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 | 33 | |
f4ce1299 TM |
34 | #include "nfstrace.h" |
35 | ||
1da177e4 LT |
36 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
37 | ||
38 | #define MIN_POOL_WRITE (32) | |
39 | #define MIN_POOL_COMMIT (4) | |
40 | ||
41 | /* | |
42 | * Local function declarations | |
43 | */ | |
f8512ad0 | 44 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 45 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 46 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 47 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 48 | static const struct nfs_commit_completion_ops nfs_commit_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 | { |
192e501b | 57 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
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 | { |
192e501b | 75 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
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); | |
9bce008b | 85 | hdr->verf = &p->verf; |
3feb2d49 TM |
86 | } |
87 | return p; | |
88 | } | |
89d77c8f | 89 | EXPORT_SYMBOL_GPL(nfs_writehdr_alloc); |
3feb2d49 | 90 | |
1763da12 FI |
91 | static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
92 | unsigned int pagecount) | |
6c75dc0d FI |
93 | { |
94 | struct nfs_write_data *data, *prealloc; | |
95 | ||
96 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
97 | if (prealloc->header == NULL) | |
98 | data = prealloc; | |
99 | else | |
100 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
101 | if (!data) | |
102 | goto out; | |
103 | ||
104 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
105 | data->header = hdr; | |
106 | atomic_inc(&hdr->refcnt); | |
107 | } else { | |
108 | if (data != prealloc) | |
109 | kfree(data); | |
110 | data = NULL; | |
111 | } | |
112 | out: | |
113 | return data; | |
114 | } | |
115 | ||
cd841605 | 116 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 117 | { |
cd841605 FI |
118 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
119 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 | 120 | } |
89d77c8f | 121 | EXPORT_SYMBOL_GPL(nfs_writehdr_free); |
3feb2d49 | 122 | |
dce81290 | 123 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 124 | { |
6c75dc0d FI |
125 | struct nfs_pgio_header *hdr = wdata->header; |
126 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
127 | ||
383ba719 | 128 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
129 | if (wdata->pages.pagevec != wdata->pages.page_array) |
130 | kfree(wdata->pages.pagevec); | |
6db6dd7d | 131 | if (wdata == &write_header->rpc_data) { |
6c75dc0d | 132 | wdata->header = NULL; |
6db6dd7d TM |
133 | wdata = NULL; |
134 | } | |
6c75dc0d | 135 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 136 | hdr->completion_ops->completion(hdr); |
6db6dd7d TM |
137 | /* Note: we only free the rpc_task after callbacks are done. |
138 | * See the comment in rpc_free_task() for why | |
139 | */ | |
140 | kfree(wdata); | |
1da177e4 | 141 | } |
89d77c8f | 142 | EXPORT_SYMBOL_GPL(nfs_writedata_release); |
1da177e4 | 143 | |
7b159fc1 TM |
144 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
145 | { | |
146 | ctx->error = error; | |
147 | smp_wmb(); | |
148 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
149 | } | |
150 | ||
29418aa4 MG |
151 | static struct nfs_page * |
152 | nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) | |
277459d2 TM |
153 | { |
154 | struct nfs_page *req = NULL; | |
155 | ||
29418aa4 | 156 | if (PagePrivate(page)) |
277459d2 | 157 | req = (struct nfs_page *)page_private(page); |
29418aa4 MG |
158 | else if (unlikely(PageSwapCache(page))) { |
159 | struct nfs_page *freq, *t; | |
160 | ||
161 | /* Linearly search the commit list for the correct req */ | |
162 | list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { | |
163 | if (freq->wb_page == page) { | |
164 | req = freq; | |
165 | break; | |
166 | } | |
167 | } | |
277459d2 | 168 | } |
29418aa4 MG |
169 | |
170 | if (req) | |
171 | kref_get(&req->wb_kref); | |
172 | ||
277459d2 TM |
173 | return req; |
174 | } | |
175 | ||
176 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
177 | { | |
d56b4ddf | 178 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 179 | struct nfs_page *req = NULL; |
277459d2 | 180 | |
587142f8 | 181 | spin_lock(&inode->i_lock); |
29418aa4 | 182 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
587142f8 | 183 | spin_unlock(&inode->i_lock); |
277459d2 TM |
184 | return req; |
185 | } | |
186 | ||
1da177e4 LT |
187 | /* Adjust the file length if we're writing beyond the end */ |
188 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
189 | { | |
d56b4ddf | 190 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
191 | loff_t end, i_size; |
192 | pgoff_t end_index; | |
1da177e4 | 193 | |
a3d01454 TM |
194 | spin_lock(&inode->i_lock); |
195 | i_size = i_size_read(inode); | |
196 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 197 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 198 | goto out; |
d56b4ddf | 199 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 200 | if (i_size >= end) |
a3d01454 | 201 | goto out; |
1da177e4 | 202 | i_size_write(inode, end); |
a3d01454 TM |
203 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
204 | out: | |
205 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
206 | } |
207 | ||
a301b777 TM |
208 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
209 | static void nfs_set_pageerror(struct page *page) | |
210 | { | |
d56b4ddf | 211 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
212 | } |
213 | ||
1da177e4 LT |
214 | /* We can set the PG_uptodate flag if we see that a write request |
215 | * covers the full page. | |
216 | */ | |
217 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
218 | { | |
1da177e4 LT |
219 | if (PageUptodate(page)) |
220 | return; | |
221 | if (base != 0) | |
222 | return; | |
49a70f27 | 223 | if (count != nfs_page_length(page)) |
1da177e4 | 224 | return; |
49a70f27 | 225 | SetPageUptodate(page); |
1da177e4 LT |
226 | } |
227 | ||
1da177e4 LT |
228 | static int wb_priority(struct writeback_control *wbc) |
229 | { | |
230 | if (wbc->for_reclaim) | |
c63c7b05 | 231 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 232 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
233 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
234 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
235 | } |
236 | ||
89a09141 PZ |
237 | /* |
238 | * NFS congestion control | |
239 | */ | |
240 | ||
241 | int nfs_congestion_kb; | |
242 | ||
243 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
244 | #define NFS_CONGESTION_OFF_THRESH \ | |
245 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
246 | ||
deed85e7 | 247 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 248 | { |
deed85e7 | 249 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
250 | int ret = test_set_page_writeback(page); |
251 | ||
deed85e7 | 252 | WARN_ON_ONCE(ret != 0); |
89a09141 | 253 | |
deed85e7 TM |
254 | if (atomic_long_inc_return(&nfss->writeback) > |
255 | NFS_CONGESTION_ON_THRESH) { | |
256 | set_bdi_congested(&nfss->backing_dev_info, | |
257 | BLK_RW_ASYNC); | |
89a09141 PZ |
258 | } |
259 | } | |
260 | ||
261 | static void nfs_end_page_writeback(struct page *page) | |
262 | { | |
d56b4ddf | 263 | struct inode *inode = page_file_mapping(page)->host; |
89a09141 PZ |
264 | struct nfs_server *nfss = NFS_SERVER(inode); |
265 | ||
266 | end_page_writeback(page); | |
c4dc4bee | 267 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 268 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
269 | } |
270 | ||
cfb506e1 | 271 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 272 | { |
d56b4ddf | 273 | struct inode *inode = page_file_mapping(page)->host; |
e261f51f | 274 | struct nfs_page *req; |
e261f51f TM |
275 | int ret; |
276 | ||
587142f8 | 277 | spin_lock(&inode->i_lock); |
074cc1de | 278 | for (;;) { |
29418aa4 | 279 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
074cc1de TM |
280 | if (req == NULL) |
281 | break; | |
7ad84aa9 | 282 | if (nfs_lock_request(req)) |
e261f51f TM |
283 | break; |
284 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 285 | * then the call to nfs_lock_request() will always |
e261f51f TM |
286 | * succeed provided that someone hasn't already marked the |
287 | * request as dirty (in which case we don't care). | |
288 | */ | |
587142f8 | 289 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
290 | if (!nonblock) |
291 | ret = nfs_wait_on_request(req); | |
292 | else | |
293 | ret = -EAGAIN; | |
e261f51f TM |
294 | nfs_release_request(req); |
295 | if (ret != 0) | |
074cc1de | 296 | return ERR_PTR(ret); |
587142f8 | 297 | spin_lock(&inode->i_lock); |
e261f51f | 298 | } |
587142f8 | 299 | spin_unlock(&inode->i_lock); |
074cc1de TM |
300 | return req; |
301 | } | |
302 | ||
303 | /* | |
304 | * Find an associated nfs write request, and prepare to flush it out | |
305 | * May return an error if the user signalled nfs_wait_on_request(). | |
306 | */ | |
307 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 308 | struct page *page, bool nonblock) |
074cc1de TM |
309 | { |
310 | struct nfs_page *req; | |
311 | int ret = 0; | |
312 | ||
cfb506e1 | 313 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
314 | if (!req) |
315 | goto out; | |
316 | ret = PTR_ERR(req); | |
317 | if (IS_ERR(req)) | |
318 | goto out; | |
319 | ||
deed85e7 TM |
320 | nfs_set_page_writeback(page); |
321 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 322 | |
deed85e7 | 323 | ret = 0; |
f8512ad0 FI |
324 | if (!nfs_pageio_add_request(pgio, req)) { |
325 | nfs_redirty_request(req); | |
074cc1de | 326 | ret = pgio->pg_error; |
f8512ad0 | 327 | } |
074cc1de TM |
328 | out: |
329 | return ret; | |
e261f51f TM |
330 | } |
331 | ||
f758c885 | 332 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 333 | { |
d56b4ddf | 334 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 335 | int ret; |
1da177e4 | 336 | |
91d5b470 CL |
337 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
338 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
339 | ||
d56b4ddf | 340 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 341 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
342 | if (ret == -EAGAIN) { |
343 | redirty_page_for_writepage(wbc, page); | |
344 | ret = 0; | |
345 | } | |
346 | return ret; | |
f758c885 | 347 | } |
7fe7f848 | 348 | |
f758c885 TM |
349 | /* |
350 | * Write an mmapped page to the server. | |
351 | */ | |
352 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
353 | { | |
354 | struct nfs_pageio_descriptor pgio; | |
355 | int err; | |
49a70f27 | 356 | |
d56b4ddf | 357 | NFS_PROTO(page_file_mapping(page)->host)->write_pageio_init(&pgio, |
57208fa7 BS |
358 | page->mapping->host, |
359 | wb_priority(wbc), | |
360 | &nfs_async_write_completion_ops); | |
f758c885 TM |
361 | err = nfs_do_writepage(page, wbc, &pgio); |
362 | nfs_pageio_complete(&pgio); | |
363 | if (err < 0) | |
364 | return err; | |
365 | if (pgio.pg_error < 0) | |
366 | return pgio.pg_error; | |
367 | return 0; | |
4d770ccf TM |
368 | } |
369 | ||
370 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
371 | { | |
f758c885 | 372 | int ret; |
4d770ccf | 373 | |
f758c885 | 374 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 375 | unlock_page(page); |
f758c885 TM |
376 | return ret; |
377 | } | |
378 | ||
379 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
380 | { | |
381 | int ret; | |
382 | ||
383 | ret = nfs_do_writepage(page, wbc, data); | |
384 | unlock_page(page); | |
385 | return ret; | |
1da177e4 LT |
386 | } |
387 | ||
1da177e4 LT |
388 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
389 | { | |
1da177e4 | 390 | struct inode *inode = mapping->host; |
72cb77f4 | 391 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 392 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
393 | int err; |
394 | ||
72cb77f4 TM |
395 | /* Stop dirtying of new pages while we sync */ |
396 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
397 | nfs_wait_bit_killable, TASK_KILLABLE); | |
398 | if (err) | |
399 | goto out_err; | |
400 | ||
91d5b470 CL |
401 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
402 | ||
57208fa7 | 403 | NFS_PROTO(inode)->write_pageio_init(&pgio, inode, wb_priority(wbc), &nfs_async_write_completion_ops); |
f758c885 | 404 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 405 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
406 | |
407 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
408 | smp_mb__after_clear_bit(); | |
409 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
410 | ||
f758c885 | 411 | if (err < 0) |
72cb77f4 TM |
412 | goto out_err; |
413 | err = pgio.pg_error; | |
414 | if (err < 0) | |
415 | goto out_err; | |
c63c7b05 | 416 | return 0; |
72cb77f4 TM |
417 | out_err: |
418 | return err; | |
1da177e4 LT |
419 | } |
420 | ||
421 | /* | |
422 | * Insert a write request into an inode | |
423 | */ | |
d6d6dc7c | 424 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
425 | { |
426 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
427 | |
428 | /* Lock the request! */ | |
7ad84aa9 | 429 | nfs_lock_request(req); |
e7d39069 TM |
430 | |
431 | spin_lock(&inode->i_lock); | |
011e2a7f | 432 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 433 | inode->i_version++; |
29418aa4 MG |
434 | /* |
435 | * Swap-space should not get truncated. Hence no need to plug the race | |
436 | * with invalidate/truncate. | |
437 | */ | |
438 | if (likely(!PageSwapCache(req->wb_page))) { | |
439 | set_bit(PG_MAPPED, &req->wb_flags); | |
440 | SetPagePrivate(req->wb_page); | |
441 | set_page_private(req->wb_page, (unsigned long)req); | |
442 | } | |
1da177e4 | 443 | nfsi->npages++; |
c03b4024 | 444 | kref_get(&req->wb_kref); |
e7d39069 | 445 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
446 | } |
447 | ||
448 | /* | |
89a09141 | 449 | * Remove a write request from an inode |
1da177e4 LT |
450 | */ |
451 | static void nfs_inode_remove_request(struct nfs_page *req) | |
452 | { | |
3d4ff43d | 453 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
454 | struct nfs_inode *nfsi = NFS_I(inode); |
455 | ||
587142f8 | 456 | spin_lock(&inode->i_lock); |
29418aa4 MG |
457 | if (likely(!PageSwapCache(req->wb_page))) { |
458 | set_page_private(req->wb_page, 0); | |
459 | ClearPagePrivate(req->wb_page); | |
460 | clear_bit(PG_MAPPED, &req->wb_flags); | |
461 | } | |
1da177e4 | 462 | nfsi->npages--; |
4d65c520 | 463 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
464 | nfs_release_request(req); |
465 | } | |
466 | ||
61822ab5 | 467 | static void |
6d884e8f | 468 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 469 | { |
61822ab5 TM |
470 | __set_page_dirty_nobuffers(req->wb_page); |
471 | } | |
472 | ||
89d77c8f | 473 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
8dd37758 TM |
474 | /** |
475 | * nfs_request_add_commit_list - add request to a commit list | |
476 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
477 | * @dst: commit list head |
478 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 479 | * |
ea2cf228 | 480 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
481 | * number of outstanding requests requiring a commit as well as |
482 | * the MM page stats. | |
483 | * | |
ea2cf228 | 484 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 485 | * holding the nfs_page lock. |
1da177e4 | 486 | */ |
8dd37758 | 487 | void |
ea2cf228 FI |
488 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
489 | struct nfs_commit_info *cinfo) | |
1da177e4 | 490 | { |
e468bae9 | 491 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
492 | spin_lock(cinfo->lock); |
493 | nfs_list_add_request(req, dst); | |
494 | cinfo->mds->ncommit++; | |
495 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
496 | if (!cinfo->dreq) { |
497 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 498 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
499 | BDI_RECLAIMABLE); |
500 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
501 | I_DIRTY_DATASYNC); | |
502 | } | |
1da177e4 | 503 | } |
8dd37758 TM |
504 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
505 | ||
506 | /** | |
507 | * nfs_request_remove_commit_list - Remove request from a commit list | |
508 | * @req: pointer to a nfs_page | |
ea2cf228 | 509 | * @cinfo: holds list lock and accounting info |
8dd37758 | 510 | * |
ea2cf228 | 511 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
512 | * number of outstanding requests requiring a commit |
513 | * It does not update the MM page stats. | |
514 | * | |
ea2cf228 | 515 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
516 | */ |
517 | void | |
ea2cf228 FI |
518 | nfs_request_remove_commit_list(struct nfs_page *req, |
519 | struct nfs_commit_info *cinfo) | |
8dd37758 | 520 | { |
8dd37758 TM |
521 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
522 | return; | |
523 | nfs_list_remove_request(req); | |
ea2cf228 | 524 | cinfo->mds->ncommit--; |
8dd37758 TM |
525 | } |
526 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
527 | ||
ea2cf228 FI |
528 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
529 | struct inode *inode) | |
530 | { | |
531 | cinfo->lock = &inode->i_lock; | |
532 | cinfo->mds = &NFS_I(inode)->commit_info; | |
533 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 534 | cinfo->dreq = NULL; |
f453a54a | 535 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
536 | } |
537 | ||
538 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
539 | struct inode *inode, | |
540 | struct nfs_direct_req *dreq) | |
541 | { | |
1763da12 FI |
542 | if (dreq) |
543 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
544 | else | |
545 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
546 | } |
547 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
548 | |
549 | /* | |
550 | * Add a request to the inode's commit list. | |
551 | */ | |
1763da12 | 552 | void |
ea2cf228 FI |
553 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
554 | struct nfs_commit_info *cinfo) | |
8dd37758 | 555 | { |
ea2cf228 | 556 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 557 | return; |
ea2cf228 | 558 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 559 | } |
8e821cad | 560 | |
d6d6dc7c FI |
561 | static void |
562 | nfs_clear_page_commit(struct page *page) | |
563 | { | |
564 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 565 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
566 | } |
567 | ||
8dd37758 | 568 | static void |
e468bae9 TM |
569 | nfs_clear_request_commit(struct nfs_page *req) |
570 | { | |
8dd37758 TM |
571 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
572 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 573 | struct nfs_commit_info cinfo; |
e468bae9 | 574 | |
ea2cf228 FI |
575 | nfs_init_cinfo_from_inode(&cinfo, inode); |
576 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
577 | spin_lock(cinfo.lock); | |
578 | nfs_request_remove_commit_list(req, &cinfo); | |
579 | spin_unlock(cinfo.lock); | |
8dd37758 | 580 | } |
d6d6dc7c | 581 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 582 | } |
e468bae9 TM |
583 | } |
584 | ||
8e821cad TM |
585 | static inline |
586 | int nfs_write_need_commit(struct nfs_write_data *data) | |
587 | { | |
465d5243 | 588 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
589 | return data->header->lseg == NULL; |
590 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
591 | } |
592 | ||
8e821cad | 593 | #else |
68cd6fa4 BS |
594 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
595 | struct inode *inode) | |
596 | { | |
597 | } | |
598 | ||
599 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
600 | struct inode *inode, | |
601 | struct nfs_direct_req *dreq) | |
602 | { | |
603 | } | |
604 | ||
1763da12 | 605 | void |
ea2cf228 FI |
606 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
607 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
608 | { |
609 | } | |
610 | ||
8dd37758 | 611 | static void |
e468bae9 TM |
612 | nfs_clear_request_commit(struct nfs_page *req) |
613 | { | |
e468bae9 TM |
614 | } |
615 | ||
8e821cad TM |
616 | static inline |
617 | int nfs_write_need_commit(struct nfs_write_data *data) | |
618 | { | |
619 | return 0; | |
620 | } | |
621 | ||
6c75dc0d FI |
622 | #endif |
623 | ||
061ae2ed | 624 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 625 | { |
ea2cf228 | 626 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
627 | unsigned long bytes = 0; |
628 | ||
629 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
630 | goto out; | |
ea2cf228 | 631 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
632 | while (!list_empty(&hdr->pages)) { |
633 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
634 | |
635 | bytes += req->wb_bytes; | |
636 | nfs_list_remove_request(req); | |
637 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
638 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 639 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
640 | nfs_context_set_write_error(req->wb_context, hdr->error); |
641 | goto remove_req; | |
642 | } | |
643 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
644 | nfs_mark_request_dirty(req); | |
645 | goto next; | |
646 | } | |
647 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
2f2c63bc | 648 | memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf)); |
ea2cf228 | 649 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
650 | goto next; |
651 | } | |
652 | remove_req: | |
653 | nfs_inode_remove_request(req); | |
654 | next: | |
1d1afcbc | 655 | nfs_unlock_request(req); |
d1182b33 | 656 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 657 | nfs_release_request(req); |
6c75dc0d FI |
658 | } |
659 | out: | |
660 | hdr->release(hdr); | |
8e821cad | 661 | } |
1da177e4 | 662 | |
89d77c8f | 663 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
ea2cf228 FI |
664 | static unsigned long |
665 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) | |
fb8a1f11 | 666 | { |
ea2cf228 | 667 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
668 | } |
669 | ||
ea2cf228 | 670 | /* cinfo->lock held by caller */ |
1763da12 | 671 | int |
ea2cf228 FI |
672 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
673 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
674 | { |
675 | struct nfs_page *req, *tmp; | |
676 | int ret = 0; | |
677 | ||
678 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
679 | if (!nfs_lock_request(req)) |
680 | continue; | |
7ad84aa9 | 681 | kref_get(&req->wb_kref); |
ea2cf228 | 682 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 683 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 684 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
685 | nfs_list_add_request(req, dst); |
686 | ret++; | |
1763da12 | 687 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 688 | break; |
d6d6dc7c FI |
689 | } |
690 | return ret; | |
fb8a1f11 TM |
691 | } |
692 | ||
1da177e4 LT |
693 | /* |
694 | * nfs_scan_commit - Scan an inode for commit requests | |
695 | * @inode: NFS inode to scan | |
ea2cf228 FI |
696 | * @dst: mds destination list |
697 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
698 | * |
699 | * Moves requests from the inode's 'commit' request list. | |
700 | * The requests are *not* checked to ensure that they form a contiguous set. | |
701 | */ | |
1763da12 | 702 | int |
ea2cf228 FI |
703 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
704 | struct nfs_commit_info *cinfo) | |
1da177e4 | 705 | { |
d6d6dc7c | 706 | int ret = 0; |
fb8a1f11 | 707 | |
ea2cf228 FI |
708 | spin_lock(cinfo->lock); |
709 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 710 | const int max = INT_MAX; |
d6d6dc7c | 711 | |
ea2cf228 FI |
712 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
713 | cinfo, max); | |
714 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 715 | } |
ea2cf228 | 716 | spin_unlock(cinfo->lock); |
ff778d02 | 717 | return ret; |
1da177e4 | 718 | } |
d6d6dc7c | 719 | |
c42de9dd | 720 | #else |
ea2cf228 | 721 | static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
722 | { |
723 | return 0; | |
724 | } | |
725 | ||
1763da12 FI |
726 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
727 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
728 | { |
729 | return 0; | |
730 | } | |
1da177e4 LT |
731 | #endif |
732 | ||
1da177e4 | 733 | /* |
e7d39069 TM |
734 | * Search for an existing write request, and attempt to update |
735 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 736 | * |
e7d39069 TM |
737 | * If the attempt fails, then the existing request is flushed out |
738 | * to disk. | |
1da177e4 | 739 | */ |
e7d39069 TM |
740 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
741 | struct page *page, | |
742 | unsigned int offset, | |
743 | unsigned int bytes) | |
1da177e4 | 744 | { |
e7d39069 TM |
745 | struct nfs_page *req; |
746 | unsigned int rqend; | |
747 | unsigned int end; | |
748 | int error; | |
749 | ||
750 | if (!PagePrivate(page)) | |
751 | return NULL; | |
1da177e4 LT |
752 | |
753 | end = offset + bytes; | |
e7d39069 | 754 | spin_lock(&inode->i_lock); |
1da177e4 | 755 | |
1da177e4 | 756 | for (;;) { |
29418aa4 | 757 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
e7d39069 TM |
758 | if (req == NULL) |
759 | goto out_unlock; | |
760 | ||
761 | rqend = req->wb_offset + req->wb_bytes; | |
762 | /* | |
763 | * Tell the caller to flush out the request if | |
764 | * the offsets are non-contiguous. | |
765 | * Note: nfs_flush_incompatible() will already | |
766 | * have flushed out requests having wrong owners. | |
767 | */ | |
e468bae9 | 768 | if (offset > rqend |
e7d39069 TM |
769 | || end < req->wb_offset) |
770 | goto out_flushme; | |
771 | ||
7ad84aa9 | 772 | if (nfs_lock_request(req)) |
1da177e4 | 773 | break; |
1da177e4 | 774 | |
e7d39069 | 775 | /* The request is locked, so wait and then retry */ |
587142f8 | 776 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
777 | error = nfs_wait_on_request(req); |
778 | nfs_release_request(req); | |
779 | if (error != 0) | |
780 | goto out_err; | |
781 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
782 | } |
783 | ||
784 | /* Okay, the request matches. Update the region */ | |
785 | if (offset < req->wb_offset) { | |
786 | req->wb_offset = offset; | |
787 | req->wb_pgbase = offset; | |
1da177e4 | 788 | } |
1da177e4 LT |
789 | if (end > rqend) |
790 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
791 | else |
792 | req->wb_bytes = rqend - req->wb_offset; | |
793 | out_unlock: | |
794 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
795 | if (req) |
796 | nfs_clear_request_commit(req); | |
e7d39069 TM |
797 | return req; |
798 | out_flushme: | |
799 | spin_unlock(&inode->i_lock); | |
800 | nfs_release_request(req); | |
801 | error = nfs_wb_page(inode, page); | |
802 | out_err: | |
803 | return ERR_PTR(error); | |
804 | } | |
805 | ||
806 | /* | |
807 | * Try to update an existing write request, or create one if there is none. | |
808 | * | |
809 | * Note: Should always be called with the Page Lock held to prevent races | |
810 | * if we have to add a new request. Also assumes that the caller has | |
811 | * already called nfs_flush_incompatible() if necessary. | |
812 | */ | |
813 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
814 | struct page *page, unsigned int offset, unsigned int bytes) | |
815 | { | |
d56b4ddf | 816 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 817 | struct nfs_page *req; |
1da177e4 | 818 | |
e7d39069 TM |
819 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
820 | if (req != NULL) | |
821 | goto out; | |
822 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
823 | if (IS_ERR(req)) | |
824 | goto out; | |
d6d6dc7c | 825 | nfs_inode_add_request(inode, req); |
efc91ed0 | 826 | out: |
61e930a9 | 827 | return req; |
1da177e4 LT |
828 | } |
829 | ||
e7d39069 TM |
830 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
831 | unsigned int offset, unsigned int count) | |
832 | { | |
833 | struct nfs_page *req; | |
834 | ||
835 | req = nfs_setup_write_request(ctx, page, offset, count); | |
836 | if (IS_ERR(req)) | |
837 | return PTR_ERR(req); | |
838 | /* Update file length */ | |
839 | nfs_grow_file(page, offset, count); | |
840 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 841 | nfs_mark_request_dirty(req); |
1d1afcbc | 842 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
843 | return 0; |
844 | } | |
845 | ||
1da177e4 LT |
846 | int nfs_flush_incompatible(struct file *file, struct page *page) |
847 | { | |
cd3758e3 | 848 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 849 | struct nfs_lock_context *l_ctx; |
1da177e4 | 850 | struct nfs_page *req; |
1a54533e | 851 | int do_flush, status; |
1da177e4 LT |
852 | /* |
853 | * Look for a request corresponding to this page. If there | |
854 | * is one, and it belongs to another file, we flush it out | |
855 | * before we try to copy anything into the page. Do this | |
856 | * due to the lack of an ACCESS-type call in NFSv2. | |
857 | * Also do the same if we find a request from an existing | |
858 | * dropped page. | |
859 | */ | |
1a54533e TM |
860 | do { |
861 | req = nfs_page_find_request(page); | |
862 | if (req == NULL) | |
863 | return 0; | |
2a369153 TM |
864 | l_ctx = req->wb_lock_context; |
865 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
0f1d2605 | 866 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
867 | do_flush |= l_ctx->lockowner.l_owner != current->files |
868 | || l_ctx->lockowner.l_pid != current->tgid; | |
869 | } | |
1da177e4 | 870 | nfs_release_request(req); |
1a54533e TM |
871 | if (!do_flush) |
872 | return 0; | |
d56b4ddf | 873 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
874 | } while (status == 0); |
875 | return status; | |
1da177e4 LT |
876 | } |
877 | ||
dc24826b AA |
878 | /* |
879 | * Avoid buffered writes when a open context credential's key would | |
880 | * expire soon. | |
881 | * | |
882 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
883 | * | |
884 | * Return 0 and set a credential flag which triggers the inode to flush | |
885 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
886 | * RPC_KEY_EXPIRE_TIMEO. | |
887 | */ | |
888 | int | |
889 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
890 | { | |
891 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
892 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
893 | ||
894 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
895 | } | |
896 | ||
897 | /* | |
898 | * Test if the open context credential key is marked to expire soon. | |
899 | */ | |
900 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
901 | { | |
902 | return rpcauth_cred_key_to_expire(ctx->cred); | |
903 | } | |
904 | ||
5d47a356 TM |
905 | /* |
906 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
907 | * the PageUptodate() flag. In this case, we will need to turn off | |
908 | * write optimisations that depend on the page contents being correct. | |
909 | */ | |
8d197a56 | 910 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 911 | { |
8d197a56 TM |
912 | if (nfs_have_delegated_attributes(inode)) |
913 | goto out; | |
81d9bce5 | 914 | if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) |
8d197a56 TM |
915 | return false; |
916 | out: | |
917 | return PageUptodate(page) != 0; | |
5d47a356 TM |
918 | } |
919 | ||
c7559663 SM |
920 | /* If we know the page is up to date, and we're not using byte range locks (or |
921 | * if we have the whole file locked for writing), it may be more efficient to | |
922 | * extend the write to cover the entire page in order to avoid fragmentation | |
923 | * inefficiencies. | |
924 | * | |
925 | * If the file is opened for synchronous writes or if we have a write delegation | |
926 | * from the server then we can just skip the rest of the checks. | |
927 | */ | |
928 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
929 | { | |
930 | if (file->f_flags & O_DSYNC) | |
931 | return 0; | |
932 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) | |
933 | return 1; | |
934 | if (nfs_write_pageuptodate(page, inode) && (inode->i_flock == NULL || | |
935 | (inode->i_flock->fl_start == 0 && | |
936 | inode->i_flock->fl_end == OFFSET_MAX && | |
937 | inode->i_flock->fl_type != F_RDLCK))) | |
938 | return 1; | |
939 | return 0; | |
940 | } | |
941 | ||
1da177e4 LT |
942 | /* |
943 | * Update and possibly write a cached page of an NFS file. | |
944 | * | |
945 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
946 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
947 | */ | |
948 | int nfs_updatepage(struct file *file, struct page *page, | |
949 | unsigned int offset, unsigned int count) | |
950 | { | |
cd3758e3 | 951 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 952 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
953 | int status = 0; |
954 | ||
91d5b470 CL |
955 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
956 | ||
6de1472f AV |
957 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
958 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 959 | |
c7559663 | 960 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 961 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 962 | offset = 0; |
1da177e4 LT |
963 | } |
964 | ||
e21195a7 | 965 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
966 | if (status < 0) |
967 | nfs_set_pageerror(page); | |
59b7c05f TM |
968 | else |
969 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 970 | |
48186c7d | 971 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 972 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
973 | return status; |
974 | } | |
975 | ||
3ff7576d | 976 | static int flush_task_priority(int how) |
1da177e4 LT |
977 | { |
978 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
979 | case FLUSH_HIGHPRI: | |
980 | return RPC_PRIORITY_HIGH; | |
981 | case FLUSH_LOWPRI: | |
982 | return RPC_PRIORITY_LOW; | |
983 | } | |
984 | return RPC_PRIORITY_NORMAL; | |
985 | } | |
986 | ||
c5996c4e FI |
987 | int nfs_initiate_write(struct rpc_clnt *clnt, |
988 | struct nfs_write_data *data, | |
d138d5d1 | 989 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 990 | int how, int flags) |
1da177e4 | 991 | { |
cd841605 | 992 | struct inode *inode = data->header->inode; |
3ff7576d | 993 | int priority = flush_task_priority(how); |
07737691 | 994 | struct rpc_task *task; |
bdc7f021 TM |
995 | struct rpc_message msg = { |
996 | .rpc_argp = &data->args, | |
997 | .rpc_resp = &data->res, | |
cd841605 | 998 | .rpc_cred = data->header->cred, |
bdc7f021 | 999 | }; |
84115e1c | 1000 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 1001 | .rpc_client = clnt, |
07737691 | 1002 | .task = &data->task, |
bdc7f021 | 1003 | .rpc_message = &msg, |
84115e1c TM |
1004 | .callback_ops = call_ops, |
1005 | .callback_data = data, | |
101070ca | 1006 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1007 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1008 | .priority = priority, |
84115e1c | 1009 | }; |
2c61be0a | 1010 | int ret = 0; |
1da177e4 | 1011 | |
d138d5d1 AA |
1012 | /* Set up the initial task struct. */ |
1013 | NFS_PROTO(inode)->write_setup(data, &msg); | |
1014 | ||
1015 | dprintk("NFS: %5u initiated write call " | |
1e8968c5 | 1016 | "(req %s/%llu, %u bytes @ offset %llu)\n", |
d138d5d1 AA |
1017 | data->task.tk_pid, |
1018 | inode->i_sb->s_id, | |
1e8968c5 | 1019 | (unsigned long long)NFS_FILEID(inode), |
d138d5d1 AA |
1020 | data->args.count, |
1021 | (unsigned long long)data->args.offset); | |
1022 | ||
8c21c62c WAA |
1023 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
1024 | &task_setup_data.rpc_client, &msg, data); | |
1025 | ||
d138d5d1 AA |
1026 | task = rpc_run_task(&task_setup_data); |
1027 | if (IS_ERR(task)) { | |
1028 | ret = PTR_ERR(task); | |
1029 | goto out; | |
1030 | } | |
1031 | if (how & FLUSH_SYNC) { | |
1032 | ret = rpc_wait_for_completion_task(task); | |
1033 | if (ret == 0) | |
1034 | ret = task->tk_status; | |
1035 | } | |
1036 | rpc_put_task(task); | |
1037 | out: | |
1038 | return ret; | |
1039 | } | |
a69aef14 | 1040 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
1041 | |
1042 | /* | |
1043 | * Set up the argument/result storage required for the RPC call. | |
1044 | */ | |
6c75dc0d | 1045 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 | 1046 | unsigned int count, unsigned int offset, |
ea2cf228 | 1047 | int how, struct nfs_commit_info *cinfo) |
d138d5d1 | 1048 | { |
6c75dc0d | 1049 | struct nfs_page *req = data->header->req; |
d138d5d1 | 1050 | |
1da177e4 LT |
1051 | /* Set up the RPC argument and reply structs |
1052 | * NB: take care not to mess about with data->commit et al. */ | |
1053 | ||
6c75dc0d | 1054 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 1055 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
1056 | /* pnfs_set_layoutcommit needs this */ |
1057 | data->mds_offset = data->args.offset; | |
1da177e4 | 1058 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 1059 | data->args.pages = data->pages.pagevec; |
1da177e4 | 1060 | data->args.count = count; |
383ba719 | 1061 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 1062 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 1063 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
1064 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
1065 | case 0: | |
1066 | break; | |
1067 | case FLUSH_COND_STABLE: | |
ea2cf228 | 1068 | if (nfs_reqs_to_commit(cinfo)) |
87ed5eb4 TM |
1069 | break; |
1070 | default: | |
1071 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 1072 | } |
1da177e4 LT |
1073 | |
1074 | data->res.fattr = &data->fattr; | |
1075 | data->res.count = count; | |
1076 | data->res.verf = &data->verf; | |
0e574af1 | 1077 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 1078 | } |
1da177e4 | 1079 | |
6e4efd56 TM |
1080 | static int nfs_do_write(struct nfs_write_data *data, |
1081 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
1082 | int how) |
1083 | { | |
cd841605 | 1084 | struct inode *inode = data->header->inode; |
0382b744 | 1085 | |
9f0ec176 | 1086 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0); |
1da177e4 LT |
1087 | } |
1088 | ||
275acaaf TM |
1089 | static int nfs_do_multiple_writes(struct list_head *head, |
1090 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
1091 | int how) |
1092 | { | |
1093 | struct nfs_write_data *data; | |
1094 | int ret = 0; | |
1095 | ||
1096 | while (!list_empty(head)) { | |
1097 | int ret2; | |
1098 | ||
6c75dc0d | 1099 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
1100 | list_del_init(&data->list); |
1101 | ||
dce81290 | 1102 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
1103 | if (ret == 0) |
1104 | ret = ret2; | |
1105 | } | |
1106 | return ret; | |
1107 | } | |
1108 | ||
6d884e8f F |
1109 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1110 | * call this on each, which will prepare them to be retried on next | |
1111 | * writeback using standard nfs. | |
1112 | */ | |
1113 | static void nfs_redirty_request(struct nfs_page *req) | |
1114 | { | |
1115 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1116 | nfs_unlock_request(req); |
d1182b33 | 1117 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 1118 | nfs_release_request(req); |
6d884e8f F |
1119 | } |
1120 | ||
061ae2ed | 1121 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1122 | { |
1123 | struct nfs_page *req; | |
1124 | ||
1125 | while (!list_empty(head)) { | |
1126 | req = nfs_list_entry(head->next); | |
1127 | nfs_list_remove_request(req); | |
1128 | nfs_redirty_request(req); | |
1129 | } | |
1130 | } | |
1131 | ||
061ae2ed FI |
1132 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1133 | .error_cleanup = nfs_async_write_error, | |
1134 | .completion = nfs_write_completion, | |
1135 | }; | |
1136 | ||
25b11dcd TM |
1137 | static void nfs_flush_error(struct nfs_pageio_descriptor *desc, |
1138 | struct nfs_pgio_header *hdr) | |
1139 | { | |
1140 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1141 | while (!list_empty(&hdr->rpc_list)) { | |
1142 | struct nfs_write_data *data = list_first_entry(&hdr->rpc_list, | |
1143 | struct nfs_write_data, list); | |
1144 | list_del(&data->list); | |
1145 | nfs_writedata_release(data); | |
1146 | } | |
1147 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
1148 | } | |
1149 | ||
1da177e4 LT |
1150 | /* |
1151 | * Generate multiple small requests to write out a single | |
1152 | * contiguous dirty area on one page. | |
1153 | */ | |
6c75dc0d FI |
1154 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1155 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1156 | { |
6c75dc0d | 1157 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1158 | struct page *page = req->wb_page; |
1159 | struct nfs_write_data *data; | |
d097971d | 1160 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1161 | unsigned int offset; |
1da177e4 | 1162 | int requests = 0; |
ea2cf228 | 1163 | struct nfs_commit_info cinfo; |
1da177e4 | 1164 | |
ea2cf228 | 1165 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
1da177e4 | 1166 | |
b31268ac | 1167 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1168 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || |
b31268ac TM |
1169 | desc->pg_count > wsize)) |
1170 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1171 | ||
1172 | ||
275acaaf | 1173 | offset = 0; |
c76069bd | 1174 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1175 | do { |
1176 | size_t len = min(nbytes, wsize); | |
1177 | ||
6c75dc0d | 1178 | data = nfs_writedata_alloc(hdr, 1); |
25b11dcd TM |
1179 | if (!data) { |
1180 | nfs_flush_error(desc, hdr); | |
1181 | return -ENOMEM; | |
1182 | } | |
30dd374f | 1183 | data->pages.pagevec[0] = page; |
ea2cf228 | 1184 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); |
6c75dc0d | 1185 | list_add(&data->list, &hdr->rpc_list); |
1da177e4 | 1186 | requests++; |
e9f7bee1 | 1187 | nbytes -= len; |
275acaaf | 1188 | offset += len; |
e9f7bee1 | 1189 | } while (nbytes != 0); |
25b11dcd TM |
1190 | nfs_list_remove_request(req); |
1191 | nfs_list_add_request(req, &hdr->pages); | |
6c75dc0d | 1192 | desc->pg_rpc_callops = &nfs_write_common_ops; |
25b11dcd | 1193 | return 0; |
1da177e4 LT |
1194 | } |
1195 | ||
1196 | /* | |
1197 | * Create an RPC task for the given write request and kick it. | |
1198 | * The page must have been locked by the caller. | |
1199 | * | |
1200 | * It may happen that the page we're passed is not marked dirty. | |
1201 | * This is the case if nfs_updatepage detects a conflicting request | |
1202 | * that has been written but not committed. | |
1203 | */ | |
6c75dc0d FI |
1204 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1205 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1206 | { |
1207 | struct nfs_page *req; | |
1208 | struct page **pages; | |
1209 | struct nfs_write_data *data; | |
c76069bd | 1210 | struct list_head *head = &desc->pg_list; |
ea2cf228 | 1211 | struct nfs_commit_info cinfo; |
1da177e4 | 1212 | |
6c75dc0d FI |
1213 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1214 | desc->pg_count)); | |
1215 | if (!data) { | |
25b11dcd TM |
1216 | nfs_flush_error(desc, hdr); |
1217 | return -ENOMEM; | |
44b83799 | 1218 | } |
6c75dc0d | 1219 | |
ea2cf228 | 1220 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
30dd374f | 1221 | pages = data->pages.pagevec; |
1da177e4 LT |
1222 | while (!list_empty(head)) { |
1223 | req = nfs_list_entry(head->next); | |
1224 | nfs_list_remove_request(req); | |
6c75dc0d | 1225 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1226 | *pages++ = req->wb_page; |
1da177e4 | 1227 | } |
1da177e4 | 1228 | |
b31268ac | 1229 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1230 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) |
b31268ac TM |
1231 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; |
1232 | ||
1da177e4 | 1233 | /* Set up the argument struct */ |
ea2cf228 | 1234 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); |
6c75dc0d FI |
1235 | list_add(&data->list, &hdr->rpc_list); |
1236 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
25b11dcd | 1237 | return 0; |
1da177e4 LT |
1238 | } |
1239 | ||
6c75dc0d FI |
1240 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1241 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1242 | { |
1243 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1244 | return nfs_flush_multi(desc, hdr); |
1245 | return nfs_flush_one(desc, hdr); | |
dce81290 | 1246 | } |
89d77c8f | 1247 | EXPORT_SYMBOL_GPL(nfs_generic_flush); |
dce81290 TM |
1248 | |
1249 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1250 | { |
6c75dc0d FI |
1251 | struct nfs_write_header *whdr; |
1252 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1253 | int ret; |
1254 | ||
6c75dc0d FI |
1255 | whdr = nfs_writehdr_alloc(); |
1256 | if (!whdr) { | |
9b5415b5 | 1257 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
6c75dc0d FI |
1258 | return -ENOMEM; |
1259 | } | |
1260 | hdr = &whdr->header; | |
1261 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1262 | atomic_inc(&hdr->refcnt); | |
1263 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1264 | if (ret == 0) |
6c75dc0d FI |
1265 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1266 | desc->pg_rpc_callops, | |
1267 | desc->pg_ioflags); | |
6c75dc0d | 1268 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 1269 | hdr->completion_ops->completion(hdr); |
275acaaf | 1270 | return ret; |
1751c363 | 1271 | } |
1751c363 TM |
1272 | |
1273 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1274 | .pg_test = nfs_generic_pg_test, | |
1275 | .pg_doio = nfs_generic_pg_writepages, | |
1276 | }; | |
1277 | ||
57208fa7 | 1278 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
061ae2ed FI |
1279 | struct inode *inode, int ioflags, |
1280 | const struct nfs_pgio_completion_ops *compl_ops) | |
1da177e4 | 1281 | { |
061ae2ed | 1282 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops, |
1751c363 TM |
1283 | NFS_SERVER(inode)->wsize, ioflags); |
1284 | } | |
ddda8e0a | 1285 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1286 | |
dce81290 TM |
1287 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1288 | { | |
1289 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1290 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1291 | } | |
1f945357 | 1292 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1293 | |
1da177e4 | 1294 | |
def6ed7e AA |
1295 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1296 | { | |
1297 | struct nfs_write_data *data = calldata; | |
ef1820f9 N |
1298 | int err; |
1299 | err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); | |
1300 | if (err) | |
1301 | rpc_exit(task, err); | |
def6ed7e | 1302 | } |
def6ed7e | 1303 | |
0b7c0153 FI |
1304 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1305 | { | |
1306 | struct nfs_commit_data *data = calldata; | |
1307 | ||
1308 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1309 | } | |
1310 | ||
1da177e4 LT |
1311 | /* |
1312 | * Handle a write reply that flushes a whole page. | |
1313 | * | |
1314 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1315 | * writebacks since the page->count is kept > 1 for as long | |
1316 | * as the page has a write request pending. | |
1317 | */ | |
6c75dc0d | 1318 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1319 | { |
788e7a89 | 1320 | struct nfs_write_data *data = calldata; |
1da177e4 | 1321 | |
c9d8f89d TM |
1322 | nfs_writeback_done(task, data); |
1323 | } | |
1324 | ||
6c75dc0d | 1325 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1326 | { |
1327 | struct nfs_write_data *data = calldata; | |
cd841605 | 1328 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1329 | int status = data->task.tk_status; |
788e7a89 | 1330 | |
6c75dc0d FI |
1331 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1332 | spin_lock(&hdr->lock); | |
1333 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1334 | ; /* Do nothing */ | |
1335 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
9bce008b TM |
1336 | memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf)); |
1337 | else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf))) | |
6c75dc0d FI |
1338 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); |
1339 | spin_unlock(&hdr->lock); | |
1da177e4 | 1340 | } |
cd841605 | 1341 | nfs_writedata_release(data); |
1da177e4 LT |
1342 | } |
1343 | ||
6c75dc0d | 1344 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1345 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1346 | .rpc_call_done = nfs_writeback_done_common, |
1347 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1348 | }; |
1349 | ||
1350 | ||
1da177e4 LT |
1351 | /* |
1352 | * This function is called when the WRITE call is complete. | |
1353 | */ | |
13602896 | 1354 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1355 | { |
1da177e4 LT |
1356 | struct nfs_writeargs *argp = &data->args; |
1357 | struct nfs_writeres *resp = &data->res; | |
cd841605 | 1358 | struct inode *inode = data->header->inode; |
788e7a89 | 1359 | int status; |
1da177e4 | 1360 | |
a3f565b1 | 1361 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1362 | task->tk_pid, task->tk_status); |
1363 | ||
f551e44f CL |
1364 | /* |
1365 | * ->write_done will attempt to use post-op attributes to detect | |
1366 | * conflicting writes by other clients. A strict interpretation | |
1367 | * of close-to-open would allow us to continue caching even if | |
1368 | * another writer had changed the file, but some applications | |
1369 | * depend on tighter cache coherency when writing. | |
1370 | */ | |
cd841605 | 1371 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1372 | if (status != 0) |
13602896 | 1373 | return; |
cd841605 | 1374 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1375 | |
89d77c8f | 1376 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
1da177e4 LT |
1377 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { |
1378 | /* We tried a write call, but the server did not | |
1379 | * commit data to stable storage even though we | |
1380 | * requested it. | |
1381 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1382 | * the server reports NFS_DATA_SYNC, but performs | |
1383 | * NFS_FILE_SYNC. We therefore implement this checking | |
1384 | * as a dprintk() in order to avoid filling syslog. | |
1385 | */ | |
1386 | static unsigned long complain; | |
1387 | ||
a69aef14 | 1388 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1389 | if (time_before(complain, jiffies)) { |
48186c7d | 1390 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1391 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1392 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1393 | resp->verf->committed, argp->stable); |
1394 | complain = jiffies + 300 * HZ; | |
1395 | } | |
1396 | } | |
1397 | #endif | |
6c75dc0d FI |
1398 | if (task->tk_status < 0) |
1399 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); | |
1400 | else if (resp->count < argp->count) { | |
1da177e4 LT |
1401 | static unsigned long complain; |
1402 | ||
6c75dc0d | 1403 | /* This a short write! */ |
cd841605 | 1404 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1405 | |
1da177e4 | 1406 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1407 | if (resp->count == 0) { |
1408 | if (time_before(complain, jiffies)) { | |
1409 | printk(KERN_WARNING | |
1410 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1411 | argp->count); | |
1412 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1413 | } |
6c75dc0d FI |
1414 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1415 | task->tk_status = -EIO; | |
13602896 | 1416 | return; |
1da177e4 | 1417 | } |
6c75dc0d FI |
1418 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1419 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1420 | /* Resend from where the server left off */ | |
1421 | data->mds_offset += resp->count; | |
1422 | argp->offset += resp->count; | |
1423 | argp->pgbase += resp->count; | |
1424 | argp->count -= resp->count; | |
1425 | } else { | |
1426 | /* Resend as a stable write in order to avoid | |
1427 | * headaches in the case of a server crash. | |
1428 | */ | |
1429 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1430 | } |
6c75dc0d | 1431 | rpc_restart_call_prepare(task); |
1da177e4 | 1432 | } |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | ||
89d77c8f | 1436 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
71d0a611 TM |
1437 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1438 | { | |
b8413f98 TM |
1439 | int ret; |
1440 | ||
71d0a611 TM |
1441 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1442 | return 1; | |
b8413f98 TM |
1443 | if (!may_wait) |
1444 | return 0; | |
1445 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1446 | NFS_INO_COMMIT, | |
1447 | nfs_wait_bit_killable, | |
1448 | TASK_KILLABLE); | |
1449 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1450 | } |
1451 | ||
f453a54a | 1452 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1453 | { |
1454 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1455 | smp_mb__after_clear_bit(); | |
1456 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1457 | } | |
1458 | ||
0b7c0153 | 1459 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1460 | { |
0b7c0153 FI |
1461 | put_nfs_open_context(data->context); |
1462 | nfs_commit_free(data); | |
1da177e4 | 1463 | } |
e0c2b380 | 1464 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1465 | |
0b7c0153 | 1466 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1467 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1468 | int how, int flags) |
1da177e4 | 1469 | { |
07737691 | 1470 | struct rpc_task *task; |
9ace33cd | 1471 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1472 | struct rpc_message msg = { |
1473 | .rpc_argp = &data->args, | |
1474 | .rpc_resp = &data->res, | |
9ace33cd | 1475 | .rpc_cred = data->cred, |
bdc7f021 | 1476 | }; |
84115e1c | 1477 | struct rpc_task_setup task_setup_data = { |
07737691 | 1478 | .task = &data->task, |
9ace33cd | 1479 | .rpc_client = clnt, |
bdc7f021 | 1480 | .rpc_message = &msg, |
9ace33cd | 1481 | .callback_ops = call_ops, |
84115e1c | 1482 | .callback_data = data, |
101070ca | 1483 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1484 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1485 | .priority = priority, |
84115e1c | 1486 | }; |
9ace33cd FI |
1487 | /* Set up the initial task struct. */ |
1488 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1489 | ||
1490 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1491 | ||
8c21c62c WAA |
1492 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1493 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1494 | ||
9ace33cd FI |
1495 | task = rpc_run_task(&task_setup_data); |
1496 | if (IS_ERR(task)) | |
1497 | return PTR_ERR(task); | |
1498 | if (how & FLUSH_SYNC) | |
1499 | rpc_wait_for_completion_task(task); | |
1500 | rpc_put_task(task); | |
1501 | return 0; | |
1502 | } | |
e0c2b380 | 1503 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1504 | |
1505 | /* | |
1506 | * Set up the argument/result storage required for the RPC call. | |
1507 | */ | |
0b7c0153 | 1508 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1509 | struct list_head *head, |
1510 | struct pnfs_layout_segment *lseg, | |
1511 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1512 | { |
1513 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1514 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1515 | |
1516 | /* Set up the RPC argument and reply structs | |
1517 | * NB: take care not to mess about with data->commit et al. */ | |
1518 | ||
1519 | list_splice_init(head, &data->pages); | |
1da177e4 | 1520 | |
1da177e4 | 1521 | data->inode = inode; |
9ace33cd | 1522 | data->cred = first->wb_context->cred; |
988b6dce | 1523 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1524 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1525 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1526 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1527 | |
1528 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1529 | /* Note: we always request a commit of the entire inode */ |
1530 | data->args.offset = 0; | |
1531 | data->args.count = 0; | |
0b7c0153 | 1532 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1533 | data->res.fattr = &data->fattr; |
1534 | data->res.verf = &data->verf; | |
0e574af1 | 1535 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1536 | } |
e0c2b380 | 1537 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1538 | |
e0c2b380 | 1539 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1540 | struct pnfs_layout_segment *lseg, |
1541 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1542 | { |
1543 | struct nfs_page *req; | |
1544 | ||
1545 | while (!list_empty(page_list)) { | |
1546 | req = nfs_list_entry(page_list->next); | |
1547 | nfs_list_remove_request(req); | |
ea2cf228 | 1548 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1549 | if (!cinfo->dreq) { |
1550 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1551 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1552 | BDI_RECLAIMABLE); |
1553 | } | |
1d1afcbc | 1554 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1555 | } |
1556 | } | |
e0c2b380 | 1557 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1558 | |
1da177e4 LT |
1559 | /* |
1560 | * Commit dirty pages | |
1561 | */ | |
1562 | static int | |
ea2cf228 FI |
1563 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1564 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1565 | { |
0b7c0153 | 1566 | struct nfs_commit_data *data; |
1da177e4 | 1567 | |
c9d8f89d | 1568 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1569 | |
1570 | if (!data) | |
1571 | goto out_bad; | |
1572 | ||
1573 | /* Set up the argument struct */ | |
f453a54a FI |
1574 | nfs_init_commit(data, head, NULL, cinfo); |
1575 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1576 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1577 | how, 0); | |
1da177e4 | 1578 | out_bad: |
ea2cf228 | 1579 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1580 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1581 | return -ENOMEM; |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * COMMIT call returned | |
1586 | */ | |
788e7a89 | 1587 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1588 | { |
0b7c0153 | 1589 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1590 | |
a3f565b1 | 1591 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1592 | task->tk_pid, task->tk_status); |
1593 | ||
788e7a89 | 1594 | /* Call the NFS version-specific code */ |
c0d0e96b | 1595 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1596 | } |
1597 | ||
f453a54a | 1598 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1599 | { |
5917ce84 | 1600 | struct nfs_page *req; |
c9d8f89d | 1601 | int status = data->task.tk_status; |
f453a54a | 1602 | struct nfs_commit_info cinfo; |
788e7a89 | 1603 | |
1da177e4 LT |
1604 | while (!list_empty(&data->pages)) { |
1605 | req = nfs_list_entry(data->pages.next); | |
1606 | nfs_list_remove_request(req); | |
d6d6dc7c | 1607 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1608 | |
1e8968c5 | 1609 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1610 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1611 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1612 | req->wb_bytes, |
1613 | (long long)req_offset(req)); | |
c9d8f89d TM |
1614 | if (status < 0) { |
1615 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1616 | nfs_inode_remove_request(req); |
c9d8f89d | 1617 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1618 | goto next; |
1619 | } | |
1620 | ||
1621 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1622 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1623 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1624 | /* We have a match */ |
1625 | nfs_inode_remove_request(req); | |
1626 | dprintk(" OK\n"); | |
1627 | goto next; | |
1628 | } | |
1629 | /* We have a mismatch. Write the page again */ | |
1630 | dprintk(" mismatch\n"); | |
6d884e8f | 1631 | nfs_mark_request_dirty(req); |
05990d1b | 1632 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1633 | next: |
1d1afcbc | 1634 | nfs_unlock_and_release_request(req); |
1da177e4 | 1635 | } |
f453a54a FI |
1636 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1637 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1638 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1639 | } |
1640 | ||
1641 | static void nfs_commit_release(void *calldata) | |
1642 | { | |
0b7c0153 | 1643 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1644 | |
f453a54a | 1645 | data->completion_ops->completion(data); |
c9d8f89d | 1646 | nfs_commitdata_release(calldata); |
1da177e4 | 1647 | } |
788e7a89 TM |
1648 | |
1649 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1650 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1651 | .rpc_call_done = nfs_commit_done, |
1652 | .rpc_release = nfs_commit_release, | |
1653 | }; | |
1da177e4 | 1654 | |
f453a54a FI |
1655 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1656 | .completion = nfs_commit_release_pages, | |
1657 | .error_cleanup = nfs_commit_clear_lock, | |
1658 | }; | |
1659 | ||
1763da12 FI |
1660 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1661 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1662 | { |
1663 | int status; | |
1664 | ||
ea2cf228 | 1665 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1666 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1667 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1668 | return status; |
1669 | } | |
1670 | ||
b608b283 | 1671 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1672 | { |
1da177e4 | 1673 | LIST_HEAD(head); |
ea2cf228 | 1674 | struct nfs_commit_info cinfo; |
71d0a611 | 1675 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1676 | int res; |
1da177e4 | 1677 | |
b8413f98 TM |
1678 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1679 | if (res <= 0) | |
c5efa5fc | 1680 | goto out_mark_dirty; |
ea2cf228 FI |
1681 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1682 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1683 | if (res) { |
a861a1e1 FI |
1684 | int error; |
1685 | ||
ea2cf228 | 1686 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1687 | if (error < 0) |
1688 | return error; | |
b8413f98 | 1689 | if (!may_wait) |
c5efa5fc | 1690 | goto out_mark_dirty; |
b8413f98 TM |
1691 | error = wait_on_bit(&NFS_I(inode)->flags, |
1692 | NFS_INO_COMMIT, | |
1693 | nfs_wait_bit_killable, | |
1694 | TASK_KILLABLE); | |
1695 | if (error < 0) | |
1696 | return error; | |
71d0a611 TM |
1697 | } else |
1698 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1699 | return res; |
1700 | /* Note: If we exit without ensuring that the commit is complete, | |
1701 | * we must mark the inode as dirty. Otherwise, future calls to | |
1702 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1703 | * that the data is on the disk. | |
1704 | */ | |
1705 | out_mark_dirty: | |
1706 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1707 | return res; |
1708 | } | |
8fc795f7 TM |
1709 | |
1710 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1711 | { | |
420e3646 TM |
1712 | struct nfs_inode *nfsi = NFS_I(inode); |
1713 | int flags = FLUSH_SYNC; | |
1714 | int ret = 0; | |
8fc795f7 | 1715 | |
3236c3e1 | 1716 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1717 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1718 | return ret; |
1719 | ||
a00dd6c0 JL |
1720 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1721 | /* Don't commit yet if this is a non-blocking flush and there | |
1722 | * are a lot of outstanding writes for this mapping. | |
1723 | */ | |
ea2cf228 | 1724 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1725 | goto out_mark_dirty; |
420e3646 | 1726 | |
a00dd6c0 | 1727 | /* don't wait for the COMMIT response */ |
420e3646 | 1728 | flags = 0; |
a00dd6c0 JL |
1729 | } |
1730 | ||
420e3646 TM |
1731 | ret = nfs_commit_inode(inode, flags); |
1732 | if (ret >= 0) { | |
1733 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1734 | if (ret < wbc->nr_to_write) | |
1735 | wbc->nr_to_write -= ret; | |
1736 | else | |
1737 | wbc->nr_to_write = 0; | |
1738 | } | |
8fc795f7 | 1739 | return 0; |
420e3646 TM |
1740 | } |
1741 | out_mark_dirty: | |
8fc795f7 TM |
1742 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1743 | return ret; | |
1744 | } | |
c63c7b05 | 1745 | #else |
8fc795f7 TM |
1746 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1747 | { | |
1748 | return 0; | |
1749 | } | |
1da177e4 LT |
1750 | #endif |
1751 | ||
8fc795f7 TM |
1752 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1753 | { | |
a8d8f02c BS |
1754 | return nfs_commit_unstable_pages(inode, wbc); |
1755 | } | |
89d77c8f | 1756 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1757 | |
acdc53b2 TM |
1758 | /* |
1759 | * flush the inode to disk. | |
1760 | */ | |
1761 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1762 | { |
1763 | struct writeback_control wbc = { | |
72cb77f4 | 1764 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1765 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1766 | .range_start = 0, |
1767 | .range_end = LLONG_MAX, | |
34901f70 | 1768 | }; |
f4ce1299 TM |
1769 | int ret; |
1770 | ||
1771 | trace_nfs_writeback_inode_enter(inode); | |
1772 | ||
1773 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1774 | |
f4ce1299 TM |
1775 | trace_nfs_writeback_inode_exit(inode, ret); |
1776 | return ret; | |
1c75950b | 1777 | } |
ddda8e0a | 1778 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1779 | |
1b3b4a1a TM |
1780 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1781 | { | |
1782 | struct nfs_page *req; | |
1b3b4a1a TM |
1783 | int ret = 0; |
1784 | ||
1b3b4a1a | 1785 | for (;;) { |
ba8b06e6 | 1786 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1787 | req = nfs_page_find_request(page); |
1788 | if (req == NULL) | |
1b3b4a1a | 1789 | break; |
7ad84aa9 | 1790 | if (nfs_lock_request(req)) { |
8dd37758 | 1791 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1792 | nfs_inode_remove_request(req); |
1793 | /* | |
1794 | * In case nfs_inode_remove_request has marked the | |
1795 | * page as being dirty | |
1796 | */ | |
1797 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1d1afcbc | 1798 | nfs_unlock_and_release_request(req); |
1b3b4a1a TM |
1799 | break; |
1800 | } | |
1801 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1802 | nfs_release_request(req); |
1b3b4a1a | 1803 | if (ret < 0) |
c988950e | 1804 | break; |
1b3b4a1a | 1805 | } |
1b3b4a1a TM |
1806 | return ret; |
1807 | } | |
1808 | ||
7f2f12d9 TM |
1809 | /* |
1810 | * Write back all requests on one page - we do this before reading it. | |
1811 | */ | |
1812 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1813 | { |
29418aa4 | 1814 | loff_t range_start = page_file_offset(page); |
1c75950b | 1815 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1816 | struct writeback_control wbc = { |
4d770ccf | 1817 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1818 | .nr_to_write = 0, |
4d770ccf TM |
1819 | .range_start = range_start, |
1820 | .range_end = range_end, | |
1821 | }; | |
1822 | int ret; | |
1c75950b | 1823 | |
f4ce1299 TM |
1824 | trace_nfs_writeback_page_enter(inode); |
1825 | ||
0522f6ad | 1826 | for (;;) { |
ba8b06e6 | 1827 | wait_on_page_writeback(page); |
73e3302f TM |
1828 | if (clear_page_dirty_for_io(page)) { |
1829 | ret = nfs_writepage_locked(page, &wbc); | |
1830 | if (ret < 0) | |
1831 | goto out_error; | |
0522f6ad | 1832 | continue; |
7f2f12d9 | 1833 | } |
f4ce1299 | 1834 | ret = 0; |
0522f6ad TM |
1835 | if (!PagePrivate(page)) |
1836 | break; | |
1837 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1838 | if (ret < 0) |
73e3302f | 1839 | goto out_error; |
7f2f12d9 | 1840 | } |
73e3302f | 1841 | out_error: |
f4ce1299 | 1842 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1843 | return ret; |
1c75950b TM |
1844 | } |
1845 | ||
074cc1de TM |
1846 | #ifdef CONFIG_MIGRATION |
1847 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1848 | struct page *page, enum migrate_mode mode) |
074cc1de | 1849 | { |
2da95652 JL |
1850 | /* |
1851 | * If PagePrivate is set, then the page is currently associated with | |
1852 | * an in-progress read or write request. Don't try to migrate it. | |
1853 | * | |
1854 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1855 | * that we can safely release the inode reference while holding | |
1856 | * the page lock. | |
1857 | */ | |
1858 | if (PagePrivate(page)) | |
1859 | return -EBUSY; | |
074cc1de | 1860 | |
8c209ce7 DH |
1861 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1862 | return -EBUSY; | |
074cc1de | 1863 | |
a6bc32b8 | 1864 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1865 | } |
1866 | #endif | |
1867 | ||
f7b422b1 | 1868 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1869 | { |
1870 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1871 | sizeof(struct nfs_write_header), |
1da177e4 | 1872 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1873 | NULL); |
1da177e4 LT |
1874 | if (nfs_wdata_cachep == NULL) |
1875 | return -ENOMEM; | |
1876 | ||
93d2341c MD |
1877 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1878 | nfs_wdata_cachep); | |
1da177e4 | 1879 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1880 | goto out_destroy_write_cache; |
1da177e4 | 1881 | |
0b7c0153 FI |
1882 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1883 | sizeof(struct nfs_commit_data), | |
1884 | 0, SLAB_HWCACHE_ALIGN, | |
1885 | NULL); | |
1886 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1887 | goto out_destroy_write_mempool; |
0b7c0153 | 1888 | |
93d2341c | 1889 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1890 | nfs_cdata_cachep); |
1da177e4 | 1891 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1892 | goto out_destroy_commit_cache; |
1da177e4 | 1893 | |
89a09141 PZ |
1894 | /* |
1895 | * NFS congestion size, scale with available memory. | |
1896 | * | |
1897 | * 64MB: 8192k | |
1898 | * 128MB: 11585k | |
1899 | * 256MB: 16384k | |
1900 | * 512MB: 23170k | |
1901 | * 1GB: 32768k | |
1902 | * 2GB: 46340k | |
1903 | * 4GB: 65536k | |
1904 | * 8GB: 92681k | |
1905 | * 16GB: 131072k | |
1906 | * | |
1907 | * This allows larger machines to have larger/more transfers. | |
1908 | * Limit the default to 256M | |
1909 | */ | |
1910 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1911 | if (nfs_congestion_kb > 256*1024) | |
1912 | nfs_congestion_kb = 256*1024; | |
1913 | ||
1da177e4 | 1914 | return 0; |
3dd4765f JL |
1915 | |
1916 | out_destroy_commit_cache: | |
1917 | kmem_cache_destroy(nfs_cdata_cachep); | |
1918 | out_destroy_write_mempool: | |
1919 | mempool_destroy(nfs_wdata_mempool); | |
1920 | out_destroy_write_cache: | |
1921 | kmem_cache_destroy(nfs_wdata_cachep); | |
1922 | return -ENOMEM; | |
1da177e4 LT |
1923 | } |
1924 | ||
266bee88 | 1925 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1926 | { |
1927 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1928 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1929 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1930 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1931 | } |
1932 |