
1406 | rpc_wait_for_completion_task(task); | |
1407 | rpc_put_task(task); | |
1408 | return 0; | |
1409 | } | |
e0c2b380 | 1410 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1411 | |
1412 | /* | |
1413 | * Set up the argument/result storage required for the RPC call. | |
1414 | */ | |
0b7c0153 | 1415 | void nfs_init_commit(struct nfs_commit_data *data, |
988b6dce FI |
1416 | struct list_head *head, |
1417 | struct pnfs_layout_segment *lseg) | |
9ace33cd FI |
1418 | { |
1419 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1420 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1421 | |
1422 | /* Set up the RPC argument and reply structs | |
1423 | * NB: take care not to mess about with data->commit et al. */ | |
1424 | ||
1425 | list_splice_init(head, &data->pages); | |
1da177e4 | 1426 | |
1da177e4 | 1427 | data->inode = inode; |
9ace33cd | 1428 | data->cred = first->wb_context->cred; |
988b6dce | 1429 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1430 | data->mds_ops = &nfs_commit_ops; |
1da177e4 LT |
1431 | |
1432 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1433 | /* Note: we always request a commit of the entire inode */ |
1434 | data->args.offset = 0; | |
1435 | data->args.count = 0; | |
0b7c0153 | 1436 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1437 | data->res.fattr = &data->fattr; |
1438 | data->res.verf = &data->verf; | |
0e574af1 | 1439 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1440 | } |
e0c2b380 | 1441 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1442 | |
e0c2b380 | 1443 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1444 | struct pnfs_layout_segment *lseg, |
1445 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1446 | { |
1447 | struct nfs_page *req; | |
1448 | ||
1449 | while (!list_empty(page_list)) { | |
1450 | req = nfs_list_entry(page_list->next); | |
1451 | nfs_list_remove_request(req); | |
ea2cf228 | 1452 | nfs_mark_request_commit(req, lseg, cinfo); |
64bfeb49 FI |
1453 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1454 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1455 | BDI_RECLAIMABLE); | |
9994b62b | 1456 | nfs_unlock_request(req); |
64bfeb49 FI |
1457 | } |
1458 | } | |
e0c2b380 | 1459 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1460 | |
1da177e4 LT |
1461 | /* |
1462 | * Commit dirty pages | |
1463 | */ | |
1464 | static int | |
ea2cf228 FI |
1465 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1466 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1467 | { |
0b7c0153 | 1468 | struct nfs_commit_data *data; |
1da177e4 | 1469 | |
c9d8f89d | 1470 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1471 | |
1472 | if (!data) | |
1473 | goto out_bad; | |
1474 | ||
1475 | /* Set up the argument struct */ | |
988b6dce | 1476 | nfs_init_commit(data, head, NULL); |
0b7c0153 | 1477 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, how); |
1da177e4 | 1478 | out_bad: |
ea2cf228 | 1479 | nfs_retry_commit(head, NULL, cinfo); |
71d0a611 | 1480 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1481 | return -ENOMEM; |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * COMMIT call returned | |
1486 | */ | |
788e7a89 | 1487 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1488 | { |
0b7c0153 | 1489 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1490 | |
a3f565b1 | 1491 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1492 | task->tk_pid, task->tk_status); |
1493 | ||
788e7a89 | 1494 | /* Call the NFS version-specific code */ |
c0d0e96b | 1495 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1496 | } |
1497 | ||
0b7c0153 | 1498 | void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1499 | { |
5917ce84 | 1500 | struct nfs_page *req; |
c9d8f89d | 1501 | int status = data->task.tk_status; |
788e7a89 | 1502 | |
1da177e4 LT |
1503 | while (!list_empty(&data->pages)) { |
1504 | req = nfs_list_entry(data->pages.next); | |
1505 | nfs_list_remove_request(req); | |
d6d6dc7c | 1506 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1507 | |
48186c7d | 1508 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1509 | req->wb_context->dentry->d_sb->s_id, |
1510 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1511 | req->wb_bytes, |
1512 | (long long)req_offset(req)); | |
c9d8f89d TM |
1513 | if (status < 0) { |
1514 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1515 | nfs_inode_remove_request(req); |
c9d8f89d | 1516 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1517 | goto next; |
1518 | } | |
1519 | ||
1520 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1521 | * returned by the server against all stored verfs. */ | |
1522 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1523 | /* We have a match */ | |
1524 | nfs_inode_remove_request(req); | |
1525 | dprintk(" OK\n"); | |
1526 | goto next; | |
1527 | } | |
1528 | /* We have a mismatch. Write the page again */ | |
1529 | dprintk(" mismatch\n"); | |
6d884e8f | 1530 | nfs_mark_request_dirty(req); |
1da177e4 | 1531 | next: |
9994b62b | 1532 | nfs_unlock_request(req); |
1da177e4 | 1533 | } |
5917ce84 | 1534 | } |
e0c2b380 | 1535 | EXPORT_SYMBOL_GPL(nfs_commit_release_pages); |
5917ce84 FI |
1536 | |
1537 | static void nfs_commit_release(void *calldata) | |
1538 | { | |
0b7c0153 | 1539 | struct nfs_commit_data *data = calldata; |
5917ce84 FI |
1540 | |
1541 | nfs_commit_release_pages(data); | |
71d0a611 | 1542 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1543 | nfs_commitdata_release(calldata); |
1da177e4 | 1544 | } |
788e7a89 TM |
1545 | |
1546 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1547 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1548 | .rpc_call_done = nfs_commit_done, |
1549 | .rpc_release = nfs_commit_release, | |
1550 | }; | |
1da177e4 | 1551 | |
84c53ab5 | 1552 | static int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
ea2cf228 | 1553 | int how, struct nfs_commit_info *cinfo) |
84c53ab5 FI |
1554 | { |
1555 | int status; | |
1556 | ||
ea2cf228 | 1557 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1558 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1559 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1560 | return status; |
1561 | } | |
1562 | ||
b608b283 | 1563 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1564 | { |
1da177e4 | 1565 | LIST_HEAD(head); |
ea2cf228 | 1566 | struct nfs_commit_info cinfo; |
71d0a611 | 1567 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1568 | int res; |
1da177e4 | 1569 | |
b8413f98 TM |
1570 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1571 | if (res <= 0) | |
c5efa5fc | 1572 | goto out_mark_dirty; |
ea2cf228 FI |
1573 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1574 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1575 | if (res) { |
a861a1e1 FI |
1576 | int error; |
1577 | ||
ea2cf228 | 1578 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1579 | if (error < 0) |
1580 | return error; | |
b8413f98 | 1581 | if (!may_wait) |
c5efa5fc | 1582 | goto out_mark_dirty; |
b8413f98 TM |
1583 | error = wait_on_bit(&NFS_I(inode)->flags, |
1584 | NFS_INO_COMMIT, | |
1585 | nfs_wait_bit_killable, | |
1586 | TASK_KILLABLE); | |
1587 | if (error < 0) | |
1588 | return error; | |
71d0a611 TM |
1589 | } else |
1590 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1591 | return res; |
1592 | /* Note: If we exit without ensuring that the commit is complete, | |
1593 | * we must mark the inode as dirty. Otherwise, future calls to | |
1594 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1595 | * that the data is on the disk. | |
1596 | */ | |
1597 | out_mark_dirty: | |
1598 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1599 | return res; |
1600 | } | |
8fc795f7 TM |
1601 | |
1602 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1603 | { | |
420e3646 TM |
1604 | struct nfs_inode *nfsi = NFS_I(inode); |
1605 | int flags = FLUSH_SYNC; | |
1606 | int ret = 0; | |
8fc795f7 | 1607 | |
3236c3e1 | 1608 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1609 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1610 | return ret; |
1611 | ||
a00dd6c0 JL |
1612 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1613 | /* Don't commit yet if this is a non-blocking flush and there | |
1614 | * are a lot of outstanding writes for this mapping. | |
1615 | */ | |
ea2cf228 | 1616 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1617 | goto out_mark_dirty; |
420e3646 | 1618 | |
a00dd6c0 | 1619 | /* don't wait for the COMMIT response */ |
420e3646 | 1620 | flags = 0; |
a00dd6c0 JL |
1621 | } |
1622 | ||
420e3646 TM |
1623 | ret = nfs_commit_inode(inode, flags); |
1624 | if (ret >= 0) { | |
1625 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1626 | if (ret < wbc->nr_to_write) | |
1627 | wbc->nr_to_write -= ret; | |
1628 | else | |
1629 | wbc->nr_to_write = 0; | |
1630 | } | |
8fc795f7 | 1631 | return 0; |
420e3646 TM |
1632 | } |
1633 | out_mark_dirty: | |
8fc795f7 TM |
1634 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1635 | return ret; | |
1636 | } | |
c63c7b05 | 1637 | #else |
8fc795f7 TM |
1638 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1639 | { | |
1640 | return 0; | |
1641 | } | |
1da177e4 LT |
1642 | #endif |
1643 | ||
8fc795f7 TM |
1644 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1645 | { | |
863a3c6c AA |
1646 | int ret; |
1647 | ||
1648 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1649 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1650 | int status; |
1651 | bool sync = true; | |
863a3c6c | 1652 | |
846d5a09 | 1653 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1654 | sync = false; |
863a3c6c AA |
1655 | |
1656 | status = pnfs_layoutcommit_inode(inode, sync); | |
1657 | if (status < 0) | |
1658 | return status; | |
1659 | } | |
1660 | return ret; | |
8fc795f7 TM |
1661 | } |
1662 | ||
acdc53b2 TM |
1663 | /* |
1664 | * flush the inode to disk. | |
1665 | */ | |
1666 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1667 | { |
1668 | struct writeback_control wbc = { | |
72cb77f4 | 1669 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1670 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1671 | .range_start = 0, |
1672 | .range_end = LLONG_MAX, | |
34901f70 | 1673 | }; |
34901f70 | 1674 | |
acdc53b2 | 1675 | return sync_inode(inode, &wbc); |
1c75950b TM |
1676 | } |
1677 | ||
1b3b4a1a TM |
1678 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1679 | { | |
1680 | struct nfs_page *req; | |
1b3b4a1a TM |
1681 | int ret = 0; |
1682 | ||
1683 | BUG_ON(!PageLocked(page)); | |
1684 | for (;;) { | |
ba8b06e6 | 1685 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1686 | req = nfs_page_find_request(page); |
1687 | if (req == NULL) | |
1b3b4a1a | 1688 | break; |
1b3b4a1a | 1689 | if (nfs_lock_request_dontget(req)) { |
8dd37758 | 1690 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1691 | nfs_inode_remove_request(req); |
1692 | /* | |
1693 | * In case nfs_inode_remove_request has marked the | |
1694 | * page as being dirty | |
1695 | */ | |
1696 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1697 | nfs_unlock_request(req); | |
1698 | break; | |
1699 | } | |
1700 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1701 | nfs_release_request(req); |
1b3b4a1a | 1702 | if (ret < 0) |
c988950e | 1703 | break; |
1b3b4a1a | 1704 | } |
1b3b4a1a TM |
1705 | return ret; |
1706 | } | |
1707 | ||
7f2f12d9 TM |
1708 | /* |
1709 | * Write back all requests on one page - we do this before reading it. | |
1710 | */ | |
1711 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1712 | { |
1713 | loff_t range_start = page_offset(page); | |
1714 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1715 | struct writeback_control wbc = { |
4d770ccf | 1716 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1717 | .nr_to_write = 0, |
4d770ccf TM |
1718 | .range_start = range_start, |
1719 | .range_end = range_end, | |
1720 | }; | |
1721 | int ret; | |
1c75950b | 1722 | |
0522f6ad | 1723 | for (;;) { |
ba8b06e6 | 1724 | wait_on_page_writeback(page); |
73e3302f TM |
1725 | if (clear_page_dirty_for_io(page)) { |
1726 | ret = nfs_writepage_locked(page, &wbc); | |
1727 | if (ret < 0) | |
1728 | goto out_error; | |
0522f6ad | 1729 | continue; |
7f2f12d9 | 1730 | } |
0522f6ad TM |
1731 | if (!PagePrivate(page)) |
1732 | break; | |
1733 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1734 | if (ret < 0) |
73e3302f | 1735 | goto out_error; |
7f2f12d9 | 1736 | } |
73e3302f TM |
1737 | return 0; |
1738 | out_error: | |
4d770ccf | 1739 | return ret; |
1c75950b TM |
1740 | } |
1741 | ||
074cc1de TM |
1742 | #ifdef CONFIG_MIGRATION |
1743 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1744 | struct page *page, enum migrate_mode mode) |
074cc1de | 1745 | { |
2da95652 JL |
1746 | /* |
1747 | * If PagePrivate is set, then the page is currently associated with | |
1748 | * an in-progress read or write request. Don't try to migrate it. | |
1749 | * | |
1750 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1751 | * that we can safely release the inode reference while holding | |
1752 | * the page lock. | |
1753 | */ | |
1754 | if (PagePrivate(page)) | |
1755 | return -EBUSY; | |
074cc1de | 1756 | |
7549ad5f | 1757 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1758 | |
a6bc32b8 | 1759 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1760 | } |
1761 | #endif | |
1762 | ||
f7b422b1 | 1763 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1764 | { |
1765 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1766 | sizeof(struct nfs_write_header), |
1da177e4 | 1767 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1768 | NULL); |
1da177e4 LT |
1769 | if (nfs_wdata_cachep == NULL) |
1770 | return -ENOMEM; | |
1771 | ||
93d2341c MD |
1772 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1773 | nfs_wdata_cachep); | |
1da177e4 LT |
1774 | if (nfs_wdata_mempool == NULL) |
1775 | return -ENOMEM; | |
1776 | ||
0b7c0153 FI |
1777 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1778 | sizeof(struct nfs_commit_data), | |
1779 | 0, SLAB_HWCACHE_ALIGN, | |
1780 | NULL); | |
1781 | if (nfs_cdata_cachep == NULL) | |
1782 | return -ENOMEM; | |
1783 | ||
93d2341c MD |
1784 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1785 | nfs_wdata_cachep); | |
1da177e4 LT |
1786 | if (nfs_commit_mempool == NULL) |
1787 | return -ENOMEM; | |
1788 | ||
89a09141 PZ |
1789 | /* |
1790 | * NFS congestion size, scale with available memory. | |
1791 | * | |
1792 | * 64MB: 8192k | |
1793 | * 128MB: 11585k | |
1794 | * 256MB: 16384k | |
1795 | * 512MB: 23170k | |
1796 | * 1GB: 32768k | |
1797 | * 2GB: 46340k | |
1798 | * 4GB: 65536k | |
1799 | * 8GB: 92681k | |
1800 | * 16GB: 131072k | |
1801 | * | |
1802 | * This allows larger machines to have larger/more transfers. | |
1803 | * Limit the default to 256M | |
1804 | */ | |
1805 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1806 | if (nfs_congestion_kb > 256*1024) | |
1807 | nfs_congestion_kb = 256*1024; | |
1808 | ||
1da177e4 LT |
1809 | return 0; |
1810 | } | |
1811 | ||
266bee88 | 1812 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1813 | { |
1814 | mempool_destroy(nfs_commit_mempool); | |
1815 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1816 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1817 | } |
1818 |